Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025].

See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Modeling of Physical Systems
Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Modeling of Physical Systems

Raul G. Longoria and Joseph J. Beaman

The University of Texas at Austin
Simulation and Control

Austin, Texas, USA

 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
This edition first published 2025
© 2025 John Wiley & Sons Ltd.

All rights reserved, including rights for text and data mining and training of artificial technologies or similar technologies. No part of this
publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, electronic, mechanical, photocopying,
recording or otherwise, except as permitted by law. Advice on how to obtain permission to reuse material from this title is available at
http://www.wiley.com/go/permissions.

The right of Raul G. Longoria and Joseph J. Beaman to be identified as the author of this work has been asserted in accordance with law.

Registered Offices
John Wiley & Sons, Inc., 111 River Street, Hoboken, NJ 07030, USA
John Wiley & Sons Ltd, New Era House, 8 Oldlands Way, Bognor Regis, West Sussex, PO22 9NQ, UK

For details of our global editorial offices, customer services, and more information about Wiley products visit us at www.wiley.com.

The manufacturer’s authorized representative according to the EU General Product Safety Regulation is Wiley-VCH GmbH, Boschstr. 12, 69469
Weinheim, Germany, e-mail: [email protected].

Wiley also publishes its books in a variety of electronic formats and by print-on-demand. Some content that appears in standard print versions of
this book may not be available in other formats.

Trademarks: Wiley and the Wiley logo are trademarks or registered trademarks of John Wiley & Sons, Inc. and/or its affiliates in the United
States and other countries and may not be used without written permission. All other trademarks are the property of their respective owners.
John Wiley & Sons, Inc. is not associated with any product or vendor mentioned in this book.

Limit of Liability/Disclaimer of Warranty

In view of ongoing research, equipment modifications, changes in governmental regulations, and the constant flow of information relating to the
use of experimental reagents, equipment, and devices, the reader is urged to review and evaluate the information provided in the package insert
or instructions for each chemical, piece of equipment, reagent, or device for, among other things, any changes in the instructions or indication of
usage and for added warnings and precautions. While the publisher and authors have used their best efforts in preparing this work, they make
no representations or warranties with respect to the accuracy or completeness of the contents of this work and specifically disclaim all
warranties, including without limitation any implied warranties of merchantability or fitness for a particular purpose. No warranty may be
created or extended by sales representatives, written sales materials or promotional statements for this work. The fact that an organization,
website, or product is referred to in this work as a citation and/or potential source of further information does not mean that the publisher and
authors endorse the information or services the organization, website, or product may provide or recommendations it may make. This work is
sold with the understanding that the publisher is not engaged in rendering professional services. The advice and strategies contained herein may
not be suitable for your situation. You should consult with a specialist where appropriate. Further, readers should be aware that websites listed in
this work may have changed or disappeared between when this work was written and when it is read. Neither the publisher nor authors shall be
liable for any loss of profit or any other commercial damages, including but not limited to special, incidental, consequential, or other damages.

Library of Congress Cataloging-in-Publication Data Applied for:

Hardback: ISBN: 9781119945048

Cover image: © Raul G. Longoria

Cover design: Wiley

Set in 9.5/12.5pt STIXTwoText by Straive, Chennai, India

Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
To Monica, for patience, love, and steadfast encouragement. – RGL
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
vii

Contents

Preface xv
About the Companion Website xvii

1 Introduction 1
1.1 System Modeling Concepts 1
1.2 General Steps in Modeling 2
1.3 Definitions of System Modeling Concepts 3
1.4 Energy Basis for Physical System Modeling 5
1.4.1 Reticulation (in the Spirit of Paynter and Russell) 6
1.4.2 Energy Conservation and Continuity 7
1.4.3 Power Balance in Multiport Systems 9
1.5 Relation to Classical Dynamics 10
1.6 Power Flow in Physical System Modeling 12
1.6.1 Using Power Flow in Modeling 13
1.6.2 Power Flow Definitions and Concepts 15
1.7 Outline of the Text 15
1.8 Problems 16

2 Kirchhoff Systems 23
2.1 Mechanical Translation 23
2.1.1 Potential Energy Storage 23
2.1.2 Kinetic Energy Storage 25
2.1.2.1 System of Particles 26
2.1.3 Dissipation 27
2.1.4 Power Sources 29
2.1.5 Dynamic and Kinematic Laws 29
2.1.6 Block Diagram Description 30
2.1.7 Summary 31
2.2 Mechanical Rotation 32
2.2.1 Potential Energy Storage 32
2.2.2 Kinetic Energy Storage 34
2.2.3 Dissipation 36
2.2.4 Power Sources 37
2.2.5 Dynamic and Kinematic Laws 37
2.2.6 Summary 39
2.3 Electric Systems 40
2.3.1 Electric Field Systems 40
2.3.2 Magnetic Field Systems 43
2.3.3 Dissipation and Imperfect Conduction 47
2.3.4 Power Sources 48
2.3.5 Kirchhoff’s Laws 48
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
viii Contents

2.3.6 Summary 49
2.4 Hydraulic Systems 50
2.4.1 Potential Energy Storage 50
2.4.2 Kinetic Energy Storage 52
2.4.3 Dissipation 53
2.4.4 Power Sources 54
2.4.5 Kirchhoff’s Laws for Hydraulic Systems 54
2.4.6 Compressibility Effects in Hydraulic Systems 55
2.4.7 Summary 56
2.5 Ideal Couplers 57
2.6 Thermal System Elements and Effects 60
2.6.1 Thermal Energy Storage 61
2.6.2 Heat Transfer Processes as Thermal Resistors 62
2.6.3 Power Sources 65
2.6.4 Kirchhoff’s Laws for Thermal Systems 66
2.6.5 Thermal Circuit Analogs 68
2.6.6 Summary 69
2.7 State-Space and Numerical Response Analysis 69
2.7.1 State-Space System Descriptions 69
2.7.2 Usage of Simulation Methods 72
2.7.2.1 Simulation of State-Space Models 72
2.7.2.2 A Pattern for Practical Use 74
2.7.2.3 Extensions and Additional Considerations 76
2.7.2.4 Simulation of Linear State Space 76
2.8 Summary 78
2.9 Problems 78

3 Physical Modeling with Bond Graphs 91

3.1 Bond Graph Variables 93
3.2 Bond Graph Elements 94
3.2.1 Energy Storing Elements 94
3.2.1.1 Generalized Potential Energy: C Element 95
3.2.1.2 Generalized Kinetic Energy: I Element 96
3.2.2 Ideal Coupling Elements 97
3.2.2.1 The Two Port Special Case 97
3.2.3 Dissipative Element 99
3.2.4 Junction Elements 100
3.2.5 Source Elements 101
3.2.6 Summary 101
3.3 Modeling Examples 102
3.4 Algorithmic Conversion of Schematics to Bond Graphs 106
3.5 Causality 109
3.5.1 Elemental Causality 110
3.5.1.1 Source Causality 110
3.5.1.2 Energy Storage Causality 110
3.5.1.3 Dissipative Element Causality 111
3.5.1.4 Coupling Element Causality 111
3.5.1.5 Junction Element Causality 112
3.5.2 Causality Assignment Procedure 112
3.6 State Variables and State Equation Derivation 113
3.6.1 State Variable Types and Selection 113
3.6.2 State Equation Form 114
3.6.3 Model Analysis with Causality 115
3.6.3.1 Dependent Causality 115
3.6.3.2 Detecting Improper Models 116
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Contents ix

3.6.4 Deriving State Equations from a Bond Graph 117

3.6.5 State Equation Derivation in Explicit Form 117
3.6.5.1 Basic Mass–Spring–Samper 118
3.6.5.2 LRC Circuit 118
3.6.5.3 Lever–Spring–Damper System 119
3.6.6 Alternative Formulations 121
3.6.6.1 Systems with Derivative Causality 121
3.6.6.2 Systems with Algebraic Loops 122
3.6.7 Summary 124
3.7 Thermal Effects in Bond Graph Models 124
3.7.1 Thermal Bond Graph Elements 124
3.7.1.1 Heat Transfer Modes 125
3.7.1.2 Thermal Energy Storage 126
3.7.1.3 Elementary Thermoelastic Effects 127
3.7.1.4 Ideal Gas 127
3.7.2 Thermal Junction Elements 128
3.7.3 Examples with Thermal Effects 129
3.7.4 Equivalent Conductive Two-Port R-Elements and Pseudo-Bond Graphs 133
3.8 Summary 134
3.9 Problems 138

4 System Model Formulation and Evaluation 155

4.1 Equilibrium Analysis 156
4.1.1 Definition of Equilibrium 156
4.1.2 Equilibrium Bond Graphs 159
4.1.3 Source-Load Modeling and Analysis 162
4.1.4 Summary 164
4.2 Linearization Techniques 165
4.2.1 Tangent Linearization 165
4.2.2 Secant Linearization 167
4.2.3 Statistical Linearization 167
4.2.3.1 Statistical Linearization Approach 167
4.2.3.2 Formulating Statistically Linearized State Equations 168
4.2.4 Summary of Linearization 170
4.3 Signals and Block Diagrams with Causal Bond Graphs 170
4.4 Transfer Functions 175
4.5 Stability of Physical Systems 181
4.5.1 Stability Concept 181
4.5.2 Stability and Eigenvalues 182
4.5.3 Physical Instabilities 184
4.5.4 Routh Stability Criterion 185
4.6 Constitutive Structure 188
4.6.1 Energy-Storing Constitutive Relations 188
4.6.1.1 Energetic Structure 188
4.6.1.2 Passivity Structure 190
4.6.1.3 Causal Structure 191
4.6.2 Dissipative Constitutive Relations 193
4.6.2.1 Entropic or Passivity Structure 193
4.6.2.2 1-Port R Element 194
4.6.2.3 Causal Structure 194
4.6.3 Summary 195
4.7 Modulation Structure 195
4.7.1 Functional Description of Physical Systems with Modulation 196
4.7.2 Energy Storing Elements Should Not Be Modulated 197
4.7.3 Modulated Coupling Elements 198
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
x Contents

4.7.3.1 Modulation in Analytical Mechanics 199

4.7.3.2 Rigid Body Dynamics and Gyroscopic Modulation 200
4.7.3.3 Electromechanical Modulation 203
4.7.3.4 Theoretical Structure of Modulated Coupling Elements 205
4.7.4 Modulated Dissipative Element 206
4.7.5 Modulation in Fluid–Structure Interaction 208
4.7.5.1 Inertial Control Volume 208
4.7.6 Dependent and Modulated Sources 210
4.7.6.1 Controlled Source Modeling of Ideal Operational Amplifiers 211
4.7.6.2 Controlled Source in Variable-Mass Systems 213
4.7.6.3 Controlled Source in Fluid Filling of a Tank 214
4.7.7 Summary of Modulation Structure 215
4.8 Summary 215
4.9 Problems 216

5 Linear System Modeling and Analysis 239

5.1 Analysis of First-Order System Models 239
5.1.1 First-Order Linear Time Invariant Solution 242
5.1.1.1 First-Order Response Solutions and Trends 243
5.1.1.2 Practical Consideration of Model Input Types 244
5.1.2 Summary 245
5.2 Analysis of Second-Order System Models 246
5.2.1 Second-Order Linear Time Invariant Solution 247
5.2.2 Response to Initial Condition and Step Input 248
5.2.2.1 Overdamped, 𝜁 > 1 248
5.2.2.2 Critically Damped, 𝜁 = 1 249
5.2.2.3 Underdamped, 0 < 𝜁 < 1 250
5.2.3 Initial Derivative Condition Response 252
5.2.3.1 Overdamped, 𝜁 > 1 252
5.2.3.2 Critically Damped, 𝜁 = 1 252
5.2.3.3 Underdamped, 0 < 𝜁 < 1 252
5.2.4 Impulse Response 253
5.2.5 Summary 254
5.3 Application of Model Response 256
5.3.1 Experiment Design and Parameter Determination 256
5.3.1.1 Undamped or “Lightly Damped” Systems 256
5.3.1.2 Logarithmic Decrement for Underdamped Systems 256
5.3.2 Time-Domain System Specifications 261
5.4 Analysis Using System Model Transfer Functions 263
5.4.1 Standard Transfer Function Forms 264
5.4.2 System Decomposition Using Partial Fractions 265
5.5 Poles and Zeros from Transfer Function Models 268
5.5.1 First- and Second-Order Systems 268
5.5.2 The Influence of Zeros 269
5.5.3 Summary 272
5.6 Response of nth-Order Linear Systems 272
5.6.1 Free Response of nth-Order Systems 272
5.6.2 Unforced Response Using Matrix Exponential 275
5.7 Forced Response of nth-Order Systems 278
5.7.1 General Step Response 279
5.7.2 Difference Equation 280
5.7.3 Approximate Response to Arbitrary Input 281
5.8 Forced Response from Transfer Functions 281
5.8.1 Geometric Interpretation of Convolution 282
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Contents xi

5.9 Chapter Summary 283

5.10 Problems 283

6 Frequency Response and Impedance-Based Modeling 299

6.1 Frequency Response 299
6.1.1 Frequency Response 299
6.1.2 First-Order System 301
6.1.3 Formulating Frequency Response Functions 302
6.1.4 Logarithmic Plots of Frequency Response 304
6.2 Basic Factors of Frequency Response Functions 305
6.3 Impedance Methods Using Bond Graphs 310
6.3.1 Basic Impedance Elements 310
6.3.2 Impedance-Based Transfer Functions 312
6.3.3 Impedance Across Coupling Elements 314
6.3.4 Summary 317
6.4 Applications of Frequency Response 317
6.4.1 Vibration Applications 317
6.4.2 Characterizing Sensors and Actuators 320
6.4.3 Frequency Response Measurement and Transfer Function Approximation 323
6.4.3.1 Measuring Frequency Response 324
6.4.3.2 Approximating a Transfer Function 328
6.5 Two-Port Models and Transmission Matrices 330
6.5.1 Definition of Linear Two-Port Elements 331
6.5.1.1 Causal Two-Port Matrices 331
6.5.1.2 Transmission or Transfer Matrix 332
6.5.2 Characteristics and Properties of Two-Ports 335
6.5.3 Deriving Two-Port Model Relations 336
6.5.3.1 Basic Two-Port Formulation 337
6.5.3.2 Coupling Element Transmission 339
6.5.4 Interconnection of Two-Ports 340
6.5.5 Summary on Use of Two-Port Models 341
6.5.5.1 Two-Port Analysis Problems 341
6.5.5.2 Experimental Testing and Characterization 343
6.6 Additional Application of Impedance Formulations 344
6.6.1 Impedance and Multiports 344
6.6.2 Impedance and Active Elements 347
6.6.3 Using Two-Ports for Distributed-Parameter Models 348
6.6.3.1 Transmission Lines from Differential Elements 349
6.6.3.2 Modeling with a W-Line 351
6.6.3.3 Modeling with an H-Line 355
6.6.3.4 Lumped-Parameter Approximation of Distributed-Parameter Elements 356
6.6.3.5 Using Infinite Product Expansion Approximations of DPMs 357
6.6.3.6 Insights into Transient Response Analysis of DPMs 359
6.6.3.7 Summary and Perspective 360
6.7 Summary 360
6.8 Problems 360

7 Modeling Feedback Control Systems 375

7.1 Feedback Control Representations 375
7.2 Modeling Control System Elements 379
7.2.1 Transfer Function Models in Control Systems 379
7.2.2 Using Block Diagram Descriptions 382
7.2.2.1 Nonlinear Elements 385
7.2.3 Measurement and Actuation Systems 386
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
xii Contents

7.2.4 Error Detection and Controller Functionality 387

7.3 Closed-Loop Feedback Control 388
7.3.1 Feedback System Relations 388
7.3.1.1 Output, y 388
7.3.1.2 Error, e 389
7.3.1.3 Control Effort, u 389
7.3.2 Beneficial Effects of Feedback 390
7.3.2.1 Force-Control Case Study 390
7.3.2.2 Effect of Feedback on Gain 391
7.3.2.3 Sensitivity of Output Relations 391
7.3.2.4 Effect of Feedback on Stability 393
7.3.2.5 Effect of Feedback in Tracking a Reference Signal 393
7.3.2.6 On the Range of Operation Over Which a System Behaves Linearly 394
7.3.3 Open-Loop and Feedforward Control 394
7.4 Linear Feedback Controllers and Compensators 396
7.4.1 Application of PID Control 397
7.4.2 Comparing Derivative Action and Rate Feedback 401
7.4.3 Error Constants and Linear Stability Analysis 402
7.4.3.1 Error Constants 402
7.4.3.2 Stability Analysis 404
7.4.4 Influence of Model Dynamics on Response and Stability 406
7.4.4.1 Relative Stability and Stability Margins 409
7.4.4.2 Relating Time and Frequency Domain Response and Stability Metrics 410
7.4.5 Controller and Compensator Selection and Design 413
7.4.5.1 Effect of Added Poles and Zeros 414
7.4.5.2 Pole-Zero Cancellation 414
7.4.5.3 Studying Model-Structure for Controller Selection 416
7.4.5.4 Using Root Locus for Controller Assessment 418
7.4.5.5 Application of Frequency Domain Methods 418
7.4.5.6 Dealing with Open-Loop Unstable and Non-Minimum Phase Plants 421
7.4.6 Ziegler–Nichols PID Controller Tuning 423
7.5 State Variable Control Methods 424
7.5.1 Controlling a System with Full-State Variable Feedback 425
7.5.2 Controllability and Observability 426
7.5.2.1 Definition of Controllability 426
7.5.2.2 Definition of Observability 427
7.5.2.3 Condition for Complete State Controllability 427
7.5.2.4 Condition for Complete State Observability 427
7.5.3 How Do Uncontrollable or Unobservable Systems Arise? 431
7.5.3.1 Redundant State Variables 431
7.5.3.2 Physically Uncontrollable States 431
7.5.3.3 Too Much Symmetry 431
7.5.4 Full-State Feedback 432
7.5.4.1 Basic Control Formulation 432
7.5.4.2 Finding Control Gain Matrix: n ≤ 3 Order Systems 433
7.5.4.3 Ackermann’s Formula for Pole-Placement 433
7.5.5 State Feedback with a Reference Input 434
7.5.6 Internal Model Approach for Tracking Reference Input 438
7.5.6.1 Simulating the Internal Model Approach 439
7.5.7 Linear Observer Design for State Estimation 441
7.5.7.1 Observer Design: Low-Order Systems 442
7.5.7.2 Ackermann’s Formula for Observer Design 443
7.5.8 Using Observers Without Feedback Control 444
7.5.9 Full-State Feedback Control with an Observer 447
7.5.10 Reference Inputs with Full-State Feedback and Observer 449
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Contents xiii

7.5.10.1 Case 1: Observer Error Dynamics Do Not Depend on r 450

7.5.10.2 Case 2: Estimator Driven by Tracking Error, y − r 451
7.5.11 Linear Quadratic Optimum Control 451
7.5.11.1 Quadratic Sums 451
7.5.11.2 Optimal Control Problem 452
7.5.11.3 Relevance of P 452
7.5.11.4 Optimal Control of State-Space System 452
7.6 Simulation of Controlled Systems 453
7.6.1 Continuous-Time State Equations and Feedback Control 454
7.6.1.1 Direct Simulation of P-Control 454
7.6.1.2 Implementing a PID Algorithm 456
7.6.1.3 Application of the PID Algorithm 457
7.6.2 Continuous-Time State Equations and Discrete Feedback Control 459
7.6.2.1 Modeling Discrete PID 459
7.6.2.2 Implementing Discrete PID in Simulation 459
7.7 Summary 460
7.8 Problems 461

8 Multiport Modeling and Energy Methods 475

8.1 Multiport Concept and Usage 475
8.1.1 Energy-Storing Multiports 476
8.1.2 Dissipative Multiports 479
8.1.3 Summary 480
8.2 Causality and Constitutive Relations for Multiports 480
8.2.1 Causality on Multiports 480
8.2.2 Multiport Constitutive Relations 483
8.3 Electromechanical Systems Modeling 488
8.3.1 EM Capacitive Multiports 488
8.3.1.1 Electroquasistatic (EQS) Multiports 488
8.3.1.2 Devices with Polarizable or Piezoelectric Material 491
8.3.2 EM Magnetoquasistatic Devices 492
8.3.2.1 Lossless EM Multiport 492
8.3.3 Magnetic Circuit Modeling and Devices 498
8.3.3.1 Properties of Magnetic Materials 498
8.3.3.2 Stored Energy 499
8.3.3.3 Magnetic Circuit Variables and Elements 499
8.3.3.4 Magnetic Circuit Examples 501
8.3.3.5 Magnetomechanical Force Interactions 502
8.3.3.6 Modeling Devices with Permanent Magnets 506
8.4 Lagrange’s Equations in System Modeling 508
8.4.1 Application of Classical Lagrange’s Equations 509
8.4.1.1 Nonconservative Effects 510
8.4.1.2 Constraint Relations 511
8.4.1.3 Example Applications 512
8.4.1.4 Subsystem Partitioning 514
8.4.2 Lagrange Subsystem in a Bond Graph 514
8.4.3 Mechanical and Electromechanical System Examples 517
8.4.4 Dealing with Variable-Mass Effects 523
8.4.5 Derivation of Lagrange Equations 528
8.4.6 Hamiltonian and CoLagrange Formulations 530
8.4.7 Summary 533
8.5 Variational and Minimum Principles 533
8.5.1 Introduction to Calculus of Variations 533
8.5.1.1 Derivation of Euler–Lagrange Equations 534
8.5.2 Hamilton’s Principle 535
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
xiv Contents

8.6 Chapter Summary 537

8.7 Problems 537

9 Thermodynamic Systems 557

9.1 Thermodynamic Systems and Relations 557
9.1.1 Equilibrium and States 558
9.1.2 Gibbs Internal Energy 560
9.1.2.1 Thermal Inertance 562
9.1.2.2 Legendre Transformations 562
9.1.2.3 Maxwell Reciprocity 563
9.2 Equations of State for Ideal Gases 565
9.2.1 Ideal Gas 565
9.2.2 Van der Waals Gas 568
9.2.3 Summary 572
9.3 Applications in Thermomechanical Systems 572
9.3.1 Thermal Systems in Motion 572
9.3.2 Thermoelastic Systems 573
9.3.3 Summary 576
9.4 Modeling Open Thermo-Fluid Systems 576
9.4.1 One-Dimensional Compressible Flow in Systems 577
9.4.1.1 Flowrate and Velocity Descriptions 577
9.4.1.2 Speed of Sound and Mach Number 577
9.4.1.3 Reference State of Fluid 578
9.4.1.4 Enthalpy, Matter Flow Rate, and Power Convection Bond 579
9.4.1.5 Steady Isentropic Flow and Stagnation Enthalpy 580
9.4.1.6 Flow Through Orifices and Nozzles 580
9.4.1.7 Flow in Conduits Such as Tubes, Pipes, and Ducts 582
9.4.1.8 Summary 583
9.4.2 Modeling Storage of Energy in Thermo-Fluid Systems 584
9.4.2.1 The Multiport Thermodynamic C 584
9.4.2.2 Pseudo-Bond Graph Approach for Gas Dynamics 586
9.4.3 Models of Open Thermo-Fluid Systems 588
9.5 Multicomponent Systems 592
9.6 Open System Effects and Diffusion 597
9.6.1 Preliminary Concepts 597
9.6.1.1 Measures of Content 597
9.6.1.2 Properties of Solutions 597
9.6.1.3 Chemical Potential of Real and Ideal Solutions 598
9.6.1.4 Membrane Processes 598
9.6.1.5 Osmosis 599
9.6.1.6 Osmotic Pressure 599
9.6.1.7 Reverse Osmosis 600
9.6.2 Energy Storage Representations 600
9.6.3 Diffusion Processes in System Models 600
9.6.4 Models with Diffusion and Stored Internal Energy 604
9.6.4.1 Coupled Diffusion Between Solvent and Solute Flow Across a Membrane 605
9.6.4.2 Extensions to Include Effect of Pressure 606
9.6.4.3 Examples of Pseudo-Bond Graphs for Simple Diffusion Cases 606
9.7 Systems with Chemical Reactions 609
9.7.1 Summary 617
9.8 Chapter Summary 617
9.9 Problems 617

References 625
Index 633
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
xv

Preface

This book introduces modeling as a tool to support design, simulation, and control of engineered and natural physical
systems. Chapters 1 and 2 introduce mechanical, fluid, electrical, magnetic, and thermal system concepts and physical laws
using an energetic basis. This introduction is meant to consolidate and unify the approach to first-principles model formu-
lation. As such, these first two chapters are suitable for an introduction to dynamic systems modeling. Those familiar with
lumped-parameter modeling may choose to begin with the bond graph introduction in Chapter 3, referring to the material
in Chapter 2 as needed. Chapter 3 introduces bond graph concepts, construction, and the use of causality in the deriva-
tion of mathematical models as state space equations. At this point, only 1-port modeling elements are used, postponing
multiport modeling to later chapters. Chapter 4 focuses on proper formulation and evaluation of models. Equilibrium and
linearization is discussed, as well as other topics that can be useful for checking and developing models for effective use
in response prediction and design of systems. This chapter supports the development and use of complex dynamic system
models in engineering practice, which has become much more common given the ability to conduct simulation studies
using widely available commercial and open-source simulation software tools.
Another need for physical system models is for conducting linear system analysis and control. Essential topics in this area
are discussed in Chapters 5–7, while maintaining a relation to proper model formulation. Chapter 5 focuses on time-domain
linear system analysis, while Chapter 6 introduces frequency response modeling and analysis. This includes the use of
impedance methods within a bond graph framework, which can provide an alternative approach for deriving system trans-
fer functions. Chapter 6 also introduces two-port impedance modeling methods, which can be appealing in a wide range of
system types. In addition, two-port impedance methods are used to introduce distributed-parameter elements into system
models.
An introduction is given to the concept of feedback in physical systems and control in Chapter 7. This also includes the
combined use of bond graphs and block diagrams. An emphasis is placed on taking proper steps in forming effective models
for use in feedback control system analysis and design. Selected concepts from both classical and state-space control are
discussed in this chapter.
More advanced modeling methods are introduced in Chapter 8, including multiport system concepts and energy methods.
This chapter builds on basic one-port modeling methods emphasized in Chapters 2–4. The discussion in Chapter 8 provides
an opportunity for showing how more complex types of processes and systems, such as electromechanical devices, magnetic
devices and circuits, can be integrated into bond graph models. In addition, the integration of Lagrange formulations
within a bond graph framework is introduced, in particular for effective modeling of nonlinear mechanical systems. Finally,
multiport energy methods also lay a foundation for Chapter 9, which provides a more complete approach to modeling ther-
modynamic systems and processes. In addition to reviewing how thermodynamic effects can be modeled within a bond
graph framework, more discussion is given on thermodynamic equations of state, open systems, as well as an introduction
to modeling chemical reactions within a bond graph framework.
The chapters of this book include over 100 worked examples and approximately 300 unsolved end-of-chapter problems. In
addition, almost 100 solved problems and case studies are included as a supplement from the book’s website. The problems
in each chapter range from basic to more complex to allow usage in both undergraduate and graduate teaching. A concerted
effort was made to “harvest” and adapt problems from classical texts in system dynamics, from as far back as the 1960s.
In addition, to encourage cross-disciplinary learning and application of system modeling, many problems are drawn from
a wide range of discipline-specific texts or technical literature.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
xvi Preface

Historical context

Henry M. Paynter developed the bond graph method during the 1950s and 1960s while a professor at MIT. His oft cited
published course notes “Analysis and Design of Engineering Systems” (ADES) [1] introduced his bond graph concept in
1961, and Paynter teased a forthcoming book Ergs and Bits: the Flow of Energy and Signals in Engineering Systems. While the
latter book never came to fruition, key theoretical and practical developments of the bond graph method were made over
subsequent decades through numerous journal and conference articles, as well as books by his former students and others.
Paynter would later initiate a collaboration with the second author of this book (J.J.B.) in the early 1980s to resurrect Ergs
and Bits. Their collaboration included transforming the core curriculum in dynamic systems and controls for mechanical
engineers at the University of Texas at Austin (UT-Austin), while Paynter served as a visiting professor during the 1980s
(after retiring from MIT). An emphasis on modeling of physical systems at both the undergraduate and graduate level at
UT-Austin has continued to date. Sadly, Professor Paynter passed away in 2002, not long after the first author (R.G.L.)
was invited to help complete the book, which by then was titled Modeling of Physical Systems. In its final form, this book
embodies much of the philosophy inspired by Paynter, modernized and expanded over decades of teaching and research
with our own students at UT-Austin. The content has also benefited from advancements in the use of bond graph methods
by many practitioners, researchers, and related textbooks. One key goal of this book has been to reinforce how engineering
and technical innovation can take advantage of the insights gained from physical system modeling. Another goal is to help
build physical system literacy that enables engineers to feel comfortable working across discipline boundaries, enabled by
Paynter’s original vision for Ergs and Bits.

August 2, 2024 Raul G. Longoria and Joseph J. Beaman

Austin, Texas
Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
xvii

www.wiley.com/go/Modeling_of_Physical_Systems/1e
This book is accompanied by a companion website.
About the Companion Website

This website includes:

Solutions Manuals
Solved Problems
●

●
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
1

Introduction

This chapter introduces some basic concepts useful in modeling physical systems. Simple examples are used to illustrate
how these concepts are helpful when developing models. Making decisions about how models are composed of more basic
elements is a key step, as is adopting discipline-specific knowledge, unified within an energetic basis for modeling physical
systems using bond graphs. A short discussion of classical dynamics is presented for historical perspective as well as to
provide a comparison with the energetic basis for system modeling to be used throughout this book. Some preliminary
concepts in bond graph methods are introduced along with the concept of a word bond graph. Lastly, problems at the end
of the chapter are meant to review these concepts as well as to motivate using modeling and analysis methods to answer
practical questions about systems of interest.

1.1 System Modeling Concepts

Modeling is central to all activities in dynamic systems and control. The goal is to generate a model to describe the functional
behavior of a system, which is needed in almost any design or analysis problem. This model may take the form of an actual
physical prototype, although for many systems the time and cost may be prohibitive. For this reason, an emphasis in this
book will be on the development of mathematical models, especially those that can be effectively solved by computer-based
analysis and simulation. Once a valid “computer model” has been developed for a system, the effect of design changes,
parameter variations, and the influence of different stimuli on system performance can be studied expeditiously.
The type of model that we need to develop is often highly dependent on the engineering task at hand. There are basically
two main subclasses for the uses of models: (i) models for understanding and (ii) models for prediction. Consider the simple
case of a rock supported with a string and under the influence of the earth’s gravitational field as shown in Figure 1.1(a).
Assuming that the swinging motion is confined to a two-dimensional plane, a system model can be composed of a point
mass with a massless inextensible support as shown in Figure 1.1(b). This idealized model of the actual system can be
used to approximate the actual motion of the rock. The schematic representation conveys physical parameters based on
geometrical and material properties that are considered important: m, mass of the rock, l, distance from the fixed point to
the center of mass, g, gravitational acceleration, and 𝜃(0), initial angular displacement at time t = 0. The identification of
useful physical parameters of a system is a critical part in understanding the system performance.
Models for understanding can be an important part of a creative design. From an experiment, one could see that the
suspended rock when released with an initial (angular) displacement will oscillate with fairly uniform period. This uniform
period could be used as a basis for keeping time, but a more thorough understanding of variation from a constant period
would be needed in order to design a pendulum clock. An empirical approach would require performing a large number of
experiments with a combination of masses, lengths, gravity accelerations, and initial displacements, tabulating the change
in observed oscillation period. Such a method can be very inefficient and time-consuming. An alternative approach would
be to develop a mathematical description, such as,
ml2 𝜃̈ + mgl sin 𝜃 = 𝜏d , (1.1)
which results from a summation of torques about the fixed pivot, where 𝜏d represents
√ unmodeled disturbance torques.
Dividing this equation by the mgl and defining a nondimensional time as, 𝜉 = t g∕l, results in,
d2 𝜃
+ sin 𝜃 = 𝜏d ∕mgl. (1.2)
d𝜉 2
Modeling of Physical Systems: Simulation and Control, First Edition. Raul G. Longoria and Joseph J. Beaman.
© 2025 John Wiley & Sons Ltd. Published 2025 by John Wiley & Sons Ltd.
Companion Website: www.wiley.com/go/Modeling_of_Physical_Systems/1e
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
2 1 Introduction

Figure 1.1 (a) Physical system: string-supported rock and (b) idealized model for
understanding: ideal planar pendulum.
g
l
θ
m
(a) (b)

Then assuming that the disturbance torque is small compared to mgl sin(𝜃), this equation implies that the period can be
determined from a single set of parameters of m, g, and l, while varying only the initial angle. Although it requires prior
knowledge from dynamics, the insight gained in this approach represents a tremendous reduction in experimental effort.
Note that neither a closed-form nor a numerical solution of the differential equation for the “rock-pendulum” was needed
in order to derive benefit from the modeling effort; however, there are also circumstances when it is appropriate and desir-
able to obtain detailed solutions. This is especially the case when using models for prediction. For systems that are too
time-consuming or expensive to build and test, one can resort to detailed simulation for prediction of the unbuilt system.
Even if the system has been built, it may be difficult to perform extended and detailed tests, especially if it is in operation,
so that simulation can provide an alternative means for insight. One particularly important example is use of simulation
for conditions that could possibly be destructive to the system, and so experimental testing would be prohibitive.

1.2 General Steps in Modeling

Modeling in many respects is a mixture of art, science, and experience. Another key element is making decisions about what
is relevant to the modeling task at hand. The detail of the model is highly dependent on its intended use, and many ideal-
izations and approximations are used in any model. The following steps are intended as guidelines and not as hard-and-fast
rules for the modeling procedure.
a) Define the system. Before one can start the modeling procedure, the object or system of interest and its inputs and
outputs need to be identified. This is often done almost subconsciously but unclear identification can be the source of
modeling errors.
b) Divide and conquer. Separate the system into its essential components. The internal structure and function of each
component should be understood.
c) Interconnect. Identify the interrelationships between the components.
d) Quantify the components. By using physical theories, the functional description of the components can be described
in mathematical terms.
e) Formulate system equations. By using the component interconnections and mathematical descriptions, the system
equations can be formulated.
f) Analyze the system. Apply mathematical techniques to study the model’s behavior.
g) Test the model. Compare the predicted system behavior of the model against actual system behavior, if available, or
engineering judgment, if not. Change the model (repeating steps a–f) as necessary.
To illustrate the steps above, consider Figure 1.2, which illustrates a mass m suspended at the end of a rubber band. The
other end of the rubber band is held between two fingers. It is desired to construct a model that can be used to predict
the vertical displacement of the block. In the first step, defining the system, we idealize the two fingers as being capable of
prescribing a motion at one end independent of the motion of the rubber band and block, 𝑣fingers . If this approximation is
made, the system consists of the rubber band and the block. The inputs to the system are the finger motion and a gravity
force on the block. The system can be naturally divided into the rubber band component and the block component. The
rubber band is primarily a compliant (or spring-like) component while the block is primarily a mass component. Any
damping in the rubber band is ignored for now.
The interconnection consists of the geometric connection between the rubber band and the block and between the band
and the fingers. We are assuming implicitly by this statement that there is no rubber band slip between the fingers, and the
connection between the block and the band remains fixed. There is also a dynamic interconnection between the gravity
force and the block. The idealized model with these assumptions is shown in Figure 1.2, which implies purely translational
motion of the mass in the vertical direction.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
1.3 Definitions of System Modeling Concepts 3

Figure 1.2 (a) Mass held at end of rubber band, and (b) an idealized υfingers
model for understanding: base-excited spring–mass system with one
degree of freedom. g

kb

υblock

(a) (b)

A quantitative description of the components is found by assuming that the rubber band force is proportional to the
extension in the band and the velocity of the block is proportional to the momentum of the block. System equations are
formulated by using kinematic and dynamic relations. From kinematics, we then know the time rate of change of the
extension in the band can be found from the difference in the velocity of the fingers and that of the block. From dynamics,
we know that the time rate of change of the block’s momentum can be found from the difference in the rubber band force
and the gravity force. This entire process can be represented in mathematical form by identifying the key variables and their
relations as follows:
Frb = krb ⋅ xrb (xrb = rubber band extension)
𝑣block = p∕m (p = block momentum)
ẋ rb = 𝑣fingers − 𝑣block (kinematic relation)
These relations can then be used in Newton’s second law (dynamics) to write,

ṗ = Frb − Fgravity . (1.3)

These equations collectively form a dynamic model of our system. If we define xblock to be the position of the block and
xfingers the position of the fingers, these equations can be put in a more classical second-order differential equation form,

m̈xblock + kb xblock = kb xfingers + Fgravity . (1.4)

An analysis of this mathematical model would yield an analytical solution in the form,

xblock = A + B cos(𝜔t). (1.5)

The model predicts that the mass will move in an oscillatory fashion. If we were to conduct physical tests, we know that the
mass would not oscillate indefinitely: the motion would eventually die out. If we want to model such a damping effect, the
model would need to be changed by including a model element to represent any significant damping forces.

1.3 Definitions of System Modeling Concepts

Before proceeding further with our discussion of modeling, it will be useful to succinctly and more precisely define some
systems concepts which can be useful in the modeling process. Reference will be made to the diagram in Figure 1.3.

Figure 1.3 Graphical description of system concepts. System

boundary
Outputs

System (S)
Inputs
↑

S E
↑

E S
Environment (E)
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
4 1 Introduction

● System: An entity separable from the rest of the universe by means of a physical or conceptual boundary.
● Environment: All that is external to the system but which may interact with it across the shared boundary.
● System variable: Any characteristic of the system which may vary with space or time.
● Input variable: A system variable which is prescribed by the environment independently of what goes on in the system.
This independence is usually only an approximation.
● Output variable: Any system variable of interest, including certain state variables (see below).
● Model of a system: A symbolic, graphical, physical, or mathematical description of a system which duplicates the char-
acteristics of the system variables.
● State variables: The set of system variables required to define the system at any time t.
● State-determined system: A system for which a set of state variables can be determined for all future times t ≥ to given
the values of the set at time t = to and the values of all the system inputs for future time t ≥ to .
Selection of system boundary and inputs. It is important that both the system and its inputs be chosen with care.
Consider a wheeled vehicle traveling down a road at a known forward velocity as shown in Figure 1.4(a). There is a need
to build a model to predict the vertical motion of the vehicle.
When traveling on the road, the system would be chosen as the vehicle since system inputs, identified as the gravity force
and known wheel/road elevation, can be specified independently of what goes on in the system (vehicle). This is valid as
long as: (i) the road can be considered rigid compared to the movement of the vehicle, (ii) road elevation is known a priori
as a function of distance down the road, for example, zg (x), and (iii) the vehicle forward velocity is specified. Note that any
attempt to use road/wheel force as a system input would lead to a modeling error since these forces are strongly dependent
on the dynamic state of the vehicle system.
If the same vehicle is placed on a flexible suspension bridge as shown in Figure 1.4(b), system boundaries should be
extended beyond just the vehicle to include at least the bridge, since neither wheel/bridge vertical motion nor force can be
prescribed independent of the vehicle or bridge systems. These two subsystems, vehicle and bridge, must be included into
an overall system to effectively analyze the vehicle’s vertical motion.1
Selection of state variables. State variables are not unique. As will be discussed later, we may choose different state
variables to model a system. Consider the mass–spring–damper system shown in Figure 1.5(a), for which one possible set
of state variables would be the position and velocity of the mass.

g g
x
υ υ

zg (x)

(a) (b)

Figure 1.4 (a) Vehicle traversing a road with known elevation, zg (x), and (b) vehicle traveling over a suspension bridge.

System boundary
Fd

Fk x
F (t)
m υd

Fd

(a) (b)

Figure 1.5 (a) Mass–spring–damper system and (b) multivalued damper force relation.

1 Theoretically, all systems should be all-encompassing since any finite system affects everything else in the environment. Thus, system
boundaries and inputs are always approximations that require engineering judgment.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
1.4 Energy Basis for Physical System Modeling 5

Figure 1.6 Two different emptying vessels: (a) a tank where pressure,
P, is predominantly due to hydrostatic effects; (b) a tank where P may
be due to hydrostatic as well as dynamic effects.
V
h

h V
Q(t) Q(t) QR
QR
P P
(a) (b)

If the damper force, Fd , had a multivalued relationship with damper velocity, 𝑣d , as shown in Figure 1.5(b), the two
proposed state variables would not be able to uniquely define the state of the system. It would be necessary to introduce addi-
tional state variables to describe which branch of the function, upper or lower, would be need to determine the value of Fd .
Now, if we define x1 = x and x2 = ẋ = 𝑣x , and if Fd is assumed to be a single-valued function, that is, Fd = Fd (x2 ), then a
set of state equations can be written as,
[ ] [ ]
ẋ 1 x2
= . (1.6)
ẋ 2 −Fd (x2 )∕m − Fk (x1 )∕m + F(t)∕m

The notation adopted here, Fd (x2 ), is used to express that Fd “is a function of” x2 . Now, if we are given initial conditions,
x1 (0) and x2 (0), as well as the force input, F(t), it is possible to solve, at least numerically, for x1 (t) and x2 (t). One approach
to doing this is presented in Solved Problem A-1-4.
Making model element decisions. Modeling systems requires making assumptions about how to represent key phys-
ical components. To illustrate, consider the fluid-filled tank or vessel which is shown in two distinct forms in Figure 1.6.
This is a common element found in many practical engineering systems. In this case, both tanks have a specified input
inflow of Q(t) and empty with an unknown (output) flow rate, QR . It is often assumed in hydraulic systems such as these
that a “tank” is a liquid-storing vessel with a well-defined amount of liquid. In addition, when the ratio of height to area is
small, as for the tank on the left, and the fluid exist via a small orifice, it is usually assumed that the level of the contained
fluid falls at a rate such that we can assume the pressure at the bottom of the tank is only dependent on the height of the
fluid; that is, pressure is due to hydrostatic effects only, that is, P = 𝜌gh.
This assumption allows a simple model for predicting volume of fluid in the tank, V – , to be derived from mass continuity,
–V̇ = Q(t) − QR , assuming incompressible flow. We would need to rely on knowledge of fluid mechanics to relate QR , the
exiting flow, to the volume state variable, V –.
Now, we might use the same assumption for the tank in Figure 1.6(b), although there may be pause to consider dynamic
effects as well as more significant losses. It is not clear whether the pressure at the bottom of this elongated tank is only due
to the hydrostatic pressure (related to fluid height). A decision needs to be made based on past experience, by comparing
the first model to available data, by building the two systems, or introducing additional fluid flow physics.
For many systems, it may be possible to rely on insight and intuition along with knowledge of physical systems to build
useful and reliable system models. Experienced system dynamicists in many disciplines are able to formulate highly com-
plex models, relying on skills and expertise to arrive at proper mathematical and computational models. This book aims
to support these approaches; however, we also introduce an approach that emphasizes structured system modeling using
bond graphs. This approach not only conveys the interconnection of system elements in a unified manner but also supports
consistent and proper equation formulation. The method is introduced with relatively simple problems, but the approach
becomes particularly helpful (if not essential) as complexity grows. The remaining sections in this chapter lay out some
foundational concepts for such an approach. These ideas form the basis for the content of Chapters 2 and 3.

1.4 Energy Basis for Physical System Modeling

A unified approach for modeling physical systems makes it possible to formulate functional representations for systems
composed of components and processes from multiple energy domains and their interactions. The diagram in Figure 1.7
illustrates how practical engineering systems can require consideration of multiple energy domains. Understanding how
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
6 1 Introduction

motors, generators, etc.

Mechanical
turbines, pumps, etc.

e/m rheological

Electrical conduction Fluid

losses
batteries heat exchangers,
friction
Chemical Thermal
exothermic reactions

Figure 1.7 Physical energy domains that will be covered in this text.

to model multi-energetic systems requires dealing with principles and methods for modeling mechanical translation and
rotation, electrical, fluid, thermal, and chemical components, as well as the devices and processes that exchange energy
between these energy domains.
Modeling these types of physical systems involves an extremely broad range of objectives, tasks, and methods. It may not
always be necessary to investigate a system at its most fundamental physical level. Indeed, many modeling studies begin
with a functional description, perhaps in the form of algebraic and/or low-order differential equations. What would follow
then might include analysis, model verification and validation, or control investigations. This book will discuss these types
of activities. To begin with, however, consideration is given to concepts and principles useful for understanding how a bond
graph approach forms the basis for modeling multi-energetic systems.

1.4.1 Reticulation (in the Spirit of Paynter and Russell)

Reticulation is the process of endowing a system with structure. – Henry Paynter2
To exhibit the structure of an object is to mention its parts and the ways in which they are interrelated. – Bertrand
Russell

Paynter referred to the process of dividing a system into its components as reticulation [1], a conceptual process that can
be described in the following three steps:
1. Separate the system from the universe. This is called the System
“prime reticulation”. boundary

System (S)

Environment (E)
2. Subdivision: divide the system into its essential elements, System
determined by the modeling task at hand. Elements boundary
System (S)

Environment (E)
3. Delineate the important bonds of interaction between System
each of the elements and the environment. (Note that there boundary
should be at least two lines for each interaction since, in System (S)
general, elements must communicate both ways with each
other and the environment.)

Environment (E)

2 Henry Paynter developed the bond graph approach as a professor at the Massachusetts Institute of Technology starting in 1959.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
1.4 Energy Basis for Physical System Modeling 7

υx
F1 F2 h
m Q(t) QR

Tank
F1 F2 Q(t) P
Q(t)
Mass Inlet P QR QR
υx υx Pin Orifice
Pout

(a) (b)

Figure 1.8 (a) Mass with two external forces and (b) simple tank with input flow supply and orifice flow.

The use of a system boundary to partition the system of interest from its environment should be a familiar practice,
especially to anyone who has studied thermodynamics. As will be discussed in more detail later in this chapter and in
other parts of this book, this step can have some critical consequences on the validity of the model. On the other hand,
there are many system modeling problems where these steps can be accomplished with very little thought. The end result
should be an abstraction formed by model elements and interconnections that are properly defined. This is what Paynter
meant by endowing a system with structure. Part of this refers to the system element interactions. The use of a signal shown
with a full arrowhead, →, is common for representing information flowing from one system to another. A bilateral com-
munication as indicated between elements in step 3, however, is a means for representing a physical connection between
two elements.
The conceptual reticulation process is demonstrated in Figure 1.8 for two simple systems. In Figure 1.8(a), a single mass
has two input forces applied along the x-axis, so a system here shows the environment specifying F1 and F2 onto the system,
which is composed of a single element, Mass. Note that we would expect the mass to move with a velocity according to
Newton’s second law. In a system diagram, the Mass element returns 𝑣x via a signal to the environment, E.
In Figure 1.8(b), the simple tank from Figure 1.6(a) is reticulated into three elements: Inlet, Tank, and Orifice. We model
the inflow Q(t) here as specified by the environment, passing through the Inlet to the Tank. On the Orifice element, the
ambient pressure is specified by the environment on one side while the Tank specifies pressure, P. Note that the orifice then
communicates QR to the environment as well as the Tank.
These conceptualizations convey modeling decisions that are readily made for familiar systems. Even causal relationships
between the two signals at each interaction should be intuitive (see Section 1.6). These examples are meant to convey steps
that can be taken in all models we build, although not necessarily in this form. The value of these steps in the modeling
process becomes evident as complexity increases. In addition, even for such a simple case as the tank in Figure 1.8(b), it
becomes evident that the bilateral communication (using two signals) between Tank and the environment at the inlet is
problematic. It should become clear after some thought that Tank could inform Orifice about P, if the model is able to
– . However, it is desirable to also know Pin , the pressure at the inlet. Is the model
determine the fluid height, h, or volume, V
complete enough to determine this quantity? Is it necessary to find Pin , and why would it be needed? A particular modeling
application would put such questions in context. In the following, we motivate the use of energy and power flow as one
basis for identifying elements and interactions that play a significant role in the behavior of a system.

1.4.2 Energy Conservation and Continuity

There remains for us one enunciation of the principle of conservation of energy: There is something which remains
constant. – Henri Poincare

The above discussion of reticulation is general to all system modeling. However, our particular study emphasizes physical
systems. A physical system is defined as one in which the primary significant interactions are associated with exchanges
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
8 1 Introduction

of energy. Recall that energy is defined as the ability to do work or the ability to push through a distance. As illustrated in
Figure 1.9, work can be represented mathematically as,
2
work = F ⋅ ds. (1.7)
∫1
The principal aspect of energy is its conservation. Energy can be transformed from one form F 2
to another, but the total amount, in all forms, is conserved. Although this concept now seems
self-evident, it was a novel idea in the mid-19th century. The principle of conservation of energy was
arrived at independently by Sadi Carnot in France, Julius von Mayer and Hermann von Helmholtz ds
1
in Germany, James Joule in England, and L. A. Colding in Denmark. A familiar expression for
energy conservation occurs as the first law of thermodynamics, namely, Figure 1.9 Work
definition.
ΔE = ΔQ − ΔW, (1.8)

where E is the total energy, ΔQ is the heat transfer, and ΔW is the generalized work transfer which
must include work in its many forms.
Equation 1.8 helps solve for the change in energy from one state of a system to another. Many practical problems can be
solved by imposing this conservation principle. However, dynamic system modeling seeks to address how physical states
of a system change continuously over space and time.
During the same period that energy conservation was put forward, a field-continuum approach emerged following the
pioneering efforts of J.B.J. Fourier, S.B. Poisson, G. Green, K.F. Gauss, G.G. Stokes, and others, it was James Clerk Maxwell
who first considered localization of field energies in the form of energy densities (or distributed energy per unit volume)
analogous to mass density (or speciated mass densities). N.A. Umov (1874), O. Heaviside (1884), and J.H. Poynting (1884)
then independently further developed this concept of localization of energy. Simply stated this implies that energy is strictly
conserved both with time and with space changes, even infinitesimal ones. Localizing first in time implies,

Ė = Q̇ − W,
̇ (1.9)

and then in space implies,

𝜕𝜀
= −∇ ⋅ S − 𝜎, (1.10)
𝜕t
where 𝜀 is the energy density or energy per unit volume, ∇ is the gradient operator, S is the power flux (Poynting vector)
or energy transport per area, and 𝜎 is the available power dissipated per volume. This extension was perhaps best stated by
Heaviside:

The principle of the continuity of energy is a special form of that of its conservation. In the ordinary understanding
of the conservation principle it is the integral amount of energy that is conserved, and nothing is said about its
distribution or its motion. This involves continuity of existence in time not necessarily in space also.
But if we can localize energy definitely in space, then we are bound to ask how energy gets from place to place. If it
possesses continuity in time only, it might go out of existence at one place and come into existence simultaneously
at another. This is sufficient for its conservation. This view, however, does not recommend itself. The alternative is
to assert continuity of existence of space also and to enunciate the principle thus:

● When energy goes from place to place It traverses the intermediate space.

This is so intelligible and practical a form of the principle, that we should do our utmost to carry it out.

Yet curiously, Heaviside made no allusion to the early aphorism of Thomas Aquinas:

– Angels in traveling from place to place pass through the intervening space

A physical system can then be viewed as a system residing in an energy flux field, S,⃗ as shown in Figure 1.10. The system
can either locally store this energy, dissipate the energy into unavailable forms, or else simply pass it through the system.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
1.4 Energy Basis for Physical System Modeling 9

Figure 1.10 Physical system in an energy flux field.

System

These now well-established principles are fundamental to many fields of science and engineering. Formulation and appli-
cation of the differential and integral forms of the energy equation as used to solve field continuum problems can be found
in fields such as electromagnetics [2], acoustics [3], fluid dynamics [4], and solid mechanics [5].
To ensure that system level models are compatible with a strict energetic basis, the energy continuity equation is first put
into integral form (using the divergence theorem [1, 4]). The integral form allows derivation of a discretized (or approximate)
form of Equation 1.10,
∑
m
di ∑
n
∑
p
= j − (d )k
i
dt j k

which forms the basis for a strict accounting of energy flow among the m lumped energy stores, n power flows, and p discrete
dissipators of power. Note that this reticulated form of the energy continuity equation has units of power, and power flows
j are positive into a system while defined dissipative power flows (d )k are positive flowing out of the system.

1.4.3 Power Balance in Multiport Systems

We now have a basis for assuming that energy interactions can be lumped into a discrete number of global interactions.
Commencing with field or continuum representations, this can always be done in a space average sense. For example, a
specific power flux can now be found,

j = S⃗ j ⋅ dA, (1.11)
∫
AK

where dA is the boundary element, S⃗ j is a particular form or mode of energy flux, and AK is the corresponding surface area
for power flow. These power interactions occur through power ports, as shown in Figure 1.11(a). These ports are usually
subdivided conceptually to distinguish between different energy domains, that is, mechanical, electrical, etc. This can be
represented diagrammatically with power bonds in the bond graph of Figure 1.11(b). In this case, the power bonds represent
an energy interaction (or power flow). The half-arrow on a power bond designates the positive power flow direction. For
example, the power flows from the system to the environment via bond 3, whereas all others show power into the system.

System System

Environment

(a) (b)

Figure 1.11 (a) Lumped description in terms of power ports and (b) word bond graph representation of a multiport system with ports
1, 2, 3 … N.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
10 1 Introduction

u1 Figure 1.12 System with input–output power pairs at multiple ports.

y1
uN
u2
System yN
y2

u3 y3

The accounting of energy in the system can be found from equating the net rate of energy into a system to the rate of
energy stored in a system, or,
∑
n
̇
j = , (1.12)
j=1

where here there is only one form of energy stored. Modeling errors can occur if significant power interactions are omitted in
the description. As discussed above, we still must communicate in both directions. This implies we need a pair of variables,
at a minimum, at each power port as depicted in Figure 1.12. In this figure, the convention is adopted that uj is an input at
port j while yj is the corresponding output.
Although not mandatory, we are always free to choose the type of variables for this pair, as long as their product is equal
to the power at the port. Thus the use of the expression power-conjugate pair. This idea of power products can be traced back
to Newton and the Principia (1686) [6] where he defines action (power) to be the product of force and velocity. Consider
the following excerpt from Motte’s 1729 translation of the Principia:

“For if we estimate the action of the agent

from its force and velocity conjunctly...”

Following Newton’s lead, we take the relation below for the net power into a physical system,
∑
n
∑
n
j = uj yj . (1.13)
j=1 j=1

Besides this Newtonian lead, a further thermodynamic justification for the use of power will be presented in Chapter 3.

1.5 Relation to Classical Dynamics

As suggested above, many of the concepts that we will use in our modeling procedure are based on ideas from classical
dynamics. In this section, a short description of some of the early concepts is given. Revisiting these concepts enables us to
balance how we’ve traditionally applied them with methods that are also applicable to a broader class of physical systems.
Classical dynamics historically traces its axioms to the pioneering work of Galileo and Newton. Newton was born in the
year of Galileo’s death, and published his acclaimed Principia in 1686 in which he states the axioms or laws of motion. Since
there have been many interpretations of Newton’s laws in texts, it seems appropriate to state his laws as he himself states
them. The Principia was written in Latin and would not be readable to many, but fortunately there was an early English
translation in 1729 due to Motte who was a contemporary of Newton. Newton’s three laws of motion as given in Motte are:

Law I – Every body perseveres in its state of rest, or of uniform motion in a right line, unless it is compelled to change that state
by forces impress’d thereon.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
1.5 Relation to Classical Dynamics 11

Law II – The alteration of motion is ever proportional to the motive force impress’d; and is in the direction of the right line in
which that force is impress’d.
Law III – To every action there is always opposed an equal reaction: or the mutual actions of two bodies upon each other are
always equal, and directed to contrary parts.
Newton also gives eight definitions before stating these laws:
Definition I – The quantity of matter is the measure of the same, arising from its density and bulk conjunctly. In modern
terminology, this defines mass to be equal to density times volume, m = 𝜌– V.
Definition II – The quantity of motion is the measure of the same, arising from the velocity and quantity of matter conjunctly.
Here, he defines motion to be what is now called momentum, p = mV. The next six definitions are related to Newton’s
meaning of force, which is slightly different than later interpretations.
Definition III – The Vis Insita or innate force of matter is a power of resisting, by which every body, as much as in it lies,
endeavors to persevere in its present state, whether it be of rest, or of moving uniformly forward in a right line.
Definition IV – An impress’d force is an action exerted upon a body, in order to change its state, either of rest or of moving
uniformly forward in a right line.
Definition V – A centripetal force is that by which bodies are drawn or impelled, or any way tend, towards a point as to a centre.
Definition VI – The absolute quantity of a centripetal force is the measure of the same, proportional to the efficacy of the cause
that propagates it from the centre, through the spaces round about.
Definition VII – The accelerative quantity of a centripetal force is the measure of the same, proportional to the velocity which
it generates in a given time.
Definition VIII – The motive quantity of a centripetal force, is the measure of the same, proportional to the motion which it
generates in a given time.
From these definitions, it appears that Newton’s concept of force was that of a general cause to change the state of the
movement of matter. This cause is then quantified in terms of absolute, accelerative, and motive components. The motive
force is proportional to the momentum change, Δp, in a time interval, Δt. Motive force is called impulse in modern parlance.
The second law can then be stated as,
Δp = F = FΔt, (1.14)
where F and F are the net impulse and force, respectively.
For the third law, Newton seems to have had more in mind than the modern meaning of the law as “equal and opposite
forces.” As mentioned in the previous section, he also seems to have power balance in mind. The third law, as noted by
Thomson (Lord Kelvin) and Tait in their Treatise on Natural Philosophy (1879):

The foundation of the abstract theory of energy is laid by Newton in an admirably distinct and compact manner in the
sentence of his scholium [to the third law] already quoted in which he points its application to mechanics. The actso
agentis, as he defines it, which is evidently equivalent to the product of the effective component of the force, into the
velocity of the point on which it acts, is simply, in modern English phraseology, the rate at which the agent works.

In modern terms this yields a power balance equation. It can be helpful to see how energy and power flow concepts can
be used to derive dynamic equations. Consider the simple mass with forces applied by the environment from Figure 1.8(a).
For this problem, we define a coordinate frame, draw a free-body diagram, and apply Newton’s laws as in Figure 1.13.
∑
Newton’s second law is often written, F = ma, but here use the more general case, ṗ = F, where p = m𝑣 is the
translational momentum.

υx
F1 F2 F1 F2
y m
m

x
(a) (b)

Figure 1.13 Mass with applied forces and associated free-body diagram and coordinate frame.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
12 1 Introduction

Figure 1.14 Energy and power flow in a system representation.

1= F1vx 2= F2vx
1 mv 2
2 x

So, from a classical dynamics perspective, with motion only in the x direction,
d𝑣 ∑
ṗ x = m x = Fx = F1 − F2 .
dt
Given known forces, the mass velocity, 𝑣x , can be found over time.
Now, define the system here as a mass which stores kinetic co-energy,3 k = m𝑣2 ∕2, and has power flowing in from the
applied forces as shown in Figure 1.14. Assume there are no dissipation effects.
To quantify the power flows, for this example, a sign must be assumed for each input force. Let F1 be applied in direction
of 𝑣x and F2 opposes the motion – so it is emphasized here that the system diagram shows a negative power flow out of the
system boundary. Now the energy continuity equation gives,
dk
= 1 − 2 ,
dt
where 1 = F1 𝑣x and 2 = F2 𝑣x . So,
( ) d𝑣 d𝑣
d 1 2
m𝑣x = m  𝑣x x = F1  𝑣x ⇒ m x = F1 − F2
𝑣x − F2 
dt 2 dt dt
which gives the same result found by using Newton’s law. This simple example illustrates how keeping track of power flows
and stored (and dissipated) energy enables us to build dynamic system models.

1.6 Power Flow in Physical System Modeling

We will graphically represent significant exchanges of energy between two systems S1 and S2 using a power bond:

S1 S2

The power bond here is simply a solid line when a positive direction for power flow has yet to be defined. Recall as in
Figure 1.11(b), a half-arrow is used to assign positive power flow. So, if we chose positive power flow being from S1 to S2 ,
this would be indicated by:

S1 S2

The power flow on a given bond can be quantified by power conjugate variables, meaning that their product equals
instantaneous power flow. It may be familiar to think of factoring power into two components, such as force–velocity,
voltage–current, or pressure–flow rate. For example, the concept of “shaft power” is one that mechanical engineers readily
relate to as the product of torque and shaft angular velocity. Along these lines, we will define effort, e, variables, and flow,
f , variables in each energy domain such that power flow is,
 = e ⋅ f units in Watts
and label power bonds as shown below:

e
S1 S2
f

to indicate positive power flow from S1 to S2 . A summary of power-conjugate variables for energy domains of interest are
summarized in Table 1.1. These variables can help us build models that are energetically correct, can be related through
physical laws to derive dynamic mathematical models, and also help us see analogies between systems. It should be clear
that each (e,f ) pair above yields power.

3 We’ll discuss how energy and co-energy relations are defined by physical variables beginning in Chapter 2.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
1.6 Power Flow in Physical System Modeling 13

Table 1.1 Power-conjugate variables for energy domains of interest.

Energy domain Effort variable Flow variable Power, 

General Effort, e Flow, f e⋅f

Mechanical translation Force, F velocity, 𝑣 F⋅𝑣
Mechanical rotation Torque, 𝜏 Angular velocity, 𝜔 𝜏 ⋅𝜔
Electrical Voltage, V or e Current, i V ⋅i
Hydraulic Pressure, P, Volumetric flow rate, Q P⋅Q
Thermal Temperature, T Entropy flow rate, fs T ⋅ fs
Magnetic Magnetomotive force, M Flux rate, f𝜙 M ⋅ f𝜙
Chemical Chemical potential, 𝜇 Molar flow rate, fN 𝜇 ⋅ fN

1.6.1 Using Power Flow in Modeling

Identifying significant power flows in a system is one way to start modeling a physical system. For some systems, this can
be done at a conceptual level by subdividing the system into component elements and identifying not only power flows
but also signal flows. This was alluded to in Section 1.4.1. Recall how the power interactions were represented by bilateral
signals, ⇄. Each signal represents one of the power conjugate variables in each interaction. A power bond is equivalent to
the bilateral signals:

S1 S2 S1 S2

Clearly the single power bond is less cumbersome and preferred. Systems may also require use of a single signal bond, →,
to show that information flows from one point to another without the flow of power. Signal flows are used for representing
control systems, for example, in the form of block diagrams. Later, however, we will see that signals play an essential role
in physical modeling and complement the use of power bonds.
Power flow within the thermal domain is quantified by thermal = T ⋅ fs , as indicated in Table 1.1. In this book, a dashed
power bond will be used to distinguish thermal power flow from that in other energy domains. Consequently, two systems
exchanging thermal power T ⋅ fs between systems S1 and S2 would be connected by a dashed power bond:

T
S1 S2
fs

Note that T ⋅ fs has units of power and is equivalent to heat transfer flow rate (as will be discussed further in Chapter 2).
Signal flows are also useful for understanding what will become one of the most important features of a bond graph
description: causality. Causality in a bond graph model refers to the relationship between the two power variables on a
given power bond. A bond graph without causality is referred to as acausal, and provides insight on the significant power
flows between system elements. The following example illustrates the development of an acausal word bond graph, a
high-level model description that emphasizes power flow between model components that are not yet fully understood.
Those components are simply represented by a word description.

Example 1.1 Power flow in a shipboard propeller drive system

Figure 1.15 shows a cutaway schematic of a shipboard drive and propeller. Develop a word bond graph that uses power flow
between key system components in this system.
Solution
Figure 1.16 shows a word bond graph for the ship drive and propeller system. This diagram can serve as a conceptual first
stage in a model development process. At this stage, we identify only key components and their high-level interconnection.
Power flow is positive from the power sources (prime movers) to the propeller.
From the propeller, power flows to the translational mass of the ship. Here we have inserted a “1” junction, as will be
defined later in Chapter 3 to represent a point in a physical system at which the connected power bonds have the same
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
14 1 Introduction

Thrust
Strut bearing
Main Sprung
shaft bearing Prime
Main mover
Water
column
reduction Prime
Strut gear
bearing Stern tube mover
Propeller bearing
Direction of Hull
reactive force
(thrust)

Figure 1.15 Cutaway schematic of a ship drive and propeller.

Ship FL F Effective
loads υL υ ship mass
Fp τ1 Power
τp Energy τd Transmission source
ω1
Propeller Bearing stored Bearing with losses
ωp in shaft ωd τ2 Power
Tb2 fsb2 Tb1 fsb1 Tt fst ω2 source

Figure 1.16 Cutaway schematic of a ship drive train.

(or common) flow, in this case velocity. Note the use of dashed power bonds to show heat transfer from the bearings and
transmission.
Chapter 2 will introduce concepts and principles needed to model components such as those in this word bond graph at
much greater detail, as needed to build mathematical models. An understanding of assumptions made in modeling basic
physical elements, such as the shaft, is essential when constructing model equations and graphical model descriptions such
as a bond graph, as will be discussed in Chapter 3.

Causality is used to specify the input–output relation between the effort and flow variables at the ports of two systems
connected by a power bond. For example, if we know that a system S1 specifies effort into a system S2 , this is indicated by
using a causal stroke as shown below.

Causal stroke

e
S1 S2 S1 S2
f

The meaning of causality can also be conveyed using bilateral signals, as shown in the diagram above to the right.
Assignment of causality on all the bonds of a system bond graph will follow very specific rules, as will be discussed in
Chapter 3. In some simple cases, the assignment is straightforward. For example, if a motor was being controlled such that
it had a known shaft speed, 𝜔, then causality could be shown on a simple word bond graph such as:

τ
MOTOR ω LOAD

where now the half arrow indicates positive power flows from the MOTOR to the LOAD. The causal stroke indicates that
the MOTOR is specifying the speed to the LOAD. These are two independently assigned pieces of information about the
connection between these two components.
When causality is used effectively, we factor in the specific way that variables are related on a bond. For example, the
example above says that the LOAD has angular velocity as an input and it returns as an output the torque to the MOTOR.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
1.7 Outline of the Text 15

Causality helps us to see that we need to find a causal functional relation for the LOAD that gives 𝜏 when given 𝜔. This is
somewhat obvious when dealing with two systems interacting with a single power bond. The true advantage of causality
will become evident when it is used to reveal the key relationships between system variables across a complex system bond
graph.

Example 1.2 Shipboard prime-mover testing

An early stage testing program fits a Prony brake onto a test shaft for a gas engine prime mover proposed for the shipboard
drive system in Example 1.1. A throttle control signal can be sent to the engine to increase speed. A preliminary word bond
graph for steady-state testing could be drawn as follows:

Throttle

τ Prony
Engine
ω brake

The Prony brake exerts a load torque \textbackslash on the engine based on speed according to, 𝜏B = 𝜏o sgn(𝜔), where 𝜏o
is an adjustable constant. So, given a set engine speed, 𝜔e , the torque load on the engine would be 𝜏e = 𝜏o sgn(𝜔e ). The
“signum” function, sgn(), is a function that gives +1 if 𝜔 > 0 and −1 if 𝜔 < 0; that is, it accounts for the sign of the velocity
and opposes the motion of the shaft. This Prony brake model function is only applicable when 𝜔 ≠ 0.

1.6.2 Power Flow Definitions and Concepts

The following are general concepts and principles about systems that make use of power flow descriptions.

● Equilibrium. A system is in equilibrium when all states, xi , no longer change with time, or ẋ i = 0. As a steady-state test,
the equilibrium performance of the system in Example 1.2 was of primary importance.
● Static equilibrium. A system in equilibrium with all power flows zero; that is, ej fj = 0, for all j power bonds in a system.
A simple example of a static equilibrium might be a simple mass sitting on spring in a gravity field. In this state, there is
no power flowing between the mass and the spring.
● Stationary equilibrium. A system in equilibrium with all power flows constant; that is, ej fj =constant, for all j power
bonds in a system. The power flow between the engine and the Prony brake in Example 1.2 for a given speed setting
would define a stationary equilibrium since T𝜔 = constant.
● Source-load system. A system in equilibrium formed by connecting a power source, S, to a load, L. If the source is
a type that specifies effort, e, then the load returns flow, f . Otherwise, the source may specify, f , so the load returns
effort, e. This concept was demonstrated by Example 1.2 for the case where the flow (speed) was specified into the load
(brake). Source–load systems and their analysis have significant practical applicability [7]. In early stage analysis and
testing, source–load analysis enables assessment of system source characteristics, power requirements, and insights into
performance capabilities that can aid design and source selection.

1.7 Outline of the Text

In modeling a physical system, an engineer should first have a basic understanding of the laws applicable in various physical
domains. Thus, in Chapter 2, we start providing a review of a restricted set of physical systems that should be familiar, but
a few new generalized concepts are introduced. Most importantly, an energetic perspective is emphasized following on the
concepts introduced in this chapter. This will form a basis for subsequent chapters.
Chapter 3 builds on the concepts and principles reviewed in Chapter 2 and formalizes and extends these to define our
basic modeling construct, the bond graph. Bond graphs will act in many ways as a “structured programming language”
for physical modeling and provide the basis for much of our discussions in the rest of this book. The important concept
of causality assignment in bond graph models is discussed in more detail. Chapter 4 builds on this basic knowledge and
introduces systematic formulation of mathematical models from bond graphs, especially taking advantage of causal bond
graphs. Chapter 4 then provides further introduction to model formulation, emphasizing the use of bond graphs to evaluate
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
16 1 Introduction

the model and to derive mathematical equations. This chapter also introduces some model analysis techniques including
solutions for equilibrium problems and linearization. Some concepts useful for setting up numerical simulations are also
discussed in Chapter 4.
Chapter 5 provides a review of linear analysis tools for systems, focusing on time-domain analysis methods. This chapter
is followed by an introduction to transfer functions, impedance methods, and frequency response analysis in Chapter 6.
Given this foundation, Chapter 7 examines the role of system models in feedback control. Useful methods for system spec-
ification and design synthesis are introduced, and then an introductory account of signal and feedback to measure and
control physical systems is provided.
The remaining chapters begin to build a more advanced and sophisticated basis for modeling of physical systems.
Chapter 8 introduces energy methods and the key role they play in multiport modeling of physical elements, including use
of Lagrange and Hamilton’s equations as well as minimization principles. Finally, Chapter 9 presents further examination
of thermodynamic system modeling, including approaches that can be useful to model open systems and chemical systems.
Throughout the text, numerical and mathematical techniques are introduced as needed. Examples in system simulation
are introduced within the text as well as in the set of Solved Problems for each chapter. Various unsolved problems in each
chapter provide additional opportunities for building expertise in these methods.
It is important to note that the subject matter of this text cannot supplant the necessity for taking basic mechanical,
electrical, fluid, chemical, and thermal courses. The intent of this book is to provide the reader a good background in tools
useful for integrating system elements from all of these domains into an overall system model. It is expected and essential
that the reader refer to background reference materials, especially to fill knowledge gaps and to solidify and enhance their
working knowledge of subject matter.

1.8 Problems
Problem B-1.1 Weather Vane In a steady wind, a weather vane points into the wind, but when the wind direction is
changing, the vane cannot be trusted to be pointing in the right direction. (a) Define and sketch your model of the
weather vane system and identify input variables, output variables, and key physical parameters that you would need
to predict the response of the weather vane to its environment. (b) What do you deem as the “hardest” part(s) about
building a model of this system? (c) Sketch how you understand the position of the vane will change over time if the
wind suddenly shifted, say, from blowing toward the south-west to blowing due south.

Figure B-1.1 Weather vane.

Problem B-1.2 Archer’s arrow The archer fires an arrow while standing 50 yards away from the intended target. If we
consider the arrow as our system and wish to know where the arrow will land: (a) What are the system input variables
and system parameters? (b) What are the system output variables? (c) What assumptions are we likely to make in
modeling the system?

Problem B-1.3 Machining process Consider the machining process shown below. It is of interest to understand how
the cylindrical bar stock being cut might vibrate for a given feedrate, 𝑣feed , which is a known input variable. Determine
whether the cutting forces (e.g., FN is one component) and the rotational rate, 𝜔m , are input or output variables. Discuss
the static and dynamic characteristics of this turning process. Presume that the cutting tool is mounted on a very rigid
tool holder and that the jaws are much stiffer than the bar stock.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
1.8 Problems 17

bar

ωm
υfeed

FN
Tool bit holder

Figure B-1.3 Cutting bit interacting

with cylindrical bar on lathe.

Problem B-1.4 Vehicle ride The ride behavior of the ground vehicle shown in Figure B-1.4 depends on vibrations that
result when the vehicle moves over a given terrain. Key concerns may include how occupants in the “suspended mass”
react to forces from the suspension. It is common to develop a “quarter-car model,” where one-fourth of the suspended
body mass is assumed to move only in the vertical direction, focusing on one wheel/tire and suspension. The suspension
is idealized by ideal spring and damper elements, suspending the body mass above a corresponding wheel/axle mass.
The wheel/axle mass, or “unsuspended mass,” connects to the ground via the tire, also commonly modeled for ride
analysis by ideal spring and damper elements.
a) Neatly sketch a mechanical system schematic as you deem suitable to convey the modeling assumptions conveyed
above. Label the ideal elements in your system schematic and identify input variables and outputs of interests.
b) How many state variables do you think you would need for this system (no need to derive the mathematical model,
just make a good guess)? List those variables.
c) Describe and provide quantitative values for two conditions: (i) the forces in the tire elements (spring and damper)
when the vehicle is not moving, sitting statically; and (ii) the forces in the tire elements when the vehicle is moving
but the tire has lost contact with the ground.

Forward speed
υx

zr Road profile zf
Reference

Figure B-1.4 Ground vehicle.

Modeling and Analysis Review Problems

Problem B-1.5 Component relations Suppose in a real mechanical component we measure the velocity and we then
calculate the power or energy stored in the component. We find that the velocity, 𝑣, is given by 𝑣 = ay3 , where “a” is a
constant and “y” is a particular variable.
a) If y is the force, F, what is the power dissipated in the component in terms of 𝑣?
b) On the other hand, if y is the linear momentum, p, what is the energy stored in the component in terms of p?
c) For each of the cases above, identify what type of ideal element describes the behavior (mass, spring, or damper).

Problem B-1.6 Force for specified motion It is desired to impart to a body of mass, m, a velocity given by the expression,
𝑣m = At − Bt2 , during the time interval from t = 0 to t = tf seconds and a velocity of zero at all other times. (a) Find an
expression for the force, Fa , that must be applied on the mass to achieve the desired velocity in terms of the parameters
given. (b) For the case where m = 2 kg, A = 0.3 m/s2 , B = 0.075 m/s3 , and tf = 4 seconds, plot both the velocity and
the force as functions of time. Present solution in SI units.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
18 1 Introduction

Problem B-1.7 Spring force for prescribed motion It is desired to impart to the end of a spring of compliance, Cs , (the
other end of the spring being fixed) the velocity 𝑣s = At + Bt2 . (a) Find an expression for the force required, Fs , in terms
of A, B, Cs , and tf . Note that the spring displacement (or deflection) is the integral of velocity. (b) For the case where
Cs = 0.05 m/N, solve for and plot the force in SI units, with A = 0.3 m/s2 , B = 0.075 m/s3 , from t = 0 to tf = 4 seconds.

Problem B-1.8 Force actuator requirements It is desired to select a force actuator that can move a body of mass m = 2
kg with a velocity as plotted in Figure B-1.8. Determine the force required by the actuator in SI units.

1.25

Velocity, m/s
1
0.75
0.5
0.25
0
−1 0 1 2 3 4 5
Time, sec

Figure B-1.8 Velocity–time profile.

Problem B-1.9 Mass sliding on a surface A standard hockey puck is thrown along a surface with a known initial
velocity, 𝑣o . Consider the friction force to be approximately Coulombic, with 𝜇 = constant. How long would it take to
stop and how far will the mass travel for a given 𝑣o ?

Problem B-1.10 Spring force from specified deflection It is desired to impart to the free end of a spring of compliance
0.05 m/N and negligible mass a velocity as in Figure B-1.8. Plot the force required to produce this velocity with the
opposite end fixed at rest.

Problem B-1.11 Numerical simulation of mass–spring–damper system Rework the simulation in Solved Prob-
lem A-1-4, but now assume the spring is initially not deflected, x1 = 0 m, and the mass has an initial velocity so
x2 (0) = 𝑣x0 = 1 m/s. Use the same parameters as in the original solution provided.

Problem B-1.12 Torque-driven automobile If the rear-axle torque for each rear wheel of an automobile varies with time
as shown in Figure B-1.12 below, determine the forward speed of the vehicle as a function of time assuming negligible
bearing friction and negligible air resistance. The weight of the automobile is 3000 lbf and the radius of the rear wheels
is 15 inches. Find the distance traveled at the end of 20.0 seconds. Based on the answers obtained, do you think that
allowing for bearing friction and air resistance would have resulted in an appreciably different answer? [HINT: Bearing
friction and air resistance are approximately proportional to the square of automobile speed, and rear-axle torque
required to drive the auto at a steady speed of 60 mph is 200 ft-lbf.]

120
Torque, ft-lbf

80
40 t = 0.5 sec

0
0 5 10 15 20 25
Time, sec

Figure B-1.12 Torque profile for automobile.

Problem B-1.13 Robot on incline Steady-state performance for systems occurs when there is no longer any change in
states with time. In a mechanical system, for example, the acceleration of a mass would be zero. Consider steady-state
performance of a robotic vehicle climbing an incline as shown in Figure B-1.13. The drive system at steady state can
produce a total effective traction force, Ft , that can be described by the force–speed curve shown in Figure B-1.13(b).
The robot is climbing an incline with constant grade, 𝛼.
a) Assuming that any drag force on the robot is given as FD = 𝑣|𝑣|, find the steady nominal operating velocity of the
robot when it is moving on a level surface (𝛼 = 0). Solve using both a graphical approach and an analytical approach.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
1.8 Problems 19

b) At other times, it encounters inclines with maximum grade, 𝛼. If the total mass of the robotic vehicle is 3 kg, deter-
mine the reduction in forward velocity when climbing a 5% grade.

g 2

υ 1.5

Force, N
1
0.5
Ft α
0
0 0.5 1 1.5 2
Velocity, m/s

(a) (b)

Figure B-1.13 (a) Robot climbing incline and (b) force–speed curve
for effective traction.

Problem B-1.14 Energy required to charge capacitor For a capacitor, the voltage required to charge the capacitor a
charge q is proportional to q; that is, 𝑣 = q∕C, where C is the capacitance. Use  = 𝑣i where i is current to determine
the energy input to the capacitor required to charge it an amount q.

Problem B-1.15 Voltage response in circuit Within a certain piece of power circuitry, there exists a branch containing
L and C in series. A current with instantaneous value as plotted in Figure B-1.15 flows in the 10-henry inductor and
the 0.5 × 10−6 farad capacitor. Plot the voltage across the inductor and across the capacitor.

1.25
Current, mA

1
0.75
0.5
0.25
0
−1 0 1 2 3 4 5
Time, msec

Figure B-1.15 Current in power circuitry.

Problem B-1.16 LRC circuit in duplex form Consider a L-R-C (inductor, resistor, capacitor) series circuit. Show that the
system can be described as a duplex communication system.

Problem B-1.17 Blocks and pulley Work the following referring to Figure B-1.17.
a) A mass of m1 = 1 kg starts from rest and moves on a frictionless, level platform subject to a force of F = 10 N. Obtain
expressions for the acceleration, velocity, and distance as functions of time, using the relations,
𝑣 = ∫ adt, and
x = ∫ 𝑣dt.
g
F
m1 m1

(a) (b)
m2

Figure B-1.17 (a) Block sliding, pulled by F, on

frictionless surface. (b) Block sliding pulled by mass
with same weight over a pulley.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
20 1 Introduction

b) The same mass m1 starts from rest and moves on a frictionless and level surface now pulled by a string that is
attached to a mass, m2 , that has the same weight, W1 = W2 (passing over a pulley at the edge of the table). Find the
velocity as a function of time.
c) How long will it take the weight to drop, and the block to move, a distance of d = 1 m, and what is the velocity of
the block at this time?
d) What is the total kinetic energy of the system at distance d = 1 m and how does this compare with the potential
energy lost by the weight in dropping by d.

Problem B-1.18 Spring combinations Two ideal springs may be interconnected to form a system of springs. The springs
may be connected in such a way that their forces are equal (series) or their deflections are equal (parallel). Sketch each
combination and find the total stiffness ks for each case.

Problem B-1.19 Spring stiffness and stored energy In static tests, various weights FW are hung on a vertical spring, and
the spring length x is measured. The results are found to be FW = [0, 2, 4, 6, 8] lbf and x = [10.1, 14.1, 18.1, 21.6, 22.7]
inches. The spring is now inverted and compressed by the several weights: FW = [0, 2, 4, 6] lbf and x = [9.9, 5.9, 4.5, 4.5]
inches.
a) Discuss the spring “constant” k and the circumstances under which its use is advisable in analyzing a system con-
taining this spring. (Use an appropriate plot.) What besides experimental error can account for the discrepancy
when FW = 0?
b) Determine the elastic potential energy stored in this spring, both in tension and in compression. (Hint: Use graphical
or numerical integration based on the smoothed results of part (a)).

Problem B-1.20 Forces and energy in mass-spring system The questions below all involve the relations between forces
and energy in mechanics.
a) A weight of 44 lb is gradually lowered onto a helical spring, and the spring is found to be compressed 2 in. What is
the compliance of the spring?
b) A mass of 2 kg is gradually lowered onto a helical spring, and the spring is found to be compressed 0.05 m. What is
the compliance of the spring? How does it compare to that of part (a)?
c) Consider the weight of part (a) to be held in contact with the uncompressed spring and then suddenly released.
Compute, from energy considerations, the maximum amount by which the spring is compressed. Can you suggest a
graphical solution for this problem in which the various energies involved would be plotted against the displacement
of the upper end of the spring?
d) Consider the weight of part (a) now to be held 3 in. above the upper end of the uncompressed spring and then
released. Compute, from energy considerations, the maximum amount by which the spring is compressed.
e) A mass M is released from a distance h above the end of an uncompressed helical spring, as in part (d). The stiffness
of the spring is K. Show that the maximum compression of the spring is given by
√
xmax = Mg∕K + (Mg∕K)2 + 2Mgh∕K.
This is obtained from a quadratic equation involving energy. Justify the neglect of the second solution of the
equation.

Problem B-1.21 Cart-bumper Figure B-1.21 shows a bumper designed to bring a coasting railroad car to a stop. The car
weighs 10 000 kg and is traveling with a velocity of 10 m/sec when it strikes the spring. If the spring is not to com-
press more than 0.1 m, what must be its compliance? What will be the maximum force exerted on the car by the
spring?

υo

k
m

Figure B-1.21 Cart with a

bumper.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
1.8 Problems 21

Problem B-1.22 Steam locomotive source characteristic The drawbar characteristic for a steam locomotive are shown
in Figure B-1.22. (a) If the locomotive is to pull a constant tractive force load of 15 thousand pounds, what would you
estimate as the steady-state speed? (b) Explain why no transmission was required on these engines.

140

Tractive force, kN
120
100
80
60
40
20
0
0 10 20 30 40 50
Speed, m/s

Figure B-1.22 Steam locomotive force–speed data.

 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
23

Kirchhoff Systems

In this chapter, we review concepts used in modeling mechanical, electrical, hydraulic, and thermal systems. The basic
material in this chapter should be familiar to a reader that has a background in engineering and physics. To those not
familiar, there is sufficient discussion to provide a good foundation for the rest of the book, and cited references should be
referred to as needed to supplement the discussion presented here. The material is presented in a slightly different format
than is classically done in individual courses focused on each particular area. The reason for this shift is to treat all energy
domains in basically the same manner. The types of systems to be studied in this chapter can be represented by circuit-
or network-like model elements and interconnections. Each domain area has two distinct types of energy storage – one
primarily due to position and one due to motion. Systems that have these characteristics will be termed Kirchhoff systems.1
Much of this chapter will deal with a taxonomy of essential elements. We wish to give more precise meaning to such
modeling components as point masses, mechanical springs, capacitors, and inductors, together with properties such as
inertia, compliance, capacitance, and inductance. The reader should be aware at the outset, in spite of the formal definitions
given in this chapter, that there are no exact physical embodiments of these idealized components. These elements and their
properties merely provide modeling artifacts with which the modeler can conceptually reticulate the system.

2.1 Mechanical Translation

Consider initially a single geometrical point translating linearly in one dimension as illustrated in Figure 2.1. The amount of
power transmitted to that point can be calculated by the product of the in-line force and velocity. This evaluation is justified
by Newton’s third law as discussed in Chapter 1.
Such a point will be considered as a system having a single power port with the energy delivered to this port calculated as

= dt = F𝑣dt. (2.1)

∫ ∫

2.1.1 Potential Energy Storage

From kinematics, we know that the velocity is equal to the time rate of change of the position or 𝑣 = dx∕dt = ẋ which
implies that an equivalent form for the energy is

E= F ẋ dt = F dx (2.2)
∫ ∫
which is nothing more than the definition of the work done on the system. This form is useful whenever the force F can
be expressed solely as a function of displacement x so that the integral ∫ Fdx can be evaluated. Note that in order for the
energy to be uniquely defined, the function F(x) cannot be a multivalued relation. All systems with this particular feature
we will label translational springs.

Translational Spring: A system in which the force is a single-valued function of the displacement.

1 Named after Gustav Kirchhoff who first studied circuit topology, thus relating to the form of “Kirchhoff’s Laws” which are directly applicable
to all circuit-like systems having energy types analogous to electric and magnetic energies.

Modeling of Physical Systems: Simulation and Control, First Edition. Raul G. Longoria and Joseph J. Beaman.
© 2025 John Wiley & Sons Ltd. Published 2025 by John Wiley & Sons Ltd.
Companion Website: www.wiley.com/go/Modeling_of_Physical_Systems/1e
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
24 2 Kirchhoff Systems

F Boundary Figure 2.1 Geometrical point translating in one dimension (left) and visualized
using power flow into a system.
υ F System
υ
x

UF = Coenergy Figure 2.2 Translational spring constitutive law.

F

Ux = Energy

The relation between the force and the displacement is the constitutive law for the spring and is determined by geo-
metric and material properties of the system. As shown in Figure 2.2, the area under the constitutive curve shown rep-
resents the energy stored in the system. This energy is considered stored because it could be recovered by reversing the
force–displacement trajectory.
The area above the curve is given the name complementary energy or simply coenergy. A relation between the energy
and coenergy is2

Coenergy = x dF = F x − Energy. (2.3)

∫
In its geometric interpretation, this relation is also known as the rectangle theorem because from Figure 2.2 we can see
that F ⋅ x = Energy + Coenergy. This relation is a special case of a Legendre transform.3 We will make valuable use for the
energy–coenergy rectangle theorems and Legendre transforms later in the text.
Letter symbols used for potential energy include the characters V, U, or P.E. Coenergy is sometimes indicated with a
*, that is, U ∗ . In this text, a U will indicate potential-type energy with a subscript to indicate either energy or coenergy.
Since x is the independent variable for translational potential energy shown above, it will be indicated as Ux ; since F is the
independent variable for translational potential coenergy, it will be indicated as UF .
Ideal springs are typically associated with spatial variation of velocity and with a uniform
distribution of force. As an example, consider the coil spring depicted in Figure 2.3. If we F1 = F = F2
can neglect the mass of the coil (perhaps by moving sufficiently slowly), then the two forces F1 F2
F1 and F2 must be equal in magnitude and opposite in direction. This can easily be derived
from Newton’s second law (the vector sum of the forces on a system equals zero for a massless υ1 υ2
system). There is no such constraint on the velocities V1 and V2 .
υ = υ1 = υ2
Although the emphasis in this course of study will be on modeling techniques which can
be used for general nonlinear systems, we will for the sake of exposition treat many systems Figure 2.3 Coil spring.
with linear constitutive laws.

Linear Translational Spring: A translational spring in which the force is proportional to the spring extension
(or compression)

For a linear translational spring, we have

F = kx, (2.4)

2 This is only strictly equal to within an additive constant for general constitutive relations (those not going through the origin of the axes). This
constant will be implied for the energy–coenergy relations in the text and not be explicitly given.
3 The French mathematician Legendre developed this transformation in order to change independent and dependent variables in the solution
of differential equations. This transformation is especially useful in Analytical Mechanics and Thermodynamics. See Lanczos [8]. Helmholtz and
Gibbs functions used in thermodynamics are essentially Legendre transformations. See Holman [9].
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
2.1 Mechanical Translation 25

where k is the proportionality constant and spring extension/compression is taken as zero when the force is zero. The stored
energy can then be expressed as

Ux = kx dx = kx2 ∕2 (2.5)
∫
and the coenergy is

UF = (F∕k) dF = F 2 ∕(2k). (2.6)

∫
Although Ux and UF have different independent variables and functional forms, the energy and the coenergy are numeri-
cally equal for the special case of a linear spring. The equality can be easily seen graphically if the constitutive curve shown
in Figure 2.2 is a straight line through the origin, in which case, the energy and the coenergy are represented by the area in
equal triangles. While this special result sometimes allows one to treat the two forms of energy interchangeably, this is not
possible for the general nonlinear case.

Example 2.1 Metal rod in tension

Consider the thin metal rod or bar indicated in Figure 2.4. It is desired to obtain a relationship for the axial force, F, as a
function of rod extension, x.
Solution
Assume that the bar is in a state of one-dimensional uniform strain and a linear elastic law applies in the form of,
𝜎 = E 𝜖 [Hooke′ sLaw],
where E is Young’s modulus, a material property. The stress can then be found from 𝜎 = F∕A, where A is the cross-sectional
area, and the strain can be approximated as 𝜖 = x∕L, where x is the extension of the shaft and L is the unstretched length.
The constitutive relation can then be given as
[ ]
F = EA∕L x = kx.
Note that k = EA∕L is solely a function of material (E) and geometry (A∕L). The units of k are force/length so that in SI
units this would give Newtons/meter (N/m).

2.1.2 Kinetic Energy Storage

For potential energy storage, we are primarily interested in the geometric variables of the system in the form of a position
variable x and the relation of this geometry to the forces that arise. For kinetic energy storage, we are primarily interested in
the motion of the system. From Newton’s second law in continuous form, we have that the change of momentum (motion
in Newton’s parlance) and force are related as F = dp∕dt. This implies that another equivalent form for the energy is
dp
= 𝑣 dt = 𝑣dp. (2.7)
∫ dt ∫
This form is useful when the velocity 𝑣 van be expressed as a function of momentum 𝑣(p). The types of systems with this
property are labeled as masses.

Mass: A system in which the velocity is a single-valued function of the momentum.

The relation between the velocity and the momentum is a geometric and material determined constitutive law. The area
under the constitutive curve shown in Figure 2.5 is the energy stored in the system.
The area above the curve is the coenergy which can be found as

Coenergy = p d𝑣 = 𝑣p − Energy. (2.8)

∫

Figure 2.4 Thin metal rod in tension. x

F
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
26 2 Kirchhoff Systems

Figure 2.5 Mass constitutive law.

Tυ = Coenergy
υ

Tp = Energy

Symbols given for kinetic energy include T and K.E. Coenergy is sometimes indicated with an asterisk superscript (*). In
this text, kinetic energy will be indicated with a T and the proper independent variable by subscript to give Tp for energy
and T𝑣 for coenergy.
Rigid body masses in pure translation are accompanied by spatial variations of force but a uniform
F1 F2
distribution of velocity. Neglecting the compressibility of the block shown in Figure 2.6 so as to
enforce the assumption of rigidity, then the two velocities 𝑣1 and 𝑣2 must be equal from kinematic
considerations, while the two forces F1 and F2 will generally differ.
A linear mass can be defined as
υ1 = υ = υ2
Linear Mass: A mass in which the velocity is proportional to the momentum.
Figure 2.6 Rigid
block.
For such a linear mass, we have
𝑣 = [1∕m] p,
where 1∕m is the proportionality constant with m the mass quantity. The energy can be expressed as

Tp = (p∕m)dp = p2 ∕(2m) (2.9)

∫
while the coenergy is

T𝑣 = m𝑣 d𝑣 = m𝑣2 ∕2. (2.10)

∫
Note that the expression for kinetic coenergy is often given as the kinetic energy in many texts. This expression, although
numerically correct for velocities small compared to the speed of light, is not equal to the energy in the general nonlinear
(relativistic) case, as treated in Example 2.2.

2.1.2.1 System of Particles

Most often we are not interested in a single point mass or particle but rather in a system of parti- υi
cles, as shown in Figure 2.7. Each of these particles has its individual force, velocity, position, and
momentum. In the general case, this would require knowledge of the motion of each particle in Fi
order to completely quantify the kinetic energy of the system. We will return to the general problem
when we address thermodynamic systems in Chapter 9, but, for now, we will only treat systems that x
are further constrained. In particular, we will assume that the relative displacement between any
two particles is fixed which is the definition of a rigid body system of particles. This implies for a Figure 2.7 System of
translating particles.
translating system of particles that all the particles have the same velocity. In this case, the system of
particles can be treated identically to a single particle with the mass taken as the total system mass
and the force taken as the sum of the external forces on the system (internal forces will cancel in
the summation).
Consider as an example the cubic system shown in Figure 2.8 which is composed of a concentration C of n molecules per
unit volume with molecular weight M (mass/molecule).
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
2.1 Mechanical Translation 27

This microphysical description of the system can be used to define the continuum property mass density as
𝜌 = MC which has units of mass per volume. The momentum of the system can then be found by integrating
over the volume, V
– , of the continuum as L

p= 𝜌𝑣 d–V . (2.11)
∫ Figure 2.8
Isotropic
If we assume that density is uniform over the volume of the cube, this reduces to rigid cube.
[ ]
p = 𝜌L3 V = mV, where m = 𝜌L3 .
Note that the mass m is just a function of material and geometry, the material constant 𝜌 has a microphysical basis, and
the geometric constant L is determined by the geometry of the continuum. The units of m are mass or force–time2 /length
(kg or N-sec2 /m).

Example 2.2 Relativistic mass

In general Tp is not equal to T𝑣 . Show that this is the case for a mass particle translating at very high speeds, approaching
relativistic effects.
Solution
The relation between the momentum and the velocity considering relativistic effects is given as,
mo 𝑣
p= √ (2.12)
1 − (𝑣∕c)2
which, after some algebraic manipulation, can be inverted to yield velocity,
p∕mo
𝑣= √ ,
1 + (p∕mo c)2
where mo is a material and geometric constant called the rest mass and c is the speed of light. The kinetic energy is then
calculated in terms of momentum as,
√
Tp = 𝑣 dp = mo c2 1 + (p∕mo c)2 .
∫
This expression can be written in terms of velocity as,
mo c2
T𝑣 = √ .
1 − (𝑣∕c)2
Defining the relativistic mass m as,
mo
m= √ ,
1 − (𝑣∕c)2
yields the famous relation due to Einstein,
Tp = Energy = mc2 . (2.13)
This derivation depends on using the correct form of energy. Neither the coenergy nor the expression m𝑣2 ∕2 would yield
the correct result. The discussion by Adler [10] provides perspective on this concept.

2.1.3 Dissipation
Besides storing available energy either in potential or kinetic form, a translating system might also dissipate energy into
unavailable forms. The amount of “lost” energy can be calculated in a purely dissipative system as

d = dt = F𝑣 dt. (2.14)

∫ ∫
A purely dissipative system is one in which the net power into the system is greater than or equal to zero. For a translating
system with one power port, this implies that
 = F𝑣 ≥ 0. (2.15)
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
28 2 Kirchhoff Systems

An element of this type will be called a damper and is defined as

Translational Damper: A system in which there is a constitutive function between force and velocity such that the
product of force and velocity is non-negative if positive power is defined to be directed into the system.

This type of system represents irreversible conversion of energy since energy is constrained to flow into the system but
not out of it.
A linear translational damper is defined as:

Linear Translational Damper: A translational damper in which the force is proportional to the velocity.

For a linear translational damper, we have,

F = b𝑣,
where b is a positive proportionality constant with units of force–time/length or N-sec/m.
Dampers are quite often nonlinear. Sliding friction as represented by Coulomb friction is often considered as constant,
but this is not strictly true since the sign of the Coulomb friction force must change as the velocity changes direction. This
is indicated in Figure 2.9 as a discontinuous change of force at the origin of the axis. Square-law friction is also prevalent in
mechanical systems and can result from turbulent drag and separation as bluff bodies move in a fluid (air) or as streamlined
bodies are moved rapidly. It is only in special cases (lubricated bearings) at low velocity that we have linear damping.
Compared to springs and masses, the constitutive relations for dampers are much less certain and often are the source of
error in predictive models.

Example 2.3 Viscous damping

A flat plate moves parallel to and above another fixed flat plate, separated by distance h as shown in Figure 2.10. The space
between the plates is filled with a viscous fluid. Find the relationship between the force required to move the upper plate
and the velocity relative to the fixed lower plate, and explain how this represents an ideal damper.
Solution
Focus on the forces due to the fluid interacting between the plates, so disregard the mass of the plate for this problem. The
plate is assumed to move with velocity 𝑣, with applied force F, and the surface the plate rests on is stationary. The fluid

Square Figure 2.9 Common translational damper and friction laws.

F Linear

Coulombic

l
w υ
F

h z
Velocity profile, u(z)

Viscous fills space Base plate is fixed

between plates

Figure 2.10 Parallel plates coupled by fluid used to form a viscous damper.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
2.1 Mechanical Translation 29

between the plate and the surface is Newtonian with viscosity 𝜇. Assume also that the fluid velocity varies linearly from
𝑣 = 0 at the fixed surface to 𝑣 at the lower side of the top plate. The shear stress can then be assumed proportional to the
velocity gradient of the fluid,
d𝑣
𝜏=𝜇 = [𝜇∕h]𝑣,
dz
where h is the distance between the plate and the surface. The force F can then be calculated as,
F = 𝜏A = [𝜇A∕h]𝑣,
where A is the area of the plate. We identify b = 𝜇A∕h as a damping constant for an ideal damper relation. Once again we
see the parameter value in this constitutive relation depends on material (𝜇) and geometry (A∕h) properties.

2.1.4 Power Sources

In order to complete the modeling procedure, it is often necessary to incorporate idealized sources into the model. We will
consider two types.

Force Source: A system that prescribes the force as a function of time regardless of the velocity.

Velocity Source: A system that prescribes the velocity as a function of time regardless of the force.

Linear motors and drive systems may be considered as sources for translating systems. This approximation will be valid
as long as the force or velocity output of the drive is not appreciably affected by the translating system. Improper use of
sources is a common modeling error and they should be used with care.

2.1.5 Dynamic and Kinematic Laws

We have now defined the basic energy storing (spring and mass), dissipative (damper), and source (force and velocity)
elements needed to model translating mechanical systems, but, in order to interconnect these elements into an overall
system model, we need to impose dynamic and kinematic constraints. The dynamic constraints are expressed in Newton’s
laws. They are often implemented by using free-body diagrams (FBDs). The use of these diagrams are essential to a good
understanding of mechanical systems, but since their use is thoroughly developed in introductory dynamics courses, we
assume the reader is familiar with their use. The kinematic constraints come from the variable geometric properties of the
system to be modeled.
The dynamic constraint can be mathematically stated as a vector summation of forces as,
∑
N
Fi = 0,
i=1

̇ and the kinematic constraint may be stated as a summation

where one of the forces may be an inertial force of a mass, p,
of velocities as,
∑
N
𝑣i = 0,
i=1

̇
where one of the velocities may be a compressive or extensive velocity of a spring, x.
These concepts are probably best described in terms of a concrete example so consider the mechanical translating system
shown in Figure 2.11.

Figure 2.11 Mechanical translational system with two sliding masses connected by linear υ2
damping with one mass restrained to ground with a linear spring.
υ1 m2 b
F(t) k
m1
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
30 2 Kirchhoff Systems

υ2 Figure 2.12 Free-body diagram of (a) body 2 and (b) body 1.

m2 υ1 Fb

Fb F(t) Fk
m1

(a) (b)

The modeling elements in this system are the masses of two rigid bodies, linear damping acting between the sliding
bodies, a linear spring, and a force source. The friction of the wheels is neglected. The constitutive relations for the energy
storage and the dissipative elements are given as,
𝑣2 = p2 ∕m2 ,
𝑣1 = p1 ∕m1 ,
Fk = kxk ,
Fb = b𝑣b .
From the FBD of mass m2 in Figure 2.12(a), we have the dynamic relation,
ṗ 2 = Fb = b𝑣b , (2.16)
where Fb is taken as positive to the left, and from a free-body analysis of mass m1 in Figure 2.12(b), we obtain momentum
rate as an explicit summation of applied forces, namely,
ṗ 1 = F(t) − Fb − Fk = F(t) − b𝑣b − kxk .
Note that the vertical forces have not been included since these do not contribute to the power interactions in the system
(vertical velocity is zero). From kinematic considerations, we have,
ẋ k = 𝑣1 = p1 ∕m1 , (2.17)

Vb = 𝑣1 − 𝑣2 = p1 ∕m1 − p2 ∕m2 . (2.18)

Using the above expressions, the following state equations can be derived:
ṗ 2 = [b∕m1 ]p1 − [b∕m2 ]p2 , (2.19)

ṗ 1 = F(t) + [b∕m2 ]p2 − [b∕m1 ]p1 − kxk , (2.20)

ẋ k = [1∕m1 ]p1 . (2.21)

2.1.6 Block Diagram Description

The physical laws help model dynamic and kinematic constraints, as discussed in the previous section. These relations
also capture cause-and-effect relations that lead to the state equations, such as equations 2.19–2.21. It can be helpful to
think about how the rates of change of each state on the left-hand side of each of these equations is caused by the physical
effects summed up on the right-hand side. These arise, for example, from the expressions for Newton’s second law based
on the FBD in Figure 2.12a, which expresses that ṗ 2 is caused by Fb , as given by equation 2.16. A block diagram provides
a graphical or symbolic representation of the system equations, whereby signal (directed) lines represent key variables and
blocks represent relations between variables in a way that conveys all the mathematical equations required to represent the
system state equations. The reader with control systems background will be familiar with block diagram usage. In modeling
physical systems, however, we allow operational relations in the blocks that can represent linear as well as nonlinear effects.
For example, in general the output from a block, y, would be a function of the input, u, by y = f (u), where f () can be a linear
or nonlinear function of u. Block diagrams also require a summing block where signal variables can be summed (with sign).
In order to model dynamic systems, we also require a time integration block, where the output y is related to an input, u
t
by, y = y(0) = ∫0 udt.
The basic set of block elements described in Figure 2.13 is sufficient to build a block diagram of the mechanical system
example from Section 2.1.5. One way to begin, given a set of state equations, is to recognize that each state is the output
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
2.1 Mechanical Translation 31

Type Block Functional

representation

u y
f (.) y = f (u)
Function

u1 y
Summation + y = u1– u2
(subtraction) –
u2

u y
y = y (0) + t u dt
Integration dt 0

Figure 2.13 Some basic block elements.

υ υ .
F(t)+ p p 1 υ
F(t) dt m
F(t)
m m
Ff
Fo
Ff

(a) (b) (c)

Figure 2.14 Sliding mass with imposed force: (a) schematic (b) free-body diagram, and (c) block diagram.

from a distinct integrator block. In turn, the input to each integrator is a state derivative. Each state derivative of course is a
sum of terms that should each be functions of the system states. In this way, the outputs from each integrator block inform
the terms that feed into the integrator blocks. Consider a mass sliding on a level surface as shown in Figure 2.14(a), with
an applied force, F(t). The FBD in (b) shows a friction force, Ff . The state equation is,
ṗ = m𝑣̇ = F(t) − Ff .
For the case where Ff takes on a nonlinear, coulomb-type friction, Ff = Fo sgn(𝑣), where sgn() is the signum function. The
system can then be represented in block diagram form using a single integrator as shown in Figure 2.14(c). Note how the
block for the nonlinear friction force is shown with a coulomb-type function symbol, as is common, but it could also be
drawn with a functional form, that is, sgn(u). If the friction force was linear and Ff = b ⋅ 𝑣, then the block could be drawn
with the parameter “b” within the block, to reflect a simple linear form. Linear relations by parameters (or coefficients) can
also be shown in block form using a triangular block, as shown for relating p and V in Figure 2.14(c). This symbol relates
back to the way “gains” were represented in classical analog computing diagrams. Either form is acceptable.

2.1.7 Summary
We have now formally defined basic modeling elements for mechanical translating systems.
1) Two energy storing elements – spring and mass
2) a dissipative element – damper
3) Two sources – force and velocity
These elements are all defined in terms of constitutive relations
F(x) – spring
𝑣(p) – mass
F(𝑣) or 𝑣(F) – damper
F(t) – force source
𝑣(t) – velocity source.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
32 2 Kirchhoff Systems

Except for the sources, which are determined external to the system, the constitutive relations are solely determined by
the geometric and material properties of the system.
The modeling elements with their constitutive relations are interconnected into an overall system model by using,
∑
N
Fi = 0 dynamics,
i=1
∑N
𝑣i = 0 kinematics.
i=1

2.2 Mechanical Rotation

Consider now a single geometric point constrained to rotate in a plane maintaining a constant distance R about a point O
where only the force and velocity in the direction of rotation contribute to the power interaction. The power transmitted to
the point could be calculated as,
 = F𝑣,
where F and V are in the 𝜃 direction, but it will be convenient instead to calculate the power as,
 = 𝜏𝜔,
where 𝜏 = RF is the torque about the point O, and 𝜔 = 𝑣∕R is the angular velocity. The power flow conveyed into this
rotating point system and quantified by the product 𝜏 ⋅ 𝜔 is represented in Figure 2.15 using a power bond, as will be adopted
later in Chapter 3. The energy delivered to this point can then be found as,

= 𝜏𝜔 dt.
∫

2.2.1 Potential Energy Storage

From the kinematics of circular motion, or pure rotation, we have 𝜔 = d𝜃∕dt, where 𝜃 is the angular displacement in
radians, which implies that an equivalent form for the energy is

= 𝜏 𝜃̇ dt = 𝜏 d𝜃.
∫ ∫
This form is useful when the torque can be expressed as 𝜏(𝜃). The types of systems with this feature will be labeled as
rotational springs (although they are usually called torsional springs in mechanics):

Rotational Spring: A system in which the torque is a single-valued function of the angular displacement.

The relation between the torque and the angular displacement is the constitutive law for the rotational spring and is deter-
mined, as for all other constitutive laws, by the underlying micro and continuum physics in the form of material and
geometric specific parameters. The area under the constitutive curve shown in Figure 2.16 represents the potential energy
stored in the rotational spring, U𝜃 .

Figure 2.15 Geometric point rotating in a plane.

System
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
2.2 Mechanical Rotation 33

Coenergy

Energy

Figure 2.16 Rotational spring constitutive law.

Figure 2.17 Rotational spring. τ1 τ2

ω1 ω2

The area above the curve is called the potential coenergy, U𝜏 . A Legendre relation between rotational energy and
coenergy is,

Coenergy = U𝜏 = 𝜃 d𝜏 = 𝜏𝜃 − U𝜃 .
∫
Rotational springs are associated with spatial variation of angular velocity and a uniform distribution of torque. Consider
the circular bar shown in Figure 2.17. If the mass of the bar is neglected, then the two torques 𝜏1 and 𝜏2 must be equal in
magnitude. The angular velocities can in general be different.
A linear rotational spring can be defined as follows.

Linear Rotational Spring: A rotational spring in which the torque is proportional to angular displacement.

For a linear rotational spring, we have,

𝜏 = K𝜃,
where K is the elastic proportionality constant. The energy can be expressed as,

U𝜃 = K𝜃 d𝜃 = K𝜃 2 ∕2,
∫
and the coenergy is,

U𝜏 = [𝜏∕K]d𝜏 = 𝜏 2 ∕2K.
∫
As in the translational case, the energy and the coenergy are numerically equal only for linear springs.

Example 2.4 Torsional shaft

Consider a circular bar fixed at one end as shown in Figure 2.18. Find a constitutive relation that relates the torque as a
function of shaft twist angle, 𝜃.
Solution
Assume the bar is in a state of pure shear and using symmetry arguments it can be shown that the shear strain 𝛾𝜃z at a
radius r in the shaft can be related to the twist angle (e.g., see Popov [11]) as,
𝛾𝜃z = [r∕L]𝜃.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
34 2 Kirchhoff Systems

z Figure 2.18 Torsional shaft.

L

Assuming a linear elastic stress–strain continuum for the shear in the form,
𝜏𝜃z = G𝛾𝜃z ,
where G is the material specific shear modulus and 𝜏𝜃z is shear stress. The torque is found from integrating the shear stress
𝜏𝜃z over the cross-sectional area of the shaft as

𝜏= r𝜏𝜃z dA,
∫A
which can be expressed as,

𝜏 = G[𝜃∕L] r 2 dA = G[Iz ∕L]𝜃,

∫A
where

Iz = r 2 dA
∫A
is called the polar area moment (“of inertia”) of the cross-sectional area about the axis of the shaft. Note that the rota-
tional spring-rate K = GIz ∕L is solely a function of material (G) and geometry (Iz ∕L). The units of rotational spring–rate are
force-length/radian or in SI units N-m/rad.

2.2.2 Kinetic Energy Storage

For a single mass point or particle constrained to rotate in a plane about a point O, as shown in Figure 2.19 we have the
following relation,
̇
𝜏 = RF = Rṗ = d(Rp)∕dt = h, (2.22)
where h is called the angular momentum of the mass point or particle about the point O.
Equation 2.22 relates torque on a single particle to the change of angular momentum of that 2
particle. F
Next, let us consider the more important engineering case of a system of particles con-
R
strained to rotate about a point O. If we further constrain the system of particles to be a rigid
body system of particles, then each particle in the system will have the same rate of change of
angular displacement about the point O or all the particles will have the same angular velocity 1
𝜔. The sum of the torques on the system can then be expressed as, 3
(N )
∑ N
∑N
d ∑
𝜏= Ri Fi = Ri ṗ i = ̇
Ri pi = h, (2.23) Figure 2.19 Rotating
i=1 i=1
dt i=1
system of particles.

where 𝜏 is now the total external torque and h is the system angular momentum about the
point O (see any introductory rigid body mechanics text, such as [12]).
The above shows that planar rotation of a rigid body has the same torque–momentum form as the single particle system.
Then for either a single mass particle or a rigid body system of mass particles in pure rotation about a point, it is convenient
to express the rotational kinetic energy in the form,

= ̇ =
𝜔hdt 𝜔dh. (2.24)
∫ ∫
This form is useful when the angular velocity can be expressed solely as a function of angular momentum. The types of
systems with this feature will be defined as rotational inertias.

Rotational Inertia: A system in which the angular velocity is a single-valued function of the angular momentum.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
2.2 Mechanical Rotation 35

Figure 2.20 Rotational inertia constitutive law.

Coenergy

Energy

The constitutive relation 𝜔(h) is geometry and material specific. The area under the curve representing this constitutive
law represents the kinetic energy stored in the rotational inertia, Th , as shown in Figure 2.20. The area above the curve is
the rotational kinetic coenergy T𝜔 .
Rotational inertias are associated with spatial variation of torque and a uniform distribution of angular velocity. A linear
rotational inertia can be defined as

Linear Rotational Inertia: A rotational inertia in which the angular velocity is proportional to the angular
momentum.

For linear inertias, we have,

𝜔 = h∕J,
where 1∕J is the proportionality constant. The energy can be expressed as,

Th = [h∕J]dh = h2 ∕(2J).
∫
The coenergy is,

T𝜔 = hd𝜔 = J𝜔d𝜔 = J𝜔2 ∕2.

∫ ∫
As for with mechanical translation, the coenergy is usually given as the energy in most texts. For rigid body systems, this
will be numerically accurate since any speed remotely approaching the speed of light would invariably destroy the rigid
body due to radial stresses.

Example 2.5 Circular bar

The circular bar shown in Figure 2.21 has mass density 𝜌. It is desired to find the relation between the angular velocity and
the momentum for this bar when it rotates about the centerline.
Solution
The angular momentum about the centerline axis can be calculated by summing over N particles,
∑
N R
h= r i pi = r(𝜌𝑣)d–V = r(𝜌r𝜔)L(2𝜋rdr) = [𝜌L𝜋R4 ∕2]𝜔 = [MR2 ∕2]𝜔 = J𝜔,
i=1
∫ ∫0

where J = MR2 ∕2, the mass moment of inertia about the centerline, is just a function of material and geometry. The units
of J are mass–length2 or force–time2 –length (kg-m2 or N-sec2 -m).

Figure 2.21 Rigid circular bar, with ri the radius to the ith particle. L

R
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
36 2 Kirchhoff Systems

2.2.3 Dissipation
Besides storing available energy, a rotating system can also dissipate energy into unavailable forms. The amount of dissi-
pated energy for a rotational system can be calculated as,

d = d dt = 𝜏𝜔 dt.
∫ ∫
As mentioned previously, a purely dissipative system (with no storage or source of energy) is one in which the net power
into the system is always greater than or equal to zero. For a rotating system with one power port, this implies that the
dissipated power is,
d = 𝜏𝜔 ≥ 0.
A rotational element of this type will be called a rotational damper and is defined as:

Rotational Damper: A rotational system in which there is a constitutive function between torque and angular velocity
and the product of torque and angular velocity is non-negative if positive power is defined to be directed into the
system.

A linear rotational damper is defined as:

Linear Rotational Damper: A rotational damper in which the torque is proportional to the angular velocity.

For a linear rotational damper, we then write,

𝜏 = B𝜔,
where B has units of force–length-sec/rad or N-m-sec/rad. Just as for translational systems, rotational dampers for real
engineering systems are often nonlinear.

Example 2.6 Rotational damper

An example of a device which is designed to function like a rotational damper is the drag cup shown in Figure 2.22 which
is constructed by immersing a thin cylinder in a cup of fluid of viscosity 𝜇. Show that this device can be modeled as an ideal
rotational damper.
Solution
Assuming that the fluid is Newtonian and the gap h is small, the shear stress in the circumferential direction at the radius
R can be approximated by,
𝜏𝜃z = 𝜇RΔ𝜔∕h,
where Δ𝜔 = 𝜔2 − 𝜔1 . The external torque, neglecting rotational inertia and spring effects, can then be calculated as,
𝜏 = R(𝜏𝜃z A) = R(𝜏𝜃z 2𝜋RL) = (2𝜋𝜇R3 L∕h)Δ𝜔.
This defines a damping coefficient as B = 2𝜋𝜇R3 L∕h, which has material parameter 𝜇 and geometric parameters R, L,
and h.

1 Figure 2.22 Drag cup.

2
2
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
2.2 Mechanical Rotation 37

2.2.4 Power Sources

The ideal sources for rotational systems are a torque source and an angular velocity source defined as

Torque Source: A system that prescribes torque as a function of time regardless of the angular velocity.
Angular Velocity Source: A system that prescribes an angular velocity as a function of time regardless of the torque.

A rotary motor can often be modeled with a source, but care must be taken to ensure that there are no loading effects on
the motor arising from the system it is driving. If the output torque or angular velocity of the motor can be set within the
range of interest and independently of this system, a simple source model of the motor may be appropriate. If this is not
the case, a more detailed model is needed to accurately describe the total system. Note that this consideration can generally
not be made without regard to the intended use of the motor.

2.2.5 Dynamic and Kinematic Laws

In order to interconnect our basic rotational modeling elements (spring, inertia, damper, torque source, and angular velocity
source), we need to impose dynamic constraints in the form of,
∑
N
𝜏i = 0,
i=1

and kinematic constraints in the form,

∑
N
𝜔i = 0.
i=1

Example 2.7 Engine propeller system

Shown in Figure 2.23 is a schematic of an engine propeller system. Develop a model that can be used to predict the dynamic
effects on the propeller due to changes in engine speed as it moves through a fluid slurry.
Solution
This system can be reticulated (i.e., divided conceptually) into an engine, shaft, bearing, and propeller/slurry drag compo-
nents. Each of these components will be modeled separately and then assembled into a total system model. The output of
interest is taken as the propeller rotational speed as a function of time.

Propeller +
Engine Shaft Bearing
slurry drag

Using simple models for the components, identified through the word bond graph above, one plausible set of elements is:
Engine: Angular velocity source
Shaft: Rotational spring
Bearing: Rotational damper
Propeller: Rotational inertia and damper

Figure 2.23 Engine–propeller system.

Bearing Slurry

Engine
Shaft
Propeller
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
38 2 Kirchhoff Systems

Propeller + Slurry drag Figure 2.24 Schematic of engine–propeller system.

Shaft
Engine
input
Bearing
Inertia

where it has been assumed that the shaft stores only potential energy and the propeller stores only kinetic energy. A model
of this simplicity would have to be tested versus either experimental data or a more detailed model in order to ascertain the
degree of its validity.
Assuming its correctness for now, a schematic of the system model is shown in Figure 2.24.
Schematics of this simplified type are used extensively in many engineering texts and can provide a useful conceptual
framework for describing system models.
The constitutive relations for the modeling elements are given in the linear form as:
Engine speed source: 𝜔e = f (t)
Shaft spring: 𝜏s = K𝜃s
Bearing damper: 𝜏B = BB 𝜔B
Propeller inertia: 𝜔p = h∕J
Slurry: 𝜏D = BD 𝜔D
where 𝜔e is the engine speed, 𝜏s is the shaft torque, 𝜃s is the shaft twist, 𝜔B is the bearing angular velocity, 𝜔p is the propeller
angular velocity, h is the propeller angular momentum, 𝜏D is the slurry drag torque, and 𝜔D is the slurry drag angular
velocity.
Along with these constitutive relations, we have additional kinematic and dynamic structure which needs to be incorpo-
rated before a complete dynamic model is specified. From kinematic considerations, we have,
𝜔B = 𝜔D = 𝜔p ,
which states that the relative angular velocity between the shaft at the bearing housing, the propeller angular velocity, and
the relative angular velocity between the propeller and the slurry are all the same. Additionally, the rate of change of twist
angle in the shaft can be expressed kinematically as,
𝜃̇ s = 𝜔e − 𝜔p .
From dynamics we know that the rate of change of angular momentum of the propeller is equal to the applied torques or,
ḣ = 𝜏p = 𝜏s − 𝜏B − 𝜏D .
These two rate equations can be expressed as,
𝜃̇ s = 𝜔(t) − h∕J, (2.25)

ḣ = K𝜃s − Bb h∕J − BD h∕J (2.26)

by using the constitutive relations given above. These two coupled rate equations can be solved for 𝜃s and h as a function
of time. Since 𝜔p = h∕J, this also implies that propeller speed as a function of time can also be determined.
Instead of two coupled first-order differential equations, a single second-order differential equation for 𝜔p can be derived
from the two coupled rate equations by first differentiating the momentum rate equation and substituting 𝜔p for h∕J as,

ḧ = J 𝜔̈ p = K𝜃s − (Bb + BD )𝜔̇ p ,

and then substituting the twist rate equation to eliminate (𝜃s ) yielding,
J 𝜔̈ p = K(𝜔e − 𝜔p ) − (Bb + BD )𝜔̇ p ,
which can be written as,
J 𝜔̈ p + (Bb + BD )𝜔̇ p + K𝜔p = K𝜔e . (2.27)
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
2.2 Mechanical Rotation 39

Figure 2.25 Idealized engine torque–speed curve.

τe

ξ3 ξ2 ξ1

ωe

Equations 2.26 and 2.27 are equivalent expressions for the same model. Either set of equations could be solved for the
propeller speed as a function of time. Solution techniques for equations of both types are discussed in Chapter 5.

Example 2.8 Engine torque–speed

Consider the same system as in Example 2.7 but now the engine is characterized by an idealized torque–speed curve as a
function of throttle opening 𝜉 as shown in Figure 2.25.
Solution
As seen in Figure 2.25, the engine speed can vary considerably with the load torque on the engine. In this case, a model
of the engine as an angular velocity source could result in considerable error. Torque–speed curves are typically gener-
ated from steady-state experimental data and therefore do not model the complicated dynamics of the engine itself. But,
assuming that the dynamics of the engine are fast and can be considered in steady state compared to the time response
of the shaft–propeller system, a phenomenological model for the engine can be expressed algebraically. In this case, the
torque–speed curve in Figure 2.25 is used to model the engine speed as,
𝜔e = 𝜔nl (𝜉) − 𝜔de (𝜏e , 𝜉),
where 𝜔nl is the no load velocity (value at 𝜏e = 0), 𝜉 is a throttle level (here three values shown), and 𝜔de () is a function used
to match a provided torque–speed curve.
With the additional dynamic constraint,
𝜏 e = 𝜏s ,
the new engine model can be used to generate a new set of rate equations as,
𝜃̇ s = 𝜔nl (𝜉) − 𝜔de (K𝜃s , 𝜉) − h∕J,

ḣ = K𝜃s − (Bb + BD )h∕J.

This set of rate equations, which is in general nonlinear, can be solved numerically.

2.2.6 Summary
We have formally defined basic modeling elements for (one-dimensional) mechanical rotating systems:
1) Two energy storing elements – spring and inertia
2) A dissipative element – damper
3) Two sources – torque and velocity.
Just as for the translational elements discussed in the last section, the constitutive relations for these elements are deter-
mined by geometric and material properties of the system.
Along with these elements, we need to impose dynamic and kinematic constraints to develop overall system models.
These constraints can be expressed in the form
∑
N
𝜏i = 0,
i=1
∑
N
𝜔i = 0.
i=1
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
40 2 Kirchhoff Systems

2.3 Electric Systems

Engendered by the fundamental contributions of Georg Ohm (1827) and Gustav Kirchhoff (1847) and fostered by the many
electrotechnical challenges of electrical communication (telegraphy, telephony, radio, and television) and electric power
(lighting, motors and generators, and transmission and switchgear), the simple familiar electric circuit concept has now
become firmly established in scientific thinking.
From our general knowledge of such electric circuits, we know that the power transmitted on a pair of wires into an
electrical circuit system as shown in Figure 2.26 can be calculated as the product of voltage, V, and current, i, where the
latter must leave by one terminal (+) and enter at another (−), with the voltage spanning these terminals.
The energy delivered to the system can then be calculated as,

= dt = Vi dt. (2.28)

∫ ∫
In the following sections, we give a basic description of voltage and current in electric circuit systems. The underlying
electric and magnetic phenomena will be introduced in terms of continuum field concepts. In particular, as stated earlier,
we will consider separable electric fields and magnetic fields and the relation of these fields to the forces and velocities of
charged particles.

2.3.1 Electric Field Systems

Consider a particle which has the physical property of charge and has been placed in a static (constant in time) electric
field at point 1.
A charge of quantity q in an electric field is acted on by a force F jointly proportional to the charge and the strength E of
the field. This force is in the direction of the local electric field as shown in Figure 2.27. This fundamental relation between
the force on a charged particle in an electric field can be expressed as,
F = qE.
One basis for the existence of such a field is the Coulombic force between charged particles. If in otherwise empty space
we consider a point charge of strength Q and a small test charge of strength q separated by a directed distance r as shown
in Figure 2.27, then the force on the test charge is given by an inverse square force law as,
qQ
F= r.
4𝜋𝜖o |r|3
The parameter 𝜖o is called the permittivity of free space and has a value of,
𝜖o = 8.854 × 10−12 coulomb2 ∕Nm2 ,
if force is measured in Newtons, distance in meters, and charge in coulombs. The smallest quantity of charge is believed
to be the charge of an electron (negative) or proton (positive) which in coulombs is 1.60219 × 10−19 and hence a negative
charge of one coulomb represents about 6 × 1018 electrons. For a material substance, the permittivity is a characteristic
property of the material. If charges are imbedded in non-conducting substances such as oil or gas the permittivity will have
a value differing from that specified in a value differing from that specified above. The permittivity of air is very nearly equal
to that of a vacuum and is often treated as such.

Figure 2.26 Electric circuit system.

V
Electric V Electric
circuit circuit

Figure 2.27 Charge in an electric field.

2
ds
F q
E 1
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
2.3 Electric Systems 41

The electric field intensity of the point charge shown in Figure 2.28 is then defined to be,
Q q
E = F∕q = r, (2.29)
4𝜋𝜖o |r|3 r
where q is the quantity of charge of the test particle.
The corresponding electric field can also be calculated by defining an electric flux density field D Q
which streams away symmetrically from a point charge.4 This field, which is often called electric
displacement or displacement flux, has the property that the total electric flux passing through Figure 2.28 Coulomb
any closed surface is equal to the total charge enclosed by that surface. This is known as Gauss’s force between charged
law and it can be expressed mathematically as, particles.

D ⋅ dA = Q, (2.30)
∮S
where the integration is taken over the area of any closed surface that completely envelops the charge Q. The electric field
intensity is related to the electric flux density through a material specific constitutive relation. In the linear isotropic case,
this is given as,

D = 𝜖E. (2.31)

For the point charge shown in Figure 2.28, we can use a spherical surface enclosing the charge since, due to the symmetry
in the system, the displacement has only a radial component. The result of evaluating 2.30 gives,

D ⋅ dA = Dr dA = 4𝜋r 2 Dr = Q,
∮S ∮S
or,
Q
Dr = , (2.32)
4𝜋r 2
where Dr is the radial component of the displacement D. Using 2.31 with vacuum permittivity then gives for the radial
component of the corresponding electric field intensity,
Q
Er = . (2.33)
4𝜋𝜖o r 2
The other components of the field must necessarily be zero due to the radial symmetry. The electric field has this simple
form only for a spherically symmetric charge, but more realistic and complex fields can be built up by superposition of this
basic field. This can either be done using a vector equation like 2.29 or Gauss’s law 2.30.
Suppose the constant charge in a general field shown in Figure 2.27 is now moved from points 1 to 2. The necessary work
required by an external force source to make this move can be calculated as,
2 2 2
W =− F ⋅ ds = − qE ⋅ ds = −q E ⋅ ds. (2.34)
∫1 ∫1 ∫1
The voltage can now be defined as the ability to do work per unit charge or,
2
V21 = W∕q = − E ⋅ ds, (2.35)
∫1
and, since energy stored represents the ability to do work, this stored energy itself can be expressed as,

 = qV21 , (2.36)

where the energy is taken as zero (the ground state), say, at point 2. For a purely electrostatic system, the voltage is inde-
pendent of the path from 1 to 2 and V21 can be considered the potential energy per charge due to the electric field. This also
implies that the line integral of an electrostatic field around a closed path is zero or

E ⋅ ds = 0. (2.37)
∮

4 An excellent reference for this concept as well as many others in this section is Hayt [2].
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
42 2 Kirchhoff Systems

In the next section, we will see that relation 2.37 is not true in the vicinity of a time varying magnetic field. In this case, the
concept of an electric field potential energy would need to be re-evaluated.
If, instead of a constant charge at point 1, we consider a quantity of charge q(t) at time t and different charge quantity of
q(t + dt) at time t + dt, then the change in stored energy can be calculated as,
( 2 )
Δ = (t + dt) − (t) = [q(t + dt) − q(t)] − E ⋅ ds .
∫1
The rate of energy change d∕dt can then be expressed,
( 2 )
q(t + dt) − q(t)
̇ = lim − E ⋅ ds = V21 i, (2.38)
dt→0 dt ∫1
where i = dq∕dt is the current. This derivation is a basis for the result expressed in 2.28.
In terms of the current, i, the energy delivered to an electric field system can then be calculated as,

= Vi dt = V dq, (2.39)
∫ ∫
where we have used the relation i = dq∕dt. This form for the energy is useful if the voltage can be expressed solely as a
function of charge. Electrical systems with this property will be defined as electrical capacitors.

Electrical capacitor: A system in which the voltage is a single-valued function of the charge.

The relation between voltage and charge is the constitutive law for the capacitor, illustrated in Figure 2.29. The area under
the constitutive curve shown represents the electric potential energy stored in the capacitor and is indicated by the symbol
Uq . The electric potential coenergy UV is the area above the curve.
A linear electric capacitor is defined as:

Linear Electric Capacitor: An electric capacitor in which the voltage is proportional to the charge.

or,
V = q∕C,
where C is the proportionality constant called the capacitance. The units of C are usually given in farads and represent a
Coulomb/volt (charge2 /force–length) and a volt has units of N-m/coulomb (force–length/charge).
The electric potential energy for a linear capacitor is,

Uq = [q∕C]dq = q2 ∕2C,
∫
and the corresponding coenergy is,

UV = CV dV = CV 2 ∕2.
∫

Figure 2.29 Electric capacitor constitutive law.

UV = Coenergy
V

Uq = Energy

q
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
2.3 Electric Systems 43

i +q 2
2
V21 1 d
x E

i –q 1

Figure 2.30 Oppositely charged flat plates.

Example 2.9 Capacitive plates

As an example of an electric capacitor consider a pair of conducting flat plates separated by a material which does not
conduct current, that is, a dielectric material or simply a dielectric, and illustrated in Figure 2.30. Determine a constitutive
relation relating voltage and charge.
Solution
Assume that there is a net charge of +q on the top plate and a net charge of −q on the bottom plate. Although the entire
system is charge neutral, there exists an electric field between the plates due to the separation of charges.5 If we neglect the
fringing field at the edge of the plates, the electric field between the plates is uniform and can be found from Gauss’s law as,

D ⋅ dA = −Dx A = q,
∮S

D = −q∕Aux ,

1 −q
E= D= u,
𝜖 𝜖A x
where ux is a unit vector in the (+) x direction, A is the area of the plates, and 𝜖 is the permittivity of the dielectric. The
permittivity of a material is often given in terms of a relative permittivity or a dielectric constant as 𝜖 = 𝜖r 𝜖o where 𝜖o is the
permittivity of free space. The voltage can be calculated from the electric field as,
2 d
q q d
V21 = ux ⋅ ds = dx = q.
∫1 𝜖A ∫0 𝜖A 𝜖A
The capacitance of the plates is C = [𝜖][A∕d] which is a joint function of material (𝜖) and geometry (A∕d).

2.3.2 Magnetic Field Systems

We have seen that an electric (or more properly an electrostatic) field is associated with one or more charges at rest. But
when these same charges are moving an associated magnetic field will result; moreover, if this motion is relatively steady this
field will be magnetostatic. In the classical case of permanent magnets – the ancient magnetos or lodestones – the charges
are spinning electrons, while for the electromagnets invented by Michael Faraday and Joseph Henry, now so necessary for
motors and generators, the moving charges are the free electrons in the conductors.
In particular, a string of charges in uniform linear motion is equivalent to a current. Consider such a charge string as in
Figure 2.31, where each charge has the value q and moves along the line with constant velocity 𝑣 and constant uniform
spacing l. The equivalent current (in amperes, A) becomes,
𝑣
I=q .
l

Figure 2.31 String of moving charges.

H q
q
q l
q l
l v

5 The charge neutrality assumption allows the definition of a single current i for this system and is a common circuit assumption.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
44 2 Kirchhoff Systems

One source for a magnetic field is a current through a conducting wire as discussed further below. The magnetic field
generated by a current can be found with Ampere’s law. Ampere’s law states that the line integral of a field called the
magnetic field intensity H about a closed contour is equal to the current enclosed by the contour. This can be expressed as,

H ⋅ ds = I, (2.40)
∮C
where I is the loop current creating the field. The magnetic field intensity H is in general related to magnetic flux density
B by B = B(H). In the linear isotropic case, this becomes,
B = 𝜇H, (2.41)
where 𝜇 is called the magnetic permeability of the material. Typically, 𝜇 = 𝜇r 𝜇o , where 𝜇r is relative permeability
(dimensionless) and 𝜇o = 4𝜋 × 10−7 coulomb/sec-m2 (or H/m) is permeability of free space (vacuum).
Consider a charged particle moving in such a magnetic field B. Unlike the electric field, a charge at rest in a magnetic
field experiences no electrical force whatsoever. Even if its movement is in the direction of the field, there is still no force.
If, however, there is a component of velocity at right angles to the field, then a force is exerted in a direction at right angles
to both the velocity and the field. The magnitude of the force is proportional to the charge, velocity component, and field
component as,
F = q(v × B), (2.42)
which is called the Lorentz magnetic force law,6 and the “×” symbol represents a cross product operation.
The magnetic field intensity produced by an infinitely long straight filament carrying a current i at a radius r can be found
by using a circular path of integration around the filament as shown in Figure 2.32.
Ampere’s law can then be used to obtain magnetic field intensity at a radius r as,
2𝜋 2𝜋
H ⋅ ds = H𝜃 rd𝜃 = H𝜃 r dr = H𝜃 2𝜋r = i,
∮C ∫0 ∫0
or,
1
H𝜃 = i,
2𝜋r
where H𝜃 is the circumferential component of the field intensity. From symmetry, we know that all other components are
zero. In the linear isotropic case, the magnetic flux density would then be given as,
𝜇
B= iu ,
2𝜋r 𝜃
where u𝜃 is a unit vector in the 𝜃 direction.
We have not as yet related the magnetic field system to our previous discussion about voltage and current in the electric
field system. Faraday’s law of induction provides this valuable link. In 1831, Faraday demonstrated experimentally that a
changing magnetic field can produce an electromotive force (emf). Such an electromotive force is a voltage that arises either
from conductors moving in magnetic field or from changing magnetic fields. This law can be expressed mathematically as,
d
H ⋅ ds = − B ⋅ dA, (2.43)
∮C dt ∫S
where the direction of integration and the positive direction of the magnetic field is shown in Figure 2.32. Note that this
implies, in the region of a changing magnetic field, the electric potential may not be uniquely defined since line integral
equation 2.37 will no longer be true and the work function may be path dependent.

Figure 2.32 Magnetic field around a wire.

i

r
r θ

6 This law can be shown to be a consequence of a Lorentz (relativistic) transformation of the electric field forces between moving charges [13].
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
2.3 Electric Systems 45

B (time-varying)
1
Vloop

2
This region is assumed to have
no changing magnetic field

Figure 2.33 Perfect conductor in a magnetic field.

As an example of Faraday’s induction law consider a thin perfectly conducting wire residing in a time varying magnetic
field depicted in Figure 2.33. The line integral of the electric field is evaluated around the path consisting of the dashed
contour, which is outside of the region of the varying field, and the contour prescribed by the wire. Since in the interior of
a perfect conductor the electric field is zero,7 Faraday’s induction law yields,
2
d
H ⋅ ds = H ⋅ ds = − B ⋅ dA. (2.44)
∮C ∫1 dt ∫S
Because the terminal points 1 and 2 are outside the region of changing magnetic field, a loop voltage can be uniquely defined
in this region as,
2
Vloop = − H ⋅ ds.
∫1
The total magnetic flux 𝜑 passing through the surface of the contour can be defined as,

𝜑= B ⋅ dA, (2.45)
∫
which gives another form of Faraday’s law as,
Vloop = 𝜑,
̇ (2.46)
The perfect conductor that connects the two terminals 1 and 2 is often wound in a coil with N loops, in which case, it is
convenient to define a quantity called the magnetic flux linkage (or simply the flux linkage or linkage) as,
𝜆 = N𝜑, (2.47)
in which case, the terminal voltage is given as,
V = NVloop = 𝜆.̇ (2.48)
Further examples of the use of Faraday induction and the meaning of flux linkage will be given later in the text. For example,
equation 2.47 helps define how we can couple the electrical and magnetic energy domains when modeling electromagnetic
devices.
Using equation 2.48, the energy delivered to a magnetic field system can be calculated as,

= Vi dt = i d𝜆. (2.49)
∫ ∫
This form for the energy is useful if the current can be expressed in terms of the flux linkage. This leads to the definition of
the electric inductor.

Electric Inductor: A system in which the current is a single-valued function of the magnetic flux linkage.

The area under the constitutive curve shown in Figure 2.34 represents the magnetic energy stored in the inductor and is
indicated by the symbol T𝜆 . The magnetic coenergy Ti is the area above the curve.

7 Otherwise an infinite current would arise.

 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
46 2 Kirchhoff Systems

Ti = Coenergy Figure 2.34 Electric inductor constitutive law.

Tλ = Energy

A linear electric inductor is defined as

Linear Inductor: An inductor in which the current is proportional to the flux linkage.

For a linear inductor, we have,

i = 𝜆∕L,
where L is the proportionality constant called the inductance. The units of inductance are given in henries; one henry
represents a volt-sec (force–length–time/charge).
The magnetic energy for a linear inductor is,

T𝜆 = [𝜆∕L]d𝜆 = 𝜆2 ∕2L, (2.50)

∫
and the coenergy is,

Ti = Li di = Li2 ∕2. (2.51)

∫

Example 2.10 Toroid inductor

Consider a ring-shaped toroid around which a conducting filament has been spirally wound. Determine the constitutive
relation for this device as an ideal inductor.
Solution i V
Evaluating Ampere’s law along a contour going once around the center of the toroid as indicated A
in the Figure 2.35 yields,

H ⋅ ds = 2𝜋RH = NI, R
∮C
B
where N is the number of turns in the winding and H is in the circumferential direction.
If the radius of the toroid is large compared to the radius of a turn, then all paths around the
toroid have approximately the same length and magnetic field intensity is approximately uni-
form across the cross section of the toroid. The magnetic field intensity can then be expressed as,
Figure 2.35 Toroidal
inductor.
N
H= i.
2𝜋R
Assuming a linear isotropic constitutive relation between B and H gives for the flux linkage of the toroid,
N 2A
𝜆 = N𝜑 = NBA = 𝜇NAH = 𝜇 i, (2.52)
2𝜋R
where A is the cross-sectional area of the toroid and B is in the circumferential direction. Equation 2.52 represents the
constitutive relation for a linear inductor with L = 𝜇N 2 A∕2𝜋R.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
2.3 Electric Systems 47

2.3.3 Dissipation and Imperfect Conduction

An electrical system can also dissipate energy into unavailable forms. In order to have a dissipative system, the net power
into the system, given by the product of voltage and current, must be greater than or equal to zero, or, for an electrical circuit
with one terminal,
d = Vi ≥ 0.
An electrical system with this constraint is called an electric resistor which is defined as:

Electric Resistor: An electrical system in which there is a constitutive function between voltage and current such
that the product of voltage and current is non-negative if positive power is defined to be directed into the system.

A linear resistor is defined as:

Linear Electric Resistor: An electrical resistor in which the voltage is proportional to the current.

The constitutive function for an electric resistor can be expressed as,

V = Ri,
where R has units of ohms (N m sec/coul2 ).

Example 2.11 Cylindrical conductor

Consider a cylindrical metallic conductor shown in Figure 2.36 with a uniform electric field imposed in the axial direction
of the cylinder. Determine the resistance of this conductor.
Solution
For a metallic conductor, a current due to the movement of free electrons in the material occurs under the influence of
an electric field. The electrons quickly reach a steady-state terminal velocity due to a balance between the electric field
force and the opposing collision forces within the material. Since there are a relatively high number of charge carriers in a
metallic conductor, it can be shown that this velocity is approximately proportional to the applied field, or,
v− = 𝜇− E,
where v− is the electron velocity and 𝜇− is a material constant called the mobility of an electron. The current density or
current per area is related to this velocity as,
J = 𝜌− v − ,
where 𝜌− is the electron charge per volume of material. The current density can then be expressed as,
J = 𝜇− 𝜌− E = 𝜎E, (2.53)
with 𝜎 = 𝜇− 𝜌− called the (electrical) conductivity of the conductor. Equation 2.53 is the continuum form of Ohm’s law as
originally expressed by Ohm. The conductivity is a material property and is generally a strong function of temperature.
The total current flowing in the axial direction can be calculated by integrating the current density over the
cross-sectional area of the cylinder as

i= J ⋅ dA = 𝜎E ⋅ dA. (2.54)
∫ ∫
For a uniform electric field, this reduces to,
i = 𝜎AE. (2.55)

Figure 2.36 Metallic conductor. q

E q
q

L
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
48 2 Kirchhoff Systems

The voltage drop in the axial direction can then be calculated as,

V =− E ⋅ ds = EL. (2.56)
∫
Using 2.54 and 2.55 then gives,
A
i = 𝜎 V. (2.57)
L
The resistance of the cylinder is then given as,
V 1L
R= = . (2.58)
i 𝜎A
Metallic conductors obey Ohm’s law quite faithfully; this is a linear law, but only for isothermal systems. If the temperature
is allowed to vary, then the conductivity will itself vary. As will be seen later, relatively strong temperature dependence holds
for most dissipative elements and provides a hint of the thermal basis of dissipation to be discussed later in Section 2.6 and
in Chapter 9. It should also be noted that the resistance law just calculated follows in the pattern of all other constitutive
laws we have seen in which the basis for continuum material constants can be traced to a microphysical model and the
geometric constants come from an integration over the geometric boundaries of the continuum.

2.3.4 Power Sources

The power sources in electrical systems can be taken as a voltage source and a current source defined as:

Voltage Source: An electric system which prescribes voltage as a function of time regardless of the current.
Current Source: An electric system which prescribes current as a function of time regardless of the voltage.

2.3.5 Kirchhoff’s Laws

For a mechanical translational and rotational systems, the modeling elements, once defined, are interconnected into a
complete set of state equations by using dynamic and kinematic constraints as expressed in Newton’s laws and the geometric
properties of the systems. Similarly for electric circuits, the interconnection of the elements is performed by using the two
Kirchhoff circuit laws. Both of these can be derived from the continuum or field description of electrical systems. The
Kirchhoff current law is derived from the requisite conservation of charge. At the field level, this constraint can be expressed
as the current density leaving an enclosed surface must equal to the decrease in charge within the interior of the surface.
This can be expressed as,
d
J ⋅ dA = − 𝜌 d–V , (2.59)
∮ dt ∫ q
where 𝜌q is the charge density or charge per volume and V
– is volume. For a charge-neutral system, the decrease in charge
is zero or,

J ⋅ dA = 0.
∮
If we restrict the current to follow conducting wires as in a circuit, 2.59 yields Kirchhoff’s current law or,
∑
N
ii = 0, (2.60)
i=1

where ii is the current flowing away on the ith wire from a node of N wires.
We have earlier introduced the voltage in equation 2.35 as the difference in electrical potential. Kirchhoff’s voltage law
arises from the field equation,

E ⋅ ds = 0,
∮
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
2.3 Electric Systems 49

Figure 2.37 LRC series circuit model. VR

L iL R iR
V (t) C VC
q

which states that the line integral of an electrostatic field around a closed loop is zero. This leads to the law that the sum of
the voltage drop around a closed loop of conducting wire is necessarily zero or,
∑
N
Vi = 0, (2.61)
i=1

where Vi is the voltage drop across the ith element within a loop with N elements.

Example 2.12 Series LRC circuit

Shown in Figure 2.37 is a series circuit model of an electric system. Note that this schematic already represents a model of
a physical system. Use the elemental constitutive relations and the circuit laws to develop a set of state equations for this
model.
Solution
Assuming linear constitutive relations for the elements gives,
iL = 𝜆∕L, (2.62)
VR = RiR ,
VC = q∕C. (2.63)
From the current law, we have,
q̇ = iR = iL , (2.64)
and from the voltage law, we have,
𝜆̇ = V(t) − VR − VC . (2.65)
Using equation 2.63 in 2.64 and 2.65 yields the two first-order state equations,
q̇ = 𝜆∕L,
𝜆̇ = V(t) − R𝜆∕L − q∕C.
The state equations for this system can also be represented in second-order form in terms of charge derivatives as,
𝜆̇ = Lq̈ = V(t) − Rq̇ − q∕C,
or,
Lq̈ + Rq̇ + q∕C = V(t).

2.3.6 Summary
The basic modeling elements for electric circuits were defined in this section as:
1) Two energy storing elements – capacitor and inductor
2) A dissipative element – resistor
3) Two sources – voltage and current
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
50 2 Kirchhoff Systems

Along with these elements, we have two Kirchhoff circuit laws:

∑
N
Vi = 0,
i=1
∑N
ii = 0.
i=1

Hopefully, the reader can now see the remarkable similarity between energy domains (mechanical translation and rotation
and electrical). This is more than just a case of analogous systems since, in each domain, energy has been the unifying
concept and, as we know, energy is the ability to do work regardless of its form. It is this unifying principle of energy that
we will generalize and formalize in the next chapter. Models for more complex electrical circuit devices (e.g., transistors)
can also be developed using methods to be developed.

2.4 Hydraulic Systems

In previous sections, we studied and classified elements for rigid body systems of particles in which the relative displace-
ment between particles is fixed. Fluid system particles are in general not constrained to have fixed relative displacements
nor easily confined to masses of fixed identity. This can greatly increase the difficulty in modeling fluid systems, but in
this section, only a restricted subset of fluid systems is considered: hydraulic systems. We define hydraulic systems as
that class of fluid systems that are hydrostatic (static pressure dominates dynamic pressure), relatively incompressible,
and isothermal. Additional treatment of fluid systems is given in Section 2.6 and again in Chapter 9. But even for the
restricted class discussed here, there exist a wide variety of practical applications. Hydraulic systems are especially useful
for applications requiring high power-to-weight ratio systems, mechanically stiff drives, and fast speed-of-response. For
such hydraulic systems, the kinetic energy convected into the system can be neglected and the power delivered to the
system can then be calculated in terms of the work done on the surface of the system.
The power delivered to a hydraulic system as in Figure 2.38 could be calculated with the individual in-line forces and
velocities as they enter the system as,
∑
= Fi ⋅ v i ,
where the integration is taken over the inlet surface area, but it is more convenient to calculate the power as,
 = PQ, (2.66)
where P is the pressure calculated as the average force per area, P = F∕A, and Q is the volume flowrate, Q = AV, where V
is the average velocity. The energy delivered to this point can then be found as,

= PQ dt. (2.67)
∫

2.4.1 Potential Energy Storage

Suppose a certain mass of hydraulic fluid is confined within a closed boundary so to contain a total volume V
– . If we allow
the boundary of a hydraulic system to expand, we can relate the rate of volume change to a flowrate,
Q = –V̇ .
Using this result in 2.67 results in

= P–V̇ dt = Pd–V . (2.68)

∫ ∫

Figure 2.38 Hydraulic system.

P Hydraulic
Q system
A
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
2.4 Hydraulic Systems 51

This form is useful if the pressure can be expressed as P(–

V ). The types of hydraulic systems with this feature will be called
hydraulic capacitors.

Hydraulic Capacitor: A system in which the pressure is a single-valued function of the volume.

The potential energy, UV– , of a hydraulic capacitor is given as,

UV– = V,
Pd– (2.69)
∫
and the coenergy, UP , is defined as,

UP = –dP.
V (2.70)
∫
A Legendre transform relation between these two energies is,
V − UV– .
UP = P– (2.71)
A linear hydraulic capacitor can be defined as:

Linear Hydraulic Capacitor: A hydraulic capacitor in which the pressure is proportional to the volume.

For a linear hydraulic capacitor, we have,

– ∕C,
P=V
where C is called the hydraulic capacitance. The energy can then be expressed as,

UV– = V ∕C]d–V = –V 2 ∕2C,

[– (2.72)
∫
and the coenergy is,

UP = CPdP = CP2 ∕2. (2.73)

∫
As in all the other energy domains, the energy and the coenergy are numerically equal only for linear elements.

Example 2.13 Circular tank

Develop the constitutive relation for the constant-area tank.
Solution
For the partially filled tank in Figure 2.39, we have the following expression for the gauge pressure8 at the bottom of the
tank,
P = 𝜌gh = 𝜌g–
V ∕A,
where 𝜌 is fluid mass density, g is gravity acceleration, and A is the tank area A = 𝜋d2 ∕4. The linear capacitance is therefore,
– ∕P = A∕𝜌g,
C=V
with units of length5 /force (m5 /N).

Figure 2.39 Circular tank in gravity field.

g
d
h
P
Q
P

8 Gauge pressure is the absolute pressure minus atmospheric pressure, which serves as the reference pressure.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
52 2 Kirchhoff Systems

2.4.2 Kinetic Energy Storage

Although we are presently neglecting the convection of kinetic energy into or out of hydraulic systems, we will still consider
kinetic energy inside the hydraulic system volume or control volume. The fluid mass particles in the volume can store energy
due to the individual momenta of the particles. In the case of one dimensional, inviscid, incompressible flow, we assume
the velocity 𝑣 is common to all particles and also parallel to the pipe-walls. The only forces contributing to the net change
of power in the system are the in-line forces.
The sum of the forces on the particles can be expressed as,

1∑ d ∑ pi
N N
1 ̇
ΔP = (F1 − F2 ) = ṗ i = = Γ,
A A i dt i A
where ΔP is the net pressure difference acting on the system, p is the total momentum in the system, and Γ is the momentum
per area (Γ is sometimes called the pressure momentum). The energy for a hydraulic system can then be expressed as,

= ̇
ΓQdt = QdΓ.
∫ ∫
This form is useful when the volume flowrate can be expressed solely as a function of the momentum per area. This leads
to the definition of a hydraulic inertia as,

Hydraulic Inertia: A hydraulic system in which the volume flowrate is a single-valued function of the momentum
per area.

The kinetic energy, TΓ , is given as,

TΓ = QdΓ,
∫
and the coenergy, TQ , is given as,

TQ = ΓdQ,
∫
with Legendre relation,
TQ = ΓQ − TΓ .
A linear hydraulic inertia is defined as,

Linear Hydraulic Inertia: A hydraulic inertia in which the volume flowrate is proportional to the momentum
per area.

For linear inertias, we have therefore,

Q = Γ∕I,
where I is the fluid inertance. The energy and coenergy are, respectively,
TΓ = Γ2 ∕2I,
TQ = IQ2 ∕2.

Example 2.14 Pipe momentum

Determine the constitutive relation for the straight pipe shown in Figure 2.40.

P1 P2
AP1 F1 F2 AP2

A L

Figure 2.40 One-dimensional inviscid incompressible flow.

 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
2.4 Hydraulic Systems 53

Solution
For the straight pipe, the translation momentum of the contained fluid in the system volume is,

p= 𝜌𝑣d–V = [𝜌LA]𝑣 = [𝜌L]Q,

∫
where 𝜌 is the fluid mass density. The momentum per area is therefore,
𝜌L
Γ = p∕A = Q,
A
so the fluid inertance is given by,

I = 𝜌L∕A.

Note that the inertance increases both with density and with pipe length as expected but decreases with pipe area. This
apparent contradiction with intuition is due to the fact the area is already factored into the volume flowrate and that Γ is
momentum per unit area. It can easily be shown that an increase in area with a constant velocity, as opposed to a con-
stant volume flowrate, implies an increase in total momentum, p, just as expected. The units of hydraulic inertance are
mass/length4 or force–time2 /length5 (kg/m4 or N-sec2 /m5 ).

2.4.3 Dissipation
A hydraulic system typically can also dissipate energy into unavailable forms. The amount of dissipated energy can be
calculated as,

d = d dt = PQ dt. (2.74)
∫ ∫
In order to have a dissipative system, d must be non-negative. Thus,

d = PQ ≥ 0.

A hydraulic system element with this property is labeled a hydraulic resistance.

Hydraulic Resistance: A hydraulic system in which there is a constitutive function of pressure and volume flowrate
for which the product of pressure and flowrate is non-negative if positive power is taken as directed into the system.

A linear hydraulic resistance is defined as:

Linear Hydraulic Resistance: A hydraulic resistance in which the pressure is proportional to the volume flowrate.

Thus for a linear resistance we have,

P = RQ,

where R has units of force–time/length5 or N-sec/m5 . For fluid flow in a conduit or pipe, linear hydraulic resistance is a
valid model when the flow is laminar. It is common to consider the flow laminar for a Reynolds number Re below about
2000 [14]. Here Re = 𝜌DU∕𝜇, with U is the average velocity, D is the diameter (or hydraulic diameter), 𝜌 is the fluid density,
and 𝜇 the fluid absolute viscosity.9 For laminar (or Pouiselle flow) in a circular pipe, the value of R for linear hydraulic
resistive effects has a value R = 128μL∕𝜋D4 , where L is the pipe length.
The relation between pressure and flowrate for hydraulic loss effects is commonly nonlinear. For example, if we con-
sider the incompressible flow of fluid through a plate orifice, an approximate expression for the pressure drop across the
orifice is,
[ ]
𝜌
P= ⋅ Q|Q|,
2Cd2 A2o

9 Recall, kinematic viscosity is 𝜈 = 𝜇∕𝜌 and has units of m2 /sec, while 𝜇 has units of N sec/m2 .
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
54 2 Kirchhoff Systems

where 𝜌 is the fluid density, Ao is the orifice area, and Cd is a geometry-specific (so-called) discharge coefficient. Note that
the form “Q|Q|” (sometimes called an “absquare” function) allows this form to be used in estimating pressure drop when
Q take positive or negative values. If we know the pressure drop, then the inverted form,
√
Q = Kd ⋅ sgn(P) |P|,
√
where Kd = Cd Ao 2∕𝜌 allows us to determine the dynamically varying Q. Valve control can be achieved by varying the
orifice area, Ao , typically by the position of a valve stem.
We note that the same form of absquare resistance holds also for most rough turbulent flow in pipes and conduits and
in their interconnected networks. The specific parameters need to be provided for a particular situation. In some cases, the
fluid parameters of density and viscosity (absolute and kinematic) may have dependence on temperature.

2.4.4 Power Sources

The elemental sources for hydraulic systems are a pressure source and a volume flowrate source defined as

Pressure Source: A system which prescribes pressure as a function of time regardless of the flowrate.

Volume-Flow Source: A system which prescribes the volume-flow as a function of time regardless of the pressure.

2.4.5 Kirchhoff’s Laws for Hydraulic Systems

From volume conservation and pressure balance considerations, we can interconnect our basic hydraulic elements into
overall system models. This is done by summing particular system flowrates and pressures as
∑
N
Qi = 0,
i=1
∑N
Pi = 0.
i=1

The use of these constraints is best demonstrated by example.

Example 2.15 Storm sewer system

Develop a model for a portion of a storm sewer system shown schematically in Figure 2.41.
Solution
The system can be reticulated as shown in Figure 2.41. One plausible model for this system would be
Rain: Flow source
Tank A: Capacitor
Long pipe: Inertia and resistance
Tank B: Capacitor
Pump: Flow source

Qrain (t)
g

A QP (t)
Long pipe B
Pump

Rain Tank A Pipe Tank B Pump

Figure 2.41 Schematic of storm sewer system.

 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
2.4 Hydraulic Systems 55

where kinetic energy storage has been neglected in the large tanks but not in the long pipe and the pump is treated crudely
as a source of flow. These model idealizations would require further justification to insure an accurate model. We will
assume linear capacitors and inertia while the pipe resistance will be considered as nonlinear. The constitutive relations
can then be given as,
Rain: Qrain (t) given
Tank A: PA = V – A ∕CA
Pipe: Γ = IQ; Pr = 𝜙r (Qr ) = K|Qr |Qr
Tank B: PB = V – B ∕CB
Pump: QP (t) given

where Qrain and QP are the rain and pump flowrates, respectively, PA and PB are the gauge pressures at the bottom of
tanks A and B, respectively, V
– A and V
– B are the fluid volumes in tanks A and B, QI is the flowrate through the pipe, Γ is
the momentum per area of the pipe, Pr is the pressure drop across the long pipe due to resistance, and Qr is the resistive
flow.
Along with these constitutive relations, we have additional flow and pressure constraints which need to be incorpo-
rated before a complete dynamic model is obtained. Since the fluid is assumed incompressible, we have the following flow
constraints:
QI = Qr ,
–V̇ A = Qrain − QI ,
–V̇ B = QI − QP . (2.75)

A pressure (force) balance across the pipe gives the dynamic constraint,

Γ̇ = PA − PB − Pr , (2.76)

where Γ̇ represents inertial pressure drop, PA and PB are applied pressures on the two ends of the pipe, and Pr is pressure
drop due to flow resistance. Using the flow constraints 2.75, the dynamic constraint 2.76, and the constitutive relations
given above yields three state equations expressed as,
–̇ A = Qrain (t) − [1∕I]Γ,
V
Γ̇ = [1∕CA ]– VB ,
VA − [K∕I 2 ]|Γ|Γ − [1∕CB ]–
–̇ B = [1∕I]Γ − Qp (t).
V

The solutions to these equations can then be used to predict the dynamic response of the sewer system as part of the design
process, so as to provide adequate drainage.

2.4.6 Compressibility Effects in Hydraulic Systems

The foregoing has primarily described systems where the fluid is considered incompressible. In such cases, it is reasonable
to assume that the fluid density, 𝜌, is independent of pressure. In some cases, such as for gases (pneumatic systems) as well as
for high-pressure liquid systems (e.g., hydraulic fluid power), it is necessary to consider compressibility effects in the fluid.
For isentropic processes (fluid is weakly dependent on temperature), fluid density is primarily dependent on the pressure,
and the variation in the two quantities is related by the bulk compressibility modulus, or modulus of elasticity, EV– [14]:
dP
EV– ≡ .
d𝜌∕𝜌
If this modulus is assumed constant, then density is only a function of pressure and the fluid is called barotropic. This
relation works well for liquids and some data is given in Table 2.1 for some common liquids.
Hydraulic systems with incompressible fluid assume the bulk modulus is infinite (since d𝜌 = 0). In reality, the higher the
bulk modulus, the less compressible or stiffer the fluid. In terms of volume, an isothermal bulk modulus is written,
( )
𝜕P
𝛽T ≡ −–V . (2.77)
𝜕–V T
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
56 2 Kirchhoff Systems

Table 2.1 Typical values of bulk modulus.

Isentropic Specific
Liquid modulus GN∕m2 gravity (–)

Lubricating oil 1.44 0.88

Kerosine 1.44 0.88
Water 2.24 0.998
Glycerin 4.59 1.26

“compliant pipe”
“rigid pipe”

g
d P1 P2 P1 P2
h
Q1 Q2 Q1 Q2
P
Q P1 P2
P P1 P2

1 1
P V ΔP ΔV ΔP ΔV
C C C
A V 2r V
C ρg C C
β Etw

(a) (b) (c)

Figure 2.42 Three basic fluid capacitors: (a) hydrostatic tank, (b) slightly compressible fluid in a rigid pipe of volume –
Vo with effective
bulk modulus 𝛽, and (c) effective fluid compliance due to compliant (thin-walled) pipe with Young’s modulus E, unstressed radius, ro ,
and wall thickness t𝑤 (refer to relations on pipes with internal pressures in [15]/McGraw Hill Professional).

This relation comes from the above relation using density, but now we are considering an instantaneous (or nominal)
amount of fluid so we replace 𝜌 by V – . To account for the fact that the volume decreases as pressure increases, it is common
to find 𝛽T defined with a minus sign as shown. For real systems, we must examine the assumption of incompressibility
carefully. A common value of 𝛽T for generic oils, for example, is 250 000 psi, or about 1.72 GN∕m2 . Compare this to the
value given in Table 2.1. Refer to the Solved Problem A-2-9 for an application of these concepts to a pressure tank that has
both elastic effects in a tank combined with compressibility effects in contained fluid.
Allowing for small variations in the volume of a contained fluid due to either effective fluid compressibility or slight
expansions of the container or pipe, at least three “fluid capacitors” can be useful in practical modeling of hydraulic systems.
These three are summarized Figure 2.42. While case (c), which captures effective “breathing” of pipes under high internal
pressure, is strictly accounting for mechanical stored elastic energy, it is convenient to introduce this effect as a hydraulic
fluid capacitor.

2.4.7 Summary
In a similar fashion to all of the systems studied in this chapter, we have defined basic modeling elements for hydraulic
systems as:

1) Two energy storing elements – capacitor and inertance

2) A dissipative element – resistance
3) Two sources – pressure and volume flowrate
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
2.5 Ideal Couplers 57

and Kirchhoff-like connection laws

∑
N
Qi = 0,
i=1
∑N
Pi = 0.
i=1

2.5 Ideal Couplers

Up to this point, we have only considered energy domains separately. As mentioned in Chapter 1, our particular interest
in this book will be those systems in which more than one energy domain is present. The subject of this section is an
introduction to the lossless transfer and transformation of energy from one domain to another.
The prototype example of such lossless transformation of energy is Archimedes’ lever as indicated in Figure 2.43.
If the lever arm is idealized as rigid and massless and the friction is neglected at the pivot, then there is neither storage
nor dissipation of energy in the lever system. This implies that, as Archimedes knew, whatever power goes into one side of
the lever must immediately appear at the other side of the lever. For the sign convention shown in Figure 2.43, we have a
constitutive law between the forces as,
l1
F2 = F,
l2 1
and the velocities as,
l2
𝑣2 = 𝑣,
l1 1
which implies that,
[ ][ ]
l1 l2
2 = F2 𝑣2 = F 𝑣1 = F1 𝑣1 = 1 .
l2 1 l1
This leads to the following definition of an ideal coupling element as

Ideal Coupler: A physical system in which there is a constitutive function between the port variables such that the
net sum of the power entering the system vanishes identically.

There are many other examples of mechanical ideal couplers including ideal gear trains, pulleys, and cams. A hydrome-
chanical coupling occurs in the piston ram shown in Figure 2.44. If we neglect all the energy storage and dissipation in the
piston and fluid system, then the inlet power must equal the outlet power or,

F𝑣 = PQ.

F3
l1 l2
υ1 F1 F2
υ2 Lever υ2
υ1
F1

Figure 2.43 Mechanical lever.

v
F P F P
Piston ram
Q v Q

Figure 2.44 Piston ram coupler.

 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
58 2 Kirchhoff Systems

B
i
V F Electromechanical V
v coupling i
F

Figure 2.45 Moving conductor in a magnetic field.

This is enforced by the dynamic and kinematic constitutive relations,

F = AP,
Q = A𝑣.
The piston ram is a key element used to model hydraulic cylinders.
As an example of an electromechanical ideal coupling consider a system consisting of two parallel conductors which has
a sliding bar at one end. This system is assumed to reside in a uniform magnetic field, B, as shown below in Figure 2.45.
If a current i is imposed on the electric port of the system as shown and if we neglect the mass and damping of the moving
bar, then, in order to have dynamic equilibrium, there must be an applied force which is equal and opposite to the Lorentz
magnetic force in equation 2.42,
F = qv × B = Bli, (2.78)
where l is the length of the moving bar. Additionally, if we impose a velocity 𝑣 as shown, then a voltage due to Faraday
induction 2.44 must appear across the electric terminals as,
d
V = (BA) = Bl𝑣, (2.79)
dt
where electric resistance and inductance have been neglected. Since, as stated, all energy storage and dissipation has been
neglected in this description of the system, the only element left must represent an ideal coupling with the power balance
relation,
in = F𝑣 = Vi = out . (2.80)
It can be easily verified that 2.80 is satisfied by the constitutive functions 2.78 and 2.79. This type of coupler is the basis for
most electric machines to be discussed later in Chapter 8.
Gyroscopic coupling occurs in spinning disks as depicted in Figure 2.46. If the disk is rapidly spinning, then the angular
momentum vector will primarily be directed along the axis of the disk (the 3 directions). Also, we know from dynamics
that the vector torque about a fixed point is equal to the time rate of change of vector angular momentum,
̇
𝝉 = H.
The magnitude of the angular momentum vector in the 3 directions is H = JΩ, and its rate of change can be obtained as,
Ḣ = JΩ(𝜔2 u1 + 𝜔1 u2 ).
where 𝜔1 and 𝜔2 are components of the angular velocity in the 1 and 2 directions, respectively and unit vectors in these
directions are u1 and u2 . The torque vector can be expressed in component form as,
𝝉 = 𝜏1 u 1 + 𝜏 2 u 2 .
Equating torque to angular momentum then gives,
𝜏1 = +JΩ𝜔2 ,
𝜏2 = −JΩ𝜔1 .

τ 1, ω 1 Ω Figure 2.46 Gyroscopic torques on a spinning disk.

1
J
2 3
τ2, ω2
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
2.5 Ideal Couplers 59

High
pressure, PS
Drain ΔPs Q
pressure, Pd
x
Servovalve
x
Servovalve
Q Q
Hydraulic v Q ΔP
cylinder
M b F
Hydraulic Load
cylinder v

Figure 2.47 Hydraulic servovalve-controlled cylinder driving a mechanical load. A word diagram indicates the key subsystems for an
initial model of this system.

The input power to the system is,

 = 𝜏1 𝜔1 + 𝜏2 𝜔2
= JΩ𝜔2 𝜔1 − JΩ𝜔1 𝜔2
= 0,
or gyroscopic coupling is a power conserving transformation between the 1 and 2 axes.10
Coupling elements such as the ones presented in this section are essential elements in modeling physical systems. They
provide the gateway between the various physical domains. In the next chapter, we will study in more detail the form these
power conserving elements may take.

Example 2.16 Hydraulic servovalve-controlled cylinder driving a mechanical load

Figure 2.47 shows a schematic of a hydraulic cylinder controlled by a servovalve with a three-way-four-land spool valve
[17]. The position of the servovalve, x, controls the flow of high pressure fluid at pressure Ps , from a hydraulic power supply
to the hydraulic cylinder, which here drives a load mass. Develop a mathematical model.
Solution
The pressure forces acting on the spool valve faces are balanced, enabling the valve to control large amounts of fluid power
flowing from the pressure supply to the hydraulic cylinder with relatively low force (and power) required to move the
servovalve.
This overall system can be broken up into three key subsystems: servovalve, hydraulic cylinder, and mechanical load. Note
that the basic piston ram coupling element is used to model the hydraulic cylinder. If a detailed model of the servovalve is
required, the forces on the spool valve faces would also require use of a coupling element. In this case, model the servovalve
ideally, with the flow resistance through the gaps represented by a hydraulic resistive element that depends on the position
of the servovalve, x. The table summarizes the initial modeling approach.
Servovalve: Displacement-modulated hydraulic resistive elements. The inputs to this subsystem will be
high-power supply pressure and low-power (zero by assumption) position signal x. We will discuss
the concept of modulation in much more detail in Chapter 4. Modulation is indicated by a signal
(no power flow) input to the servovalve.
Hydraulic cylinder: Ideal piston ram coupling element
Load: translational mass and damper

The constitutive relations for the elements are summarized as follows:

Servovalve resistance: P𝑣 = f (Q𝑣 , x)
Hydraulic cylinder coupler: Fc = (Pc − Pd )A; Qc = A𝑣c
Load inertia: 𝑣 = p∕m
Load damper: Fb = b𝑣b

10 Gyroscopic coupling represents the Ω × H portion of the general torque-momentum equation, 𝝉 = Ḣ r el + Ω × H. See, for example, Crandall
et al. [16].
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
60 2 Kirchhoff Systems

where P𝑣 is the pressure drop across the servovalve, Q𝑣 is the valve flowrate, Pc is the cylinder pressure, Pd is the drain
pressure which is also assumed to be equal to cylinder drain pressure, Fc is the cylinder piston force, 𝑣c is the cylinder
piston velocity, A is the area of the cylinder piston, Qc is the cylinder flow, 𝑣 is the mass velocity, p is the mass momentum,
m is the load mass, Fb is the load damper force, b is the damping coefficient, and 𝑣b is the damper velocity. Note how the
effect of servovalve position, x, appears in the function for the pressure–flowrate relation. In this way, x is said to “modulate”
the resistance.
The kinematic constraints in the system are,
Qc = Q𝑣 ,
𝑣c = 𝑣b = 𝑣,
and the dynamic constraints are,
Ps = P𝑣 + Pc ,
ṗ = Fc − Fb .
These relations can be sorted into the form
ṗ = A(Ps − f (Ap∕m, x) − Pd ) − bp∕m.
This hydraulic cylinder actuator system has a single dynamic state, p, and three inputs: supply pressure, drain pressure,
control input x. In some cases, it is necessary to include the effect of fluid compressibility for fluid contained within the
cylinder. With this compliant effect in the fluid, the model would be able to account for any observed oscillations. More
information on modeling high performance hydraulic control systems can be found in [17].

In the following chapters, we will develop systematic techniques for developing models of the type formed in the previous
example. The intent is to adopt a more systematic manner, including ways to guide the algebraic manipulation of equations
required to obtain a final set of state equations. As evident in this example, the complexity and nonlinear effects necessary
to account for realistic effects can make it difficult to reliably reduce the equations to final form.

2.6 Thermal System Elements and Effects

Incorporating thermal effects in a system enables consideration of the influence of temperature deviations from the nom-
inal, especially when there is heat generated by dissipation effects such as from friction and electrical resistance. While
it is often necessary to consider the spatially distributed nature of thermal energy, lumped-parameter models similar to
those used in other energy domains can be effective in many practical problems. These models can be used to estimate
how thermal energy is stored within a system and how heat transfer flows into and out of a system and between system
elements. Unlike the other physical domains discussed in this chapter, there is no kinetic energy storage element in the
thermal domain. It is sufficient to account for thermal internal energy as a type a potential energy storage, an abstraction
that will be more fully explored in Chapter 9. More complex thermal systems that involve processes such as mass transfer,
phase changes (boiling, condensation), mixing, or multiple component processes, are out of the scope of this chapter. Some
of these topics will be discussed in Chapter 9. For now, we deal with basic heat transfer processes and how to account for
storage of internal energy in system components. The approach taken here follows the same one taken for other Kirchhoff
systems and with a strict set of power conjugate variables that define power flow. This is preferred when making the transi-
tion to the use of bond graphs in Chapter 3. This is in contrast to how “thermal circuits” are typically introduced (see, e.g.,
any introductory heat transfer textbook such as Incropera and DeWitt [18] or Mills [19]). Thermal circuits commonly adopt
a circuit analog approach where heat transfer flow rate, Q,̇ is analogous to current and temperature to voltage. A short dis-
cussion on this approach is given at the end of this section.
Recall that heat, often designated by Q (to be distinguished from volumetric flowrate, which is also designated by Q),
is defined as energy in transit across a boundary separating a system from its surroundings. Heat transfer results from a
temperature difference and has units of energy, but heat is not stored in a system. Rather, thermal internal energy is stored
in the form of macroscopic kinetic or potential energy.11 This should be distinguished from the type of kinetic energy due

11 Chapter 9 further reviews these concepts.

 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
2.6 Thermal System Elements and Effects 61

to motion or potential energy arising from external fields. In thermal systems, total energy in a system is often accounted
for in three distinct forms,

Δ = ΔU + ΔKE + ΔPE.

The first law of thermodynamics relates this total energy of a closed system to heat Q transferred to the system less any work
W done by the system (on environment or surrounding), namely,

Δ = ΔU + ΔKE + ΔPE = Q − W.

Changes in the thermal internal energy due to heat transfer and generation, and other processes that impact energy in this
form, need to be accounted for separately from the kinetic and potential energy forms reviewed in previous sections of this
chapter. Gibbs equation of state can be used to relate internal energy to entropy, S, and volume, V
– , in the form [20],

dU = TdS − Pd–V , (2.81)

in order to account for the influence of certain types of thermal processes. Thus, the internal energy is a function of entropy,
S, volume, V– , and the moles of constituent elements or chemical species, or U = U(S, – V , Ni (for i = 1, … , n), but here the
attention is restricted to a single substance and takes advantage of key properties of U (see Callen [21]), so that for a single
component, we define specific internal energy,

u = u(s, –𝑣), (2.82)

where u = U∕N (per unit mole or per unit mass), s = S∕N, and 𝑣
–=–
V ∕N. Thus, for a pure substance, the equation of state
can be expressed,

du = Tds − Pd–𝑣. (2.83)

𝑣 = 0 so heat transfer between components is assumed to

This introductory review is further restricted to cases, where d–
have no significant mechanical expansion. Now we see that only considering heat and internal energy,
̇ = U̇ = T Ṡ

and we can adopt temperature, T, as an effort variable for the class of thermal systems to be discussed. We identify and adopt
the product of T and Ṡ as thermal power flow, being equivalent to heat transfer flow rate Q̇ itself. This variable set will be
extended in Chapters 3 and 9. Entropy flowrate, fs = dS∕dt, is thus the flow variable of choice to form a power product with
temperature. Thermal systems can then be modeled in a way that is energetically consistent with how systems from other
energy domains have been treated in this chapter. In some cases, other variable sets may be preferred. In particular, many
heat transfer textbooks commonly adopt an electric circuit analogy with temperature analogous to voltage and heat transfer
flow rate to current [18]. Such an approach can be effective when modeling and analyzing systems with heat transfer.

2.6.1 Thermal Energy Storage

With entropy flowrate fs adopted as a flow variable, entropy S is recognized as a displacement state variable. Recall that
potential energy is generally found by taking ∫ edq, where e is the effort and q is the displacement. With temperature and
entropy variables in our model basis, now U = U(S) = US = ∫ TdS defines the thermal internal energy. A thermal capacitive
element can now be defined in a way consistent with other energy domains discussed in this chapter.

Thermal Capacitor: A system in which the temperature is a single-valued function of the entropy.

For a given thermal substance, we can identify a thermal capacitor when there is a relation, T = T(S). This assump-
tion typically assumes that this substance can be partitioned into regions with negligible resistance to heat flow, primarily
serving as a store for thermal internal energy.
A general thermal capacitor relation is shown in Figure 2.48(a), in contrast with a linear relation in (b). Under certain
conditions a linear thermal capacitor can be defined as:

Linear Thermal Capacitor: A thermal capacitor in which the temperature is proportional to the entropy.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
62 2 Kirchhoff Systems

T f(S ) T T S/Ct
T
Us Internal ΔT
energy
ΔS
S S
Ct–1 ΔT
ΔS

(a) (b)

Figure 2.48 (a) General thermal capacitor relation. (b) Linear thermal capacitor relation.

This suggests the linear relation,

T = S∕Ct ,
[ ]
where Ct is a proportionality constant called the thermal capacitance. In this context, the units of Ct are Ct = J/K2 .12

Example 2.17 Thermal energy storage in a solid

For a substance having heat capacity c = 𝜕u∕𝜕T (which assumes constant volume), find an expression for linear thermal
capacitance, Ct .
Solution
For solids and liquids, the specific heats at constant volume and constant pressure are about the same, c𝑣 ≈ cp = c. The
stored internal energy is dU = 𝜌–Vdu, where u = U∕– V and du = c𝑣 dT. From equation 2.83, du = c𝑣 dT = Tds − Pd𝑣, and
since d–𝑣 = 0, we can solve for T as a function of s; that is,
T [ ] s
dT T 1 s − so
= ln = ds = .
∫To T To c𝑣 ∫so c𝑣
This gives,
[ ]
s − so
T = To exp .
c𝑣
where s = S∕𝜌–V , so we can also write,
[ ]
S − So
T(S) = To exp ,
𝜌c𝑣 –V
where So is a constant reference entropy. A thermal capacitance can now be defined about an operating point, say So , as
follows:
[ ]
𝜕T To
Ct−1 = = .
𝜕S S=So 𝜌c𝑣 –V
See Table 2.2 at the end of this section for some typical materials and values of specific heat.

2.6.2 Heat Transfer Processes as Thermal Resistors

All materials offer some resistance to heat flow, which is evidenced by the fact that the temperature drops in the direction
of heat flow through a material.13 The three fundamental modes of heat transfer [18] are summarized in Figure 2.49.
Conduction heat transfer occurs through varied phenomena at the microscopic level, including molecular collisions in
gases, lattice vibrations in crystals, and flow of free electrons in metals. Engineers make use of macroscopic phenomeno-
logical laws, such as those proposed by J.B. Fourier in 1822. Consider a plane, homogeneous solid with a surface area A and
thickness L, having one face at x = 0 at temperature T1 and another face at x = L at T2 . With reference to Figure 2.50, the

12 Note, in a different analogy where T = (1∕Ct )Q, using Q rather than S, the units of Ct will be J/K.
13 At very low temperatures, some materials may exhibit superconductivity, and heat flows with near zero temperature difference.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
2.6 Thermal System Elements and Effects 63

Table 2.2 Representative properties for basic thermal system modeling (drawn from
various sources, for example, adapted from [18, 19]).

Material c, J/(kg-K) k, W/m-K

Aluminum 903 204

Brass (yellow) 4380 111
Carbon steels 460 43
Stainless steel (AISI 302) 480 15
Cast iron 420 51
Copper 385 401
Iron 447 80.2
Lead 129 35.3
Silicon 712 148
Silver 235 429
Wrought iron 500 60

Concrete 750–1000 1.1–1.4

Pyrex glass 750 1.090
Water ≈4160 at room T (refer to tables) 0.611
Neoprene rubber 1930 0.190
Engine oil 1900 at 20∘ C 0.145
White pine (against grain) 2800 0.10
PVC 900 0.092
Cork 1680 0.043
Fiberglass (medium density) 835 0.04
Polystyrene (e.g., styrofoam) 1100 0.028
Air 718 at 20∘ C 0.027

Conduction
(Solid or fluid) Convection Radiation

T1 T2 Moving fluid, Surface, T1

q″
q″ Ts
q″
1
T1 > T2 Ts > T Surface, T2
q″
2

Figure 2.49 Three fundamental heat transfer modes.

T1 A
T2 Q

x x Δx
Thermal
Δx conductivity, k

Figure 2.50 Conductive heat transfer through a plane wall.

 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
64 2 Kirchhoff Systems

heat flow Q̇ through the wall is in the positive x direction if T1 > T2 . Fourier’s law of heat conduction states that the local
heat flux, q′′ , is proportional to the negative of the local temperature gradient,
Q̇ dT
= q′′ and q′′ ∝ − .
A dx
When dT∕dx < 0, the heat flux is in the positive x direction. The thermal conductivity, k, is defined as a constant of propor-
tionality so that,
dT
q′′ = −k . (2.84)
dx
where k is the thermal conductivity, a property of the material (defined by this relation) with units W/m-K (see table at end
of this section), and can vary with temperature.
Energy conservation applies to the closed elemental volume in Figure 2.50, and if we assume negligible storage of energy
(i.e., ̇ = 0) in steady state, then we can assume Q̇ is the same at x and at x + Δx. Therefore,
Q̇ L T2
dx = − kdT.
A ∫0 ∫T1
For negligible changes in k with T,
kA
Q̇ 12 = (T − T2 ).
L 1
Recognizing that the heat transfer is the power flow on each side,

T1 fs1 = Q̇ 12 = T2 fs2 ,

where fsi indicates the entropy flowrate. Thus, entropy flowrates for conduction are,
[ ]
kA T1 − T2
fs1 = , (2.85)
L T1
and,
[ ]
kA T1 − T2
fs2 = . (2.86)
L T2
These relations don’t fit the electrical resistor analogy commonly used in heat transfer textbooks, since the temperature and
entropy flowrate variables are not equal on each side of the element. Electrical resistor models assume that current is the
same coming in as it is going out. Consequently, when drawing heat transfer model schematics, a conductive heat transfer
element will be represented here using a two-port word bond graph element, as shown below:

T1 T2
COND
f1 f2

f1 ≠ f2

̇ that is the same entering and leaving a conduction (COND) element. A direct
To be clear, it is the heat transfer flow rate, Q,
analogy to electrical circuits is commonly taken in introductory heat transfer textbooks by adopting temperature analogous
to voltage and heat transfer flow rate as analogous to current. In this way, circuit analogs can be used to model basic heat
transfer problems. While such an analogy has some advantages it should be pointed out that the product of temperature
and heat transfer flow rate, T ⋅ Q̇ does not give power. Heat transfer flow rate itself has units of Watts.
A conduction parameter for a wall defined as H𝑤 = kA∕L is a useful relation for modeling many practical problems.
Another useful common geometry is a cylindrical body, for which a radial conduction parameter is Hr = 2𝜋kL∕ ln(r2 ∕r1 ),
where r1 and r2 represent the inner and outer radii at temperatures T1 and T2 , respectively. In this case, L is the length of
the cylinder.
Convective heat transfer takes place between a surface and a moving fluid. The flow may be forced, as when it is being
pumped or driven by motion of a surface relative to a fluid, or it may be natural (or free), as when driven by buoyancy effects
in the fluid. The flows may be internal or external, and conditions may be laminar or turbulent. Heat transfer rates tend to
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
2.6 Thermal System Elements and Effects 65

be higher for turbulent flows. For an external flow, the rate of heat transfer depends on the surface and fluid temperatures,
prompting a defining relation for the convective heat transfer coefficient, hc (units of W/m2 -K),
q′′s = hc (Ts − Tf ), (2.87)
where Ts is the surface temperature and Tf refers to the fluid free stream temperature. This relation is also sometimes
referred to as Newton’s law of cooling. This relation can also become nonlinear for cases where there is turbulent flow or
free convection. For system modeling applications, we can consider a given surface area, As , and,
Q̇ = hc As (Ts − Tf ),
so, we can define the entropy flowrates by,
[ ]
Ts − Tf
fss = hc As , (2.88)
Ts
and,
[ ]
Ts − Tf
fsf = hc As . (2.89)
Tf

The value of hc can be influenced by surface and fluid characteristics, flow velocities, etc. A similar word bond graph can
be drawn for convection as for conduction.
Thermal radiation is emitted and absorbed from surfaces, a process that can be thought of as occurring by electromagnetic
waves or photons. A distinction is typically made between radiation energy incident (irradiation) and reflected (radiosity)
from a body. A black body is a surface that absorbs all incident radiation, reflecting none, so all the radiation that leaves
is given by the Stefan–Boltzmann law as Eb = 𝜎T 4 , where 𝜎 is the Stefan–Boltzmann constant (≈ 5.67 × 10−8 W/m2 K4 ).
To find heat transfer rate between neighboring surfaces (including completely enclosed) requires consideration of sur-
face properties and geometry. Real, or gray, surfaces emit and absorb less than black surfaces and are modeled as having
equal emittance, 𝜖, and reflectance, 𝜌. The net heat transfer rate by radiation between two real surfaces 1 and 2 follows the
Stefan–Boltzmann law, but considers the difficulty in determining exactly how much leaving surface 1 is intercepted by
surface 2, and vice versa. Thus, it is commonly written,
Q̇ 12 = 𝜎A1 12 (T14 − T24 ),
where T1 and T2 are the absolute temperatures of bodies 1 and 2. The transfer factor, 12 , will depend on emittances and
geometry. Only in special cases, such as when 1 is surrounded by 2 and A1 << A2 or surface 2 nearly black is 12 ≈ 𝜖1 ,
allowing,
Q̇ 12 = 𝜖1 𝜎A1 (T14 − T24 ),
indicating a nonlinear form of heat transfer with temperature. Similar relations for entropy flow rate can now be formulated,
[ ]
T14 − T24
fs1 = 𝜖1 𝜎A1 , (2.90)
T1
and,
[ ]
T14 − T24
fs2 = 𝜖2 𝜎A2 . (2.91)
T2

Given these three fundamental modes of heat transfer, models for how thermal energy is stored and transferred can be
formulated. Next we define the corresponding power sources useful for this energy domain.

2.6.3 Power Sources

As in other energy domains, we can define power sources for use in developing our thermal system models. To begin with:

Temperature Source: A thermal system that prescribes temperature as a function of time regardless of the entropy
flow rate.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
66 2 Kirchhoff Systems

Temperature boundary conditions are intuitive and readily recognized throughout a system model. Identifying temperature
“nodes,” whether specified as inputs or as key points in a system, is an essential step in thermal system modeling.
Within the context of Kirchhoff systems, the complement (or dual) to the effort power source is usually the flow source.
In the current context, this requires defining a source of entropy flowrate. Often these types of sources arise due to coupling
with dissipative elements and other energy domains.
All dissipative (resistive) elements inherently generate heat which is equal to the power dissipated; that is,
d = Q̇ s = T ⋅ fs = e ⋅ f ,
where always, d ≥ 0 and fs ≥ 0 (a second law of thermodynamics restriction). Recall that we will generally have a resistive
constitutive relation, e = 𝛷R (f ). In problems that are of interest for thermal systems modeling, it may also be the case
that the resistive relation also depends on temperature: e = 𝛷R (f , T). From these relations, an entropy flow source can be
defined as,
e⋅f
fs = . (2.92)
T
Either e or f may be known and then e = 𝛷R (f ) can be used to determine the other variable. Thus, we define:

Entropy flow source: A thermal system that prescribes entropy flow rate as a function of effort, e, or flow, f , and
temperature, T.

A familiar case is the electrical heater, where voltage and current are related by Ohm’s law. In the linear case, V = Ri,
so Q̇ s = T ⋅ fs = V ⋅ i. Assume the voltage across the resistor is applied as Vs (t), so i = Vs (t)∕R, then fs = (Vs (t))2 ∕(RT). If the
resistance was a known function of temperature, or R = R(T), then fs = (Vs (t))2 ∕(TR(T)). This provides the entropy flow rate
source input for a thermal system as a function of the temperature T and in this case Vs (t), the input to the resistor. In turn,
the associated electrical system would “see” the influence of thermal effects through the resulting changes in resistance,
namely, i = Vs (t)∕R(T). Similar relations can be found for entropy flow rate source inputs from mechanical and hydraulic
dissipative effects.

2.6.4 Kirchhoff’s Laws for Thermal Systems

From power conservation and temperature constraints, we can interconnect our basic thermal elements into overall system
models. This is done by summing particular system entropy flow rates, fs , and temperatures,
∑
N
(fs )i = 0,
i=1
∑N
Ti = 0.
i=1

Unlike for the other energy domains discussed where two types of energy storage elements were define, in thermal sys-
tems, only the internal energy storage element is required and thus only one type of rate equation is needed to write an
equation for the entropy, that is,
∑
N
Ṡ c = (fs )i , (2.93)
i

where Sc is the entropy state of a thermal capacitor. Typically, the sum of the entropy flow rates indicated would be taken
with appropriate sign into the storage element.
If necessary, the validity of a lumped-parameter approximation for thermal system models can be assessed. Consider
a homogeneous solid as a lumped thermal capacitor subject to combined conduction and convection processes, as in
Figure 2.51. A lumped capacitor model is considered most valid when the Biot number is, Bi = hL∕k << 1. The graph
in Figure 2.51 illustrates that the temperature distribution is essentially uniform in this case. Recall that the Biot number
is a dimensionless number that represents the ratio of conductive to convective heat transfer resistances, so a very small Bi
means that the body has essentially no resistance to conduction and thus looks like a capacitor. Figure 2.51 also illustrates
the influence of Fourier number, Fo = 𝛼t∕L2c , as well, when the Biot number is large. In this context, 𝛼 = k∕𝜌c, where c is
the specific heat and Lc = –V ∕As , –V being the volume of the body and As the surface area (see [18]).
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
2.6 Thermal System Elements and Effects 67

10 3
T(x,0)

10 2 Fo << 1 Steady
Semi-infinite Fo >> 1 T∞
T∞
solid Bi >> 1
–L L
10 1

Bi = hL/k
10 0
T(x,0)
Lumped
10 −1 capacitance
Bi << 1
T h T h T∞ T∞
–L L
−2
10
10 −2 10 −1 10 0 10 1 10 2 10 3
L L 2
x F o = αt/L

Figure 2.51 Lumped-parameter modeling is most valid when the Biot number, Bi, is very small, Bi << 1, and Fourier number
Fo = 𝛼t∕L2 , a ratio of thermal penetration distance to a characteristic length, such as L, is large.

Example 2.18 Thermal system with temperature source, two storage elements
The system below has a specified wall temperature, T(t), that drives heat conduction into a pure thermal storage element
(PTSE) at T1 . The wall separating regions 1 and 2 is assumed to have surface temperatures at T1 and T2 , respectively. The
outer walls are assumed to be insulated and the hashed regions represent conductive barriers. Determine expressions for
the entropy flow rates induced by the voltage source and the resulting temperatures.
Solution
The word bond graph shown in (b) illustrates the flow of heat using dashed power bonds to represent thermal power flow,
T ⋅ f , between the elements. The PTSEs are assumed to have known T − S equations of state.

Conduction barrier
Temperature
controlled Temp T (t) COND T1 PTSE T1 COND T2 PTSE
wall PTSE PTSE
T (t) source f HA fa S1, T1 fb HA S 2, T 2
S1,T1 S2,T2 T

(a) (b)

Each PTSE has an entropy state, as indicated, and the state equations are given by: Ṡ 1 = fa − fb and Ṡ 2 = fc , where,
T(t) − T1
fa = HA ,
T(t)
T − T2
fb = HB 1
T1
and,
T1 − T2
fc = HB ,
T2
where each “H” is defined using the relation introduced under the discussion on conduction, H = kA∕L. Each PTSE can
be modeled by a temperature–entropy relation,
[ ]
S − Sio
Ti (Si ) = Toi exp i .
𝜌i ci V i
where i = 1, 2, and assuming the T − S relation for a generic solid as derived in Example 2.17.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
68 2 Kirchhoff Systems

The following example shows how to incorporate a resistive element coupling into a thermal system and providing a
source of entropy flow rate, as described in equation 2.92

Example 2.19 Thermal system with heater source, two storage elements
Consider a system similar to that in Example 2.18, except now there is a resistive heater powered by a voltage source, Vs (t),
that generates heat. Assume also that the heater resistance may depend on this surrounding temperature, Rs = Rs (T1 ). Find
the dynamic state equations.
Solution
A thermal word bond graph is shown in (a). The resistive losses are converted into a heat flow rate, Q̇ s = Vs (t) ⋅ is = T1 fs .
Note that here the thermal side of the electrical resistor uses T1 as the reference temperature.

is
T1 Vs (t) T COND T2
Electrical 1 PTSE T1 PTSE
Vs (t) Rs T2 is Resistor fs S1, T1 fb HA fc S 2, T 2

(a) (b)

The entropy flow source is,

Vs (t)is R (T )V (t)2
fs = fs (Vs (t), T1 ) = = s 1 s ,
T1 T1
Now, the dynamic equation for the first PTSE is,
Ṡ 1 = fs (Vs (t), T1 ) − fb ,
with,
T1 − T2
fb = HB .
T1
The equation for the second PTSE is as in Example 2.18,
Ṡ 2 = fc .
For an electrical AC source, the root-mean-square value of the voltage can be used to study the transient response of this
system.

2.6.5 Thermal Circuit Analogs

This section introduced modeling basic thermal systems using an approach analogous with every other system, adopting
the use of entropy flowrate and temperature as a set to track power. In this way, entropy flowrate can be seen as analogous to
current. Thermal circuit analogs, however, are more commonly formulated by taking heat transfer flow rate (Q) ̇ as analogous
̇
to current. In this case, the product of temperature and Q is not power as in the other energy domains. This circuit analog
approach can be very intuitive in many cases and is typically the way thermal circuit modeling is presented in introductory
heat transfer books and courses.
The modeling of thermal circuits requires making some adjustments. To begin, power sources are now temperature and
heat flow sources, and the basic Kirchhoff laws for writing equations rely on,
∑
N
Q̇ i = 0, (2.94)
i
∑N
Ṫ i = 0. (2.95)
i

It can be helpful to actually draw equivalent circuit analogs for systems under study. For example, consider the system in
Example 2.18, shown in circuit analog form in Figure 2.52. In this case, we use the relation on each thermal capacitance,
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
2.7 State-Space and Numerical Response Analysis 69

Figure 2.52 An electric circuit analog for the system in Example 2.18. R1 R2

Q1 Q2
T (t)
Ct1 Ct2

̇ t , taking now Q̇ as the net heat transfer into the

Ct Tt = Q, where Q is the heat “content,” thus a state equation is Ṫ = Q∕C
capacitance. Now, using the heat transfer flows summing into the capacitance,
Ṫ t1 = Q̇ t1 ∕Ct1 = Q̇ 1 − Q̇ 2 ,
Ṫ t2 = Q̇ t2 ∕Ct2 = Q̇ 2 .
In this case, we can write Q̇ 1 = (T(t) − Tt1 )∕R1 and Q̇ 2 = (Tt1 − Tt2 )∕R2 , where R1 and R2 are linear thermal resistances. For
conduction across a planar wall of area A, for example, recall R = L∕kA, where L is the wall thickness and k is the thermal
conductivity. The use of thermal circuits in this way is commonly introduced in introductory textbooks on heat transfer
[18, 19].

2.6.6 Summary
The basic modeling elements for thermal systems were defined in this section as: (i) a single energy storage element – a
thermal capacitor, or PTSE, (ii) a dissipative element – thermal (two-port) resistor, and (iii) two types of sources: a tempera-
ture and a heat flow rate (typically from a dissipative process). For these model elements, we have one Kirchhoff circuit law,
∑N
i=1 fsi = 0, that can guide formulation of dynamic system models for basic thermal systems. In some cases, an alternative
set of variables can be used, which may be familiar from introductory heat transfer studies.

2.7 State-Space and Numerical Response Analysis

Previous sections have described how dynamic and kinematic laws in different energy domains are used to express the
time rate of change of variables associated with energy storing elements in a system model. These variables represent the
energetic states of a system. More will be said in the following chapters about determining the minimal set needed to form
a state vector to quantify the energy content of a system. Without the aid of causality, which will be further introduced in
Chapter 3, most methods in system modeling rely on classical methods and principles such as reviewed in this chapter,
along with insight and intuition, to identify system states. This section reviews state-space representations for systems as
well as methods useful for time response analysis using basic computational tools. These methods and techniques will be
used and expanded upon throughout the rest of this book.

2.7.1 State-Space System Descriptions

A system with n state variables is said to be of order n. The state equations are defined by n ẋ equations with the state
variables defining an n × 1 state vector, x. The system may also have r inputs that comprise an input vector u. For example,
recall the mechanical system in Section 2.1.5 which resulted in a third-order system (n = 3) with the equations,
ṗ 1 = −F(t) + [b∕m2 ]p2 − [b∕m1 ]p1 − kxk ,
ṗ 2 = [b∕m1 ]p1 − [b∕m2 ]p2 ,
ẋ k = [1∕m1 ]p1 ,
[ ]T
where x = p1 p2 xk (note, “()T ” indicates transpose), and u = F (r = 1). These equations can be expressed in a linear
state-space form,
⎡ ṗ 1 ⎤ ⎡ −b∕m1 +b∕m2 −k ⎤ ⎡ p1 ⎤ ⎡ −1 ⎤
⎢ ṗ ⎥ = ⎢ b∕m −b∕m 0 ⎥ ⎢ p2 ⎥ + ⎢ 0 ⎥ F(t).
⎢ 2⎥ ⎢ 1 2
⎥⎢ ⎥ ⎢ ⎥
⎣ ẋ k ⎦ ⎣ 0 0 1∕m1 ⎦ ⎣ xk ⎦ ⎣ 0 ⎦
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
70 2 Kirchhoff Systems

This is a linear set of state equations. A general form of linear state equations is defined by the form,
ẋ = Ax + Bu (2.96)
or,
ẋ 1 = a11 x1 + · · · + a1n xn + b11 u1 + · · · + b1r ur (2.97)
⋮
ẋ n = an1 x1 + · · · + ann xn + bn1 u1 + · · · + bnr u (2.98)
where A is an n × n matrix and B is an n × r matrix. Any m system variables of interest can be related to the states and
inputs and expressed as output equations. The linear case takes the form,
y = Cx + Du (2.99)
or,
y1 = c11 x1 + · · · + c1n xn + d11 u1 + · · · + d1r ur
⋮
ym = cm1 x1 + · · · + cmn xn + dm1 u1 + … + dmr u (2.100)
where C is an m × n matrix and D is an m × r matrix. For example, in the example given above, it may be desirable to define
an output variable related to the velocity difference between the two masses, y = 𝑣r = 𝑣1 − 𝑣2 = p1 ∕m1 − p2 ∕m2 . For one
output (m = 1), the state-space form is,

]⎡ 1 ⎤
p
[
y = 𝑣r = 1∕m1 −1∕m2 0 ⎢ p2 ⎥ + [0] F(t).
⎢ ⎥
⎣ xk ⎦
In the examples just given, the ABCD matrices are all formed by constant parameter values. However, linear state-space
systems allow for these matrices to also be time varying, but systems with such a form will be out of the scope of this book
(see Wiberg [22]).
State-space forms can also be used to express state equations that are not linear. For example, recall the hydraulic storm
sewer system in Example 2.15, which resulted in the equations,
–V̇ A = Qrain (t) − [1∕I]Γ,

Γ̇ = [1∕CA ]–VA − [K∕I 2 ]|Γ|Γ − [1∕CB ]–VB ,

–̇ B = [1∕I]Γ − Qp (t).
V
The term with |Γ|Γ in the Γ̇ equation is not linear so these equations cannot be expressed in linear (ABCD) state-space
form. A more general form of the state equations is the vector form,
ẋ = f(x, u), (2.101)
where f() is an n-dimensional rate vector that is a function of the system states, x and inputs, u. This vector expression is
short hand for the expanded form,
ẋ 1 = f1 (x1 , … , xn , u1 , … , ur ) (2.102)
⋮
ẋ n = fn (x1 , … , xn , u1 , … , ur ) (2.103)
In addition, a nonlinear form for outputs of interest as functions of the states and inputs is,
y = h(x, u). (2.104)
Again, this vector form represents the output relations,
y1 = h1 (x1 , … , xn , u1 , … , ur ) (2.105)
⋮
ym = hm (x1 , … , xn , u1 , … , ur ) (2.106)
and y is an m-dimensional output vector and t is the time.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
2.7 State-Space and Numerical Response Analysis 71

The storm sewer system equations can now be expressed in nonlinear state-space form,
⎡ –V̇ A ⎤ ⎡ Qrain (t) − Γ∕I ⎤
⎢ ⎥ ⎢ ⎥
̇
ẋ = ⎢ Γ ⎥ = ⎢ –VA ∕CA − K|Γ|Γ∕I − V2 – B ∕CB ⎥ .
⎢ –V̇ ⎥ ⎢ Γ∕I − Qp (t) ⎥
⎣ B⎦ ⎣ ⎦
As in the case for linear state-space, output equations can be defined for system variables of interest either as linear or
nonlinear functions of the states and inputs.
The linear case is an important special case of the more general form 2.102 and 2.104. Its importance is due to the large
number of linear analysis methods and design tools that have been developed for equations in the form (2.96) and (2.99). The
more general description shown in 2.102 and 2.104 forms the basis for studying nonlinear systems, which are less amenable
to general methods of solution. As will be discussed in the next subsection, however, many methods for numerically solving
ODEs rely on system equations in this form.
Because of these useful formulations, the state-space concept is widely adapted in dynamic system studies. While the
concept of state as used in this book refers to physical states, a system state can also be used to understand changes
in a system due to discrete or discontinuous events or behaviors. Examples include systems for which connections
between elements are “turned on and off,” or behaviors that are best modeled within a signal context (e.g., purely
computational, or control algorithms). System models that attempt to consolidate continuously varying physical states
and discrete events/behaviors are referred to as hybrid systems [23]. In those contexts, a state may also refer to what
we might call a mode of operation (i.e., a given mode would define a specific configuration of a system). Such systems
are also referred to as switched systems, especially when physically switched model elements that convey significant
power may be conveyed. Liberzon [24] presents a control-theoretic perspective of such systems. For the most part, we
will be careful to define the context of any such behaviors so that proper modeling decisions and analysis methods can
be applied. Switch-like behaviors arise frequently all types of practical systems. While some of the generalized methods
presented in the cited references may be beyond the scope of this book, examples will be included later in this book for
dealing with some of these behaviors on a case-by-case basis, using ad hoc approaches that allow answering practical
questions.

Example 2.20 Mass impacting a damper element

Consider a mass launched along a horizontal surface at velocity, 𝑣o , and which initially sees only resistive forces from
aerodynamic effects. As illustrated in Figure 2.53, eventually the mass comes into contact with a linear damper element.
Let the damper rod that comes into contact with the mass also have a stiffness, k (assume the base of the damper is very
stiff). Develop a model of this system.
Solution
Assuming the aerodynamic force is Fa = Kd 𝑣2 , one state equation prior to impact can be obtained from Newton’s law,
ṗ m = m𝑣̇ m = −Fa = −Kd 𝑣2 ,
where m is the mass. This model holds right before the system comes into contact with the damper rod. To keep track of
the front end of the mass, another state equation can be defined by,
ẋ m = −𝑣m ,
using a negative sign since xm is decreasing with the mass velocity, 𝑣m . Note that xm is not associated with an energy storage
element, and we will later refer to such states as informational. Once xm = L, the system will change. There will be an

Figure 2.53 Mass coming into contact with damper. xm

υo
b
m
k

L
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
72 2 Kirchhoff Systems

impact event. A decision will need to be made about how to model impact, which may occur over a time period that is very
small compared to the time scale of the full system dynamics. Assume the impact is inelastic and that negligible losses occur
(an efficient “catch” mechanism). After the impact, a new state equation is required. With b a (linear) damping coefficient,
the state equation might be taken as,

ṗ = m𝑣̇ m = −Kd 𝑣2 − kxk ,

where we identify that the deflection of the damper rod is xk . It is key to see that the forces in the damper rod and the
damping element are the same, Fk = Fb . The damper itself has a distinct velocity 𝑣b , which can be determined from the
spring–damper force relation, 𝑣b = kxk ∕b. Finally, we now have an energetic state for the rod deflection,

ẋ k = 𝑣m − kxk ∕b.

The system has changed from n = 2 (prior to impact) to n = 3, the latter case now be written in state-space form as,

⎡ 𝑣̇ m ⎤ ⎡ −Kd 𝑣2m ∕m − kxk ∕m ⎤

̇x = ⎢ ẋ m ⎥ = ⎢ −𝑣m ⎥.
⎢ ⎥ ⎢ ⎥
⎣ ẋ k ⎦ ⎣ 𝑣m − kxk ∕m ⎦
Two outputs of interest might be the force in the damper and the power dissipated in the damper, so the output equation
becomes,
[ ] [ ]
Fb kxk
y= = .
b (kxk )2 ∕b
Note that one output is linear but the power variable requires the nonlinear form. This example also demonstrates a simple
way to deal with changes in system modes of behavior.

State-space formulations are used throughout this book to express physical system model equations. The following section
summarizes methods for response analysis in the time domain that rely on state-space representations.

2.7.2 Usage of Simulation Methods

Models in state-space form, as introduced in this chapter, are especially suited for studying time response, and many of the
ODE solvers available in software packages assume this form. This was demonstrated with an Euler simulation method in
Chapter 1 problems and solutions. While some system models, especially those that are low order (n < 3) and linear, can
be studied using analytical solutions (as will be discussed in Chapter 5), it is practical and convenient to employ numerical
analysis techniques even for these cases. Euler and Runge–Kutta formulations make it possible to make practical use of
the models we build, and readily allow nonlinear effects to be accounted for in these analyses. The goal of this section
is to introduce simulation for time–response analysis. We can then use simulations to evaluate the influence of changes
in system model parameters, for example, during a design analysis. Numerical simulations can also be used as “virtual”
experiments for an existing system, providing simulated data for estimating physical parameters. It can be much easier
and less costly in many situations to virtually experiment with a system in this manner prior to conducting actual physical
experiments.
The following discussion focuses on practical use of simulation methods, introducing any computational methods only
as needed. It is assumed that the reader is familiar with scientific and engineering computation. If not, a brief review
of background theory for solving ordinary differential equations can be found in Bronson [25], while a more complete
description of the history and theory of popular algorithms for solving ODEs numerically can be found in Hairer et al. [26].
A more general overview of numerical methods can be referred to in texts such as Chapra and Canale [27].

2.7.2.1 Simulation of State-Space Models

The most basic idea of a physical system simulation was illustrated by problems in Chapter 1.14 Those problems demon-
strated how the time response of a state-determined system can be estimated using an Euler simulation technique. The

14 See Problems A-1-4 and A-1-5.

 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
2.7 State-Space and Numerical Response Analysis 73

Euler method can be understood by considering a single state equation, ẋ = f (x, u, t). We solve for the state at discrete time
steps, ti , where the state at some time ti+i is,
xi+1 = xi + f (xi , ui , ti ) ⋅ Δt, (2.107)
with time step, Δt = ti+1 − ti . Euler’s method can be used as long as this time step is kept sufficiently small [26]. Unlike
Euler’s method which projects forward in time using a single function evaluation, the most extensively employed algorithm
uses four function evaluations. The fourth-order Runge–Kutta method15 estimates the change in state from xi to xi+1 by,
xi+1 = xi + Δ4 (xi , ti , Δt), (2.108)
where,
Δt
Δ4 (xi , ti , Δt) = (k + 2 ⋅ k2 + 2 ⋅ k3 + k4 ) (2.109)
6 1
and,
k1 = f (xi , ti ), (2.110)
( )
Δt Δt
k2 = f xi + ⋅ k1 , ti + , (2.111)
2 2
( )
Δt Δt
k3 = f xi + ⋅ k2 , ti + , (2.112)
2 2

k4 = f (xi + Δt ⋅ k3 , ti + Δt). (2.113)

Figure 2.54 illustrates graphically the difference between the two approaches. The Euler method is indicated by  1 , a linear
progression from the initial value across Δt. Two of the intermediate values indicated by  2 and  3 are evaluated at the
half-step. These additional evaluations help RK4 improve the estimation of x1 relative to a “true” state trajectory16 compared
to the Euler method. One can think of the Euler method as a first-order RK (or RK1) method. The fixed-step RK4 algorithm
performs very well and is considered a workhorse among simulation algorithms.

Example 2.21 Sphere falling in fluid

Simulate the dynamic equations modeling a sphere falling in a fluid and plot the velocity response.
Solution
For a sphere with mass m falling vertically through fluid, Newton’s law gives,
ṗ = m𝑣̇ = −Fd − Fb + mg,
where Fd is a drag force and Fb a buoyant force. This is one state equation, and we may also introduce another to track the
position: ż = 𝑣. Note that this takes the vertical axis to be positive downward, compatible with the use of the gravitational
force above as +mg. Solving for z could be useful if it was necessary to account for how fluid properties change with vertical
position.

x
x state RK4
trajectory
Euler
3
1 2
x0 Δt
2 Δt
t0 t1 t

Figure 2.54 Comparison of an Euler and fourth-order RK state trajectory simulation.

15 Developed by C. Runge in 1895 and extended by W. Kutta in 1901.

16 This figure shows a trajectory for a simple analytical result to illustrate the concept; often the true trajectory is not analytically predictable.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
74 2 Kirchhoff Systems

We want to simulate this system from an initial time, t = 0, and with the mass released from rest. Thus, the initial
conditions are: 𝑣(0) = 0 and we assume z(0) = 0, so we will track the position from the point of release. Assume,
Fd = 𝜌CD As 𝑣2 ∕2 = Kd 𝑣2 and Fb = 𝜌g–Vs , where –Vs is the volume of the sphere. The state equations become,

𝑣̇ = a − b𝑣2 , (2.114)

ż = 𝑣, (2.115)

where a = g − 𝜌g–Vs ∕m and b = Kd ∕m, and this model holds for 𝑣 ≥ 0. Tracking the position, z, allows us to also assign
position-dependent values to the fluid density if needed.
√ since we can solve for 𝑣(t) analyti-
As derived, this model is useful for illustrating the accuracy of numerical integration,
√
cally. It is left as an exercise to show the analytical solution is: 𝑣(t) = (a∕b) tanh( ab ⋅ t).
Both Euler and fourth-order Runge–Kutta (RK4) simulation results are used with the indicated initial conditions to find
the results compared with those from the analytical prediction. Figure 2.55 shows a graph with plots for both the Euler and
RK4 simulations computed with a time step of 0.25 seconds. The analytical solution is also plotted for comparison. It is
clear that the Euler algorithm cannot accurately predict the transient region of the velocity state trajectory unless the time
step is reduced.
Once a working simulation is available, it is possible to experiment with changes. For example, consider the case where
the sphere is attached to a very light shaft meant to constrain the motion vertically so there is additional damping introduced
due to friction in the guides. The dynamic equation now needs to incorporate a force that models the damping, which can
be expressed as a function of the velocity, Ff (𝑣); that is,

ṗ = m𝑣̇ = −Kd 𝑣2 − 𝜌g–Vs − Ff (𝑣) + mg.

The simulation code could be readily updated to include this change in the model, which although minor has likely rendered
the original model equations analytically unsolvable.

2.7.2.2 A Pattern for Practical Use

Example 2.21 can be used to recommend an approach for formulating a simulation study for many basic physical
system problems. Computer code will not be provided since a particular software adopted for this purpose will require
specific syntax and function calls. However, there are common elements to most packages. While the reader may have
an established methodology, many may find it helpful to summarize the following steps to setting up a dynamic system
simulation:

1) Formulate the ODEs into state-space form, ẋ = f(x, u, t) and define a function that will accept the value of time and
̇ It may also be desirable to define other output variables that
return the values of the n rates of change of the states, x.
need to be computed which depend on the system states. Those variables can be computed with the define function as
well or in post-processing once the simulation is completed.

1.5

1 Exact
Velocity

Euler
RK4
0.5

0
0 0.5 1 1.5 2 2.5 3
Time (sec)

Figure 2.55 Comparison of simulation results to analytical solution for the sphere velocity while falling in a fluid.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
2.7 State-Space and Numerical Response Analysis 75

2) Collect all information required to parameterize the model elements and formalize all constitutive relations. This can
be a time consuming process for some systems and may require that experiments be conducted to determine some
parameters.17
3) Interpret the simulation scenario and determine the initial conditions for the n states, x(𝟎). Arbitrarily setting ini-
tial conditions can lead to large initial errors in the ODE solver results. In some cases, it can be helpful to assume
the system begins “at rest” where all the state derivatives are zero, ẋ = 𝟎; that is, equilibrium. In this case, the state
equations themselves can be used to solve for a consistent set of initial conditions from the n algebraic equations:
f(x(𝟎), u(𝟎)) = 𝟎.
4) Determine the time span, or range, of the simulation and estimate the step size. When using a fixed-step solver, the step
size needs to be specified. If there is some knowledge about the natural motion of the system (e.g., a natural frequency
of oscillation), a good rule is to begin with at least 100 point with one period. Once a simulation is running, it is a good
rule to begin cutting the step size in half until results computed after the step size has been reduced do not change from
the previous case. Once the step size is known, for fixed-step cases we relate the initial time, t0 , final time, tf , time step,
Δt, and number of steps, N, by: N = (tf − t0 )∕Δt (in integer value).
5) Set up the simulation over time by a functional call or setting up an iterative step algorithm, in the case of a basic Euler
approach. The functional call for an RK4 or more advanced numerical solver will require specifying a function name,
where the ODE equations have been defined. In addition, the solver will also require the time span and depending on
the type of solver information about the number of steps and/or step size. Adaptive ODE solvers accept an initial time
step estimate or will estimate the initial time step based on error control relations.
The ODE solver algorithm will return the time and computed state(s) for plotting and further analysis. As mentioned
above, there may be post-processing once the states have been computed in order to compute output variables of interest
in a system. The five steps need to be completed in every simulation in one form or another, and a well-documented code
can help communicate the method to other users and supports code reusability.

Example 2.22 Sphere falling in fluid – simulation construction

The following illustrates how the five steps were completed for Example 2.21, with pseudo-code used where necessary.
Step 1: The model equations in Example 2.21 are provided in state-space form. These equations are used to construct a
function definition that accepts time and state as inputs and returns the values of the rates of change of the states. For
example, a function named “xdot = sphere(t,x)” can be used to specify that time ‘t’ and the state vector “x” are inputs
and the state derivatives are returned as a vector in “xdot.” It’s helpful to recognize that this simple function is simply
a subroutine that accepts values of the states and returns values of the rates. If we were interested in, say, computing at
every time step the value of the drag force then we could define an output, y = Fd = Kd V 2 , which is clearly a nonlinear
function of the state, V.
Step 2: Script-like code has traditionally served as a way to archive the parameters used in computational models. In this
case, the parameters were defined as follows:
g = 9.81 (gravitational constant, m/s2 ) Cd = 0.6 (drag coefficient)
rho = 999 (fluid density, kg/m3 ) ms = 7.28 (sphere mass, kg)
Ds = 0.216 (sphere diameter, m) Vs = pi*Ds*Ds*Ds/6 (sphere volume)
a = g-rho*g*Vs/ms (coefficient) b = (0.5*rho*Cd*pi*Ds*Ds/4)/ms (coefficient)

Step 3: “…with the mass released from rest” specifies that the initially velocity is zero. The point of release can either be
z = 0 or an absolute value could be given.
Step 4: This problem has an analytical solution so some insight can come from that case, but it is unusual to have that type
of information. If there was some experimental data or if the sphere was observed to fall that can be used to set a final
time. Here the final time of 3 seconds and step time of 0.25 seconds were set by iteration.
Step 5: The simulation was conducted by and Euler approach as well as by RK4. The call to an ODE solver usually takes
the form:

17 In some cases, the simulation itself may be used as a tool to estimate system parameters.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
76 2 Kirchhoff Systems

Solution (t,X) = RK4(function “sphere,” time space = (0,3), N = (3 − 0)∕0.25 = 12)

where X is an n × N state vector array containing N values of the n states predicted by the solver at each time in the time
array, t.

Solving initial value problems in this way represents a fairly large percentage of the types of simulations that need to be
conducted. It is important to emphasize that the state equations should be properly defined and that initial conditions be
identified. When the right-hand side of the state equations represents continuous functions of the states, most ODE solver
approaches work fairly well. This can be the case even when there are some types of discontinuous behaviors, such as in
Example 2.20, where it is possible to determine clearly when there is a change in the model (i.e., when the front of the
mass contacts the damper rod). Of course, some ODE solvers may perform better than others, and the extent to which the
model represents reality will depend on the needs of the modeling task. Making sure valid models with proper functional
representations are constructed is the subject of subsequent chapters.18 For now, it is assumed that the model has been
constructed and a numerical simulation needs to be performed.

2.7.2.3 Extensions and Additional Considerations

Many ODE solver functions implemented in open-source or commercial software include some form of an adaptive step
size algorithm in order to manage the local (truncation) error. A balance must be struck, however, between too small a
time step and one that is so large that error will be significant or important phenomena might be “missed.” For details
about these algorithms refer to [26, 28–30]. The documentation available in most software packages, as well as the many
examples readily available through online resources, can be helpful in learning how to use these ODE solvers. Examples
will also be provided of their application in the latter chapters of this book.

2.7.2.4 Simulation of Linear State Space

The foregoing has emphasized simulation of physical system models using the general nonlinear state-space form in
equation 2.101. The methods discussed (Euler, RK4, etc.) can also be used to solve linear state-space equations in the
form of equation 2.96. However, there are methods for solving these systems analytically, although the actual numerical
solution is based on a discrete-time approximation (see, e.g., Wiberg, Chapter 5 [22]). In many cases, it is reasonable
to solve these systems with RK4 techniques (and their extensions) as well. Alternatively, one can adopt some of the
solvers that have been developed in available software packages for linear system analysis. The term “lsim” has become
common for naming a linear simulation function in many different software packages. To use these simulation functions,
it is assumed: (i) the system has been put into ABCD form, (ii) initial conditions have been identified for all of the
states, (iii) any input forcing function(s) have been defined, (iv) a proper time step has been identified, and (v) the
initial and final time values have been defined. It is assumed that all the parameter values defined the system matrices
are known.

Example 2.23 Model and simulation of linear two degree-of-freedom mass–spring–damper system
Develop the state equations for the system shown in Figure 2.56, then simulate the response for a step input force, F(t) = Fo
that turns “on” at 0.25 sec. Assume m1 = m2 = 0.1 kg, k = 10 N/m, b1 = b3 = 0.05 Ns/m, and b2 = 0.1 Ns/m. Define two
outputs of interest as the damping forces induced by b2 and b3 .

υ2 υ1 Figure 2.56 Coupled-mass system for simulation example.

k
F (t)
m2 m1
b3 b1

b2

18 Chapter 4 will discuss how we can make sure to have proper constitutive forms, for example, which ensure we can construct computationally
solvable models.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
2.7 State-Space and Numerical Response Analysis 77

Solution
Begin this system by writing applying Newton’s law on the two masses, giving two state equations for the velocities V1
and V2 . An additional state equation is required for spring deflection, xk , defined by the difference between the two mass
velocities.
Use linear element constitutive relations: Fk = kxk , Fb2 = b2 (𝑣1 − 𝑣2 ), and Fb3 = b3 𝑣2 . The final state equations in linear
state-space form become,

⎡ 𝑣̇ 1 ⎤ ⎡ −(b1 + b2 )∕m1 +b2 ∕m1 −k∕m1 ⎤ ⎡ 𝑣1 ⎤ ⎡ 1∕m1 ⎤

ẋ = ⎢ 𝑣̇ 2 ⎥=⎢ +b1 ∕m2 −b2 ∕m2 +k∕m2 ⎥ ⎢ 𝑣2 ⎥ + ⎢ 0 ⎥ F(t)
⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥
⎣ ẋ K ⎦ ⎣ 1 −1 0 ⎦ ⎣ xk ⎦ ⎣ 0 ⎦

The output relations are given by the damping forces induced by b2 and b3 , thus,

[ ] [ ]⎡𝑣 ⎤ [ ]
Fb2 b2 −b2 0 ⎢ 1 ⎥ 0
y= = 𝑣 + F(t).
Fb3 0 b3 0 ⎢ 2 ⎥ 0
⎣ xk ⎦

These equations define the ABCD matrices. These matrices with the given numerical values can be used with a linear
simulation function (e.g., lsim() is a common function name in several software packages). Linear simulation functions
require a description of the system as well as a time array and definition(s) of the input function(s). The time array defines
the discrete time step, Δt, which is used to approximate the linear response. Thus, the time step should be chosen small
enough to achieve a good approximation.19
For this example, the force F(t) was defined by a pulse function with smooth transitions to the peak value, Fo = 1 N/m,
as shown in the plots below. Simulation was conducted using an lsim() function with a time step of 0.001 seconds from 0 to
4 seconds as in the plots below. The states are collectively plotted in the middle graph, with the damper forces (the output
variables) plotted in the lower graph.

1
Force (N)

F (t)
0.5

0
0 0.5 1 1.5 2 2.5 3 3.5 4
Velocity (m/s) and

0.15
Displacement (m)

V1
0.1 V2
10xk
0.05
0
0 0.5 1 1.5 2 2.5 3 3.5 4

0
–0.02
Force (N)

Fb2
–0.04 Fb3

–0.06
0 0.5 1 1.5 2 2.5 3 3.5 4
Time (sec)

Use of lsim() and other linear system analysis tools are popular in linear system analysis of vibration and control sys-
tems. For more general physical system analysis, it is preferred to adopt nonlinear simulation tools which readily allow
incorporating the type of functional relations needed to model complex behaviors common in practical applications.

19 Chapter 5 provides some discussion on response of linear nth-order systems of this type.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
78 2 Kirchhoff Systems

2.8 Summary
This chapter reviews five separate energy domains: mechanical translation, mechanical rotation, electric systems,
hydraulic, and thermal systems. Although admittedly only the simplest cases of these systems were considered, that is,
one-dimensional mechanical elements, electric circuits, hydrostatic fluid systems, and lumped thermal systems, there is a
definite unifying pattern which has emerged in this chapter.
In each energy domain, two distinct types of energy storage were present (except for thermal): one primarily due to
position and one due to motion, and each domain has a dissipative component that can be used to model conversion of
energy into unavailable forms (i.e., heat), a pair of sources to input energy into a system, and a set of physical laws by which
to “join” the basic elements, which we categorized as dynamic and kinematic. To model the connection of systems with
various energy domains, we have discussed a variety of power conserving coupling elements. We discussed how dissipative
effects couple our systems to a thermal domain, allowing our models to track the influence of temperature and its possibly
limiting influence on systems.
In the next chapter, these foundational concepts are used to help form a more generalized modeling approach using bond
graphs. Bond graphs help form a framework for a unified modeling of physical systems.

2.9 Problems

Mechanical translation

Problem B-2.1 Sliding Friction Consider a mass moving on a horizontal surface with velocity 𝑣. If the mass experiences
a friction force of magnitude |Ff | = 𝜇N = 𝜇Mg, where 𝜇 is a friction coefficient and N = Mg is the normal force, show
in two FBDs the direction of the friction force for positive and negative velocities. What type of element would you use
to model this effect and sketch the shape of the constitutive law (properly label). Give an algebraic expression for Ff as
a function of 𝑣.

Problem B-2.2 Three mechanical systems Study the three mechanical system models shown schematically in
Figure B-2.2 composed of ideal mass, spring, and damper elements connected in different ways. Note that the indi-
cated rigid frame has negligible mass, and motion is constrained to the vertical direction. Assume that the three ideal
elements have linear constitutive relations. Include effects of gravity in your model. For each case, do the following:
a) Build a FBD
b) Identify the forces applied on the mass
c) Show that the three systems have the same mathematical model, which you should derive first in state-space form
(first-order equations)
d) Convert the state equations into a single second-order ODE.

g g g
F (t) F (t)
F (t)
Rigid m Rigid
frame frame m
m
k b k b k
b

(a) (b) (c)

Figure B-2.2 Three different mass–spring–damper

schematics.

Problem B-2.3 Hanging mass–spring–damper systems Develop mathematical models for the two mechanical system
models shown in Figure B-2.3. Assume all ideal elements have linear constitutive relations. (a) Derive first-order state
equations. (b) Convert the first-order equations to second-order form.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
2.9 Problems 79

g g
F (t) F (t)

k b
k
b
vm vm
m m

(a) (b)

Figure B-2.3 Two hanging mass–spring–damper

systems.

Problem B-2.4 Comparing damper configurations Two spring–damper configurations are shown in Figure B-2.4, pro-
posed to be inserted in a machine to provide damping of the applied input force, F. A damping force is induced by the
viscous friction between the damper piston element and its housing (hatched region). Assume that the damping force
is linearly related to damper velocity, 𝑣d , or Fd = b𝑣d , and that the spring force also follows a linear relation with spring
displacement, xk .

Viscous friction Viscous friction

F (t) F (t)

(a) (b)

Figure B-2.4 Two spring damper configurations.

Build a model for each system and derive a mathematical model that will allow you to determine the damper force,
Fd , in terms of system inputs, state(s), and system parameters. Then address the following questions:
a) Does it matter which configuration is used from the standpoint of how much force is conveyed to the indicated
“ground”? Answer yes or no from intuition.
b) Find expressions that allow you to compare the power dissipated in the damper for each design. Use a value of b
for the damper, assuming linear damping.
c) Based on your modeling, which design would you recommend and why?
d) Sketch analogous electrical and hydraulic systems for each case.

Rotational systems

Problem B-2.5 Two rotational systems – 1 For each rotational system shown in Figure B-2.5, identify the ideal rota-
tional model elements you would use to model the system dynamics. In both cases, the shaft velocity is an input, 𝜔(t).
Case (a) has all linear ideal elements; however, Case (b) has a rotational brake component. Assume this brake has a
torque–speed function given by, 𝜏B = ΦB (𝜔B ).

Fluid Fluid
coupling Brake
coupling

(a) (b)

Figure B-2.5 Two rotational systems.

 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
80 2 Kirchhoff Systems

For each case, identify the proper state variables, then derive a mathematical model in state-space form. Represent the
brake torque, 𝜏B , in functional form for Case (b).

Problem B-2.6 Two rotational systems – 2 For each rotational system shown in Figure B-2.6, identify the ideal rota-
tional model elements you would use to model the system dynamics.

J1 J1 J2
Φ (ω 2)
B1 K1 B2 τ (t) K

ω (t) Fluid ω1 ω2 ω1 ω2
coupling

B1 B2
(a) (b)

Figure B-2.6 Two rotational systems.

For each case, identify the proper state variables, then derive a mathematical model in the state-space form. Represent
the brake torque, 𝜏B , in the functional form for Case (b).

Problem B-2.7 Two coupled gears in a drive system In the rotational system shown in Figure B-2.7, a gear train couples
a torque applied at shaft 1, 𝜏(t), to a load attached to shaft 2, indicated by a load torque, 𝜏L . As indicated, the elements in
the system are assumed to follow linear constitutive relations with the parameters given. The gears have mass moments
of inertia J1 and J2 . Determine the states required to model this system and derive the state equations.

K1 J1
τ (t)
Shaft 1
ω1
τL
B
Shaft 2
ω2
J2

Figure B-2.7 Gear train drive

system.

Electrical systems

Problem B-2.8 Series and parallel resistors Find an expression for the equivalent resistance of an arbitrary number, n,
of linear resistive elements that are either in series or in parallel, R1 to Rn . Determine an expression for the equivalent
resistance between terminals A and B, RAB , for the resistive network given in Figure B-2.8.

R2
A R1 B
R3

Figure B-2.8 Equivalent resistance.

Problem B-2.9 Toroid inductor design The constitutive relation for a toroid inductor is reviewed in Example 2.2.10.
Conceive of a toroid that would fit in the palm of your hand (say it has a mean circumferential diameter of about 1 inch)
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
2.9 Problems 81

and estimate all the necessary parameters for inductance, L = 𝜇N 2 A∕2𝜋R. Refer to a handbook for suitable values of 𝜇.
Also, estimate how much resistance is in the coiled wire. What is the value of L as well as the coil resistance, R.

Problem B-2.10 Electric capacitor Find the energy stored in the electric field associated with the charged capacitor
shown in Figure B-2.10. Determine the capacitance in farads as the first step in the calculations. Compare the capaci-
tance, C, for the case when the region between the plates is air versus dielectrics such as glass or mica.

2 cm
i
1 cm
V = 10 volts
Air 3 cm
i

Figure B-2.10 Electrical capacitor plates.

Problem B-2.11 Two electrical circuits For each system in Figure B-2.11, identify state variables, assume all linear con-
stitutive relations for the elements and derive a mathematical model in state-space form. Assume there is no load in
(b); that is, treat as an open circuit. Also, for each case, write an output equation for the voltage, 𝑣2 , across resistor R2 .

L R1
L
Vi (t)
R2 R2 Load
Vi (t) C
R1

(a) (b)

Figure B-2.11 Two electrical circuits.

Problem B-2.12 RCC filter system The circuit shown in Figure B-2.12 is designed to filter the input signal V1 (t) and the
output is taken across C2 as V2 . Model this system and derive a dynamic equation in terms of V2 . Note: While there are
two capacitors, only a single first-order differential equation is needed to model this system.

R1

V1 C1 V2
C2

Figure B-2.12 Two electrical

systems.

Problem B-2.13 Electric circuit with algebraic loop Study the circuit given in Figure B-2.13. Identify the states needed
to represent the system and then derive the state equations. The derivation will require solving algebraic relations in
order to write the equations in proper state-space form (this system has an algebraic loop, a concept that will be studied
further in Chapter 3).

L R1 R3

Vi (t) R2 C

Figure B-2.13 Electrical circuit with an

algebraic loop.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
82 2 Kirchhoff Systems

Hydraulic systems

Problem B-2.14 Spherical and conical tanks An open tank with constant cross-sectional area, A, as shown in
Example 2.2.13 is a commonly used model element for many types of fluid distribution systems. In that case, the
pressure–volume relationship is linear, P = –V ∕C, where C = A∕(𝜌g), with 𝜌 the density of the incompressible fluid
and g the gravitational constant. It is not uncommon, however, to encounter tanks with more complex geometries,
such as the conical and spherical tanks shown in Figure B-2.14(a) and (b).

g g
V h H R V h
P P P
Q d Q P
(a) (b)

Figure B-2.14 Two tanks with varying area.

Beginning with the fundamental assumptions that the pressure at the base of the tank is hydrostatic and given by
P = 𝜌gh, and that the fluid is incompressible, determine a relation between pressure and stored fluid volume, V
– , for
each case, solely in terms of the indicated tank geometry parameters, 𝜌, and g.

Problem B-2.15 Resistive model for pipe flow The pressure drop P across a pipe of length L due to a steady flowrate
Q of an incompressible fluid can be modeled by a resistive element model. Review the relationship between P and Q
for cases of laminar and turbulent flow as presented in Section 2.4.3. Reference a fluid mechanics textbook or other
references as needed.
For a certain hydraulic fluid, the absolute viscosity, 𝜇, and the density, 𝜌, are found to vary with temperature according
the following models: 𝜇 = A exp(𝛽), where 𝛽 = (B∕T)3 , and 𝜌 = A + B ⋅ T. For the viscosity model, A = 0.3938 and
B = 483.03, and for density A = 1066.9 and B = −0.7095, with T absolute temperature (in Kelvin), and units of viscosity
in centipoise and density in kg∕m3 .
a) Plot the pressure drop across a pipe versus flowrate, if the pipe is 2 meter long, 2 centimeters in diameter. Let Q
vary from 0 to 1000 cm3 ∕sec and generate a graph with one plot of P versus Q for 20 deg C and another for 0 deg
C. Generate a separate plot of the power required to drive flow through the pipe for each case as a function of
flowrate Q.
b) Based on results from item (a), would you conclude that it is necessary to account for temperature changes in this
fluid when making predictive calculations of performance? Explain the bases for your conclusion.
Note: 1 centipoise (cP) = 10−2 poise = 0.1 kg∕m∕sec.

Problem B-2.16 Viscous damping model Two parallel flat plates are placed as shown in Figure B-2.16(a). Recall the
related problem in Example 2.3 where the lower plate was fixed. The small gap between the plates is filled with oil or
some other viscous fluid. The upper plate moves with a constant velocity, 𝑣1 , and the lower plate has velocity, 𝑣2 . If the
plates are parallel and no pressure variations occur inside the gap between the plates, the fluid velocity can be assumed
to vary linearly from 𝑣2 at the lower plate to 𝑣1 at the upper plate.
τ1
l ω1
w v1 h
F1
d
h z F2
v2 ω2
τ2
Fluid fills space Velocity profile, u(z)
between plates D

(b)
(a)

Figure B-2.16 (a) Parallel plates with viscous fluid and (b) rotational drag cup.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
2.9 Problems 83

From the definition of fluid viscosity, the shear stress, 𝜏, acting on the surface of one of the plates is 𝜏 = 𝜇 dudy
, where
𝜇 is the viscosity, u is the velocity of the fluid, and y is the distance measured normal to the plates. This assumption
holds for Newtonian fluids.
a) Derive an expression relating the force F and the velocities 𝑣1 and 𝑣2 . What type of element represents this relation-
ship?
b) If the velocities become very high or if the gap, h, becomes large, the velocity profile, u(y) will become distorted and
the flow between the plates will become turbulent. Under these conditions, the shear stress on the plates varies as
the square of the velocity difference: 𝜏 = c(𝑣1 − 𝑣2 )|𝑣1 − 𝑣2 |. Can the behavior of the plates (F versus 𝑣1 and 𝑣2 ) still
be represented by linear damping element?
c) For the rotational drag cup shown in Figure B-2.16(b), assume that 𝜔1 , 𝜔2 are inputs and that h does not change
and is small relative to the diameter D. Use the results from items (a) and (b) to derive a relationship between 𝜔1 ,
𝜔2 , and 𝜏 = 𝜏1 = 𝜏2 . Explain how these assumptions hold as long as the rotational inertias of the rotating elements
are negligible.

Problem B-2.17 Hydraulic reservoir system The hydraulic reservoir system shown in Figure B-2.17 is to be modeled
assuming the linear hydraulic elements indicated. Identify state variables, assume all linear constitutive relations for
the elements, and derive a mathematical model in state-space form. Confirm your understanding of how the junction
relations relate to basic Kirchhoff voltage law (KVL) and Kirchhoff current law (KCL) relations for electrical systems.

I1, R1 C1 R2 C2
P (t)

Figure B-2.17 Hydraulic reservoir system.

Problem B-2.18 Circulation pumping system In the system in Figure B-2.18, a circulation pump with controlled
flowrate, Q(t) at point C, can drive flow between tanks A and B. Assume that there is fluid inertance and resistance in
pipes 1 and 2 as indicated. Determine the states required for a model of this system, assume all linear elements, and
derive the state equations.

I1, R1
CA CB

I2, R2
Circulation
C pump, Q (t)

Figure B-2.18 Hydraulic fluid circulation

system.

Problem B-2.19 Elastic energy in a pressure vessel Solved Problem A-2-9 shows how energy is stored in a pressure tank
due to fluid compressibility. It is also possible to account for the effect of mechanical elastic expansion of the vessel
due to internal pressure. Approximate relations for thin-walled vessels are described in the diagrams and relations of
Figure B-2.19.
a) Use the relations provided to estimate the change in volume, Δ– V of a pressure vessel under uniform radial pressure
(Case 1).
b) Determine the effective hydraulic capacitance for a thin-walled pressure vessel, Ctw = P∕Δ– V , using the parameter
values given for the pressure tank in Problem A-2-9.
c) Since the tank given in Problem A-2-9 has capped ends, a better approximation for the effective hydraulic capaci-
tance would be given using the relations for Case 2 in Figure B-2.19. Repeat steps (a) and (b) for this case.
d) Compare the values of Ctw found in steps (b) and (c). Explain whether they compare in the way you would expect.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
84 2 Kirchhoff Systems

Figure B-2.19 Thin-walled pressure vessel relations: P = pressure, 𝜎1 = meridional stress,

𝜎2 = circumferential or hoop stress, R = inner radius, t = wall thickness, ΔR = change in
radius, ΔL = change in length or height, E = modulus of elasticity, 𝜈 = Poisson’s ratio.

Analogy between systems

Problem B-2.20 Analogous systems Study the mechanical and electrical systems shown in Figure B-2.20. First confirm
that the mechanical system in (a) can be modeled by the analogous electrical circuit shown in (b). Then derive the
system equations for each in state-space form, using analogous system elements and state variables.

v1 v2
R1 C1
k1

L1
k2 b2 C2 R2
b1 m1
(a) (b)

Figure B-2.20 Mechanical and electrical system analogs.

Coupling Elements and Mixed-Energy Systems

Problem B-2.21 Link torsion spring A torsional element is constructed by combining an ideal linear spring with a lever
as shown in Figure B-2.21. The unstressed length of the spring is l, and the lever radius is also l. The lever is vertical
and the spring is horizontal when the torque T is zero.
a) Determine and sketch an expression for T versus 𝜃 for 0 < 𝜃 < 𝜋∕2. What element is represented by this function?
b) On your sketch show graphically the area representing the energy stored when 𝜃 = 𝜋∕4.
c) Develop an analytical expression of the energy stored in this element as a function of the angle 𝜃.

θ l

Figure B-2.21 Torsional spring.

Problem B-2.22 Mass–spring-link system The system shown in Figure B-2.22 was worked in Solved Problem A-2-4.
Study the derivation of equations and either re-write the nth-order mathematical model equations in terms of
first-order state equations or re-derive the equations in the first-order form. Since this system is linear, so write the
state equations in linear ABCD form, including an output equation for the velocity across the damper, b1 .
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
2.9 Problems 85

FBD:
b2 FC
vC
b c m2 b c
B Fb2
C
vB
k1 FB
Fk2
k2
A
b1

Figure B-2.22 Linkage coupling mechanism.

Problem B-2.23 Spring–mass–piston tank system For the system shown in Figure B-2.23: (a) separate the system into
basic elements identified in the schematic (mechanical, hydraulic), (b) apply analysis to key elements, such as FBDs,
control volumes, etc., toward the goal of building a mathematical model of this system, and (c) derive a mathematical
model sufficient to describe the dynamics of this system, and express as state equations. Make sure to state all of your
assumptions in completing these steps. You may express any unknown constitutive relations as nonlinear functions of
proper state variables.

Short
restriction

Figure B-2.23 Hydromechanical system.

Problem B-2.24 Reduced model of analog meter A model of an analog voltmeter was described in Solved Problem
A-2-8.
Write two state equations that result when the meter inductance is neglected, and write an explicit relation for current,
im , as an output equation.

Problem B-2.25 PMDC motor model A schematic of a permanent-magnet dc (pmdc) motor is shown in Figure B-2.25.
The ideal electromechanical torque is 𝜏m = rm im , where rm is the motor constant (units of torque/current). The cor-
responding back emf relation is Vm = rm 𝜔m . Write a state equation for the flux linkage state of the inductor, Lm , and
another for the angular velocity, 𝜔m , of the motor rotor inertia, Jm . Include the effect of losses from the electrical
resistance, Rm , and rotational damping, Bm .

Figure B-2.25 Permanent-magnet dc (pmdc)

motor.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
86 2 Kirchhoff Systems

Problem B-2.26 Field-excited dc servomotor A field-excited dc servomotor drives a rotational load through a gear pair
as shown in Figure B-2.26. The motor has inertia, J1 , and the load has inertia, J2 .

constant

Shaft 1

Gear 2

Figure B-2.26 Field-excited dc servomotor.

Assume there are rotational losses (not shown) both in the motor and on the load. The input voltage is Vs (t). Assume
the servomotor torque can be given by 𝜏m = rm im , where rm is a constant (for a constant field circuit current).20 The
“back emf,” Vm = rm 𝜔1 . There is also a load torque, 𝜏L , applied on J2 .
a) To simplify the model, determine an effective rotational inertia, Jeff , that combines J1 and J2 .
b) Derive the system equations (in first order state-space form).
c) Write an output equation for the motor speed, 𝜔1 , and for the motor torque, 𝜏m .

Problem B-2.27 Battery-powered spark ignition system A diagram of a classical battery-powered spark ignition circuit
is shown in Figure B-2.27(a). The spark plug requires a very high voltage, which is achieved by inducing a dynamic
ac voltage across the ignition secondary coil. One way this can be accomplished is by opening and closing the breaker
switch points mechanically as shown. Of course, this requires the points and distributor mechanisms to be tuned so
that the primary current is built-up while the points are closed, as shown by the circuit in B-2.27(b). The breaker switch
points are indicated by ‘SW’. When SW is opened as in B-2.27(c), the condenser helps induce a dynamically varying
voltage across the primary inductance (𝑣p = 𝜆̇ p ) that is transformed into the secondary coil to sparking the plug. The
large transformer ratio helps achieve the very high voltage require on the primary side.
a) Formulate a mathematical model (state equations) for the case where SW is closed ((b), dwell time) and then another
model for the case when SW is open ((c), spark duration). The state during dwell time should only be primary
current, ip , while both ip and condenser charge, qc (or voltage, Vc ) is needed during spark duration.
b) For each set of state equation(s), write the output equation for spark voltage, Vspark .
c) Set up a simulation to solve for the primary current, ip , and the spark voltage, Vspark , for the ignition circuit system
with the following parameters: Vb = 12 volts, Lp = 5 mH, C = 20 μF, Rp = 2 Ω, and n2 ∕n1 = 100. Set the dwell time
to 𝜏d = 5Lp ∕Rp , which is five times the effective time constant for the circuit in (a), in which C does not play a role.
Experiment with the time period for spark duration, allowing the current and voltage oscillations to damp out.

Distributor

Ignition Ignition
coils coils
Ignition Secondary Breaker
switch points Spark
Spark
plug plug
Primary
Ignition coils SW SW
Cam
Battery Condenser
Spark
plug
Breaker switch closed Breaker switch opens
(a) (b) (c)

Figure B-2.27 (a) Battery-powered spark ignition system. (b) Effective circuit during dwell time. (c) Circuit during spark duration.

20 Note, generally a field-excited dc motor may have a controllable field current.

 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
2.9 Problems 87

Thermal systems and thermal effects in systems

Problem B-2.28 Heat from rotational friction Solved Problem A-2-11

describes how power dissipated in a damper is used to estimate the heat transfer flow rate, Q̇ d from that element.
a) Consider a linear rotational damper with 𝜏d = B𝜔d . Write an expression for the heat transfer flow rate Q̇ d as well as
for the entropy flow rate, fsd , assuming the reference temperature is taken as the damper case temperature, Td .
b) Consider rotational damping due to a nonlinear friction torque, 𝜏d = 𝜇d (Td )sgn(𝜔d ), where 𝜇d (Td ) is a
temperature-dependent coefficient of friction. Find Q̇ d and an expression for the entropy flow rate fsd .

Problem B-2.29 Rotational shaft in journal bearing A rotational shaft is supported by a journal bearing formed by a
bearing support (B) in Figure B-2.29(a). The shaft has diameter D and transmits power  = 𝜏𝜔 while loaded vertically
by FN .
a) Study and confirm a first thermal word bond can be formed as in Figure B-2.29(b). Complete the word bond graph
by annotating each thermal power bond with temperature and entropy flow rate variables. For example, it is assume
that the bearing support (B) can be modeled as an ideal thermal capacitance (PTSE) with entropy SB and tempera-
ture TB .
b) Assume that the shaft velocity 𝜔 is an input and that FN is known. Determine an approximate expression for the
friction torque, 𝜏, induced by the bearing. If needed, assume the shaft-bearing contact area remains constant at Ac .
c) Write expressions for the heat transfer flow rate Q̇ f and entropy flow rate fsf due to the bearing friction.
d) Write all the expressions and parameters that would be needed to model the indicated heat transfer processes.

Temperature
Bearing support, B source,

CONV

Friction heat PTSE (B)

generation

COND

Frame, F PTSE (F)

(a) (b)

Figure B-2.29 (a) Rotationally-driven shaft in a journal bearing. (b) Thermal word bond graph.

Problem B-2.30 Ingot quenching Consider an ingot immersed in a bath of constant temperature, To , as shown in
Figure B-2.30 during a quenching process.
a) Treat the ingot as a lumped thermal capacitance and sketch a thermal word bond graph, assuming the primary
resistance to heat transfer occurs at the interface between the ingot and the bath. Indicate the temperature and
entropy flow rates as power variables (quantifying heat transfer flow rate). Define parameters you would need to
determine heat transfer rates.
b) Write the entropy flow rate expressions needed for the heat transfer model.
c) With the entropy of the ingot, Si as the state variable, derive a single differential equation. Write the equation for
the ingot temperature, Ti , as a function of this state.

Bath

Steel
ingot

Figure B-2.30 Thermal

quenching of a steel ingot.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
88 2 Kirchhoff Systems

Problem B-2.31 Two-capacitance ingot quenching thermal model Describe a model of the ingot quenching that takes
into account the thermal capacitance of the bath. This model has two lumped capacitance elements. Use a word bond
graph to convey this model and annotate the thermal bonds. Assume the external temperature is To . Write the state
equations for this system.

Problem B-2.32 Three-capacitance ingot quenching thermal model The ingot quenching model can be further
extended by assuming that the ingot can be idealized as two concentric cylinders with defined thermal capacitance
(based on material, geometry). Discuss this model extension as in the previous problem.

Problem B-2.33 Thermal packaging You’re asked to estimate the temperature variation inside an instrumentation pack-
age that will be dropped into a marine environment that has a prescribed temperature variation with depth, To (z), where
z is the depth from the surface. Assume you will be able to estimate or control the rate of depth, 𝑣z . Describe a thermal
model with a single lumped capacitance, using a sphere as an estimate of the geometry for the package (assume a
radius, rp ) and an approximate value for thermal conductivity, kp . Explain the model formulation and how you would
go about building a model that can be used to design how the package should be designed to maintain the volume, V p .
Describe the role of thermal insulation and how you might introduce an internal electrical heat source, Q̇ s = Tp fsi , to
manage the temperature of the package, Tp . If Q̇ s is from a resistive heater having resistance Rs , explain also how your
model would allow you to estimate the power required given voltage Vs and known Rs .

Problem B-2.34 Brushed dc motor thermal model A typical brushed dc electric motor uses internal commutation,
allowing it to be powered from a direct current power source [31, 32]. As illustrated in Figure B-2.34(a), contact brushes
are used to allow current to flow into the armature (rotor) coil windings. Contact with the commutator is ensured by
a normal force induced using preloaded springs (not shown).
a) A simplified schematic of the permanent magnet dc (pmdc) motor is shown in Figure B-2.34(b). To study the thermal
effects in this motor, consider the coil resistance, Rm , and effective linear damping, Bm , model elements as the
principal sources of heat. For each element, write expressions for the heat transfer flow rate Q̇ and the entropy flow
rate fs .
b) Using the construction inferred by the simplified diagram of Figure B-2.34(a), recommend a first-cut model using a
thermal word bond graph construction. Make sure to integrate the sources of entropy flow rate from (a) and suggest
PTSE and heat transfer model elements to include. Identify key thermal states and variables on elements and on
thermal power bonds.

Rotor coils
Commutator Lm Rm
Jm
im
Shaft
Vs (t) Vm
Bm
Stator magnets

Vs (t) Brushes

(a) (b)

Figure B-2.34 (a) Typical brushed permanent magnet dc (pmdc) motor schematic. (b) Electrical
and mechanical schematic.

Problem B-2.35 Friction fire in a nutshell Consider a simple model to explain the ignition of wood due to
friction-induced heating as illustrated in Figure B-2.35(a). A wooden shaft (S) is positioned orthogonally rela-
tive to a flat wooden base (B), and the contact geometry is assumed to remain uniform and flat. The shaft is driven
rotationally with a known angular velocity, 𝜔(t), and a normal force is maintained axially along the rod to create, FN ,
the normal contact force. The coefficient of friction between the shaft and the wood base is taken as 𝜇. Assuming a
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
2.9 Problems 89

uniform contact area, A, the shear stress due to friction induced at the interface can be estimated by 𝜏f = 𝜇FN ∕A, as
shown in Figure B-2.35(b), taken here as independent of the velocity magnitude.
a) Consider the differential torque induced due to the annular element of width dr and area dA = 2𝜋rdr is given by
d𝜏 = r ⋅ 𝜏f ⋅ dA and find the total torque 𝜏 by integrating over the total contact area.
b) If the wood base will be assumed to be a lumped thermal capacitance (or PTSE), take its temperature as TB . Write
an expression for the heat transfer flow rate and the entropy flow rate generated by the friction contact given input
𝜔(t).
c) Sketch out a thermal bond graph that couples the rotational power port to the thermal elements and that would be
needed to begin constructing a friction fire model. Make assumptions about how and where you would represent
storage of internal energy and significant heat transfer flows as T − fs thermal power bonds. Explain assumptions
for which heat transfer modes would be most significant. Assume an ambient temperature is TA and state any other
factors and assumptions.

Shaft (S)

rad

cond
Base (B)

(a) (b)

Figure B-2.35 (a) Rotationally driven shaft with end contact on a surface. (b)
Contact detail for determining friction torque.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
91

Physical Modeling with Bond Graphs

We are now in a position to consider the task of modeling real devices and systems using the elements introduced in
Chapter 2. In this chapter, we begin to adopt a more unified and integrated treatment of practical engineering systems in
terms of the interconnection of the storage, transfer, and conversion (or transduction) of energy, both within and across
various domains. We will define and discuss the basic variables and elements of bond graph modeling. In this context,
we will also introduce the important modeling concept of causality, which will play a central role in the generation of
state equations. This chapter also discusses additional structure that should be placed on a bond graph model to insure
self-consistent and physically realizable systems.
Bond graphs model the significant power flow using power bonds which can be, albeit incompletely, represented by a
single line between two model elements. Sign convention (e.g., direction of positive power flow) is then indicated by a
half-arrow. The classical convention of using a full-arrow signal line in system and block diagrams is reserved for flow
of information, not power. Thus, the half-arrow power flow bond implies there are two (conjugate) power variables. A
summary of generalized and domain specific power conjugate variables is given in Table 3.1.
Consider the permanent magnet DC motor shown in Figure 3.1(a), commonly used both for sensing and actuation in
many systems. The key power interactions are depicted by the word bond graph of the electric motor shown in Figure 3.1(b).
The power interactions of this system are the electrical power represented by a voltage- current pair of variables,
mechanical power represented by a torque–angular velocity pair, and power dissipation in the system represented by the
temperature–entropy production rate pair. These variables are: 𝑣 = voltage, i = current, 𝜏m = motor torque, 𝜔m = motor
angular velocity, Tm = motor temperature, and fsm = Ṡ m = motor entropy production rate. For the motor as our system of
interest, the power flows sum as,
∑
̇
 = Vi − 𝜏𝜔 − Tm fsm = ,
where  represents the stored energy.
Next, consider the hydraulic pump in Figure 3.2(a) (such as used in fluid power systems), where the important power
interactions are the shaft power, discharge hydraulic power, and thermal losses as indicated in the pump bond graph in
Figure 3.2(b). The in-take hydraulic power can be assumed negligible. If the electric motor depicted in Figure 3.1(a) is used
to drive the pump, the direct coupling of the real motor to the real pump can be represented abstractly by bonding the
shaft of the motor model to that of the pump model. Functionally this means that 𝜏m = 𝜏p and 𝜔m = 𝜔p and shaft power
leaving the motor enters the pump through its shaft where 𝜏p = pump torque, 𝜔p = pump angular velocity, PD = discharge
pressure, Qd = discharge flowrate, Tp = pump temperature, fsp = pump entropy production rate. Input hydraulic power is
indicated by inlet pressure PI and flowrate QI .
It is re-emphasized that the half-arrow in bond graphs indicate the direction of positive power. For example, the half
arrows indicate that power is considered positive if it enters the motor in Figures 3.1 via the electrical port and exits the
motor through the mechanical rotation and thermal ports. These half-arrows just indicate a sign convention – they do not
require that the power be positive. Power could flow in the opposite direction to the arrow in which case the power will
be considered as negative. For example, if the motor were run as an electric generator then the mechanical and electric
power would be negative with the given sign convention. More discussion on sign convention is given when we discuss the
derivation of equations from bond graph models later in this chapter. Once again, the use of a full arrow is reserved for signal
bonds, which have historically been used to represent solely conveyance of information – that is, no power flow. For this
reason, signal bonds are distinguished in our models from power bonds, and both are used extensively in communicating
a model’s structure.
Modeling of Physical Systems: Simulation and Control, First Edition. Raul G. Longoria and Joseph J. Beaman.
© 2025 John Wiley & Sons Ltd. Published 2025 by John Wiley & Sons Ltd.
Companion Website: www.wiley.com/go/Modeling_of_Physical_Systems/1e
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
92 3 Physical Modeling with Bond Graphs

Table 3.1 Generalized and domain-specific power conjugate variables.

Kirchhoff Systems

Generalized Translation Rotation Electric Circuit Hydraulic Magnetic Circuit

e effort F force 𝜏 torque V voltage P pressure M magnetomotive

force
f flow 𝑣 velocity 𝜔 angular velocity i current Q volume flowrate 𝜑̇ magnetic flux rate
p momentum p linear h angular λ magnetic flux Γ momentum per n/a
momentum momentum linkage area
q displacement x displacement 𝜃 angular q charge – volume
V 𝜑 magnetic flux
displacement

Thermodynamic Systems

Generalized Thermal Chemical

e effort T temperature 𝜇 chemical potential

f flow fs = Ṡ entropy rate fN = Ṅ molar flow rate
p momentum n/a n/a
q displacement S entropy N moles

Permanent magnet Permanent magnet

dc motor dc motor

(a) (b)

Figure 3.1 (a) Permanent magnet DC motor and (b) a word bond graph depiction.

Hydraulic Hydraulic
pump pump
Fluid
Fluid discharge
intake

(a) (b)

Permanent magnet Hydraulic

dc motor pump

(c)

Figure 3.2 (a) Pump schematic. (b) Pump word bond graph. (c) Word bond graph for motor–pump system.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
3.1 Bond Graph Variables 93

Hydraulic Hydraulic
pump pump
Fluid
Fluid discharge
intake

(a) (b)

Permanent magnet Hydraulic

dc motor pump

(c)

Figure 3.3 Motor–clutch–pump system.

For example, suppose it is proposed to have the motor-pump system previously described fitted with a clutch to control
engagement of the pump drive shaft. As shown in Figure 3.3(a), a control signal can be used to switch the clutch between
its operative modes. A high-level word bond graph indicates the need to incorporate not only the rotational power flows
from the motor to the pump but also quantify how the clutch control signal modulates this power flow. It is also expected
that the clutch will dissipate power, so a thermal port is also included as shown in Figure 3.3(b).

3.1 Bond Graph Variables

As discussed in Chapter 1, each power bond of a bond graph has an associated pair of variables. These variables are chosen
such that their product is equal to the power transmitted by the bond. If the pair of variables are labeled e, for effort, and f ,
for flow, then, for a single-port system or sub-system, we have the power relation,
 = e ⋅ f.
It is also advantageous to define the time integration of the effort as the generalized momentum p and the time integration
of flow as the generalized displacement q,
p = ∫ edt. (3.1)
q = ∫ fdt. (3.2)
Paynter’s tetrahedron of state is a useful graphical depiction of how the bond graph variables are related [1]. Two of the
lines on this tetrahedron, shown in Figure 3.4, can be used to represent the fundamental physical laws studied in Chapter 2:

Effort
Ca
Resistive

pa
cit
ive

Momentum, Displacement
Ine
rti
ve

Flow

Figure 3.4 Bond graph variables and Paynter’s tetrahedron of state. Source: Adapted from [1].
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
94 3 Physical Modeling with Bond Graphs

ṗ = e and q̇ = f , while the remaining three represent basic element constitutive relationships. A dashed line connects p and
q, the state variables used to identify independent energy storage in physical systems.
The relations between the port variables in the various energy domains and the generalized variables are summarized
in Table 3.1.1 Note that for the thermal and chemical domains the integrated efforts (temperature and chemical potential)
have not been defined. The reason for this omission is based on the second law of thermodynamics as will be discussed in
Chapter 9.

3.2 Bond Graph Elements

Bond graphs depict the energy flow and storage in physical systems. In order to develop state determined functional rela-
tionships for systems and subsystems such as those shown in Figures 3.1 and 3.2, a set of bond graph modeling elements
are needed that have precise functional descriptions. These essential modeling elements can be either:
1) store energy,
2) transform energy into available forms,
3) dissipate energy into unavailable forms,
4) join or add subsystems to form total systems, and
5) in order to describe the effects between the environment and the system, be a source of energy.
It is convenient to create bond graph modeling elements that exactly fit the taxonomy in the above list. For convenience,
Tables 3.4–3.6 (found at the end of this chapter) summarize key relations for C, I, and R elements for Kirchhoff systems.

3.2.1 Energy Storing Elements

Since one of our goals is to develop state-determined functional descriptions of physical systems, we assume that the energy
can be found as a function of a set of state variables. As discussed in Chapter 1, knowing the state variables at time t = to and
the values of the inputs to the system for t > to determines the state variables for all t > to . This implies that for an energetic
state-determined system, the energy of the system is also determined. For any energy storing element (S) of an energetic sys-
tem, we can equate the input power to the rate of change of the stored energy. For a single-power or one-port system element,
̇ 𝜕
 = (x) = ̇
x, (3.3)
𝜕x
where  is the stored energy for this assumed ideal lossless, energy storing element, and x is the state variable. This
formulation can readily be generalized to n power ports, but for now the one-port bond graph is shown in Figure 3.5(a),
with the half-arrow power bond indicating power is positive into the system.
An associated input–output form can also be conceived as shown in Figure 3.5(b), where input u and output y variables
quantify the power entering the system,
̇
 = y ⋅ u = . (3.4)
Equating the input u with ẋ and the output y with 𝜕∕𝜕x produces a desired product form for the power and is an energetic
(thermodynamic) justification for using power product pairs of variables. This will be defined as a general integral form for
an energy storing element and is expressed as,
rate ∶ ẋ = u, (3.5)
energy ∶ (x) = y ⋅ u dt = y ⋅ dx, (3.6)
∫ ∫
constitutive ∶ y = Φs (x), (3.7)

u
S
Figure 3.5 (a) Single-port system bond graph and (b) input–output form.
S y

(a) (b)

1 These variables are defined and useful for lumped-parameter modeling. See Paynter [1] or Brown [7] for discussion on how these variables
relate to those used to model physical systems where continuum and field effects must be considered.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
3.2 Bond Graph Elements 95

where Φs (x) is a constitutive relation in the form of a proper function and in general u and y represent power flow variables
(i.e., effort and flow) on the bonds.
e
3.2.1.1 Generalized Potential Energy: C Element q̇ = f
C
As we found in Chapter 2, the types of energy storage elements (ESEs) are commonly divided into two
Figure 3.6
types2 : one in which the state is generalized displacement and the other in which the state is generalized Single or
momentum. The first type is given the symbol C as shown in Figure 3.6. one-port C
This element is functionally defined as, element.

rate relation ∶ q̇ = f , (3.8)

constitutive relation ∶ e = Φs (q), (3.9)
where Φs (q) is a time-invariant constitutive relation for the C element. The C element can then be defined as:

C element: A bond graph energy storing element in which the effort is a single-valued function of the displacement
such that an energy function is uniquely defined.

The energy stored in a C element can be found from a generalized work relation as,

= e ⋅ dq = Uq , (3.10)
∫
where Uq represents generalized potential energy. A complimentary coenergy, Ue , can also be defined as,

Ue = q ⋅ de = e ⋅ q − Uq . (3.11)
∫
The C element is used to model mechanical compliance and electric, hydraulic, thermal, and chemical capacitors, such as
the basic Kirchhoff potential energy elements discussed in Chapter 2. An example of a single port C element is a mechanical
translational spring with one end fixed, as shown in Figure 3.7.
In this case, it is clear that power flows into the system at one port. Many other examples will be given later in this
chapter. Although we will discover that there are many important engineering systems that can be modeled with single
port energy storage, there are numerous others that do not allow this simplification, and therefore circuit-like techniques
cannot be directly applied to them. In those cases, a more general multiport C element can be adopted. An example where
a two-port C may be needed is when modeling potential energy stored in a cantilevered beam. Two different configurations
are shown in Figure 3.8. In Figure 3.8(a), the total stored potential energy can only be determined by defining a function
of both displacement x and angular displacement 𝜃. Stored potential energy can be changed either by pushing vertically

Figure 3.7 Translational spring as a one-port C element. v

F F
x v
C

Power flow
is zero
(a) (b)
Figure 3.8 (a) Cantilevered beam as a 2-port C element (b) 2-port with zero torque on rotational port.

2 In actuality only one type is needed as will be discussed later. Moreover, as we shall learn, Hamiltonian dynamics merges these separate types
into a single total energy function.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
96 3 Physical Modeling with Bond Graphs

(a)

0
(b) (c)

Figure 3.9 (a) A translational spring with distinct end velocities. (b) Proposed use of two-port mathbfC element. (c) Valid one-port C
usage for the translational spring.

downward on the beam to effect a change in displacement x or, independently, by twisting the end of the beam to cause a
change in angular displacement 𝜃 (relating to the slope of the beam).
In the case where 𝜏 = 0 as shown in Figure 3.8(b), we only push vertically on the beam while its end is free to rotate.
This is a clamped-free beam, in reference to the support conditions. The two-port C could be reduced to a one-port C, since
power is zero on the rotational side, thus appearing equivalent to the purely translational spring of Figure 3.7. The two-port
C would also reduce to a one-port if the angular velocity 𝜔 = d𝜃∕dt = 0. In this case, we could model a clamped–clamped
beam as a simple translational spring. By this example, we can see that our definition of a C element can be more general
than the circuit-like springs (and capacitors) discussed in Chapter 2, and it is only in the special one-port case that the
concepts are equivalent.
Multiport elements are extremely useful in modeling practice but should only be used as needed. When first introduced
to this concept, one may be tempted to use a two-port C element to model the translational spring shown schematically in
Figure 3.9(a) using the form in (b). This representation, however, does not recognize that, in this case, energy can only be
stored due to changes in the spring displacement, x = ∫ (𝑣1 − 𝑣2 )dt, which quantifies spring force through a constitutive
relation, F = Φ(x). Unlike the beam example described earlier, this spring does not require two unique states to quantify
the stored energy nor are there two relations for the forces, since F1 = F2 = F for this pure translational spring.3 Recall,
ideal springs only account for stored potential energy and typically it is assumed that mass and dissipation effects are either
negligible or can be accounted for separately by other elements.
It can be helpful to first gain practice using bond graphs with single port elements, which are applicable to a very wide
range of physical systems, prior to adopting multiport modeling methods. For this reason, modeling with multiport ele-
ments will be more fully considered in Chapter 8. The use of two-port elements, however, will be introduced for the special
case of thermal system modeling later in this chapter in Section 3.7, since this will allow treating many problems of practical
interest.

3.2.1.2 Generalized Kinetic Energy: I Element

Consider now the case where the generalized momentum, p, is the state variable for the energy storing element, given the
bond graph symbol I as in Figure 3.10.
The functional description of this I element is then, ṗ = e
I
f
rate relation ∶ ṗ = e. (3.12)
Figure 3.10
constitutive relation ∶ f = Φs (p). (3.13)
Single or
The I element can now be defined as: one-port I
element.
I element: A bond graph energy storing element in which the flow is a single-valued function of the momentum
such that an energy function is uniquely defined.

The generalized kinetic energy, Tp , stored in the I element can be expressed as,

= f ⋅ dp = Tp , (3.14)
∫

3 The term pure is used here to refer to elements that model one form of energy storage or dissipation [33], which results from our modeling
abstraction through reticulation as discussed in Chapter 1.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
3.2 Bond Graph Elements 97

Figure 3.11 (a) Lossless electrical transformer and (b) modeled as a two-port I element.

(a) (b)

and the kinetic co-energy, Tf , is

Tf = p ⋅ df = f ⋅ p − Tp . (3.15)
∫
The I element is used to model mass, inertia, and inductance effects in physical systems.
Similar to the C element, all of the Kirchhoff kinetic ESEs discussed in Chapter 2 are single-port I elements. An example
of the need for a two-port I element would be to represent how energy is stored in the two-winding transformer shown
in Figure 3.11(a). A two-port I as shown in Figure 3.11(b) would represent the energy storage but not the losses in a real
transformer, which would need to be modeled separately (using resistive elements). Note how bond graph sign convention
dictates positive power flow is into each port of the two-port.

3.2.2 Ideal Coupling Elements

An ideal coupling element is characterized by the feature that it neither creates, stores, nor dissipates energy. It simply
converts energy into other available forms. This implies that a coupling element is power conserving,
∑
n
= yi ⋅ ui = 0, (3.16)
i=1

or the net power into the device is identically zero. The power conserving constraint 3.16 is enforced if there exists a relation
between input and output such that,
y = Hu, (3.17)
T
where H = −H defines a skew symmetric matrix and in general H can be a function of other system variables. Conversely,
it can be shown that any time a system has a power conserving constraint 3.16 then the input–output function for the system
can be expressed as the skew symmetric relation 3.17 between the input and output variables.4

3.2.2.1 The Two Port Special Case

If we now concentrate on just two-port coupling elements, then it is conventional to assume power is positive into one port
and positive out of the other. In this case, the power constraint 3.16 can be expressed alternatively,
1 = y1 u1 = y2 u2 = 2 , (3.18)
and the input–output constraint relation 3.17 can then be expressed as the relation,
[ ] [ ][ ]
y1 0 h u1
= . (3.19)
y2 h 0 u2
This expression can also be directly obtained by considering a general linear input–output relation of the form,
y1 = h11 u1 + h12 u2 .
y2 = h21 u1 + h22 u2 .
The power conserving constraint 3.18 then implies that,
y1 u1 = h11 u21 + h12 u2 u1 = h21 u1 u2 + h22 u22 = y2 u2 .
In order for this expression to be true for general independent inputs u, this requires that,
h11 = h22 = 0,
h12 = h21 ,
which is identical to 3.19.

4 See Wyatt and Chua [34].

 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
98 3 Physical Modeling with Bond Graphs

Note that 3.19 implies that an input on bond 1 (alone) determines an output on bond 2 (alone) and vice versa. Since y
and u can in general be either efforts or flows, there are two possibilities of input–output pairs of variables which satisfy
the power constraint:

1) the effort on port 2 is related to the effort on port 1 and the flow on port 2 is related to the flow on port 1, or
2) the effort on port 2 is related to the flow on port 1 and the effort on port 1 is related to the flow on port 2.

There are thus two distinct types of power-conserving two-ports defined to match these two possibilities. One is the
transformer or T element which functionally satisfies a relation in the form,
f1 = nf2,
e2 = ne1, (3.20)

where n is the transformer modulus. The T element can also be defined as:

T element: A two-port bond graph power-conserving element in which the efforts on the two ports and the flows on
the two ports are directly related.

The other type of power-conserving two-port is the gyrator or G element which functionally satisfies a relation in the
form,
e1 = rf2 ,
e2 = rf1 , (3.21)

where r is the gyrator modulus. The G element can then be defined as:

G element: A two-port bond graph power-conserving element in which the effort on a port is directly related to the
flow on the other port.

The T element models the ideal power transformation in devices such as levers, gears, electric transformers, fluid
pistons, and convective couplings. Similarly, the G element is used to model electromechanical (EM) couplings, gyro-
scopic forces, and dynamic turbomachines. Figure 3.12 shows two common coupling elements. A model for a lever
in Figure 3.12(a), for example, assumes the bar is rigid and has negligible mass. Thus, 𝑣1 ∕l1 = 𝑣2 ∕l2 , where l1 and l2
are the respective lengths on each side of the pivot. This identifies a modulus, n = l1 ∕l2 , relating the velocities (flows)
indicating a transformer. The corresponding force relation can be found from the power constraint: F1 𝑣1 = F2 𝑣2 . The
other example is a gyrator element, used to model the EM coupling when a conductor moving through a magnetic field
(cf. Figure 2.44).
The power-conserving property of ideal coupling elements has also been referred to as isentropic. From thermodynamics,
a process is isentropic when there is negligible heat transfer (adiabatic) and it is reversible, and thus entropy remains con-
stant. As such, realistic effects that arise in coupling effects that store or dissipate energy should be accounted for separately
from the ideal T and G elements. For example, a hydraulic piston can be represented by an ideal T element as in Figure 2.45.

(a) (b)

Figure 3.12 (a) Lever as a transformer coupling element. (b) Electromechanical force as gyrator coupling element.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
3.2 Bond Graph Elements 99

However, a more realistic model may include leakage, friction in the seals, compressibility in the fluid, and inertia effects.
Any specific modeling study would need to weigh the benefits of incorporating such effects.
We also point out that modeling of some physical devices and processes may require moduli n and r that vary as functions
of system variables while retaining the power conserving property in ideal coupling elements, i.e., e1 f1 = e2 f2 . In this case,
the elements are said to be modulated transformers or modulated gyrators. A separate discussion on the role of modulation
in physical system modeling is postponed until Chapter 4. Finally, it can be argued that only the G element is strictly
necessary in modeling. The main point is that one can construct a T by using two cascaded G elements. This point is not
purely theoretical and has been shown to have practical value.5 For the most part, however, there is practical value when
clearly identifying and categorizing power conserving effects in systems using the two distinct coupling elements.

3.2.3 Dissipative Element

A distinguishing feature between a general and ideal power transforming element (i.e., the e T
transformer or gyrator) and a purely dissipative element is the inherent entropy production in f
R fs
the latter. The power dissipated by dissipative elements is equal to the entropy production, d =
T Ṡ = T ⋅ fs > 0. A general dissipative element or process is represented by an R as shown in Figure 3.13 Single- or
Figure 3.13. Note that this element includes at least one port with effort and flow variables as one-port R element with
thermal port.
well as a thermal port. The thermal port is used to represent the flow of dissipated power, d ,
into heat. The thermal bond is indicated by a dashed line to convey the energy lost to unavailable
forms. The flow variable is entropy rate, fs , which when taken with temperature defines an energetically consistent set of
power-conjugate variables, since e ⋅ f = T ⋅ fs (units of power). The dashed line as shown in Figure 3.13 will be used to model
power flow in the thermal domain.
If the entropy production bond or thermal port is suppressed, as it commonly is for isothermal systems, then for a
single-port dissipative element,
e ⋅ f ≥ 0, (3.22)
if power is assumed positive into the R element. For non-isothermal cases where the entropy production port is not sup-
pressed we have, as stated earlier,
e ⋅ f = T ⋅ fs , (3.23)
where fs = Ṡ is the rate of entropy production. With the physical constraint that the (absolute) temperature,
T ≥ 0, (3.24)
this in turn imposes the constraint,
fs ≥ 0, (3.25)
in order to satisfy the second law of thermodynamics.
If the temperature–entropy port is suppressed, the R element can be defined as,

R element: A bond graph dissipative element in which there is a constitutive function relating effort and flow such
that the product of effort and flow is greater than or equal to zero if positive power is directed into the element.

This R element can be used to represent the losses and nonequilibrium effects in a physical system: friction, resistance,
diffusion, chemical reaction, etc. This two-port form of the R element would be used to model the entropy flow rate source
element as introduced in Chapter 2.
Consider the general case where u is a variable causally specified as the input to an R element (either e or f ) and y is the
output variable. The general dissipative or R element then has a general constitutive relation of the form,
y = Φd (u, T), (3.26)

5 For example, you can show that a torque converter is effectively two cascaded turbomachines of a G type [35].
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
100 3 Physical Modeling with Bond Graphs

(the subscript “d” here referring to a dissipative type element). The corresponding entropy production relation can then be
found from,
fs = (y ⋅ u)∕T, (3.27)
where u and y are the power-conjugate variables for the specific R element related by the constitutive function Φd (). These
relations were introduced in Chapter 2 for modeling power sources for basic thermal systems. Note that in order to be
dissipative, the constitutive relation 3.26 must be constrained such that  = y ⋅ u ≥ 0. The single port version of the R
element is thus suitable when thermal effects are not significant.
Additional discussion of the R element will be given in Section 3.7, and again in Chapter 9 on system thermodynamics.
Reference tables of the C, I, and R elements are provided at the end of this chapter.

3.2.4 Junction Elements

We have defined, in the preceding sections, 3.2.1–3.2.4, the energy storing elements which functionally can be considered
as integrating elements, and the coupling and dissipative elements which imply proportional and direct functional
relationships. The joining or addition of these elements to form system and component models is provided by the junction
elements. Functionally, the junction elements represent addition and subtraction in physical models. Physically, they
represent the summation of forces and velocities in mechanical systems as dictated by kinematics, dynamics, and Newton’s
laws, the summation of currents and voltages in circuits as expressed by Kirchhoff’s laws, the summation of pressure drop
and volume flows in hydraulic systems, the conservation of matter in pneumatic and chemical systems, and the addition
of entropy production in thermal systems.
There are two basic bond graph junction elements6 : the common flow or 1-junction, and the common effort or 0 junction,
as shown in Figure 3.14. Both of these elements represent power conservation.
A 1-junction is used to model physical interconnections where attached bonds have flows that are all the same. Since the
element is assumed power conserving, then (with f the common flow),
( n )
∑n
∑ ∑ n
ei ⋅ fi = ei f = 0 ⇒ ei = 0, (3.28)
i=1 i=1 i=1

or efforts sum. The 0-junction is used to model interconnections for which the efforts on attached bonds are all the same
and therefore we have (with e the common effort),
( n )
∑n
∑ ∑n
ei ⋅ fi = fi e = 0 ⇒ fi = 0, (3.29)
i=1 i=1 i=1

or flows sum.
The 1-junction can be used to represent system joining processes like summation of forces and torques and series circuit
connections. The 0-junction can be used to represent joining processes like summation of velocities, parallel circuits, and
conservation of matter.
As an example, consider when a translational spring and damper are connected first as shown in Figure 3.15(a). With
this topological connection, the force is common (or the same) to both of the elements and the velocities sum such that the
total velocity 𝑣 is equal to the spring velocity, 𝑣s , plus the damper velocity 𝑣d . If the elements are connected as shown in

1 1 Figure 3.14 (a) Common effort junction. (b) Common flow junction elements.

2 2
0 1
3 3

(a) (b)

6 As for energy storing and coupling elements only one is necessary.

 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
3.2 Bond Graph Elements 101

0 1

(a) (b)

Figure 3.15 Physical examples of junction elements (a) common effort (force) and (b) common flow (velocity).

Figure 3.15(b), then the velocities are common and the forces sum. Note that in general the half-arrow power sign can be
directed both into or away from the junction elements.7 This allows for representing both addition and subtraction of effort
and flow variables as,
∑
n
𝜎i ⋅ ei ⋅ fi = 0, (3.30)
i=1

where 𝜎i = 1 if the half arrow is directed into the junction and 𝜎i = −1 if it is directed away from the junction.
The junction elements can be defined as:

0 element: a bond graph power-conserving element which has the same effort at all of its ports.

1 element: A bond graph power-conserving element which has the same flow at all of its ports.

3.2.5 Source Elements

Two source elements are defined in the basic bond graph set. A source of effort or an Se element and a source of flow or an
Sf element. The source elements specify either the effort or the flow as a function of time and can be defined as follows and
shown in bond graph form in Figure 3.16:

Sf element: A bond graph element in which the flow variable is specified as a function of time.

Se element: A bond graph element in which the effort variable is specified as a function of time.

Source elements can be used to represent the power flowing into or out of a system from the environment. They are nec-
essary but unlike other physical modeling elements they do not conserve energy. As such, they are a potentially dangerous
source of modeling errors, and even experienced practitioners take care in their use and application.

3.2.6 Summary
We have now defined the basic bond graph modeling elements. They represent a modeling assembly language by which we
can build realistic and complex engineering physical system models. The generalized elements are summarized in Table 3.2.
Their use is best learned by example, which is the subject of Section 3.3. Explicit summaries of the energy storage and
dissipative elements for four of the basic energy domains studied at the end of this chapter are provided in Tables 3.4–3.6.
These tables include bond graph representations along with the Kirchhoff circuit complement.
It should be noted that the quantities following the elements in the bond graph description in these summary tables (e.g.,
m, C, B, R, and Rf ) represent a constitutive parameter for the special case of a linear element. They do not represent the
element itself, which can in general have nonlinear constitutive relations and require more than one parameter.

Figure 3.16 Source elements. e(t)

Sf Se
f(t)

7 Note, however, that we generally always assign positive power into C, I, and R elements.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
102 3 Physical Modeling with Bond Graphs

Table 3.2 Summary of basic bond graph elements.

Energy storing Coupling Dissipative Junction Source

e e1 e2 e T 1 Sf
C T R f (t)
q̇ = f f1 f2 f fs
y ⋅ u = Tfs f (t) given
(q) e2 = ne1 2 n
0
e = Φ(q) f1 = nf2 For u = e, y = Φ(e, T) ·
For u = f , y = Φ(f , T) 3 · ·

e 1 = e 2 = · · · = en
∑ n
fi = 0
i=1

ṗ = e e1 e2 1 e(t)
I G Se
f f1 f2
(p) e1 = rf2 2 n e(t) given
1
f = Φ(p) e2 = rf1 ·
3 · ·

f1 = f 2 = · · · = f n
∑n
ei = 0
i=1

3.3 Modeling Examples

This section includes a series of modeling examples demonstrating the use of bond graph construction.

Example 3.1 Sliding mass with spring

Consider a mass that slides on a surface with friction as shown in Figure 3.17(a). The mass is also constrained by a spring
at one end. Relate the free-body diagram in Figure 3.17(b) to the bond graph in (c).
Solution
An I element is used to represent storage of kinetic energy in the mass, a C element models the potential energy storage in
the spring, and the R element accounts for the dissipation due to sliding. The 1-junction represents the common velocity
of the mass, the spring, and the sliding frictional element. In addition, the junction element represents the summation of
forces by Newton’s law (nominally along a relevant degree of freedom for a mass).
The sign convention given in the bond graph, Figure 3.17(c), is consistent with that shown on the free-body diagram,
Figure 3.17(b). All of the forces, including the inertia force, FI , are directed opposite to the velocity and, since power
is the product of force and velocity in a translating system, the half-arrows are directed away from the 1-junction in
Figure 3.17(c). The summation of efforts at the 1-junction can be expressed as ṗ = FI = −Fs − Fd where p is the mass
momentum and the common flow at 1 implies ẋ s = 𝑣d = 𝑣m , where xs is the extension of the spring and 𝑣d is the damper
(sliding) velocity. It should be noted that the sign convention on the bond graph would not change if all the vector quantities

(a) (b) (c)

Figure 3.17 Translating mechanical system.

 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
3.3 Modeling Examples 103

(velocities and forces) in the free-body diagram were given opposite orientations. The only thing that would change would
be the sign of the initial mass momentum or velocity and the sign of the initial spring extension or force. It is customary to
assume positive power into all of the ports of the I and C elements and into all the available energy ports of the R elements
(not the T − Ṡ port) as this will, in general, be the desired sign convention. Although this is not a requirement, this custom
will be strictly followed in this text, as it will make discussion of the allowed forms for the constitutive function for these
elements simpler and it does not hinder the development of general models. The constitutive laws for the elements in this
example can be expressed as

Constitutive

Element General Linear

I 𝑣m = Φm (p) 𝑣m = p∕m

C Fs = Φs (xs ) Fs = kxs

R Fd = Φl (𝑣d ) Fd = b𝑣d

Example 3.2 LRC circuit

A series LRC circuit is shown in Figure 3.18(a). Although an electrical circuit already represents a well-defined model of a
physical system, it is instructive to interpret the circuit in terms of a bond graph.
Solution
The 1-junction represents the common current in all the elements, q̇ C = iL = iR , where qC is the charge on one plate of the
capacitor. This junction also enforces Kirchhoff’s voltage law as 𝜆̇ L = −VR − VC , where 𝜆L is the magnetic flux linked in the
inductor. Note that this bond graph and the one for the translating mass in Example 3.1 have the same topological structure.
This implies that these systems could be used as analogs of each other, where the analysis of one system can be used for
studying the properties of the other.

The schematic and circuit diagrams for Examples 3.1 and 3.2 represent fairly unambiguous descriptions of models that
can be interpreted in the same way by different observers. Consider the mechanical system shown in Figure 3.19(a), which
might represent part of a physical device.
In order to complete the modeling process several assumptions must be stated. Since a bond graph can be used to give
a unique description of a physical system, it is instructive to interpret the model in terms of a bond graph. The bar is
assumed to be a rigid element rotating about pivot with negligible friction. The bar stores only rotational kinetic energy
and is represented by the I element in the graph. The I element is bonded to a coupling T (transformer) element which
models the lossless transformation between rotational and vertical movement of the tip (A) of the bar.
Assuming small angular displacement, the modulus of the transformer is n = L, where L is the length of the bar from the
pivot to the point of connection with the spring. This transformer represents the flow (velocity) constraint 𝑣 = L𝜔 and also
the effort (torque) constraint ḣ = LFA , where ḣ is the rate of change of angular momentum of the bar and FA is the sum of
the forces at the bar’s tip.
The right end of the transformer is bonded to a 1-junction, which represents the common velocity of the bar’s tip (at point
A), the velocity of the applied force F, and the velocity of the upper end of the spring. This 1-junction also represents the
summation of forces at the tip, FA = F − Fs , with the sign convention implied on the graph.

Figure 3.18 (a) Series circuit. (b) Bond graph.

(a) (b)
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
104 3 Physical Modeling with Bond Graphs

Se

F
L
ḣ ·· FA
I ω T 1

0 C
ẋs

vd

R
(a) (b)

Figure 3.19 (a) Lever–spring–damper mechanical system. (b) Bond graph.

Seal Leakage
friction resistance
Seal
friction

Leakage to Cylinder
exterior pressure

(a) (b)

Figure 3.20 Loss effects in a hydraulic piston (a) can be integrated with an ideal T as in (b).

The 1-junction is bonded to the 0-junction which represents the common force of the upper and lower ends of the spring
and the damper force. This also represents the summation of velocities which gives ẋ s = 𝑣 − 𝑣d , where ẋ s is the stretch rate
in the spring and 𝑣d is the damper’s velocity. Note that these are two unique velocities.

Example 3.3 Leakage and friction in a hydraulic piston

The schematic for a hydraulic piston model is shown in Figure 3.20(a). Hydraulic pistons can have friction in seals between
the piston and fixed cylinder inner wall, and there can also be leakage from the cylinder to an exterior pressure. Extend the
ideal model for a hydraulic piston shown in Figure 3.12(c) to incorporate these dissipative effects to build a more realistic
model of a hydraulic piston.
Solution
Begin by identifying that the seal friction force will be associated with the same velocity as the piston, since the cylinder
is assumed to not be moving. Recall the relation between the piston velocity and the fluid flowrate is, Q = A𝑣, with A
the piston area. As conveyed in Figure 3.20(b), a free-body diagram gives the force relation, F = Fs + Fp , where now the
subscript “p” is used to distinguish the ideal piston force. Likewise, the piston velocity is indicated by 𝑣p , and with all forces
imposed with this common velocity a 1-junction is used on the mechanical side to connect to the seal friction R element as
shown in the bond graph of Figure 3.20(b).
Leakage occurs on the “hydraulic” side of the T element. In this case, we assume uniform cylinder pressure, Pp , the
cylinder, and assume there is negligible loss of pressure at the connection, so P = Pp . Thus, a 0-junction is used to connect
to the output (free) bond and to the R element that represents the leakage resistive effect. As shown, these pressures are gage
pressures (relative to common exterior pressure). Note that the 0-junction enforces the continuity relation, Q = Qp − Ql ,
where Qp = A𝑣p is the ideal flowrate from the piston motion and Ql is the leakage flowrate. If the pressures need to be
associated with different reference pressures, this model would require reformulation.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
3.3 Modeling Examples 105

Engine Shaft Bearing Propeller/slurry

(a)

0 1 1

Figure 3.21 (a) Word bond graph and (b) corresponding bond graph for engine–propeller system.

Example 3.4 Engine–propeller system

Recall the engine–propeller system discussed in Example 2.7. The first step in the modeling process was the reticulation
of the system into components. This reticulation is now represented as a word bond graph in Figure 3.21(a). Develop a
bond graph for this system and include a model for the engine that considers a known torque–speed curve. Discuss the key
relationships among the interconnections and constitutive relationships that will be needed.
Solution
Assuming the same simple model that was used Chapter 2, this word bond graph can be further reticulated into the ele-
mental bond graph given in Figure 3.21(b).
This model assumes that the engine is an angular velocity source (Sf ), the shaft is solely a compliance (C), the bearing is
a damper (R), and the propeller/slurry can be modeled as inertia (I) and dissipation (R). The junction 0 enforces constant
torque from the engine through the shaft to the bearing and accounts for difference in angular velocity between the engine
speed and bearing speed due to shaft twist. The two 1-junctions (which could be simplified to one 1-junction) enforces the
same angular velocity for the bearing, propeller, and slurry drag and accounts for the summation of shaft torque, bearing
torque, inertia torque, and drag torque (ḣ = 𝜏s − 𝜏B − 𝜏D ).
In Chapter 2, all of the constitutive relations for this system were assumed to be linear, but the bond graph given in
Figure 3.21(b) does not require this restriction. In fact, the constitutive relation for the R elements would most likely be
required to be nonlinear in order to accurately model a real bearing and slurry drag. In this case, the constitutive relation
for the R elements would take the form,
𝜏B = 𝜙B (𝜔B ),
𝜏D = 𝜙D (𝜔D ),
where the functions 𝜙B (⋅) and 𝜑D (⋅) are (nonlinear) constitutive relations as determined by material and geometry. The
constitutive relations for the C and I can be expressed as,
𝜏s = K𝜃s ,
𝜔p = h∕J,
if assumed to be linear.
If the engine is characterized by a steady-state torque–speed curve as shown in Figure 3.22(a) rather than an ideal angular
velocity source as given above, then a phenomenological bond graph model for the engine can be formulated by combin-
ing an ideal flow or effort source with a resistive element which accounts for losses. It is a useful and common modeling
practice to represent the steady-state behavior of real sources using such an approach. Real flow sources can be modeled
by using a Norton equivalent, as shown in Figure 3.22(b), while real effort sources can be modeled using a Thevenin equiv-
alent as in Figure 3.22(c). This approach is a generalization of how Norton and Thevenin equivalents are used to construct
equivalent representations of electrical networks (e.g., see Hambley [36] or Chua et al. [37]). In this case, the constitu-
tive relation for the R element could be chosen so that the speed or torque fit known performance, as may be given from
provided data.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
106 3 Physical Modeling with Bond Graphs

Figure 3.22 (a) Torque–speed curve for an engine and equivalent

representation as (b) Norton and (c) Thevenin equivalent source models
in bond graph form.

(a)

(b) (c)

The Norton equivalent model accounts for the drop in engine speed by subtracting from the no load (nl) speed 𝜔nl an
amount 𝜔de which is a function of the engine torque. The Thevenin equivalent model is similar in structure but it accounts
for the drop in engine torque by subtracting from the no speed (or stall) torque 𝜏ns an amount 𝜏de which is a function of
the engine speed 𝜔e . Both of these models will give the torque–speed curve of the engine and both of these models are
empirically based. Empirical models of this sort, although useful for a particular fixed engine, will not account for design
or environmental changes likely to occur in the general situation. A more detailed physical model of the engine would be
needed in order to account for changes of this sort. One should bear in mind, of course, that these models do not account
for transient effects either, as they are based on a steady-state characterization of the source.

These modeling examples demonstrate an approach to modeling that may be new to many readers. While the end result is
a bond graph, it should be clear that this is only the first stage. However, it should be recognized that by having a well-defined
library of modeling elements,8 it is possible to transition from a conceptual model to a graphical model representation that is
energetically correct. In addition, by using the model elements defined, it can be assured that the bond graph representation
also incorporates the underlying constitutive behavior. This will enable us to derive equations by taking advantage of a bond
graph that incorporates critical assumptions made in forming the model. In Section 3.4, some techniques are introduced
that can support converting system schematics into bonds. These methods can be especially helpful when it is difficult to
construct a bond graph purely by inspection.

3.4 Algorithmic Conversion of Schematics to Bond Graphs

The systems that we seek to model are usually first represented as schematic diagrams. Mechanical, hydraulic, and electric
circuit diagrams are typically formed using conventions that need to interpreted and converted into mathematical models.
Depending on the type of schematic, this conversion process requires making additional assumptions. Electrical circuits
convey the most the structure of a model of a physical system, and, since a bond graph also contains this information, there
should be an algorithmic method by which schematics can be converted to bond graph form. This algorithm can be stated
in six steps as9 :

1) Choose a positive reference direction in the schematic. This is usually done in terms of voltages and pressures in electrical
and hydraulic systems and terms of velocity and angular velocity for mechanical translation and rotation systems.
2) Indicate the efforts on the electrical and hydraulic schematics and flows on the mechanical schematics.
3) Represent efforts by 0-junctions and flows by 1-junctions.
e1 e2 f1 f2

0 0 1 1

8 Refer also to tables of C, I, and R elements at the end of this chapter.

9 Adapted from Karnopp et al. [38].
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
3.4 Algorithmic Conversion of Schematics to Bond Graphs 107

4) Identify in the schematic all the modeling elements that are to be used and the effort and flow differences needed to
connect these elements. Construct differences as

ea eb
0 1 0 1 0 1
fa fb

eab = ea − eb fab = fa − fb

0 1

and use coupling elements if needed.

5) Connect all elements to the corresponding junctions.
6) Simplify the bond graph using the following graphical rules.

Note also that some special cases will require maintaining the 1- or 0-junction when these indicate a reaction point in
the system; that is,

You can show that there is a change in sign of the effort in the first case (ea = −eb ) and a change in sign of the flow in
the second (fa = −fb ).
Suppose we try the procedure just described on the circuit of Figure 3.23(a). The positive reference is indicated by the +
and − symbols on the graph and the node voltages are indicated as Vo (for reference) and Va through Vd . In Figure 3.23(b),
the node voltages are represented as 0-junctions and the voltage differences (Vab , Vac , Vbc , Vcd ) needed to connect the ele-
ments have been constructed.

(a) (b)

Figure 3.23 (a) Circuit schematic and (b) identified effort and flow junctions.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
108 3 Physical Modeling with Bond Graphs

(a) (b)

Figure 3.24 Conversion of circuit schematic to bond graph. (a) Bond graph with elements and (b) simplified bond graph.

Conductive walls

is Ta
T
V (t) Rs
Ideal
Qs is
gas Tw

(a) (b)

Figure 3.25 (a) Schematic of rotating disk with applied force and (b) bond graph representing key flow junctions and flow differences.

The Se , R, C, and I modeling elements needed have been attached to their appropriate effort junction in Figure 3.24(a).
In Figure 3.24(b), the bond graph has been simplified using the graphical rules described earlier.
Another example of using schematic conversion can be demonstrated with the rolling disk shown in Figure 3.25(a), which
has a force, F, applied about the axle, which has radius r. Assume the disk can roll and slip on the surface. In Figure 3.25(a),
positive velocity of the center of gravity, 𝑣, is indicated to the right and positive angular velocity, 𝜔, is taken as clockwise.
These two flows are indicated on the schematic and represented by the 1-junctions in Figure 3.25(b). The flow differences
needed to connect elements are constructed with 0 and 1-junctions, and transformers are used to represent the relation
between translation and rotation as shown in Figure 3.25(b).
In Figure 3.26(a), all the elements have been connected to the junctions. The force source is represented by a Se ele-
ment, and the distinct translational and rotational inertias are represented by individual I elements. The slip friction is
modeled here by a R, which has a slip velocity, 𝑣s = 𝑣 − R𝜔. The bond graph in Figure 3.26(b) shows the result of applying
simplification rules.
Algorithmic techniques such as those discussed in this section can be useful aids in obtaining bond graphs from schemat-
ics. Individual steps from such approaches can be helpful when initiating part of a system bond graph, and it is not necessary
to follow the full step-by-step procedure presented in these examples. Indeed, the reader should be cautioned that overuse
of these techniques can hamper building insight into how physical systems can be modeled in bond graph form.
The investment made in constructing a bond graph pays forward with insight into how a system can be represented
in a structured form with the physical laws and constitutive relations we’ve discussed in Chapters 1 and 2. One of
the most valuable features in a system bond graph, however, arises in the embedded causal structure. Assignment of
causality on all the bonds reveals how variables are related on each bond and ultimately helps determine the system
state variables. Further, the causal structure can help guide the derivation of the state equations themselves. Section 3.5
introduces the concept of causality within the context of bond graph modeling and elaborates on these concepts
further.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
3.5 Causality 109

(a) (b)

Figure 3.26 (a) Connecting elements and (b) simplified bond graph.

3.5 Causality

Causality is one of the more important concepts in physical system modeling. It is directly related to automatically obtaining
a set of state equations from a bond graph model. Causality is the determination of input and output variables of elements
on a bond graph from model structure and environment. The primary restriction of causality on a bond graph can be
stated as:

either the effort or the flow must be an input for a port but both cannot be input or output simultaneously on the same port.

As illustrated in Figure 3.27, if effort is the input then flow is the output, and if flow is the input then effort is the output.
The allowed causality implies two way communication as discussed in Chapter 1. Consider the implications of violating
this restriction in the examples shown in Figure 3.28.
In the example of Figure 3.28(a), it is impossible to simultaneously specify the force and the velocity of a translational
mass because the mass relates the force and velocity as F = d(m𝑣m )∕dt. If the velocity is really a constant 10 m/s, then the
force must necessarily be zero and cannot equal 1 N. In the circuit example, Figure 3.28(b), the voltage and the current
cannot be both simultaneously specified on a resistor since VR = Ri = (10 Ω)(1 A) ≠ 1 V.
On a single bond between two components A and B, there are only two allowed causalities. This can be indicated on a
port by a causal stroke on the bond. If the effort is an input to A, then the flow must be an output and vice versa. This can be
indicated on a port by a causal stroke on the end of the bond, in which the effort signal is directed as shown in Figure 3.29.
A mnemonic to remember this convention is effort pushes, ⊣, and flow points, ⊢.

Figure 3.27 (a) Allowed and (b) disallowed causality at a port.

(a) (b)

(a) (b)

Figure 3.28 Examples of how physical laws disallow arbitrary assignment of causality. (a) Simultaneously assigning arbitrary force
and velocity on a mass is constrained by Newton’s law. (b) Ohm’s prevents imposing both voltage and current on a resistor.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
110 3 Physical Modeling with Bond Graphs

Causal stroke Figure 3.29 Causal stroke definitions on a power bond and also explained using bilateral signals.

e
A B A B
f
f
A B A B
e

3.5.1 Elemental Causality

Along with the primary causal restriction imposed on a given bond, the bond graph elements have certain allowed and
preferred causalities. Given the allowed and preferred causalities of the elements along with the constitutive relations for
these elements, fundamental properties of the model can be studied and state equations can be generated in a systematic
manner.

3.5.1.1 Source Causality

Consider first the source elements. By element definitions, the flow source (Sf ) specifies the flow variable as a function of
time, so its causality must be flow out. By similar reasoning the effort source (Se ) must be effort out. Figure 3.30 depicts the
only allowed causalities for the source elements.

3.5.1.2 Energy Storage Causality

If the flow variable is prescribed as the input at the port of a C element then effort must necessarily be the output
variable. As shown in Figure 3.31 (lower left quadrant), the effort is obtained by first integrating the flow f to obtain the
displacement q = ∫ fdt and then using the constitutive relation for the C element e = Φs (q). This type of input to a C
defines integral causality due to the integration involved. Integration is a storage operation by which “bits” of flow are
accumulated over time to obtain displacement. But since the energy is a function of displacement for the C, the integration
also mathematically represents the energy storage process in the element. For this reason, a C element with integral
causality is termed an independent energy storage element (ESE). Integral causality also implies that a differential equation
must be solved (by integration) in order to obtain the energy stored in the element. This will be made clear when we obtain
state equations from bond graphs later in this chapter.
The word independent in the expression above suggests that there must also be a dependent ESE. If effort is the input
into a C, then flow must be the output. This flow is found by using the inverse of the constitutive relation to obtain the
displacement, q = Φ−1 ̇ This causality is called derivative causality.
s (e), and then differentiating this result to obtain f = q.
Since we are instantaneously determining q from e, the energy as a function of q is statically (or algebraically) dependent
on the input. Since no storage or state variable is needed in this case to determine the energy residing in the element, a C
with derivative causality is termed a dependent energy storage element.
An I element is the causal dual of the C element. If we interchange e for f and replace q with p, then all the causal
statements for the C above also apply to the I. This is shown in Figure 3.31.
As a particular example of integral and derivative causality consider a translational mass. Integral causality implies that a
force is imposed on this mass which is then integrated over time to find the change of momentum. The kinetic energy can be
found algebraically from the momentum as can the velocity. Note that with a finite force, the momentum is a continuous
function of time which implies that the energy is also a continuous function of time. On the other hand, for derivative
causality, a velocity is imposed on the mass. The momentum can be found algebraically from the velocity and therefore
the kinetic energy is determined by the input. If, for example, the input were prescribed to be a step change in velocity,
this implies a step change in momentum and an inherent discontinuous change in energy would occur. A discontinuous
change in energy implies an infinite power transfer, and since infinite power levels are not physically realizable, inputs of
this type cannot be physically imposed on masses. The same argument can be made for step changes in force on springs,
voltages on capacitors, and currents on inductors.

e(t) e Figure 3.30 Allowed causalities for the sources: effort (a) and flow (b).
Se Sf
f f (t)
(a) (b)
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
3.5 Causality 111

Integral causality Derivative causality

ṗ = e e = p˙
I Constitutive I Inverse constitutive
f f
p =Φ−1I (f)
f =ΦI (p)

e e
edt dp/dt
p

p
f f −1
ΦI (p) Φ I (f )

e e Inverse constitutive
C Constitutive C
q̇ = f f = q˙ −1
e =Φ C(q) q =Φ C (e)

e e
Φ C(q) Φ−1
C (e)
q q

f f
fdt dq/dt

Figure 3.31 Causality for energy storing elements: left column illustrates integral causality and right column illustrates derivative
causality.

Figure 3.32 Causality for dissipative R element. Resistance causality Conductance causality
e T e T
R R
f fs f fs

e = ΦR (f, T ) f = ΦG (e, T )

3.5.1.3 Dissipative Element Causality

Causality for the dissipative R element can be imposed, in general, in either direction. As illustrated in Figure 3.32, flow
as an input with effort as an output will be called resistance causality and effort as an input and flow as an output will be
termed conductance causality. Some restrictions to this “causal indifference” will be discussed in Chapter 4 (in section on
constitutive structure).

3.5.1.4 Coupling Element Causality

Causality on two-port coupling elements is constrained due to the constitutive relations between the efforts and flows
inherent in these elements. For a T element, once the effort on one bond is determined, then the effort on the remaining
bond is specified. Conversely, once the flow on one bond is determined then the flow on the remaining bond is specified. For
a G element, once an effort on a bond is determined, the flow on the other bond is determined. Allowed causality for these
elements is shown in Figure 3.33. Note that power flow direction is not specified (no half arrows), and the power bonds are
shown only with causal strokes. A positive power flow convention may be indicated independent of causal assignment and
may be in either direction.
As an example consider a massless frictionless lever. Once the velocity at one end if the lever, 𝑣1 , is specified, then the
velocity at the other end 𝑣2 , is determined since it is proportional to 𝑣1 . Conversely, the force F1 is proportional to the force
F2 . Note that if 𝑣1 is an input to the lever, F1 must be an output, 𝑣2 must be an output, and F2 must be an input.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
112 3 Physical Modeling with Bond Graphs

n r Figure 3.33 Allowed causality for transformers and gyrators follow from their
e1 ·· e2 e1 ·· e2 respective model relationships.
T G
f1 f2 f1 f2

n r
e1 ·· e2 e1 ·· e2
T G
f1 f2 f1 f2

e1 en e1 en
1 0
f1 fn f1 fn

e2 f2 e2 f2

Only one bond can specify flow Only one bond can specify effort

Figure 3.34 Allowed causality for the junction elements.

3.5.1.5 Junction Element Causality

Causality for the junction elements is also constrained. As shown in Figure 3.34 for the 0-junction, since the effort is com-
mon to all the bonds, the effort can be specified by only one connected bond. For the 1-junction, the flow is specified by
only one connected bond. Again, power flow is not indicated in these descriptions as power flow definitions are assigned
independently from causality.
One example of an 0-junction is a parallel electric circuit. Once the voltage is specified on one element of the circuit, the
voltage of all the elements in parallel is determined.

3.5.2 Causality Assignment Procedure

Causality assignment on a completed bond graph follows a definite hierarchy. At the highest level of assignment are any
source elements since they have a prescribed causal form. Since the source elements model interaction between the envi-
ronment and the system, we can state that causality flows from the environment to the system. Source elements are physical
idealizations, and therefore causal input–output statements are an abstraction imposed on the physical system by the mod-
eler.10 Since the coupling and junction elements, which comprise what is called the junction structure of a model, have
only certain allowable forms, once a source element is assigned, the implications of this assignment can often be extended
into the junction structure.
The next elements in the causal hierarchical level are the energy storing elements. Although the bonds on these elements
can have both causal assignments, the preferred causality is to have integral, independent assignment if possible. The
dissipative elements are indifferent to causal assignment and are assigned last.
The sequential causality assignment procedure (SCAP) can be stated as [39]:
1) Assign the only allowed causality on the source elements and then extend the implications of these assignments through
the junction structure.
2) Assign integral causality on an energy storing element and extend the causality through the junction structure. Repeat
this procedure until all energy storing elements are assigned.
3) Assign a causality on the dissipative elements consistent with the allowed forms for the junction structure.
4) Although causality assignment is typically complete in well-formulated models at this point, any remaining junction
structure elements are consistently assigned in the last step of the procedure.
The physical reason for step 1 in the procedure is obvious from the allowed causality on the sources and the junction
structure. The SCAP is designed to determine the number of independent ESEs in the model. Step 2 assigns integral or
independent causality on as many ESEs as possible.
The assignment procedure is probably best described by considering an example. Shown in Figure 3.35 are the graphical
results of the procedure applied to the bond graph model of the mechanical system from Figure 3.19.

10 Of course, the model itself is already an abstraction.

 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
3.6 State Variables and State Equation Derivation 113

Se Se Se

F F F
L L L
ḣ ·· FA ḣ ·· FA ḣ ·· FA
I ω T 1 I T 1 I ω T 1
ω

0 C 0 C 0 C
ẋs ẋs ẋs
vd vd vd

R R R
Step 0: Unassigned Step 1: Assign source Step 2a: Assign an energy
(begin) – storage element (I)
Se Se Se

F F F
L
L
·· FA ḣ ·· FA L
·· FA
ḣ I T 1 ḣ
I ω T 1 ω I ω T 1

0 C
0 C ẋs 0 C
ẋs ẋs
vd
vd vd

R
R R

Step 2b: Extend Step 2c: Assign an energy Step 2d: Extend
storage element (C)

Figure 3.35 Sequential causality assignment applied to the bond graph of Figure 3.19.

After the bond graph model has been formulated as described in Figure 3.19, step 1 is to assign the allowed causality on
the effort source. This results in an effort input to a 1-junction. It is unclear which of the remaining two bonds of the 1
determines its flow so no further extension of causality assignment through the junction structure is possible at this point
in the procedure. At step 2a, the energy storing I element is assigned integral causality. Choosing the I element rather than
the C element is an arbitrary decision. In step 2b, the causal implications of the assignment in 2a are extended through the
junction structure. Once the flow is determined on the left bond of the T element the flow is determined as flow out of the
bond on the right. This flow determines the flow of the 1-junction, and therefore the lower bond of the 1 must be flow out.
This flow is input to the 0-junction. Further extension can be determined at this time. At step 2c, the remaining energy
storing element, the C element, is assigned integral causality. This assignment determines the effort on the 0-junction and
the extension of this assignment in step 2d completes the causality assignment procedure.

3.6 State Variables and State Equation Derivation

3.6.1 State Variable Types and Selection

A bond graph approach to modeling physical systems emphasizes those features with significant energy storage and power
flow. As we know, the state equations (rate relations) for I and C elements quantify the energy content in a system and are
possible state equations of a system. Causality in a bond graph reveals the following principles:

1) The number of energy states required to define the energy content of a bond graph model is uniquely determined by the
SCAP.
2) The number of states is equal to the number of independent energy storing one-port elements, which is the number of
bonds with integral causality assigned on energy storing elements.
3) The order, n, of a system model is the number of state variables which is equal to the number of independent ESEs.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
114 3 Physical Modeling with Bond Graphs

Table 3.3 State variable pairs.

State variables

Potential Kinetic Name of set

e f Power pairs
q f Lagrange pairs
e p Co-Lagrange pairs
q p Hamilton pairs

Now, even though the number of independent states can be uniquely determined by the causality assignment procedure,
the specific choice of the state variables is not. The potential energy of a system as accounted for by the basic C element can
be instantaneously determined by either an effort e or displacement q variable since there is a static constitutive relation
between these two variables, e = Φs (q). Likewise, the kinetic energy of a system can be found from the basic I elements by
either flow f or momentum p variables, since there is a constitutive relation, f = Φs (p).11 Since these relations are implied,
several pairs of variables could be chosen as a complete set of state variables to define the energy content of a system.
Table 3.3 summarizes some commonly used variables and their traditional designation. Note that each set adopts a variable
to associate with the storage of potential and kinetic energy.
The power pair sets of state variables are used extensively in circuit methods and analysis but will not be emphasized
in this text. The Lagrange and Complementary Lagrange (Co-Lagrange) pairs of variables will be discussed in more detail
in Chapters 8 and 9 when we take up energy methods. For basic energy storing elements, the selection of state variables
can somewhat be a matter of choice, sometimes adopting a discipline-specific preference. In a bond graph approach, the
Hamilton pairs provide a unified treatment and unique determination of the energy content of a system. This means that
definitions of energy and coenergy are uniquely defined through state-determined constitutive relations.12 These advan-
tages make the Hamilton pairs a preferred set for general physical systems modeling (and explains the preference in how
modeling elements are defined in Chapter 2).

3.6.2 State Equation Form

As previously discussed in Chapter 2, a given set of state variables can be expressed as an n dimensional state vector, x. Now,
in the context of bond graph modeling, this state vector will typically be formed only by momentum and displacement type
variables from independent ESEs (see Section 3.5).13 For example, for a system where two I elements and one C element
[ ]
have been identified as independent by causality assignment, the state vector would be: xT = p1 , p2 , q , where p1 , p2 , and q
represent the respective momentum and displacement variables. The right-hand side of the state equations represents the
̇ As previously discussed, the nonlinear form of state equations take the form,
rates of change of the states, that is, x.

ẋ = f(x, u), (3.31)

where f() is the n dimensional rate vector and u is the r dimensional input vector function of time. The linear state-space
form is,

ẋ = Ax + Bu, (3.32)

where A is an n × n matrix and B is an n × r matrix.

In addition to the state equations, there are usually outputs of interest which are expressed as functions of state and
time as,

y = h(x, u, t), (3.33)

11 This interchange between variables is not possible for what we will call composite energy storing elements, which will be discussed in
Chapter 4.
12 And as mentioned for composite elements in which the inverse relation Φ−1 s () may not be unique, there is advantage.
13 In Chapter 4, we will discuss how we may also need information states as part of the state vector.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
3.6 State Variables and State Equation Derivation 115

where y is the m dimensional output vector and t is the time. The output equation in linear form can be expressed as,
y = Cx + Du, (3.34)
where C is an m × n matrix and D is an m × r matrix.

3.6.3 Model Analysis with Causality

Before actually writing any equations, causality can also be used to evaluate the bond graph model. In addition to revealing
which ESEs are independent and thus the source of a state variable, causality can be used to: (i) determine if there are
any inconsistencies in the bond graph, especially those that can prevent the derivation of consistent state equations, and
(ii) provide some insight into how state equation derivation should proceed.

3.6.3.1 Dependent Causality

A first examination of a system and construction of a model readily reveals a need for representing energy storage and thus
the identification of potential state variables. The interconnections, or structure, in a bond graph may cause interactions
that force dependent causality in an ESE. In order to see how this can arise, consider the model of a hydromechanical system
shown in Figure 3.36(a). In this model, a pump is represented as an ideal pressure source, the pipe conveys incompressible
fluid that stores kinetic energy and has fluid losses. The fluid couples to a piston ram, and this piston–fluid interaction is
modeled using a transformer (see Figure 3.12(c)). The piston ram interacts with a load mass (representing combined piston
and load inertia). Attached to the load mass is a spring with a vibration shaker modeled as an ideal velocity source, 𝑣(t).
The complete bond graph model is shown in Figure 3.36(b).

Pump Pipe
pressure,

Piston ram

R (a)

P (t)
Se 1 T 1 0 Sf
v(t)

Dependent
(fluid inertia) → I I C
(b)

R R ← leakage

P (t)
Se 1 0 T 1 0 Sf
v(t)

Independent
(fluid inertia) →I I C
(c)

Figure 3.36 Hydromechanical system illustrating how addition of leakage influences causality. (a) System schematic, (b) complete
bond graph with no leakage and dependent fluid inertia, and (c) model with leakage resulting in independent fluid inertia.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
116 3 Physical Modeling with Bond Graphs

When we apply causality in a sequential fashion, it is discovered that it is impossible to have independent causality on
both of the inertia elements. This implies that the state of one of the I elements is dependent on that of the other, so only one
state is necessary to quantify their contribution to the total kinetic energy in a system. In Figure 3.36(b), the fluid inertia
element is indicated as dependent, but this is an arbitrary decision. A consistent causality assignment could be completed
by taking the translational inertia as dependent and the fluid inertia independent. The reason for the dependency between
the two states can readily be seen by examining the schematic of the physical system. If the fluid is incompressible and
there is no compliance in the ram’s shaft nor leakage of flow then the volume flow rate of the fluid and the velocity of the
load mass are not independent.14
The dependence between system states that can arise in a model are often a result of modeling assumptions. If it is an
observed experimental fact that the fluid flow and the load mass velocity can be separately determined, then this can be seen
as an error in the model of this system example. One plausible explanation in this case is the need to include fluid leakage
between the piston and the fluid chamber enclosing the piston (e.g., see Example 3.3). This is modeled by the addition of
a 0-junction to account for the common cylinder chamber pressure and the summation of the pipe inflow, volume flow
rate of the piston, and the flow rate due to leakage. The leakage can be modeled by an R element since it is considered
an energy loss to the system. A model that includes leakage is shown in Figure 3.36(c). Re-assigning causality shows that
all the energy storing elements can now be assigned integral causality. This system will be revisited later in this chapter to
illustrate how derivative causality impacts equation derivation.
Note that it should not be construed from this discussion that models with dependent causality are improper. There are
many cases, especially in the design of linkages, where dependent causality is the dominant causality assignment. Only
if there is evidence to indicate that dependent causality on an element is incorrect, as there was in the last example (e.g.,
experimental measurements), should an adjustment be made to the model (e.g., by adding other elements the break the
dependence). While adjustments can be made to a model to avoid dependent elements, be mindful that: (i) the complexity
of the model will be increased, and (ii) some systems might be more accurately modeled by retaining the dependent nature
in some ESEs.

3.6.3.2 Detecting Improper Models

Constructed bond graphs represent an abstraction of a system, and as is often repeated: all models are approximations.
Inaccurate models can still provide answers and bad predictions of reality, but these are not the same as improper models
and bond graph constructions. The latter cannot provide solvable equations. Causality can be used to identify physically
improper models. If, for example, we had a flow pump in the above example system but had also considered a velocity source
attached to the load mass, the bond graph model of the system with applied causality would be given as in Figure 3.37.
Assignment of causality results in a causal bind, indicating an improper model. The bond graph shows that all of the
flow sources cannot be assigned their only allowable causality, because of a causal bind or conflict as shown. This conflict
arises due to the dependency between the fluid flowrate and the load mass velocity (i.e., they cannot be assigned flows
independently) in the model as constructed. The model indicates that the two flow sources can only be specified if there
is leakage or compressibility in the model. It should be pointed out that this does not mean such a system could not be

R Sf Velocity source

Causal bind

Sf 1 T 1 0 Sf
Flow pump

I I C

Figure 3.37 Improper model identified through causal bind.

14 It is helpful to begin recognizing this type of bond graph structure. Whenever each bond of an ideal transformer has 1-junctions with
attached I elements, the flows will be related and thus one of these is dependent.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
3.6 State Variables and State Equation Derivation 117

physically realized, only that the model as formulated is not correct. Additional insights from causality analysis will be
discussed in this and Chapter 4. Next the important role causality plays in deriving state equations is introduced.

3.6.4 Deriving State Equations from a Bond Graph

State equations can be derived directly from a causally complete bond graph, which means every power bond must be
assigned causality. Section 3.6.3 addressed how causality helps identify and address issues that influence how or whether
system equations can be derived. For example, a causal conflict indicates the model must be corrected. Other such cases
will be discussed in Section 3.6.6. This section examines system models for which causality can be completely assigned
on a bond graph during the SCAP. Consider the elementary example of an electrical capacitor, C, joined with a resistive
load, R, as in Figure 3.38(a). The capacitor could have a specified initial charge, qC (0). As drawn, one could take the bond
graph with a 0-junction as well. For the system as shown, causality assignment has been completed to show there is one
[ ]
independent ESE, the capacitor, so system order is n = 1 and the state vector is x = qC .
To derive the state equation, we make use of the information that is included in the causally complete bond graph. First,
note that the rate of change of the charge (the state equation) can be expressed: q̇ C = iC . Since the C and R are on a 1-junction,
then one could see that iC = iR ; however, it is important going forward to use not only the structure of the bond graph but
also the causality assigned to the bonds. The ability to use this information may seem trivial in this example but for more
complex systems the ability to identify causal paths between variables on a bond graph can significantly expedite equation
derivation. In this case, we already know that iC and iR have the same current. However, causality assignment indicates that
the current at the 1-junction is causally-determined by the current on the R element. Thus, the causally-correct relation
is: iC = iR = VR ∕R. We must further recognize that at the 1-junction, VR = −VC = qC ∕C. It is good practice to begin writing
these relations left-to-right to express causal dependence. Thus, to complete the state equation, we have,
q̇ C = iC = iR = VR ∕R = −VC ∕R = −qC ∕(RC). (3.35)
The main point here, of course, is to demonstrate steps that will be followed moving forward and not that this circuit
equation could not be derived without the use of causality.
It can be useful to address some questions that often arise with the use of causality to identify system states. It can seem
arbitrary. In the SCAP, preference is given to assigning integral causality to ESEs when permitted by the model structure.
What if the causality assignments in Figure 3.38(b) were swapped? By our convention qC would not be a state, and it would
suggest that VR determines VC , which would imply that VC could be changed regardless of its initial state (qC (0)). This is not
a valid physics perspective. While it might be physically possible to do so, such a scenario would not be correctly represented
by the model shown in Figure 3.38(a).
Causality provides a systematic way to build models that can be solved in terms of state variables and inputs of a system.
It is a procedure uniquely formulated using bond graphs by Prof. Henry Paynter [1]. When asked what motivated causality,
Paynter pointed to the problems he was trying to solve during the 1950s and 1960s, particularly using analog computing.
If you want an op-amp integrating circuit to produce the integral of a state variable, you want to feed it signals that are
all dependent on the states and inputs it depends on. This paraphrases Paynter, and while the idea may have genesis in
“computability” of states causality has gone far beyond this one usage. More on the genesis of bond graphs from Paynter’s
perspective can also be found in [40].

3.6.5 State Equation Derivation in Explicit Form

When bond graphs with complete causality reveal that all ESEs have integral causality, the states of all the ESEs are states
of the system. For these cases, the resulting state equations can be expressed explicitly in terms of the other states. For the

initial state
(a) (b)

Figure 3.38 A simple RC circuit (a) schematic and (b) causal bond graph.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
118 3 Physical Modeling with Bond Graphs

Indep ESE

Indep ESE

(a) (b) (c)

Figure 3.39 Basic mass–spring–damper: (a) schematic, (b) free-body-diagram, and (c) causal bond graph.

most part, the state equations for these systems can be derived in a systematic manner from the bond graph, especially
if causality is used effectively to guide the algebraic manipulation. This section reviews some examples that demonstrate
these concepts and methods.

3.6.5.1 Basic Mass–Spring–Samper

As shown in Figure 3.39, causality on the bond graph for a classic mass–spring–damper identifies two independent state
variables: p and xk . The causality on the bond graph shows that the velocity at the 1-junction, which is common for all the
elements, is determined by the mass velocity, 𝑣m .
To write equations directly from the bond graph, focus on the rate relations from the independent ESEs so the two
equations are: ṗ = F(t) − Fk − Fb and ẋ k = 𝑣k = 𝑣m . Thus, we can write a final form for the two equations as:
𝑣̇ m = = −b𝑣m ∕m − kxk ∕m + F(t)∕m, (3.36)

ẋ k =𝑣m , (3.37)
where we switched to 𝑣m as the state for the mass, as may be preferred in some cases. An experienced bond graph user
can effectively read equations directly from the causal bond graph for simpler systems as demonstrated here. As systems
become more complicated, more systematic procedures will be demonstrated.

3.6.5.2 LRC Circuit

A series LRC circuit with a voltage input is shown in Figure 3.40(a). For simplicity, assume that all the passive elements
(i.e., the LRC elements) are linear. The following describes how the bond graph is used to derive state equations.
The bond graph for the LRC circuit is shown in Figure 3.40(b). Independent states are identified as the states of ESEs
with integral causality. The causality on the LRC system bond graph shows n = 2 independent ESEs, thus the state vector
[ ]T
is x = 𝜆 q .
The first state equation is the rate of change of flux linked in the inductor, 𝜆̇ L , which is the effort imposed on the I element
shown in Figure 3.40(b). This effort is determined in a causal sense at the end of the bond attached to the I opposite from
the stroke or, more simply stated, 𝜆̇ L is determined by the efforts Vs , VR , and VC imposed on the 1-junction. Since we know
that efforts sum (with the appropriate sign as determined by the power arrows) at a 1, then
𝜆̇ L = Vs (t) − VR − VC ,

I:L

λ̇L iL
Vs(t) VR
Se 1 R: R
iR
VC q̇C

C: C
(a) (b)

Figure 3.40 L–R–C circuit. (a) Series circuit and (b) bond graph.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
3.6 State Variables and State Equation Derivation 119

where VR = RiR = RiL = R𝜆∕L and VC = q∕C, thus,

𝜆̇ L = −R𝜆∕L − q∕C + Vs (t).

A state equation is not complete until it is determined in terms of the independent state variables and/or system inputs.
The charge rate accumulated in the capacitor (C element), q̇ C , is causally indicated as the flow at the 1-junction, which
in turn is causally determined by iL from the inductor bond. Thus we have,

q̇ C = iL = 𝜆∕L.

Now we can be put these equations in the vector–matrix form,

[ ] [ ]
−R∕L −1∕C 1
ẋ = x+ u, (3.38)
1∕L 0 0
where x1 = 𝜆L , x2 = qC , and u = Vs (t). If we take as the outputs the current in the circuit, y1 = iL , and the capacitor voltage,
y2 = VC , we can express the outputs in the vector–matrix form as,
[ ] [ ]
i 1∕L 0
y= L = x. (3.39)
VC 0 1∕C
The output equation y is constructed based on which state-determined system variables are desired to be known for a given
application.

3.6.5.3 Lever–Spring–Damper System

Recall the bond graph in Figure 3.19 for which causality was completed in Figure 3.35. Causality assignment showed that
the states are the displacement, xs , of the C element and the momentum, h, of the I element. These states are sufficient
to describe the energy content in this system. Since causality was completed explicitly, we expect the form for our state
equations will be ẋ s = f1 (xs , h, t), and ḣ = f2 (xs , h, t).
̇ the rate of the angular momentum of the bar,
Starting with the I element, the causal stroke indicates that the effort or h,
is the input into this element. Looking next to the T element, we find that Fa is the input effort or force to the right side of
the bar. Using the constitutive law for the T element, we have

ḣ = LFA ,

where L is the distance from pivot to spring connection, and we have assumed small angular displacement of the bar.
Proceeding in a sequential fashion to find FA , we note that FA is determined at the 1-junction by F and Fs , and the sign
convention specified by the half arrows such that,

FA = F − Fs .

This equation is the summation of forces at the bar’s tip. Since F is a known input function of time, as given by the effort
source, the only unknown in the above equation is Fs , the spring force. From the causality on the graph, we note that Fs
is determined by the effort of the spring element. Assuming a linear constitutive law for the spring, the force Fs can be
expressed as

Fs = Kxs .

The above equations can then be used to write a single rate equation for the momentum state as,

ḣ = L(F − Kxs ),

as one of the two-state equations. A second-state equation is the rate equation for the spring displacement xs . The displace-
ment rate ẋ s is causally determined from the 0 element using the relation,

ẋ s = 𝑣 − 𝑣d ,

where 𝑣 is the velocity of the bar’s tip and 𝑣d is the damper velocity. Both 𝑣 and 𝑣d are presently unknown in our procedure.
The velocity 𝑣d is determined by the R element through its constitutive law

𝑣d = Φl (Fd ),
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
120 3 Physical Modeling with Bond Graphs

which has been assumed to be a general nonlinear function. The damper force Fd is determined by the spring force at the
0-junction as
Fd = Fs .
Returning now to the variable 𝑣, we see that it is determined by the T from its input 𝜔, which is the angular velocity of the
bar, as
𝑣 = L𝜔,
and finally 𝜔 is determined by the I element from its constitutive relation
𝜔 = h∕J,
where a linear constitutive relation has been assumed. Using the above equations interpreted as causal statements with the
right-hand side of the equations as input and the left-hand side as output, the rate equation for the displacement can be
expressed as
ẋ s = Lh∕J − Φl (Kxs ).
This is the second-state equation. In summary,
ḣ = −KLxs + L ⋅ F, (3.40)

ẋ s = Lh∕J − Φl (Kxs ), (3.41)

where F is an input and Φl () is a nonlinear function.

Example 3.5 Storm sewer system

Develop a bond graph for the storm sewer system discussed in Chapter 2 and shown in Figure 2.41 and then identify the
independent states and derive state equations.
Solution
A bond graph for the storm sewer system is shown in Figure 3.41. Causality assignment indicates three independent states:
[ ]
–VA , Γ, –VB . The system has inputs defined by the rain flow, Qrain , and the flow into the head of the sewer pipe, QO .
Begin with the equation for state V – A , which has a rate of change defined by two flows,
–̇ A = Qrain (t) − QO ,
V
where we can see that QO is determined causally by the flow at the adjoining 1-junction, which in turn is specified by
QI = Γ∕I, which is state determined.
Next, the state equation for the Γ is determined by the sum of pressures at the 1-junction,
Γ̇ = PO − PR − PL ,
where each of these pressures relates to states as follows: PO = PA = – VB ∕CB , and PR = K|QR |QR =
V ∕CA , PL = PB = –
K|Γ∕I|(Γ∕I).
Lastly, the state V
– B is determined by the sum of flows,
–̇ B = QL − QP (t),
V

C I C Figure 3.41 Bond graph of storm sewer system shown in Figure 2.41.

–˙ A
PA V Γ̇I QI –˙ B
PB V

PO PL
Sf 0 1 0 Sf
Qrain QO QL QP

PR QR

R
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
3.6 State Variables and State Equation Derivation 121

where QL = QI = Γ∕I. The equations can now be put in state-space form,

⎡ 0 −1∕I 0⎤ ⎡1 0 ⎤
ẋ = ⎢ 1∕CA −K∕I −1∕CB ⎥ x + ⎢ 0 0 ⎥ u, (3.42)
⎢ ⎥ ⎢ ⎥
⎣ 0 1∕I 0⎦ ⎣ 0 −1 ⎦
where x1 = –VA , x2 = Γ, x3 = –VB , and u has the elements u1 = Qrain (t), u2 = QP (t).

3.6.6 Alternative Formulations

For systems where causality assignment reveals that some ESEs are dependent, the states of those elements are not states
of the system. There are also some systems where there exists one or more algebraic loops, which can require resolving
additional algebraic relations. Both of these situations can lead to a more complicated derivation of the state equations,
but using causality not only helps identify when these cases arise but also provides guidance in the process. This section
reviews how these situations can arise and some ways for dealing with the equation derivation.

3.6.6.1 Systems with Derivative Causality

Systems with fully-independent causality are those that result in all ESEs having integral causality. Systems with this type
of causality generally present the most straightforward cases in deriving complete state equations since: (i) each indepen-
dent ESE identifies a state rate relation, ẋ = u, that represents one of the system state equations, and (ii) causality on the
bond graph “fits”; that is, there are clear causal paths between any inputs and states to each state equation such that the
state equations can be systematically derived in a reliable fashion. There are many systems, however, in which it is not as
straightforward to derive the state equations in the explicit form of equation 3.31. This can occur, for instance, when there
are dependent ESEs in the system (i.e., those where we are forced to assign derivative causality). Dependent energy storage
arises when an ESE is configured within a system in such a way that the storage energy is determined by the causal input;
that is, a C element has the effort specified and a I element has the flow specified. These causal conditions are illustrated in
Figure 3.42.
Causality reveals that the state of a dependent element is not a state variable of the system. While dependent elements
will not contribute to the states of a system, they will influence the system dynamics of the system by virtue of the derivative
of that element’s state. For example, a spring with stiffness k that is in derivative causality with imposed force, F, responds
̇
with velocity, 𝑣k = ẋ k = F∕k. Similarly, a mass prescribed to move by a velocity source, 𝑣(t) will react with inertial force,
F = ṗ = m𝑣(t).
̇ These causal insights can be useful when building a model. Further, there are consequences on the process
of deriving state equations when there is at least one derivative causality in a system. In such cases, additional algebraic
manipulation may be required to formulate state equations into the explicit form of equations 3.31.
Recall the hydromechanical system from Figure 3.36 and shown in Figure 3.36(b) as a model with no leakage around the
piston (represented here by the transformer with modulus A). That bond graph with variable labels is shown in Figure 3.43.
In this case, integral causality was assigned to the I element representing the mass, m, resulting in a derivative causality
assignment on the I element representing the fluid inertia, I. The result is that the system has two independent states (p, xk )
and ΓI is not a state variable, so there is no state equation to derive for ΓI .
Consider linear constitutive relations for all the elements in this equation derivation. The parameters of the constitutive
relations are load mass m, spring stiffness k, flow resistance R, flow inertia I, and piston area A.
First, note that by causality, FO = Fk = kxk , so the state equation for p is,

ṗ = Fp − FO = APp − Fk = APp − kxk , (3.43)

where Pp = Ps (t) − PR − PI . Here note that, PR = RQR , with QR = Qp = A𝑣p , and the piston velocity 𝑣p = 𝑣m = p∕m. Thus,

PR = RAp∕m.

Figure 3.42 Dependent ESEs have derivative causality and have an energy state e e = ṗ
that depends on the system, and q and p are not states of the system. C I
(a) Dependent C-element and (b) dependent I-element. f = q̇ f

(a) (b)
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
122 3 Physical Modeling with Bond Graphs

R:R

PR QR
Ps (t) Pp A Fp FO
Se 1 T̈ 1 0 Sf
Qp vp vO v(t)

PI = Γ̇I QI ṗ vm Fk ẋk
Derivative
causality → I : I I:m C

Figure 3.43 Hydromechanical model with dependent storage due to derivative causality, so ΓI is not a system state.

The pressure PI is determined by first finding ΓI in terms of p, thus, ΓI = IQI = IQp = IA𝑣p = IAp∕m. Then,

PI = Γ̇ I = IAp∕m.
̇

Now ṗ from equation 3.43 above is,

[ ]
ṗ = APp − kxk = A Ps (t) − RAp∕m − IAp∕m
̇ − kxk ,

̇
which needs to be simplified by solving for p,

(1 + IA2 ∕m)ṗ = −RA2 p∕m − kxk + APs (t),

or,
[ ]
ṗ = −RA2 p∕m − kxk + APs (t) ∕(1 + IA2 ∕m).

Now ṗ is expressed explicitly as a function of system states and time, as in equation 3.31.
Lastly, the equation for state xk is found from the sum of flows at the 0-junction,

ẋ k = 𝑣O − 𝑣(t) = 𝑣m − 𝑣(t) = p∕m − 𝑣(t),

giving us our second-state equation.

The simplification of system equations when there exists a derivative causality will typically require additional manipu-
lation to achieve a desired explicit form. In some cases, it may not be possible to resolve in such a form and so it is necessary
to formulate the system equations in an implicit set of equations,
̇ x, u) = 𝟎,
g(x, (3.44)

where g() represents a vector of coupled equations. These equations might then only be solved numerically. Thus, numerical
solution of the system state equations requires iteration as well as integration. These cases can take more time to solve than
when the equations take explicit forms, which are amenable to the simulation methods discussed in Chapter 2.
It is sometimes possible to make modeling decisions to avoid derivative causality, allowing equations to be directly derived
in explicit form. Adding model elements can change the model structure, and thus alter causality assignments. This can, of
course, increase the system complexity. This was illustrated in Figure 3.36 by adding leakage in the hydraulic drive, which
resulted in a dependent fluid inertia becoming independent. An alternative approach for dealing with derivative causality
is to employ energy methods for combining elements (which will be discussed more in Chapter 8). In some cases, it makes
sense to “eliminate” the derivative causality, while in others, it may be best to resolve the derivative causality in the equation
derivation.

3.6.6.2 Systems with Algebraic Loops

Deriving state equations in explicit form can also be complicated when systems have an algebraic loop. An algebraic loop
arises when a system variable (effort or flow) is a function of itself. Functionally, this concept can be illustrated by a block
diagram as shown in Figure 3.44. In the diagram shown, the only way to determine the variable, y, is by solving the implicit
relation, y − f (u, y) = 0. If the function f () is linear, then the algebraic loop can be readily resolved. For example, if f = a ⋅
u + b ⋅ y, where a and b are constant parameters, then y − au + by = 0 and thus, y = au∕(1 + b).
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
3.6 State Variables and State Equation Derivation 123

Figure 3.44 An algebraic loop arises when a system variable in a system model is a
function of itself.

I:m R

Check valve ṗ vm PO QO
F (t) Fp A Pp
Se 1 0 Arbitrary
Outlet vp T̈
Qp
Piston PI QI
area, A

Sump R

(a) (b)

Figure 3.45 (a) Positive displacement pump and (b) associated bond graph with algebraic loop.

Algebraic loops are not unusual. When they arise in a system model, symbolic manipulation of the equations can become
more involved and, in some cases, these state equations need to be expressed in the differential-algebraic form as,
ẋ = f(x, z, u), (3.45)
z = g(x, z, u), (3.46)
in which the algebraic equation 3.46 needs to be solved for z in terms of x and u in order to eliminate z from 3.45 to obtain
a completely explicit form.
Fortunately, it is possible to detect the presence of algebraic loops in a bond graph model during the successive causality
assignment procedure. Either the model can be evaluated and modified or proper steps can be taken to derive and/or solve
the state equations. An algebraic loop is identified during SCAP when an “arbitrary” assignment is required to complete
causality on a bond graph. This concept is demonstrated in the following example.

Example 3.6 Positive displacement pump

A common positive-displacement pumping mechanism is shown in Figure 3.45(a). The force F(t) causes back and forth
motion of the piston, causing fluid to be drawn into the piston chamber and then forced from the outlet. Model each check
valve as a nonlinear hydraulic resistive element with a constitutive relation of the form,
P = Φl (Q),
where P is the pressure drop across the valve and Q is flow through the valve. Assume the fluid is incompressible and the
pipes have negligible fluid inertia. Assume that the sump and external pressure are about the same.
Solution
A simplified bond graph is shown in Figure 3.45(b).15
Sequential causality assignment proceeds as follows: (i) assign causality due to source, F(t), (ii) assign integral causality
on I ∶ m element, (iii) 𝑣p is causally determined into T, and (iv) Qp is thus specified into the 0-junction. At this point, there
is an arbitrary choice required to complete the assignment of causality on the last two bonds at the 0-junction. Arbitrary
assignment refers to the fact that there is no required or preferred causality, such as for sources and ESEs, respectively. The
two R elements representing the valves have no preferred causality. Arbitrarily assign the upper R (exit valve) as specifying
effort onto the 0-junction as shown. The final bond thus has a specified causality as indicated. The arbitrary causality
indicates there is an algebraic loop to resolve.
Begin by expressing the state equation,
ṗ = F(t) − Fp = F(t) − APp = F(t) − APO .

15 Sump and ambient pressures could be explicitly shown, but assuming they are essentially equal allows the simplified form shown.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
124 3 Physical Modeling with Bond Graphs

Here, PO = Φl (QO ), but to find QO we need to see that at the 0-junction, QO = Qp − QI . Continue with Qp = A𝑣p = (A∕m)p
and QI = Φ−1l
(PI ), where Φ−1
l
() is the inverse of the function Φl (), and then note PI = PO . This last relation is the key: we
began the process of finding PO and ended up requiring PO itself. This is the essence of an algebraic loop! Making all the
indicated substitutions allows us to find an algebraic expression for finding PO ,
( )
PO = Φl (QO ) = Φl (Qp − QI ) = Φl Ap∕m − Φ−1 l
(PO ) , (3.47)
which needs to be solved to find PO . The state equation needs to expressed in a differential-algebraic form,
ṗ = F − APO ,
( )
PO = Φl Ap∕m − Φ−1
l
(PO ) ,
which is of the form shown in equations 3.45 and 3.46. In general, equations of the form 3.47 cannot be solved in the closed
form but require numerical iteration.16 Although not true in general, there are certain constitutive laws Φl () for which 3.47
can, in fact, be solved analytically. These include the important special case of a linear constitutive law,
P = Φl (Q) = RQ.
In this case, 3.47 can now be written,
PO = R(Ap∕m − PO ∕R) = RAp∕2m,
so the state equation can be expressed in the explicit form,
ṗ = F(t) − RA2 p∕2m.

3.6.7 Summary
As we have seen in this and Sections 3.6.5 and 3.6.6, causality can be used to aid in the derivation of state equations and to
perform preliminary analysis of our model even before any equations are written. Before closing this chapter, section 3.7
reviews how to incorporate thermal effects into bond graph models, extending the concepts introduced in Chapter 2.

3.7 Thermal Effects in Bond Graph Models

It is often necessary to decide whether thermal considerations must be taken into account when modeling a system. The
environment or the system’s behavior or operation may not cause significant changes in temperature, for example, which
can prompt an isothermal assumption. Such a decision would forgo thermodynamics. Alternatively, the methods of this
section offer a preliminary way to initiate accounting for thermodynamic effects in bond graph models of systems. The
emphasis here is on how to include heat sources, heat transfer, and storage of thermal internal energy for closed systems.
Closed thermodynamic systems are those for which mass (or matter) does not change (i.e., does not cross the system bound-
ary). Thermodynamic system modeling will also be taken up in Chapter 9, including modeling of open thermodynamic
systems effects.

3.7.1 Thermal Bond Graph Elements

In Section 3.2.3, it was shown how the power dissipated in R elements is transformed into heat flow via a thermal port,
̇ as power conjugate variables (see Table 3.1).
which is equal to the product of temperature, T, and entropy flow rate, fs = S,
This two-port form of the R is adopted, when thermal effects are to be considered, and thus provides a form of coupling
to the thermal energy domain from any other energy domain. As mentioned in Chapter 1, power flow within the thermal
domain, thermal = T ⋅ fs , is represented by a dashed power bond, distinguishing its unique nature compared to power flow
in other energy domains. Extending bond graphs in this energetically-consistent manner into the thermal domain enables
us to model heat flows through physical elements as well as how thermal internal energy is distributed and stored.

16 A discussion of numerical techniques that might be used can be found in [27].

 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
3.7 Thermal Effects in Bond Graph Models 125

T1 Heat T2 T1 T2 T1 T2
R Se R Se
fs1 transfer fs2 fs1 fs2 fs1 fs2

fs1 = fs2
(a) (b) (c)

Figure 3.46 (a) Two-port word bond graph for heat transfer, (b) two-port heat transfer R, and (c) causally imposed temperatures on
two-port R.

3.7.1.1 Heat Transfer Modes

One of the primary dissipative effects in engineering thermal systems is transfer of heat from place to place. The three modes
of heat transfer commonly used for basic heat transfer models were reviewed in Section 2.6.2. All heat transfer modes17
can be represented by a two-port R element as shown in word bond graph form in Figure 3.46(a), where the heat flowrate
common to each port requires T1 fs1 = T2 fs2 and thus fs1 ≠ fs2 . The bond graph element is a two-port R as in Figure 3.46(b).
The specific mode of heat transfer will dictate the specific constitutive relation for the R element.
The basic example shown in Figure 3.46(c) has causally imposed temperature boundary conditions for which the two
entropy flow rates, f1 and f2 , each depend on both temperatures, that is,
fs1 = Φs1 (T1 , T2 ). (3.48)
fs2 = Φs2 (T1 , T2 ). (3.49)
For example, for conduction transfer, the basic macroscopic Fourier heat conduction law is,
kA
Q̇ 12 = T1 fs1 = (T − T2 ), (3.50)
L 1
where Q̇ 12 is the heat transfer rate from place 1 to place 2, k is the thermal conductivity, A is the transfer area, L is the
transfer distance, and T1,2 are the temperature at places 1 and 2. The R element in Figure 3.46(b) is a power conserving
element such that,
Q̇ 12 = T1 fs1 = T2 fs2 , (3.51)
where fs1 and fs2 are entropy flow rates, that is, fsi = Ṡ i at each place. Using 3.50 and 3.51, the following constitutive relations
for the conduction R element result:
kA T1 − T2
Ṡ 1 = fs1 = . (3.52)
L T1
kA T1 − T2
Ṡ 2 = fs2 = .
L T2
Note that net flow of entropy out of this element is positive for positive k as required by the second law and is an additional
constraint, beyond power constraints, on possible constitutive relations for thermal R elements.
For simple convection, rather than conduction, we have constitutive relations for the entropy flow rates at the surface
and the fluid, respectively, as:
Ts − Tf
Ṡ s = fss = hA ,
Ts
Ts − Tf
Ṡ f = fsf = hA ,
Tf
where h is (temperature and fluid flow sensitive) convection heat transfer coefficient, A is the surface area, Ts is the surface
temperature, and Tf is the fluid temperature.
Finally, for basic radiation, we have,
T14 − T24
Ṡ 1 = fs1 = 𝜎A1 F12 ,
T1
T14 − T24
Ṡ 2 = fs2 = 𝜎A2 F21 ,
T2

17 For review, see Incropera and DeWitt [18] or similar textbooks.

 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
126 3 Physical Modeling with Bond Graphs

where 𝜎 is the Stefan–Boltzmann constant (𝜎 = 5.67 × 10−8 W/m2 K4 ), A1 and A2 are surface areas, and F12 and F21 are
radiation shape factors. A shape factor Fij is defined to be the fraction of radiation leaving surface Ai that reaches Aj [41].
Shape factors satisfy a reciprocity relation,
A1 F12 = A2 F21 ,
which ensures power conservation, T1 fs1 = T2 fs2 . In a bond graph model, we can use the same two-port R from
Figure 3.46(b) to represent any of the three basic modes of heat transfer. This approach is illustrated through the examples
in Section 3.7.3.

3.7.1.2 Thermal Energy Storage

Thermal internal energy storage due to changes only in entropy was introduced in Chapter 2. For these cases, a single-port
C for which T = T(S) is the constitutive relation, with entropy as state S, is sufficient for models focused on the thermal
domain. We now seek to account for the influence of changes due to volume as well, as inferred by the Gibbs internal energy
relation [20],
dU = TdS − Pd–V . (3.53)
This relation suggests that a two-port C element is needed to account for changes in the internal energy at a thermal port
T − Ṡ and a mechanical port P − –V̇ . This requires that the thermal energy storing C is extended to include a fluid-mechanical
port with power flow P − –V̇ as shown in Figure 3.47. The negative sign used on P is consistent with certain types of processes
to be modeled, and as implied by equation 3.53. This also justifies the use of positive power flow convention into the C at
both ports in Figure 3.47 since the stored internal energy changes as, U̇ = T Ṡ − P– V̇ .
The two-port thermal C is a natural extension of the generalized potential energy storing
element discussed in Section 3.2.1. The more general use of multiport elements will be dis- −P
T
cussed in Chapter 8, but this preliminary and limited discussion will enable thermal effects C
to be incorporated into system models. Only closed thermodynamic systems for which the Ṡ –V˙
amount of mass (or moles of matter) does not change will be considered (until Chapter 9). Figure 3.47 Two-port
Further, introducing ways to model a generic solid, as introduced in Chapter 2, or an ideal thermodynamic C.
gas provides sufficient insight into how a two-port C can be used for many practical system
modeling purposes.
Different causal conditions on the two-port C are shown in Figures 3.48(a)–(d). For example, up to this point, the thermal
C has been used to model a generic solid with no thermal expansion. This is an isometric (constant volume) assumption,
which can be more fully conveyed with the causal two-port in Figure 3.48(a). Including the mechanical port makes it clear
that while –V̇ = 0 is imposed, the pressure −P acts on the system enforcing the isometric constraint. In order to determine
this pressure, we need an equation of state that generally depends on both S and –V (the latter a constant in this case). Since
this discussion is restricted to closed systems, it is common to adopt specific values of entropy s = S∕m and volume 𝑣 –=– V ∕m
variables, here in terms of mass, m.
The four cases in Figure 3.48 include two cases, (b) and (c), where the bond indicates an imposed effort onto the two-port
C. This indicates a derivative causality on that bond so the associated state is not a state of the system. For example, for the
isothermal case (b), the entropy would not be a state and T = To would be a known input. The two states on the two-port C
are independent of one another and causality assignments are made on each port without consideration of the conditions

T −P To −P
C Sf Se C
Ṡ –V˙ = 0 Ṡ V˙
–
(a) (b)

T −Po T −P
C Se Sf C
Ṡ –˙
V Ṡ = 0 V˙
–

(c) (d)

Figure 3.48 Four causal conditions on a two-port C element related to associated thermodynamic processes: (a) isometric/isochoric
(–V = constant); (b) isothermal (T = constant); (c) isobaric (−P = constant), and (d) isentropic (S = constant).
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
3.7 Thermal Effects in Bond Graph Models 127

(a) (b) (c)

Figure 3.49 (a) Uniform bar expanding when unconstrained with temperature raised to T from To , with 𝛼 the (linear) coefficient of
thermal expansion. (b) Bar from (a) constrained at length Lo with induced thermal stresses, and E = Young’s modulus. (c) Bond graph
for the isometric case (Lo = constant).

on the other. This concept will be demonstrated in examples to follow as well as in Chapter 8 when multiports are taken
up across all energy domains.

3.7.1.3 Elementary Thermoelastic Effects

A simple rod can be used to illustrate how thermal effects couple into the mechanical translational domain. The rod in
Figure 3.49(a) has uniform area, A, initial length, Lo , and will expand in length by 𝛿 = Lo 𝛼(T − To ) when the tempera-
ture changes to T, for 𝛼 the coefficient of thermal expansion [42, 43]. The case where the rod is inserted between two
rigid walls as shown in Figure 3.49(b) enforces the isometric case. From the causality, there is one independent state
–V which is constant when T is a specified input. As such, any dynamic variation in this system would only arise from
changes in T.
It is implied in this case that the equation of state is dictated by the given relation, 𝛿 = Lo 𝛼(T − To ), which is the special
case for linear (1D) expansion. Values for the linear coefficient of thermal expansion, 𝛼 can be found in reference books
(e.g., see [43]) or manufacturer catalogs. This relationship arises directly from the definition for the coefficient of thermal
expansion [21],
( ) ( )
1 𝜕–V 1 𝜕–𝑣
𝛼≡ = , (3.54)
–V 𝜕T P –𝑣 𝜕T P
expressed using either volume, V – , or specific volume, 𝑣
–. In this latter case, it is also referred to as volume expansivity [20, 44].
This relationship defines 𝛼 as the fractional increase in volume per unit increase in temperature of a system maintained at
constant pressure. It can be assumed that values of 2𝛼 and 3𝛼 can be used to model area and volume expansions, respectively,
unless otherwise indicated [43]. Isothermal compressibility, 𝛽T , provides a measure of volume when there is a change in
pressure under constant temperature,
( ) ( )
1 𝜕–V 1 𝜕–𝑣
𝛽T = − =− , (3.55)
–V 𝜕P T –𝑣 𝜕P T
This parameter is the inverse of isothermal bulk modulus, BT (which was defined in Chapter 2).
These coefficients provide a basis for coupling the thermal energy domain to the mechanical domain with the two-port
energy storing C element. The example above showed how the force (or pressure) can be determined explicitly in terms
of the temperature and volume, which in this case was a constant. Since temperature is dictated, there is no need for the
corresponding relation on the thermal side. In general, both constitutive relations for a two-port C may be needed in a
system model.

3.7.1.4 Ideal Gas

A contained ideal gas that is part of a system of interest can be modeled as a reversible process in a wide range of cases.
Figure 3.48(a)–(d) can be used to represent practical causal conditions when the thermodynamic substance is a contained
ideal gas. For gases at high temperature and low pressure, the ideal gas relation is applicable [20]:
RT
P= Mechanical equation of state, (3.56)
–𝑣
du
= c𝑣– Constant volume specific heat, (3.57)
dT
where u = U∕m is the specific internal energy (per unit mass or per N if molar), R is the universal gas constant and c𝑣– is
the molar constant volume-specific heat capacity. For an ideal gas, internal energy and therefore c𝑣– are solely functions of
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
128 3 Physical Modeling with Bond Graphs

Isometric
Isobaric
Isentropic Isometric

Isothermal Isobaric

Isothermal

Isentropic

(a) (b)

Figure 3.50 𝑣 plots.

Summary of trends for key polytropic processes on (a) T − s and (b) P − –

temperature and in many practical cases c𝑣– can be considered a constant. These two equations along with equation 3.53 can
be reduced to, u = u(s, –𝑣), showing how the stored internal (potential) energy is a function of two states as required for a
two-port energy-storing C element. This function will be derived in Chapter 9 and relates to the temperature and pressure
relations:
( –𝑣 )𝛾−1 ( )
s − so
T = To o exp , (3.58)
–𝑣 c𝑣
( –𝑣 )𝛾 ( )
o s − so
−P = −Po exp , (3.59)
–𝑣 c𝑣
where s and 𝑣– are the per unit mass or molar-specific entropy and volume, respectively. Recall that there is also a constant
pressure specific heat capacity, cp , and the ratio of cp to c𝑣– is defined as 𝛾. These are variables commonly used with reversible
polytropic processes for an ideal gas, where P–V n = constant, and n is referred to as the polytropic index or exponent.
Various cases for trends in T − s and P − –𝑣 for given values of n are shown in Figure 3.50. These can be associated with the
various causal conditions as described by the two-port C elements in Figures 3.48(a)–(d). After describing junction elements
in Section 3.7.2, which are used to interconnect model R and C elements in the thermal domain, examples are reviewed to
demonstrate application of these relations.

3.7.2 Thermal Junction Elements

An essential element for constructing models for the thermal domain is a common temperature 0-junction. The 0-junction
can represent spatial locations/regions of physical elements over which temperature is substantially uniform. This type
of assumption enables lumped-parameter modeling of thermal systems as discussed in Chapter 2 and as conveyed in
Figure 2.51. Recall that a 0-junction enforces power conservation and can thus represent how entropy flowrates sum to
zero at a common temperature 0-junction, as shown in Figure 3.51. Note that entropy flow rate at each bond has a positive
sign when directed toward the 0-junction. The sign of entropy flow should be assigned based on the specific application
need.

T1 Tn
0
fs1 fsn n
i fsi = 0, i = 1...n
T2 fs2
Entropy flow rates sum to zero:

Figure 3.51 A thermal 0-junction represents the assumption that a region within a physical element has a uniform temperature.
Such a “temperature node” also provides a way to represent the sum of entropy flowrates.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
3.7 Thermal Effects in Bond Graph Models 129

Figure 3.52 A thermal 0-junction is typically associated with the thermal side of a T1 T −P
two-port thermodynamic C, and a bond from the 0-junction is directed into the C. The 0 C
entropy flow rate on that bond indicates the Ṡ for that element. fs1 Ṡ –˙
V

Tn fsn

A useful way to construct the thermal system part of a model is to identify distinct points of temperature and to associate
a 0-junction with this point in the system. For example, if a thermal storage element is to be included, then a 0-junction is
typically associated with that element. As shown in Figure 3.52, the 0-junction would imply the relation,
∑
n
fsi − Ṡ = 0.
i
∑n
If causal assignment indicates that Ṡ is specified as an input to the C then S is state and a state equation results, Ṡ = i fsi .
The entropy flow rates into and out of this 0-junction would be associated with corresponding heat transfer flow rates.
It is not very common that a 1-junction is needed in the thermal domain. A 1-junction imposes equal entropy flow rates
on connected thermal elements. As pointed out by Brown [7], one use for such a construction is for models with ideal
machines. Ideal engines and pumps are commonly used in high-level models in thermodynamics textbooks [20]. In these
models, an ideal engine or pump would have one thermal port and one mechanical port. The thermal port would have a
net heat transfer flow equal to the product of a temperature difference (typically defined by two temperature reservoirs)
and an entropy flow rate. These relations can be constructed by a 1-junction as shown in Figure 3.53. While ideal, such
models demonstrate how bond graphs can be applied to provide practical insight into these systems (see examples in Solved
Problems).

3.7.3 Examples with Thermal Effects

This section reviews modeling systems with thermal system effects. The examples emphasize use of the two types of
two-port R elements as well as thermal energy storage with a one- or two-port C.

Example 3.7 Ideal gas in a cylinder with a movable piston

Consider a cylinder with a movable piston that contains a fixed amount of an ideal gas, as shown in Figure 3.54(a). Various
processes may be induced by imposing different conditions, such as illustrated by the P − –
V plots in Figure 3.54(b). Consider
that a quasi-static adiabatic process (no heat transfer) in which volume is increased from V – A to V
– B proceeds along the
P3 –V 5 = constant path.
a) Use quasi-static analysis to find the work done in the system along ADB as well as the heat transfer. Assume heat Qs
is added while the pressure is held constant (isobaric) and the piston expands along AD. Further, the process DB is a
constant volume (isometric) cooling process for which pressure drops from PD to PB .
b) Formulate bond graph models to analyze the processes described in part (a).

Solution (a): A quasi-static thermodynamic analysis of the process ADB can be conducted as follows:
1) Determine the total change in internal energy from A → B along the adiabat, so UAB = − ∫ Pd–
V , with P = PA (– V )k ,
VA ∕–
where k = 5∕3.

Figure 3.53 Use of a 1-junction to create the temperature difference 𝛥T = TH − TL and High T reservoir
common entropy flow rates fsH = fs = fsL .

TH fsH

ΔT Ideal e
1
fs machine f

TL fsL

Low T reservoir
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
130 3 Physical Modeling with Bond Graphs

Walls either insulated

or conductive

Pressure,
Gas
constant
is external
temperature Moveable piston
Volume,
(a) (b)

Figure 3.54 (a) Gas in cylinder with a moveable piston. (b) Various processes on a P − –V diagram.

2) Determine work done A → B using WADB = −PA (VB − VA ) + WDB , where WDB = 0.
3) Determine heat transfer over ADB by, QADB = UAB − WADB .
For the specific cases illustrated in Figure 3.54(b), the steps above yield: UAB = −112.5 J, WADB = −700 J, and thus, QADB =
UAB − WADB = 587.5 J.

Solution (b): We can describe the two processes in part (a) using bond graphs as shown below:

TD T T −PA TB T T −P
Se R 0 C Se Se R 0 C Sf
fsw Ṡ –V˙ fsw Ṡ ˙
–V = 0
Isobaric heating (AD), P = PA (constant) Isometric cooling (DB), –V = VD (constant)

In the isobaric heating case, note that the two-port C has a specified pressure PA held constant while heat transfer flows
from the ambient environment which is held at TD . The volume in the cylinder will increase from VA until it reaches VD .
This system has only one state, S, with state equation,
Ṡ = fs𝑤 = H(TD − T)∕T,
where H is an effective heat transfer coefficient (e.g., H = kA∕L, if those parameters are given). Note that here,
( )𝛾−1 ( )
–VA S − So
T = TA exp ,
–V mc
where 𝛾 = 5∕3, m is the total mass of the contained gas, and c is the gas specific heat. The volume will expand according to,
[ ( )]
S − So 1∕𝛾
V = VA exp .
mc
The system will follow the isometric model over DB, with states (S, – V̇ = 0 and the entropy state now follows:
V ), but here –
Ṡ = −fs𝑤 = −H(T − TB )∕T,
where now the external temperature has been changed to TB to force cooling to state B. In a dynamic simulation, condi-
tions must be set to allow switching from the isobaric to isometric models according to when states D and B are achieved.
For example, for the isobaric case, heating would be imposed only until V – reached VD , and this would correspond to a gas
temperature, TD = PD –VD ∕(mR), where R is the universal gas constant. Problem B-3.45 explores developing a dynamic sim-
ulation of this model. In principle, it is not necessary to build unique bond graph models for each case of interest, as shown
above. The intent here is to illustrate how causal assignments can be used to define the imposed process conditions. Note
also that in this example, the mass of the piston and any friction has been neglected. The influence of these mechanical
effects will be illustrated through other examples and problems.

The next example reviews the application of four different two-port R elements.

Example 3.8 Ideal gas in a heated chamber with heat transfer

An ideal gas in a chamber contains a resistance heater that has appreciable conduction heat transfer through the chamber
walls and convection and radiation from the chamber surface. Develop a bond graph model for this system and obtain state
equations.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
3.7 Thermal Effects in Bond Graph Models 131

(a) (b)

Figure 3.55 Resistance heater in a chamber (a) schematic and (b) bond graph.

Solution
A bond graph model is given in Figure 3.55(b). The leftmost R element models the heater as irreversible conversion of avail-
able electric power, V(t) ⋅ is , into “unavailable” thermal power, TfsR . The remaining R elements embody heat conduction
through chamber walls and convection and radiation from chamber surface. Constitutive relations for the elements are:
V(t) V(t)2
is = , fsR = (Heater R),
R(T) TR(T)
with R(T) the temperature sensitive heater resistance,
kA (T − T𝑤 )
fscond = .
L T
kA (T − T𝑤 )
fs𝑤 = (Conduction R).
L T𝑤
hATA
T𝑤 = (Convection R).
hA − fsconv
T𝑤4 − Ta4
fsrad = 𝜎A (Radiation R).
T𝑤
( )
s − so
T = To exp (Gas thermal C; with –𝑣 = –
𝑣o ).
c𝑣
Note that an arbitrary assignment of causality on one of the heat transfer R elements is required to complete causality
assignment. This implies a dissipative field and an algebraic loop. State equations (and loop) are
V(t)2 kA (T − T𝑤 )
Ṡ = fsR − fscond = −
R(T) ⋅ T L T𝑤
with the following equations defining an algebraic loop for T𝑤 :
hATa
T𝑤 = .
hA − fsconv
fsconv = fs𝑤 − fsrad .
T𝑤4 − Ta4
fsrad = 𝜎A .
T𝑤
kA (T − T𝑤 )
fs𝑤 = .
L T𝑤
Thus T𝑤 is found from solving the algebraic equation:
[ ]−1
kA (T − T𝑤 ) T𝑤4 − Ta4
T𝑤 = hATa hA − + 𝜎A ,
L T𝑤 T𝑤
where chamber temperature T = T(S). The resulting differential-algebraic equation could be solved numerically for S(t)
and therefore T(t).
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
132 3 Physical Modeling with Bond Graphs

Conductive barrier Insulated walls

(a) (b)

Figure 3.56 (a) Electrically-heated thermal system and (b) bond graph.

Example 3.9 Electrically-heated element with conduction

The system shown in Figure 3.56(a) was initially modeled in Example 2.19. Develop a bond graph of the complete system,
assign causality and identify state variables, and then derive any state equations.
Solution
A bond graph is shown in Figure 3.56(b). From causality assignment, both thermal C elements are found to have integral
[ ]T
causality, so we have n = 2 states, x = S1 S2 . In this model, the P − – V̇ port is ignored since –
V̇ = 0, although the pressure
varies.
As described in Section 3.2.3, a two-port resistive element represents how electrical power is dissipated by the resistor
and transformed into heat, that is, d = V(t)i = Q̇ R = T1 fsR . A relation for fsR can be determined as follows:
V(t)i V(t)2
V(t)i = T1 fsR ⇒ fsR = = .
T1 T1 R(T1 )
Note how fsR is a function of V(t) and T1 (S1 ). A voltage source is directly connected on the electrical side of this two-port
resistor, thus the voltage (effort) causality is imposed on that side of the R. Note also how a 0-junction is used to represent a
uniform temperature node, T1 , assumed for region 1. This 0-junction serves as a basis for summing the three entropy flow
rates with implied sign convention: fsR − Ṡ 1 − fs1 = 0.
A two-port heat transfer R is used to represent conduction from regions 1 to 2, and all the heat transfer from conduction
flows into region 2 so Ṡ 2 = fs2 . The walls are assumed to be insulated.
The two-state equations can now be expressed as,
𝑣(t)2 T (S ) − T2 (S2 )
Ṡ 1 = fsR − fs1 = − Hc 1 1 ,
T1 R(T1 ) T1 (S1 )
T1 (S1 ) − T2 (S2 )
Ṡ 2 = fc = Hc ,
T2 (S2 )
where Hc = kA∕L, k is the thermal conductivity of the divider, A the surface area, and L the thickness. Each region is a
[ ]
generic solid with Ti (Si ) = Toi exp (Si − Soi )∕𝜌i ci –Vi , 𝜌i representing the density, ci the constant volume specific heat, and
–Vi the constant volume of material i = 1, 2. Each solid has initial condition Soi , corresponding to Toi .

The following example now considers a contained ideal gas as well as chamber walls that can conduct heat transfer.

Example 3.10 Electrically heated ideal gas chambers as two-port C with conduction and non-insulated walls
The system studied in Example 3.9 is now shown in Figure 3.57(a), where each region is filled with an ideal gas. There is
also a change to the walls around region 2 so that there can be heat transfer to the ambient environment. Develop a model
for this system as in Example 3.9.
Solution
The bond graph in Figure 3.57(b) reflects the changes made to the previous example system. First, we indicate the coupling
to a fluid-mechanical domain with the two-port C, although in this case the chambers cannot expand. Even though – V̇ 1 =
–V̇ 2 = 0, as causally enforced by the flow sources, these equations represent two additional and valid state equations for
– 1 and V
states V – 2 . Of course, the volumes stay constant at the initial values. Second, there is now a 0-junction representing
the assumption that the temperature in region 2 is uniformly at T2 and heat can flow from this region through the walls
to Tamb .
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
3.7 Thermal Effects in Bond Graph Models 133

Conductive barrier

Insulated walls Non-insulated walls

(a) (b)

Figure 3.57 (a) Electrically heated ideal gas chambers and (b) bond graph using two-port C.

The state equation for S1 remains the same as before,

V(t)2 T (S ) − T2 (S2 )
Ṡ 1 = − Hc 1 1 ,
T1 R(T1 ) T1 (S1 )
but now the state equation for S2 accounts for the additional entropy flow due to heat transfer via the walls to the ambient
environment,
T1 (S1 ) − T2 (S2 ) T (S ) − Tamb
Ṡ 2 = fs2 − fs𝑤 = Hc − Hwalls 2 2 .
T2 (S2 ) T2 (S2 )
Further, the constitutive relations for T1 and T2 now change to be functions of the respective entropy and volume states
and are of the form given by equation 3.58. Note that while the volumes of each chamber are constant, they are still state
variables of the system. Further, the pressure in each chamber can be computed from equation 3.59.

3.7.4 Equivalent Conductive Two-Port R-Elements and Pseudo-Bond Graphs

It is common and convenient when building lumped-parameter models for heat transfer to adopt an electrical circuit analog.
In this approach, temperature is taken as analogous to voltage and heat transfer flow rate, Q, ̇ to current (see, e.g., [18, 45]).
̇
In this way, heat transfer relations define a heat resistance, Q = 𝛥T∕R in the same way as in electrical circuits. Aside from
the utility of the circuit analogy, a key advantage of such an approach is in forming equivalent models for heat transfer
resistive elements configured in series or parallel.
Consider a common situation with two different conductive elements, as shown in Figure 3.58(a), Q̇ = (T1 − T3 )∕Req ,
where Req = R12 + R23 . For conduction through a plane wall with surface area A, thickness L, and thermal conductivity
k, for example, R = L∕kA. Thus, the two conducting walls are easily combined into a single heat transfer two-port as in
Figure 3.58(c). It can be useful to first analyze heat transfer paths in this way, thus enabling the construction of true bond
graph models using two-port R and C elements as discussed in this section. This approach reinforces using an energetically
consistent approach for thermal system modeling with entropy as a state variable. As previously mentioned, this approach
will be extended in Chapter 9.
As discussed in Chapter 2, it is also very common to adopt the heat transfer flow rate Q̇ as a flow variable in thermal
systems modeling, so the conjugate pair becomes (T, f = Q). ̇ Since the product of this conjugate pair does not equal power,

combine

(a) (b) (c)

Figure 3.58 (a) Two conductive materials with electrical analog circuit, (b) bond graph with 2 two-port R elements, and
(c) an equivalent two-port heat transfer R.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
134 3 Physical Modeling with Bond Graphs

Figure 3.59 A pseudo-bond graph form of the system in Example 3.10.

care must be taken when used in combination with regular or “true” bond graphs. In many cases, a pseudo-bond graph can
be convenient, particularly when not necessarily coupling into other energy domains (and so the issue of a power mismatch
on the bonds is irrelevant). When using a pseudo-bond graph for heat transfer modeling, for example, the 1-junctions will
represent common heat transfer paths. As such, the analogy with electrical circuits is much more direct. The equivalent
“resistance,” for example, as discussed above is much simpler to perceive as two R elements in “series” and thus the
resistances simply add. This is the analogy most often taken when using electrical analogs in introductory heat transfer
courses. Some accommodations need to be made. For example, revisiting Example 3.10, rather than using a two-port R to
model the heat coming from the electrical system, as shown in Figure 3.59, one simply uses a ‘flow source’ equivalent for
the Q̇ from the resistor. Modeling of the thermal capacitive elements typically changes when adopting pseudo-bond graphs.
Temperature rather than entropy is used as a state. The state equation, say, for a (linear) capacitive element can be expressed
as [19, 33],
∑
U̇ = Ct Ṫ = ̇
Q,
where the right-hand side is the net heat transfer flows, say, at a 0-junction. Here Ct = mc, with c the specific heat of the
material. Using this approach is often very convenient when deriving linearized models, say, for thermal control system
modeling and analysis.

3.8 Summary
We have defined basic modeling elements18 and concepts that we will use throughout the remainder of the text. This chapter
has also discussed methods for formally deriving system models given a causally complete bond graph. An emphasis has
been placed on deriving models in the form of state-space equations. We have found that:
1) A Hamilton set of momentum and displacement state variables, (p, q), will be the preferred set of energy states in this
text for all energy domains.
2) Explicit sets of state equations ẋ = f (x, u) are the preferred form, particularly for efficient numerical analysis.
3) Causal assignment with causal strokes can be used to systematically identify independent states and to guide derivation
of the most explicitly possible set of state equations.
4) A foundation is established for extending bond graphs to dealing with physical effects that required multiport represen-
tations. In the special case of thermal system modeling, a preliminary introduction is given in this chapter to enable an
energetically consistent approach for incorporating thermal effects in our system models.
In Chapter 4, we will provide additional discussion on formulating models and building simulation programs. We will
also show that bond graphs provide a convenient path to formulating computational models in the form of block diagrams
as well. Chapter 5, 6, and 7 will introduce methods for using system state equations to analyze the linear and fully nonlinear
performance of physical systems. It is important to not forget that most physical systems embody nonlinearity which can
be readily incorporated in a bond graph model. While linear analysis represents yet another approximation that must be
made, it should not deter the engineer from applying linear tools. The results can be very useful and insightful and one
should understand the limitations and bounds of validity of the methods. Design based on linear analysis should always be
checked with experimentation and/or simulation.

18 See Tables 3.4–3.6 at end of this chapter.

 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
3.8 Summary 135

Table 3.4 Summary of C-elements by energy domain.

Physical element Rate and constitutive relations Bond graph and energy

Generalized potential energy State: q = displacement Energy: Uq = ∫ e ⋅ dq

storage element e = effort Coenergy: Ue = ∫ q ⋅ de
e Rate: q̇ = f
q̇ = f
C Constitutive: e = Φ(q)
Linear case:
Energy: Uq = q2 ∕2C
capacitive element Linear: e = q∕C
C = capacitance Coenergy: Ue = Ce2 ∕2

Mechanical translation State: x = displacement F

F = force C
ẋ = v
Rate: ẋ = 𝑣
Energy: Ux = ∫ F ⋅ dx
Constitutive: F = F(x)
Coenergy: UF = ∫ x ⋅ dF
F1 = F2 = F Linear: F = kx
k = stiffness = 1∕compliance, C Linear case:
𝑣1 − 𝑣2 = 𝑣 spring Energy: Ux = kx2 ∕2
Coenergy: UF = F 2 ∕2k

Mechanical rotation State: 𝜃 = angular displacement τ

𝜏 = torque C
θ̇ = ω
Rate: 𝜃̇ = 𝜔
Energy: U𝜃 = ∫ 𝜏 ⋅ d𝜃
𝜏1 = 𝜏2 = 𝜏 Constitutive: 𝜏 = 𝜏(𝜃)
Linear: 𝜏 = K𝜃 Coenergy: U𝜏 = ∫ 𝜃 ⋅ d𝜏
𝜔 1 − 𝜔2 = 𝜔
K = stiffness = 1∕compliance, C Linear case:
rotational spring
Energy: U𝜃 = K𝜃 2 ∕2
Coenergy: U𝜏 = 𝜏 2 ∕2K

Electrical State: q = electrical charge V

C
V = voltage q̇ = i
Rate: q̇ = i Energy: Uq = ∫ V ⋅ dq
Constitutive: V = V(q) Coenergy: UV = ∫ q ⋅ dV
Linear: V = q∕C Linear case:
C = capacitance Energy: Uq = q2 ∕2C
Coenergy: UV = CV 2 ∕2
V1 = V2 = V
i1 − i2 = i
electrical capacitor

Hydraulic State: V
– = volume P
P = pressure C
–˙ = Q
V
V̇ = Q
Rate: –
Energy: UV– = ∫ P ⋅ d–
V
Constitutive: P = P(–
V)
Coenergy: UP = ∫ –
V ⋅ dP
Linear: P = –
V ∕C
Linear case:
C = hydraulic capacitance
V 2 ∕2C
Energy: UV– = –
P1 = P2 = P Coenergy: UP = CP2 ∕2
Q1 − Q2 = Q
hydraulic capacitor
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
136 3 Physical Modeling with Bond Graphs

Table 3.5 Summary of I-elements by energy domain.

Physical element Rate and constitutive relations Bond graph and energy

Generalized kinetic energy State: p = momentum Energy: Tp = ∫ f ⋅ dp

storage element f = flow Coenergy: Tf = ∫ p ⋅ df

I
ṗ = e Rate: ṗ = e Linear case:
f Constitutive: f = Φ(p) Energy: Tp = p2 ∕2I
inertive element Linear: f = p∕I
Coenergy: Tf = If 2 ∕2
I = inertance

Mechanical translation State: p = momentum ṗ = F

𝑣 = velocity v I
Rate: ṗ = F Energy: Tp = ∫ f ⋅ dp
Constitutive: 𝑣 = 𝑣(p) Coenergy: T𝑣 = ∫ p ⋅ d𝑣
Linear: 𝑣 = p∕m
Linear case:
F1 − F2 = F m = mass
Energy: Tp = p2 ∕2m
𝑣1 = 𝑣2 = 𝑣 Coenergy: T𝑣 = m𝑣2 ∕2
mass

Mechanical rotation State: h = angular momentum ḣ = e

𝜔 = angular velocity ω I
Rate: ḣ = 𝜏 Energy: Th = ∫ f ⋅ dh
Constitutive: 𝜔 = 𝜔(h) Coenergy: T𝜔 = ∫ h ⋅ d𝜔
Linear: 𝜔 = h∕J
Linear case:
𝜏 1 − 𝜏2 = 𝜏 J = rotational inertia Energy: Th = h2 ∕2J
𝜔1 = 𝜔2 = 𝜔
Coenergy: T𝜔 = J𝜔2 ∕2
rotational inertia

Electrical State: 𝜆 = flux linkage

λ̇ = V
i = current I
i
Rate: 𝜆̇ = V
Energy: T𝜆 = ∫ i ⋅ d𝜆
Constitutive: i = i(𝜆)
V1 − V2 = V Coenergy: Ti = ∫ 𝜆 ⋅ di
Linear: i = 𝜆∕L
i1 − i2 = i L = inductance Linear case:
electrical inductor Energy: T𝜆 = 𝜆2 ∕2L
Coenergy: Ti = Li2 ∕2

Hydraulic State: Γ = fluid momentum Γ̇ = P

Q = flowrate I
Q
Rate: Γ̇ = P
Energy: TΓ = ∫ Q ⋅ dΓ
Constitutive: Q = Q(Γ)
Linear: Q = Γ∕I Coenergy: TQ = ∫ Γ ⋅ dQ
P1 − P2 = P
I = fluid inertia Linear case:
Q1 = Q2 = Q
Energy: TΓ = Γ2 ∕2I
hydraulic inertia
Coenergy: TQ = IQ2 ∕2
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
3.8 Summary 137

Table 3.6 Summary of R-elements by energy domain.

Physical element Rate and constitutive relations Power dissipation relations

Generalized dissipative element Resistive law: e = ΦR (f ) Instantaneous power dissipation:

R
e T Conductive law: f = Φ−1
R (e)
d = e ⋅ f = T ⋅ fs = heat
f Ṡ Linear: e = R ⋅ f f = ΦR (f ) ⋅ f

resistive element
or, f = e∕R e = e ⋅ Φ−1
R (e)
R = linear resistance

Mechanical translation General: F = Φ(𝑣) F

or, 𝑣 = Φ−1 (F) v R
Linear: F = b ⋅ 𝑣 Content: 𝑣 = ∫ F ⋅ d𝑣
or, 𝑣 = F∕b
F1 = F2 = F Coenergy: F = ∫ V ⋅ dF
b = linear damping
𝑣1 − 𝑣2 = 𝑣 Dissipation: d = 𝑣 + F
damper Linear case dissipation:
d = b𝑣2 = F 2 ∕b

Mechanical rotation General: 𝜏 = Φ(𝜔) τ

or, 𝜔 = Φ−1 (T) ω R
Linear: 𝜏 = B ⋅ 𝜔 Content: 𝜔 = ∫ 𝜏 ⋅ d𝜔
𝜏1 = 𝜏2 = 𝜏 or, 𝜔 = T∕B Coenergy: 𝜏 = ∫ 𝜔 ⋅ d𝜏
𝜔 1 − 𝜔2 = 𝜔 B = linear rotational damping Dissipation: d = 𝜔 + 𝜏
rotational damper Linear case dissipation:
d = B𝜔2 = 𝜏 2 ∕B

Electrical General: V = Φ(i) V

or, i = Φ−1 (V) R
i
Linear: V = R ⋅ i Content: i = ∫ V ⋅ di
or, i = V∕R
V1 − V2 = V Coenergy: V = ∫ i ⋅ dV
R = linear resistance
i1 = i2 = i Dissipation: d = i + V
electrical resistor Linear case dissipation:
d = Ri2 = V 2 ∕R

Hydraulic General: P = Φ(Q) P

or, Q = Φ−1 (P) R
Q
Linear: P = R ⋅ Q
Content: Q = ∫ P ⋅ dQ
or, Q = P∕R
P1 − P2 = P Coenergy: P = ∫ Q ⋅ dP
R = linear hydraulic resistance
Q1 = Q2 = Q Dissipation: d = Q + P
hydraulic resistor Linear case dissipation:
d = RQ2 = P2 ∕R
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
138 3 Physical Modeling with Bond Graphs

3.9 Problems

Word bond graph construction

Use the schematics to guide construction of a word bond graph for the systems described. Do not include detailed bond
graph elements (e.g., R, C, I, G, or T) but do identify 0- and 1-junctions as needed to} convey any assumptions about how
model elements should be connected.

Problem B-3.1 Motor–pulley–ballscrew actuator A linear motion actuator is composed of a ballscrew mechanism
driven by a permanent-magnet DC (PMDC) motor and pulley system as shown in Figure B-3.1. The tip of the ballscrew
casing is assumed rigid and interacts with a load force. The second pulley drives the ballscrew at velocity, 𝑣b , and the
belt is assumed to be very stiff. Note that the PMDC motor is shown with no coil inductance, assumed to be negligible
in this case. Develop a word bond graph of this system that includes the key elements/subsystems needed for modeling
this actuator system.

ballscrew
Pulley velocity load velocity
radius,
load force

Belt

Pulley
radius,

Figure B-3.1 Linear ballscrew actuator with PMDC-pulley drive.

Problem B-3.2 Aquarium air pump A battery-powered aquarium air pump schematic is shown in Figure B-3.2(a). This
pump uses a single-piston driven by a battery-powered DC (PMDC) motor through a crank-slider mechanism, which is
shown in side view in Figure B-3.2(b). Pumping of air into and out of the cylinder is achieved by a positive-displacement
mechanism, as studied in Example 3.6. Develop a word bond graph showing key elements and primary power flow
interactions.

Long hose
Side-view detail of
crank-slider-piston
Piston
Motor shaft
velocity, vp
speed,
Porous air
Motor shaft Check valve stone
speed,

crank radius, rc
Air inlet connecting rod length, lc
Piston mass,
and area,

(a) (b)

Figure B-3.2 (a) Battery-powered air pump system. (b) Detail of the crank-slider-piston
mechanism.

Problem B-3.3 Vacuum cleaner model You’ve been tasked with modeling a traditional vacuum cleaner to better under-
stand how key components affect overall performance. The model results will be compared with a new robotic vacuum
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
3.9 Problems 139

cleaner configuration. In contrast to the robotic system that would use a pmdc motor or brushless motor, the vacuum
cleaner in Figure B-3.3 uses an AC universal motor to drive both a centrifugal fan and a roller brush. The latter is fit
into the nozzle area as shown, where dirt and other debris enters the pumping/vacuum chamber. Develop a word bond
graph as a first-cut model of this system. Indicate key elements and primary power flow interactions.

Electric
motor

Nozzle
Filter bag catches
dirt particles
allows air to exit

Dirt particles
Belt
Centrifugal fan/pump Brush

Figure B-3.3 Schematic of floor vacuum cleaner with centrifugal fan/pump, front roller
brush, and filter bag.

Problem B-3.4 Hydrostatic drive for tracked vehicle The schematic diagram in Figure B-3.4 is the drive system for a
tracked vehicle that consists of two independent hydrostatic transmissions [46, 47]. Both have hydraulic pumps that
are driven by an engine, and each pump has a swashplate that can rotate by using the control lever. If the lever is not
turned, the system is in neutral and no torque is delivered to the drive axles. If the control lever is pushed forward,
both swashplates are tilted and both motors create torque. By turning the control lever the tilt of each swashplate can
be adjusted to steer the vehicle.

Front of
vehicle
Drive sprocket Drive sprocket
Final reduction
gears

Axle Axle

Fixed swashplate
Hydraulic motor
Oil reservoir Pistons
Internal oil passages
Port plate (stationary)

Hydraulic pump
Cylinder block (rotating)
Tilting swashplate
(neutral position) Engine
Tilting swashplate
(drive position)

T-bar control lever

(shown in left turn position)

Figure B-3.4 Tracked-vehicle hydrostatic drive schematic.

 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
140 3 Physical Modeling with Bond Graphs

Develop a word bond graph that conveys the key elements and power interactions needed in a first step when develop-
ing a model for this system. This system may also require signal bonds to model control, say, of the tilting swashplate,
which modulates the hydraulic pump output. A signal bond would be used when a power interaction does not need to
be modeled. More will be said about information flow and modulation use in system modeling in Chapter 4.

Combining model elements with bond graphs

Problem B-3.5 Equivalent resistive element Two R elements are connected as shown below (Figure B-3.5).
a) Use the 1-junction relations to justify an equivalent R representation as shown and determine the equivalent linear
resistance, Req .
b) Can an equivalent be found for two nonlinear R elements with constitutive relations e1 = Φ1 (f1 ) and e2 = Φ2 (f2 )?
Show how or explain why it may not be possible.

R : R1

e1 f1

e0 e2 e0
1 R : R2 ⇒ Req
f0 f2 f0
(a) (b)

Figure B-3.5 (a) Bond graph of two interconnected linear resistive

elements and (b) equivalent form.

Problem B-3.6 Equivalent resistive element from delta-form Three R elements are connected in a delta-connection as
shown in Figure B-3.6.
a) Neatly sketch a bond graph for this resistive system.
b) Assume all of the resistors in the circuit given have a square-law, Vi = ii |ii |, i = 1, 2, 3. Obtain a plot of the voltage
versus current at the input terminal, V1 versus i1 .
c) Determine an expression for V as a function of i, and plot V versus i.

Figure B-3.6 Three resistive

elements in a delta connection.

Problem B-3.7 Equivalent capacitive element Three C elements are connected as shown below. Use linear constitutive
and junction relations to determine an equivalent C element as shown and determine the equivalent linear capacitance,
Ceq (Figure B-3.7).

C : C1 C : C2

e1 f1 e3 f3

e0 e2 e4 e0
0 1 C : C4 ⇒ Ceq
f0 f2 f4 f0

(a) (b)

Figure B-3.7 (a) Bond graph of interconnected linear capacitive elements and
(b) equivalent form.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
3.9 Problems 141

Problem B-3.8 Stepped rotational shaft A stepped rotational shaft is shown in Figure B-3.8(a). Justify the proposed
bond graph shown in (b) which only accounts for the compliance in each of the two sections with the indicated lengths
and diameters. Determine the compliance for each section in terms of physical parameters (see Chapter 2). Assume
the two free bonds represent the end connections and find an equivalent C element and its constitutive relation.

C : C1 C : C2

(a) (b)

Figure B-3.8 (a) Stepped shaft schematic and (b) associated bond graph.

Problem B-3.9 Extended stepped rotational shaft model Extend the model of the stepped rotational shaft in Problem
B-3.8 to include rotational inertia. Sketch a schematic that describes your model and build an equivalent bond graph.
State all assumptions made.

Problem B-3.10 Pressure tank submodel Hydraulic capacitive models due to both fluid compressibility and mechanical
elastic expansion of a cylindrical pressure vessel such as shown in Figure B-3.10 are examined in Solved Problems
A-2-9 and B-2.19.
a) Sketch a subsystem bond graph showing how both effects can be incorporated, indicating one power bond with Q1
flowrate and another power bond with Q2 flowrate.
b) How would you decide whether you can use linear capacitance for these models?
c) How would you decide whether you can neglect one effect over the other?
d) Optional. Work Problem B-2.19 and use the results to address item (c) in a quantitative manner for that specific
case.

Rigid ends

Figure B-3.10 Pressure tank with fluid compressibility and tank

elastic energy storage.

Problem B-3.11 Equivalent transformed resistive element An ideal electrical transformer is loaded by a linear resis-
tor with resistance R, as shown in Figure B-3.11(a). Use the bond graph proposed in Figure B-3.11(b) to justify the
equivalent form in Figure B-3.11(c) and then derive an expression for Reff .

(a) (b) (c)

Figure B-3.11 (a) Linear resistor loading a transformer, (b) bond graph
model, and (c) equivalent model.

Problem B-3.12 Equivalent inertive element Study the bond graph models as shown in Figure B-3.12.
a) Sketch an electrical system that is represented by the bond graph provided in part (a) of the diagram.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
142 3 Physical Modeling with Bond Graphs

b) Derive an equivalent I element by combining the two linear I elements that are coupled through the transformer
with modulus n.

I : I1

e1 f1
n
e0 e2 ·· e3 e0
0 T I : I3 ⇒ Ieq
f0 f2 f3 f0

(a) (b)

Figure B-3.12 (a) Bond graph and (b) equivalent form.

Problem B-3.13 Rack-and-pinion equivalent sub-model Study the rack and pinion system shown in Figure B-3.13(a) and
complete the following.
a) Sketch a bond graph, treating the input to the shaft as an open bond (i.e., no specified torque or velocity). Include
only ESEs and power-conserving elements in this model. Assume the shaft is very rigid.
b) Show how the rack and pinion system can be modeled by the equivalent rotational inertia Jeff shown in (b). Express
this equivalent inertia in terms of Jp , mr , and the rack/pinion gearing ratio based on pinion radius rp .

Pinion with rotational Equivalent rotational

inertia, inertia,

Rack of mass

Pinion radius,

(a) (b)

Figure B-3.13 (a) Rack and pinion subsystem and (b) equivalent inertia referred to
rotational shaft.

Bond graph modeling problems

Problem B-3.14 Driven mechanical systems The two basic mechanical systems shown in Figures B-3.14(a) and (b) are
driven in different ways. Study the systems and construct a bond graph for each case. Assume there is also linear
damping between the mass and the sliding surface as shown.

(a) (b)

Figure B-3.14 (a) Force driven system. (b) Force and velocity
driven system.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
3.9 Problems 143

Problem B-3.15 Mass–spring link system Develop a bond graph of the system in Problem A-2-4.

Problem B-3.16 Vibrational system with nonlinear friction Draw a bond graph for the system shown in Figure B-3.16.
Note that each mass has sliding friction in contact with the ground.

Figure B-3.16 Multi-mass vibrational system nonlinear friction.

Problem B-3.17 Analogous mechanical and electrical systems – 1 Two sets of analogous mechanical and electrical sys-
tems are shown in Figure B-3.17. For each case, study the systems, confirm they are analogous, and then sketch a bond
graph that should apply for either the mechanical or the electrical system.

(a)

(b)

Figure B-3.17 Analogous (a) mechanical and (b) electrical

systems.

Problem B-3.18 Analogous mechanical and electrical systems – 2 Confirm that the mechanical and electrical systems
shown in Figure B-3.18 are analogous and then sketch a corresponding bond graph that should apply for either
system.

(a) (b)

Figure B-3.18 Analogous (a) mechanical and (b) electrical

systems.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
144 3 Physical Modeling with Bond Graphs

Problem B-3.19 Driven rotational drive systems For the two rotational systems in Problem B-2.6, develop a bond graph
and apply causality. Identify state variables from causality assignment.

Problem B-3.20 Parallel-elastic rotational drive An electric drive is equipped with a parallel-elastic system, as shown
in Figure B-3.20. Matching to a dynamic load, 𝜏L (t), is accomplished by engaging a spring with stiffness, K, using a
clutch. Assume the clutch has a dissipative characteristic, 𝜏c = Φc (𝜔c ), where 𝜔c is the clutch slip velocity. Assume the
electric motor is a permanent magnet dc motor with resistance, Rm , and negligible inductance. The motor constant is
rm . Assume the gear ratio for the gears having inertias J1 and J2 is GR, with 𝜔1 = GR ⋅ 𝜔2 . Develop a bond graph for
this system. Assign causality and identify state variables and then derive the state equations. At least, one of the states
should be the rotational speed 𝜔2 (or h2 ). Write an output equation for the motor current, im .

Electric
motor
Dynamic load

Clutch

Figure B-3.20 Parallel-elastic drive system.

Problem B-3.21 Electrical networks Refer to the electrical circuits in Figure B-3.21. Sketch a bond graph for the electrical
networks in (a)–(d). Apply causality, identify any state variables, and derive state equations.

(a) (b)

(c) (d)

Figure B-3.21 Miscellaneous electrical networks.

Problem B-3.22 Resistive–capacitive bridge network Figure B-3.22 shows a bridge circuit with two resistive and two
capacitive arms.
a) Develop a bond graph considering a known input voltage, Vs (t). Model the connection at the output using a flow
source with io = 0 (an ideal voltage sensing device draws negligible current).
b) Apply causality and identify the system states.
c) Derive state equations.
d) Write an output equation for Vo .
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
3.9 Problems 145

Figure B-3.22 Resistive–capacitive

bridge network.

Problem B-3.23 Hydraulic fluid circulation system The pump in the hydraulic circulation system shown in Figure B-3.23
can control the flowrate between tanks A and B at Q(t).
a) Develop a system bond graph accounting for the tank capacitances and the fluid inertance and resistance effects in
pipes 1 and 2. Assume all linear elements.
b) Apply causality and determine the states.
c) Derive the state equations.

Circulation
pump,

Figure B-3.23 Hydraulic fluid circulation

system.

Bond graph modeling and state equation derivation

For the next four problems, do the following: (i) develop a bond graph model, (ii) assign power convention, number and
label the bonds, (iii) assign causality, (iv) use causality to identify the number of independent states and write the state
vector, x, and (v) derive state equations and write in state-space form (linear or nonlinear form). Assume model elements
have linear constitutive relations only if a parameter is explicitly indicated (e.g., bearing friction, b, means bearing imparts
torque, Tb = b𝜔), otherwise adopt a nonlinear constitutive relation.

Problem B-3.24 Mechanical system vibration A vibrational source of velocity, 𝑣(t), is imposed on the mechanical system
as shown in Figure B-3.24, which is also loaded by a dynamic force, F(t).

Rigid,
negligible mass

Figure B-3.24 Mechanical system driven by velocity and force

sources.

Problem B-3.25 Mass moving within a case Figure B-3.25 shows a force F(t) is applied onto a mass that is contained
within a case and there is dry friction between the case (mass m2 ) and ground. Build a bond graph, apply causality and
identify the system states, and then derive the state equations.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
146 3 Physical Modeling with Bond Graphs

Rigid rod

Figure B-3.25 Rod-driven mass within sliding case.

Problem B-3.26 Suspended mass within a case A force F(t) is applied onto case of mass m2 that contains a mass m1
that is suspended by two springs of stiffness k1 . Follow instructions given above (Figure B-3.26).

Figure B-3.26 Mass suspended

within a supported case with
applied force F(t).

Problem B-3.27 Dual-geared rotational drive system For the system shown in Figure B-3.27, assume the gears have a
total effective inertia, J1 , referred to the drive gear velocity, 𝜔1 . Assume each gear ratio is different as indicated by n2
and n3 . Follow instructions given above.

drive gear

Load torque

load inertia

Figure B-3.27 Dual-geared rotational drive

system.

Problem B-3.28 Electrical power network For the electrical circuit shown in Figure B-3.28, in addition to the state
equations, write two output equations: one for the voltage Vout , and one for the total power supplied by the source,
Vs (t). Follow instructions given above.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
3.9 Problems 147

Figure B-3.28 Electrical power network.

Problem B-3.29 Pushrod–lifter–valve

The schematic in Figure B-3.29(a) shows a mechanical pushrod–lifter–valve arrangement as might be found in some
internal combustion engines (e.g., see Taylor [48]). The bond graph in Figure B-3.29(b) is proposed for studying the
dynamic response.
a) Verify and complete the bond graph shown in part (b) of the figure, explaining the assumptions that are implied.
b) Apply causality and identify the state variables as the xrod (pushrod compression), x𝑣 (valve compression), and 𝑣𝑣
(valve mass velocity).
c) Show that the three differential equations are as follows:
Compression/extension of the pushrod: ẋ rod = 𝑣cam (t) − n𝑣𝑣
Velocity of the valve mass: 𝑣̇ 𝑣 = nkrod xrod ∕m𝑣 − k𝑣 x𝑣 ∕m𝑣 − b𝑣 𝑣𝑣 ∕m𝑣
Compression of valve spring: ẋ 𝑣 = 𝑣𝑣
where n = a∕b = 𝑣p ∕𝑣𝑣
d) Write the final state equations in state-space form.

Rocker
arm Valve
Pushrod spring

Follower

Cam Valve

(a) (b)

Figure B-3.29 (a) Pushrod-lifter-valve system and (b) bond graph.

Problem B-3.30 L-frame linkage An L-frame structure shown in Figure B-3.30 forms part of a sensing system used to
detect pressure changes on the surface of an underwater vessel. A rotational sensor would be attached at the pivot
point, P, to detect the angular rotation of the L-frame, 𝜃.

L-frame
Pivot, P

Figure B-3.30 L-frame subsystem.

 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
148 3 Physical Modeling with Bond Graphs

For this part of the system, develop a bond graph model from which you can identify the state variables: angular velocity,
̇ of the L-frame (about P), deflection, xL , of the spring kL , and deflection, xP , of the pivot spring, kp . Note there is
𝜔 = 𝜃,
one input, 𝑣(t), at the base of damper bL . Assume that the angular displacement of the L-frame, 𝜃, is constrained to be
small (< 10∘ ).

Problem B-3.31 Surge tank with nonlinear area–height variation The diagram in Figure B-3.31 represents a hydro-
electric power generation system. Develop a bond graph model of the hydraulic system up to and including the valve
which empties to a known pressure, Pe . Use a (nonlinear) pressure–volume relation, P = P(–
V ), for the surge tank.

Surge
tank

Constant height
reservoir

Pressure tunnel

values refer to elevation Nonlinear

Pressure valve
from a datum (not shown)
pipe

Figure B-3.31 Surge tank in hydroelectric system. Source: Cellier [49]/Springer Nature.

Problem B-3.32 Analog meter The analog voltage meter modeled in Solved Problem A-2-8 is an EM device that is
designed so that the voltage applied at the input terminal, Vin , is visually displayed by the response of an indicator
needle’s angular deflection. For example, the meter needle might deflect 90∘ given an input dc voltage of 15 V.
a) Use the model from Solved Problem A-2-8 to develop a bond graph model that includes electrical resistance, meter
inductance, Lm , meter needle and coil rotational inertia, Jm , rotational damping, Bm , and meter rotational spring
stiffness, Km . Assume the meter input voltage is Vin .
b) Apply causality and identify state variables.
c) Derive state equations. You should find that there are three states in this system and the system is linear.
d) Express the model in linear state-space form including an output equation for needle angular position, 𝜃.

Problem B-3.33 PMDC motor model Revisit Problems B-2.25 and B-2.34 on modeling of the basic permanent-magnet
DC (PMDC) motor and incorporation of thermal effects. The model should include thermal model elements as devel-
oped Problem B-2.34. Assume that the electrical resistance and the mechanical friction have dependence on effective
motor temperature, Tm .
a) Develop the complete bond graph with thermal effects, using a single thermal C to model the thermal capacitance.
Assume an effective linear thermal capacitance, Ct .
b) Apply causality to show that the motor flux linkage, 𝜆m , the motor shaft speed, 𝜔m , and the entropy, Sm , for the
effective thermal capacitance are state variables. Derive the state equations for these states.
c) Change the model by neglecting the motor inductance. Reapply causality to show that only 𝜔m and entropy
are system states. Derive the state equations for these states and also write an output equation for the armature
current, im .
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
3.9 Problems 149

Causal conflicts, derivative causality, and algebraic loops

Problem B-3.34 System with dependent C element Show that the bond graph in Figure B-3.34 has one C element in
derivative causality, so the system is of order n = 1. Derive the state equation for this system, assuming all elements
are linear.

e1 (t)
Se 1 C
f1

C 0 R
Figure B-3.34 Bond graph with
a dependent C element.

Problem B-3.35 Driven rack-and-pinion with assist An initial design for a rack and pinion driven by a shunt dc motor
has revealed a need for additional assist at startup. A controlled velocity 𝑣a (t) is proposed to be applied to the rack as
shown in Figure B-3.35. The initial model provided includes: the ideal motor voltage, Vs (t), a shunt resistance, Rm ,
the motor ideal EM conversion that relates torque to motor current, 𝜏m = rm im , a rigid shaft connecting the motor and
pinion inertias Jm and Jp (known), the indicated rack mass, mr , pinion radius, rp , and a support stiffness, k. You can
also assume there is friction between rack and the surface it moves on.

Motor pinion rotational

inertia,

Pinion radius,

Rack of mass

Controlled
assist

Figure B-3.35 Coupled gears in a drive.

Build a bond graph of the model as proposed, apply causality, and explain whether any changes should be made to the
model and why. For your final model, use the applied causality to determine the number of state variables. Apply state
equations as completely as possible.

Problem B-3.36 Coupled gears in a drive The system in Figure B-3.36 has an input torque, 𝜏(t) and a specified load
torque 𝜏L . In this system, assume the gears have a gear ratio GR = 𝜔1 ∕𝜔2 .
a) Develop a bond graph and from causality identify there is a derivative causality when two distinct I elements are
used in the model.
b) To eliminate the derivative causality, combine the two inertias into a single I element on the output side (at 𝜔2 ).
Reapply causality and identify the state variables.
c) Derive the state equations.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
150 3 Physical Modeling with Bond Graphs

Shaft 1

Shaft 2

Figure B-3.36 Coupled gears

in a drive.

Problem B-3.37 Three connected reservoirs Figure B-3.37 shows two ways that three reservoirs in a fluid distribution
system might be interconnected. Assume that each reservoir has additional flow connections (not shown in figure)
that can supply or remove fluid at a controlled flowrate.

(a) (b)

Figure B-3.37 (a) Three fluid reservoirs in Δ-connection and

(b) Y-connection. The connecting pipes are labeled A, B, and C.

Develop a bond graph for each system. For each case, consider that each connecting pipe has inertia and resistive
effects due to fluid flow. Apply causality and derive state equations for each case. There are no inputs specified but one
or more can be added as flow rates into one or more of the tanks.

Problem B-3.38 Torque-driven wheel The wheel in Figure B-3.38(a) is driven by a known torque, 𝜏d (t) (as from a
motor or drive shaft). Assume planar motion along the x-direction and that the wheel remains in contact with a flat,
non-deformable surface as it rotates, as shown. The normal contact force, Fz , from the free-body diagram of diagram
(b) is taken solely as the total weight, Fz = W = mg.
a) Sketch a bond graph for the no slip case, 𝑣x = r𝑤 𝜔, showing that the translational and rotational momenta are
related so one of these elements must be in derivative causality.
b) For case (a), define an effective mass, meff , that includes the rotational inertia J referred to the translational veloc-
ity. Redraw the bond graph with the effective mass at the translational velocity. Introduce a translational resistive
element to model total “road loads,” FL (rolling resistance, grade, etc.).
c) Apply causality to the bond graph of (b) and derive a state equation in terms of the translational velocity, 𝑣x . Model
total road loads FL as the sum of rolling resistance force, Frr = fr Wsgn(𝑣x ), and also a force, Fg , due to an incline
(grade) of angle 𝛼, in the surface along x. Note that FL is applied at the center of gravity of the wheel along the x
axis. Write the state equation for 𝑣x . Include an output equation for the traction force, Ftx (derived from the bond
graph).
d) When there is significant slip at the wheel-surface contact, 𝑣x ≠ r𝑤 𝜔, the rotational momentum and translation
velocity will be independent states. Develop a bond graph for this case, introducing a traction R element to
model the traction force, Ftx (e.g., see Figure 3.26 for guidance). Apply causality and derive state equations
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
3.9 Problems 151

for the slip case. Model the traction force using a nonlinear function, Ftx = Φ(𝑣s ), where 𝑣s = r𝑤 𝜔 − 𝑣x is the
slip velocity.

FBD:

(a) (b)

Figure B-3.38 (a) A torque 𝜏d (t) drives a wheel and

(b) free-body diagram.

Algebraic loop problems

Problem B-3.39 Mechanical/electrical analog systems with algebraic loop Develop a bond graph for each of the systems
shown in Figure B-3.39 to confirm they are analogous systems. Then apply causality to the bond graph to identify that
an algebraic loop exists. Identify state variables and derive state equations for the general system (use generalized
(ei , fi , pi , qi ) variables).

Figure B-3.39 Mechanical/electrical analogs with

algebraic loop.

Problem B-3.40 Bond graph with an algebraic loop Use causality to show that the bond graph in Figure B-3.40 has order
n = 1 and that there is an algebraic loop. Derive the state equation for this system. Assume all elements are linear and
use a modulus n for the given transformer, T.

I R 1 Se

Se 1 T 0

R 1 Se

Figure B-3.40 System with algebraic loop.

Problem B-3.41 Algebraic loop in gyrator-coupled systems Show through assignment of causality that an algebraic loop
exists for the bond graph given in Figure B-3.41. Complete causality to identify state variables for the system and derive
state equations in explicit form. Use generalized variables (ei , fi , pi , qi ) with subscripts based on the bond numbering
provided. Assume all elements are linear and that the gyrator has modulus r.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
152 3 Physical Modeling with Bond Graphs

I : I2 R : R5 I : I7

2 5 7
R : R1 0 G 0 1 R : R8
1 3 4 6 8

Figure B-3.41 Bond graph for gyrator-coupled systems with algebraic loop.

Problem B-3.42 Electrical circuit with algebraic loop Develop a bond graph for the electrical circuit shown in
Figure B-3.42. Use causality to determine if there is an algebraic loop. Identify state variables and derive the state
equations in explicit form.

Figure B-3.42 Electrical circuit with

resistive field.

Problem B-3.43 Real source modeling and simulation Consider the bond graph in Figure B-3.43 for a linear resistive–
capacitive (RC) system driven by a source that has been characterized experimentally with effort-flow data. A Thevenin
real source model is proposed and shown within the dashed box.
a) Confirm that the causality assignment as given on this bond graph indicates there is an algebraic loop.
b) Derive the state equations, including an expression for the algebraic loop flow variable, fR , assuming a given relation
is available for the source, eo = eo (fo ).
c) Consider a model where the Thevenin source is replaced with an ideal effort source having a value eT = 5. Compare
a simulation of this system with the case where the source is modeled by the Thevenin source where the following
data is provided for the real source characteristics:
e_o = [5, 4.1, 3.500, 2.600, 1.000]
f_o = [0, 0.005, 0.010, 0.015, 0.020]
Take R = 200 and C = 0.01 and assume qC (0) = 0. Plot the results from the ideal source case against those from the
real source case. Generate three plots with both cases on each comparing: (i) the input effort, eo , (ii) the flow, fR ,
and (iii) the effort, ec .

Rt R:R

eRT eR fR

eT eo eC
Se 1 1 C:C
fo q̇C

Figure B-3.43 RC system driven by Thevenin real

source model.

Problem B-3.44 Geared motor-driven split drive with load Consider the system shown in Figure B-3.44 and assume the
gears have negligible inertia and gear ratio is GR.
a) Develop a bond graph and identify the algebraic loop during causality assignment. Derive state equations for this
case.
b) If the coupling damper BL is eliminated, show that there is no algebraic loop and rederive the state equations.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
3.9 Problems 153

Gear ratio,

Load torque

load inertia

Figure B-3.44 Geared motor-driven split drive with load.

Problem B-3.45 Gas in a cylinder with movable piston Refer to the case of a cylinder with contained ideal gas and
a movable piston as discussed in Example 3.7. That example describes quasi-static analysis and development of bond
graph models for different process paths. Use the models provided to develop a dynamic simulation that solves how the
system evolves over time from state A to D and then to B. To model the heat transfer, assume an effective heat transfer
coefficient H = 1 (SI units). Take the contained mass as m = 1 kg. Plot on a single figure the following variables over
̇ pressure in the cylinder, P (Pa), volume of the contained gas, V
time: heat transfer flow rate, Q, – , and the total heat in
the system, Q. Compare with the results from the quasi-static end-point analysis.

Case studies

Problem B-3.46 Motor-driven web/film windup drive system A PMDC motor drives a windup process as shown in
Figure B-3.46. Tension is maintained in the web using pinching rollers, which control the web feed rate at 𝑣s (t). Develop
state equations for this system, explaining any assumptions needed to model this system. Assume the controlled motor
voltage is Vs (t) and for the purposes of this model assume that JL remains constant.

Gearbox, GR

Wind-u
roller

In tension

(Effective stiffness)

Web/film Side Pinch

Top rollers
material feed view view

Figure B-3.46 Tension-control system for a winding process.

Problem B-3.47 Closed-cylinder with heating element For the system shown in Figure B-3.47, a circular piston of mass
mp is placed onto a cylinder, trapping the contained volume of air, which is initially at (Pa ,Ta ). There is negligible
leakage, but the piston (diameter dp ) can move within the cylinder (diameter dc ) and settles at a point where the
weight of the mass balances the pressure of the contained air, Pc .
a) Develop a bond graph model of this system, accounting for damping/friction between the piston and cylinder,
as well as heat transfer through the cylinder walls and the piston (assume all relevant material and geometric
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
154 3 Physical Modeling with Bond Graphs

properties can be determined). Assume an ac voltage source drives the resistive heater which has resistance R(Tc ),
where Tc is the temperature of the contained air.
b) Apply causality and identify state variables, two of which should be the air volume, –Vc , and entropy Sc .
c) Formulate the constitutive equations for a two-port thermomechanical C model for the contained air relating the
pressure and temperature of the air contained in the cylinder, Pc and Tc , to volume and entropy states. Assume the
amount of air remains constant (no leakage).
d) Derive the system state equations.

Figure B-3.47 Closed piston-cylinder

with resistive heater.

Problem B-3.48 Bellows-driven suction-cup foothold A small robotic glass wall climber uses a driven bellows (B) to con-
trol the air pressure, Ps , at suction cups (S) used as footholds, as shown in Figure B-3.48. The suction cup is sufficiently
compliant to allow uniform contact, but assume it does not change the shape significantly during suction.
a) Develop a bond graph model for the bellows driven with a controlled velocity 𝑣b (t). The tube connecting the bellows
and the suction cup should be assumed rigid. The suction cup geometry defines a fixed volume, and the seal made
with the glass surface prevents leakage. Assume isothermal operation and assume the tube only introduces resistive
effects between B and S.
b) Apply causality and identify state variables.
c) Formulate the constitutive equations required to estimate the suction pressure.
d) Derive state equations.
e) Develop an expression to estimate the holding force capacity, Ft , assuming the coefficient of friction between the
suction cup material and glass is 𝜇.

Rigid
tube

Bellows
Glass

Figure B-3.48 Bellows-driven suction-cup

foothold.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
155

System Model Formulation and Evaluation

It is helpful to be familiar with different modeling approaches and representations commonly used in a particular field of
study. Relations between broadly used physical and functional representations are summarized in Figure 4.1. The relations
suggest some model representations can be readily integrated or transformed from one form to another. The directions of
arrows indicate that it is possible in some cases to transform one representation into another, but only in some cases such a
transformation is uniquely reciprocal. For example, state equations can be derived from a causally complete bond graph as
shown in Chapter 3, but the inverse does not have a unique result – there may be many bond graph forms that can represent
the same set of state equations.
A block diagram of a system can be directly generated from derived state equations, or directly from a causal bond graph.
There may also be occasions where a bond graph integrated with a block diagram subsystem is the best way to graphically
convey a complete or cohesive system model. And depending on the end result desired, either state equation form or a
fully rendered block diagram may be formulated. Indeed, a block diagram description may be preferred when using some
computational environments that support simulation, control analysis, and design. Some examples of these approaches
will be provided later in this chapter, notably to show how it can be helpful to use a bond graph to guide block diagram
construction.
Figure 4.1 also illustrates the relationship between nonlinear state equations and linear state space. Once a system is
linearized, it is possible to transform into transfer function or linear nth-order ordinary differential equation (ODE) form.
Chapter 5 reviews methods useful in analyzing such linear system model representations. Naturally, if a system is linear,
the bond graph leads directly to linear system models. Further, a bond graph can be used to directly derive linear transfer
functions using impedance relations, an approach to be discussed in Chapter 6.
The first part of this chapter reviews methods that can play a key role in model formulation and evaluation. Methods for
determining system equilibrium are first discussed, including ways to interpret and use equilibrium conditions of a system
for analyzing a system. Different ways of linearizing nonlinear functions and systems are then introduced, and examples
are provided of application of computational tools for this purpose. Equilibrium and linearization are essential concepts
for discussing concepts of system stability, which also plays a role in model evaluation. Then, ways that we can integrate
bond graph models with block diagrams are discussed. Block diagrams are useful for representing purely functional model
descriptions and flow of information by signals in a system, so they can help us represent parts of a system that we can-
not easily model from first principles. Thus, block diagrams of control systems can be integrated with a bond graph, or
we can convert a bond graph purely into a functional block diagram representation, as commonly done in some commer-
cial computer-aided engineering software. Since they are commonly used in block diagrams, transfer functions are also
introduced in this chapter, although their derivation and use will be more thoroughly discussed in Chapters 5, 6, and 7
as well.
The latter part of this chapter discusses two foundational modeling concepts that are especially useful when dealing
with systems or system elements that are not as well defined or familiar as the classical Kirchhoff systems introduced in
Chapter 2. The concept of constitutive structure provides guidance and imposes restrictions on the way we define and use
functional relationships to model energy storing and dissipative elements. Then we introduce modulation, a concept that
allows us to represent physical effects in our models that are influenced by signal or information flow; that is, not by direct
power interactions. We find that modulation is essential for modeling many practical devices and processes, so we describe
how to identify modulation and how it should be used without violating the underlying energetic principles in bond graph
models.

Modeling of Physical Systems: Simulation and Control, First Edition. Raul G. Longoria and Joseph J. Beaman.
© 2025 John Wiley & Sons Ltd. Published 2025 by John Wiley & Sons Ltd.
Companion Website: www.wiley.com/go/Modeling_of_Physical_Systems/1e
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
156 4 System Model Formulation and Evaluation

Bond graph

Nonlinear state space order linear ODE

Linear
state
space

Block diagram Linear transfer function

Figure 4.1 Physical bond graph representations and functional representations.

4.1 Equilibrium Analysis

The concept of system equilibrium is a useful first step in evaluating and analyzing a physical system model. Equilibrium
relations are defined by the system mathematical model and can be used to analyze or satisfy known or desired conditions.
For example, equilibrium states can be used to identify valid initial conditions for a system under study. Equilibrium states
can be determined analytically or numerically, depending on the intent of the analysis. For generally nonlinear systems, an
equilibrium analysis is a first step in studying the linear behavior “about equilibrium states.” A linearization process is used
to formulate a linear approximation of the nonlinear state equations. The linear system can be used to study how the system
behaves under small excursions from the equilibrium state. Stability of the system can also be examined. Linearization and
stability will be further discussed in Sections 4.2 and 4.5, respectively.
Equilibrium states define a “destination” of a system state when the system is either unforced (zero-input equilibrium)
or forced (a reference or nominal operating point equilibrium). In many cases, knowledge about the equilibrium state of
a system can be useful if not essential to understanding its dynamic behavior. To illustrate, consider the familiar case of a
pendulum, which will achieve a different equilibrium state depending on whether it is unforced, as shown in Figure 4.2(a),
or forced with a steady wind as in Figure 4.2(b).

4.1.1 Definition of Equilibrium

A system or a process is in an equilibrium or steady state if the state variables are no longer changing with time,
ẋ = f(x0 , u0 ) = 𝟎.
Solving the resulting linear or nonlinear algebraic equations yields the equilibrium state(s), x0 , for inputs, u0 . These
equations clarify how an equilibrium can be classified based on whether there is no forcing of the system (zero-input) or if
there is forcing, and power would be continuously “pumped” into the system.
Consider again the simple pendulum now with a torque 𝜏a applied at the pivot to drive the pendulum motion. The rate
of change of momentum is,
∑
ḣ = ml2 𝜔̇ = 𝜏 = −mgl sin 𝜃 + 𝜏a .

Steady
wind

Unforced equilibrium Forced equilibrium

at at
(a) (b)

Figure 4.2 Simple pendulum under (a) unforced and (b) forced conditions.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
4.1 Equilibrium Analysis 157

We would write the state equations,

𝜃̇ = 𝜔
𝜔̇ = −(g∕l) sin 𝜃 + u,
where u = 𝜏a ∕ml2 . These equations give equilibrium conditions,
(g∕l) sin 𝜃0 = u0 and 𝜔0 = 0.
If u0 = 0, then 𝜃0 = n ⋅ 𝜋, for n = 0, 1, 2, …, and 𝜔0 = 0. The 𝜃0 are the expected vertical positions (up and down). If u0 ≠ 0,
different equilibrium positions can be achieved for 𝜃, as when the pendulum is exposed to a torque induced by a steady
wind, but always with 𝜔 = 0.
Since the equilibrium condition is determined mathematically by f(x0 , u0 ) = 𝟎, we should expect to solve n equations that
are either linear or nonlinear algebraic equations. Since there are n unknown states and possibly m inputs, it is necessary
to specify additional equations or either known states or inputs to solve for the n + m unknowns. For example, it may be
desirable to determine the value of an input given that one of the states is known.
In many cases, the idea is to solve for the n states given known m inputs, and this is not a problem. On the other hand,
the equilibrium relations could be used to find the input(s) required to achieve a certain steady-state condition. There can
also be systems where it is found that an equilibrium condition or state can not exist, or that there may be as many as there
are “settings” on the input combinations.
Consider the case of a single sphere totally submerged and falling in a fluid as illustrated in Figure 4.3(a).
From Newton’s law,
∑
ṗ = m𝑣̇ = F = mg − Fd − Fb .
Equilibrium is defined by the single dynamic equation ṗ = 0, thus giving, Fd + Fb = mg. This equation says equilibrium is
achieved when the weight, mg balances drag, Fd , and buoyancy forces, Fb . An equilibrium velocity, 𝑣o , will then depend on
how we model the drag force, in particular. Consider three possible cases:
1) Fd = 0, no equilibrium
2) Fd = B𝑣 (linear), a unique equilibrium
3) Fd = (1∕2)𝜌Cd 𝑣2 , with Cd = f (Re), where Re is the velocity-dependent Reynolds number, multiple equilibrium velocities.
If a thrust force, F, were imposed on the sphere as in Figure 4.3(b), there can be as many equilibrium states as we have
throttle settings. The following examples illustrate how equilibrium problems are set up using state-space models.

Example 4.1 Linear elastic system

The mechanical system shown in Figure 4.4(a) is to be studied to determine equilibrium conditions. This system is com-
posed of interconnected elastic elements and mass elements. Use the bond graph in Figure 4.4(b) to derive the equilibrium
relations.
Solution
The bond graph indicates there are five states: p1 , x1 , p2 , x2 , x3 , and x4 , with the state equations:
ṗ 1 = −k1 x1 − k3 x3 − k4 x4 + F1 (t)
ẋ 1 = p1 ∕m1
ṗ 2 = k3 x3 + k4 x4 − k2 x2 + F2 (t)
ẋ 2 = p2 ∕m2
ẋ 3 = p1 ∕m1 − p2 ∕m2
ẋ 4 = p1 ∕m1 − p2 ∕m2 .

Figure 4.3 (a) Equilibrium of a falling sphere and (b) with a thrust force, F. Thrust
force,

(a) (b)
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
158 4 System Model Formulation and Evaluation

Figure 4.4 Equilibrium of a linear elastic system.

Equilibrium is found by setting all the derivatives to zero, so in this case, all the “x” derivatives become zero indicating
that for equilibrium,

p1o = p2o = 0,

Also note that x3o = x4o , thus we are left to find the equilibrium values for x1o , x2o , and x3o (same as x4o ). This leaves two
equations,

k1 x1o + (k3 + k4 )x3o = F1o

−(k3 + k4 )x3o + k2 x2o = F2o .

This can’t be solved without another equation. Identify that the system is constructed such that the three spring lengths
are related by: L1 + L3 = L2 , where Li = li − xi , with i = 1, 2, 3. Here li is the free length and xi is the spring deflection
(positive in compression). Thus, a third equation arises between the three spring deflections,

l1 − x1o − l2 + x2o + l3 − x3o = 0,

or,

x1o − x2o + x3o = l1 − l2 + l3 = l.

In an ideal case (mass 1 has no size), l1 + l3 = l2 and l = 0 (size of mass 1). Then the three equations can be solved for
the three unknown values, given F1o , F2o , and known stiffness values. Let, k1 = 3k, k2 = 2k, k3 = k, and k4 = k, where k is
a known value, and also, F1o = F, and F2 = 2F, F a nominal force. For this case,

3x1o + 2x3o = F∕k

−2x3o + 2x2o = 2F∕k

x1o − x2o + x3o = 0,

which is readily solved for the three equilibrium values in terms of F∕k.
This problem demonstrates an approach to solving for the equilibrium states of a linear system. Introducing nonlinear
effects due to friction or in the elastic behavior of the springs would require solving the equations using a nonlinear equation
solver. There would be no change in how the third equation was introduced.

Example 4.2 Hydraulic tanks in equilibrium

The two tanks shown
√ in Figure 4.5 have connection valves for which the flowrate is related to the pressure drop by the
relation, Qi = Ki ΔPi , where Ki is a constant for each valve. Develop a model and derive the equilibrium condition.
Solution
The system has two state equations that can be written directly from mass continuity,
√ √
V̇ 1 = Q𝑣1 − Q𝑣2 = K1 Pin (t) − –V1 ∕C1 − K2 –V1 ∕C1 − –
V2 ∕C2
√ √
V̇ 2 = Q𝑣2 − Q𝑣3 = K2 –V1 ∕C1 − –V2 ∕C2 − K3 – V2 ∕C2 − Pout (t).
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
4.1 Equilibrium Analysis 159

Nonlinear
valve

Figure 4.5 Equilibrium of coupled tanks.

The equilibrium condition for this system can be found by,

√ √
0 = K1 Pin,o − –V1o ∕C1 − K2 –V1o ∕C1 − –
V2o ∕C2
√ √
0 = K2 –V1o ∕C1 − –V2o ∕C2 − K3 –V2o ∕C2 − Pout,o .
Solving for the equilibrium state (V1o , V2o ) requires specifying equilibrium pressure inputs (Pin,o , Pout,o ) and use of a non-
linear equation solver.

4.1.2 Equilibrium Bond Graphs

Examples 4.1 and 4.2 illustrate how equilibrium conditions can be solved from the state equations of a system. Another
way to study the equilibrium of a system is by examining the state of power flow between system elements. Paynter [1]
referred to an equilibrium condition where all the power bonds have zero power flow as a static equilibrium, as shown in
Figure 4.6. In contrast, a stationary equilibrium refers to a condition when all the power flows are constant. As an example,
consider the simple pendulum of Figure 4.2. The unforced pendulum has a static equilibrium at 𝜃 = 0. On the other hand,
the pendulum exposed to a fluid cross-flow would achieve a nonzero angle equilibrium that can be classified as stationary,
since power is flowing in that case. These concepts suggest another way to analyze the equilibrium state of a system based
on a bond graph model.
If we can define the equilibrium state(s) of a system through the mathematical model, it stands to reason there may be a
way to examine equilibrium states directly from a bond graph. Recall that equilibrium leads to two different states of power
flow on the bonds between system elements as described in Figure 4.6, that is,
static equilibrium: ⇒ ei fi = 0, stationary equilibrium: ⇒ ei fi = constant
This suggests that a dynamic system bond graph can be converted into an equilibrium or steady-state bond graph by
imposing an equilibrium condition on energy-storing elements (ESEs) with causal conditions as follows:

Equilibrium I causality Equilibrium C causality

e = ṗ = 0 e
I C
f f = q̇ = 0
I imposes no effort, e = 0 C imposes no flow, f = 0

Note that power flow is thus zero into ESEs at equilibrium, ṗ ⋅ f = 0 and e ⋅ q̇ = 0. In addition, the corresponding stored
energy in I and C elements at equilibrium is constant, Tpo and Uqo , respectively.

Static equilibrium:
all

Stationary equilibrium:
all constant

The
can be elements, devices, junctions, etc.

Figure 4.6 Static and stationary equilibrium.

 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
160 4 System Model Formulation and Evaluation

This information can be used to analyze systems in equilibrium at the bond graph level, providing insight before or in
lieu of deriving full dynamic state equations. If a system is to be analyzed specifically for equilibrium, however, a slightly
modified causality assignment procedure (SCAP) should be followed. Equilibrium causality as defined above should be
enforced on ESEs. Explicit equilibrium causality then refers to when there are as many algebraic equations (rather than
differential equations) as you have equilibrium conditions on ESEs. If assignment of causality over the system requires an
arbitrary assignment then, as in the case of a standard algebraic loop, an additional algebraic equation is required to define
the equilibrium state of a system.
Recall that the concept of derivative causality arose when an I or C element in a system could not be assigned integral
causality. If equilibrium is to be achieved by a system then equilibrium causality must be possible on all the ESEs. If not,
this implies a system as modeled has no equilibrium state.1
Consider a suspended mass–spring–damper in Figure 4.7. Causality assignment for equilibrium is made as indicated
numerically. The source causal assignments are made first. We begin by imposing equilibrium on the mass at  3 , which then
requires causality assignments  4 and 5 . Another equilibrium causality assignment is then made on the spring with  6 and
with 7 the system causality is completed. For two ESEs, there were two assigned equilibria so this is explicit equilibrium
causality, meaning there should be two algebraic relations that result from the model. In a dynamic system model, the
equations are the state derivatives, here ṗ and x.
̇ However, now for equilibrium, the causality indicates we want to find the
causal outputs at the I and C elements, which are 𝑣m and Fk , respectively.
From the bond graph, 𝑣m = 𝑣4 = f (t) − ẋ = f (t), and Fk = e = mg. If we consider the case where f (t) = 0, then 𝑣mo = 0.
Further, Fk = kxo = mg, or xo = mg∕k. This equilibrium would be classified as static given the inputs provided.
Consider the mechanical system shown in Figure 4.8 in which a mass is supported by two linear damping elements, one
of which is attached to ground while the end of the other is driven with known velocity, 𝑣1 (t). Conduct an equilibrium
analysis using an equilibrium bond graph.
Causality is assigned on the two sources, and then equilibrium causality is applied on the mass  3 . At this point, the
model structure does not force any causality assignment so an arbitrary assignment must be made and here  4 is indicated.
The remaining causality assignments follow.
The equilibrium velocity of the mass, 𝑣mo , can be seen from the bond graph causality to be 𝑣mo = 𝑣b2o = Fb2o ∕b2 . But
Fb2o = Fo = Fb1o , a second unknown (signaled by the arbitrary assignment at  4 ). So we write, Fb1o = b1 (𝑣1o − 𝑣mo ). Thus,
we must solve,
𝑣mo − Fb1o ∕b2 = 0
b1 𝑣mo + Fb1o = b1 𝑣1o ,

Explicit equilibrium causality

static equilibrium

Figure 4.7 Suspended mass–spring–damper in equilibrium.

1 arbitrary
assignment

Figure 4.8 Equilibrium analysis of a damper-supported mass.

1 Systems with no equilibrium can be relevant in some applications, such as chaos-based circuit design [50] but are outside the scope of the
current discussion on physical system equilibria.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
4.1 Equilibrium Analysis 161

Figure 4.9 Equilibrium analysis of spring–damper combination.

for two variables (𝑣mo , Fb1o ) given the input condition 𝑣1o . We find, 𝑣mo = b1 𝑣1o ∕(b1 + b2 ), and Fb1o = b1 b2 𝑣1o ∕(b1 + b2 ). This
is classified as a static equilibrium only when 𝑣1o = 0, otherwise power flows are constant on the bonds and the equilibrium
is stationary.
Figure 4.9 illustrates two springs and a damper in equilibrium. In this case, the input flow is specified at  1 and the
remaining causal assignments follow to give explicit equilibrium causality. We directly find that 𝑣b4o = 𝑣1o . Practically, we
also want to find the corresponding force, Fb4o = b4 𝑣1o , which would then dictate the force in the springs as well as the
force transmitted into the wall. This system is in stationary equilibrium with constant power flowing into and dissipated
by damper b given known 𝑣1o .

Example 4.3 Equilibrium bond graph analysis of two tank system

Recall the two tanks problem from Example 4.2. Solve for the equilibrium states using an equilibrium bond graph approach.
Solution
In the case where known (constant) pressures are imposed at the input and output, an equilibrium bond graph can be
constructed as in Figure 4.10. An equilibrium analysis using the bond graph can quickly provide insight into the equilibrium
behavior, which is stationary in this case. That is, there is a constant finite power flow at each bond, except for the ESEs
where power flow is zero.
On this bond graph, after applying equilibrium causality, we end up with one arbitrary assignment at  5 , applied to the R
element for valve 1. Once this assignment is made, causality is dictated on the remaining bonds. This arbitrary assignment
indicates that we can write a single equation to determine the equilibrium flow Q𝑣1o . First, note that at equilibrium all the
flows are equal,
Q𝑣2o = Q𝑣3o = Qout,o = Q𝑣1o .
This may be intuitive to some readers but it can also be read from the causality. Also note how the pressure at the bottom
of tank 1, P1o , is the sum of the pressures as shown on the bond graph; thus,
P1o = P𝑣2o + P𝑣3o + Pout,o .
But note that each valve pressure drop depends on the corresponding valve flowrate. For example, P𝑣2o = (Q𝑣2o ∕K2 )2 , and
all the valve flowrates are equal to Q𝑣1o . Thus, P𝑣1o becomes,
P𝑣1o = (1∕K22 + 1∕K32 )Q2𝑣1o + Pout,o .
Since,
√
Q𝑣1o = K1 Pin,o − P1o ,

Figure 4.10 Equilibrium bond graph for coupled tanks.

 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
162 4 System Model Formulation and Evaluation

a single nonlinear algebraic equation in terms of Q𝑣1o becomes,

√
Q𝑣1o = K1 Pin,o − (1∕K22 + 1∕K32 )Q2𝑣1o + Pout,o ,

which depends on the valve coefficients and the input pressures. Once Q𝑣1o is found: (i) all flowrates are known, and
(ii) key pressures can be determined. Note that this includes the pressures in each tank at equilibrium, P1o and P2o . These
values allow solving for the respective equilibrium volumes, –V1o and V– 2o , and the corresponding stored potential energies
at equilibrium.
Contrast the approach taken here with that implied in Example 4.2. In the latter, the final state equations are needed and
two nonlinear algebraic equations are solved simultaneously for V– 1o and –V2o . From those equilibrium states, similar results
can be obtained.

Ultimately, whether to take a classical approach for determining equilibrium conditions as in Section 4.1.1 as opposed to
using an equilibrium or steady-state bond graph depends on the intent of the analysis. The former is essential in analysis
for linearization, to be discussed in Section 4.2. Sometimes, however, the end goal is to simply analyze an equilibrium
operation for design purposes, for example. In such cases, an equilibrium bond graph can provide useful insight and can
often provide an answer more quickly. Analyses based on these methods are demonstrated in Section 4.1.3.

4.1.3 Source-Load Modeling and Analysis

Consider a source S1 coupled to a load S2 , described as a word bond graph:

S1 S2

Such systems in equilibrium or steady state can be described using effort-flow source characteristics, such as shown in
Figure 4.11.
As illustrated in Figure 4.11, a source (supply) with a falling trend coupled to a load (demand) that is rising will have a
stable equilibrium point at the intersection of these two characteristics. To see how the operating condition is stable consider
the flow variable is increased slightly. The load would then produce a larger effort in response, forcing operation back to
the intersection. On the other hand, if flow decrease the source load being greater than the load in that case would cause
the flow to increase and a cause a return to the stable equilibrium.
Certain systems can have intersecting characteristics but the equilibrium may be unstable as in Figure 4.11(b). For this
case, if the flow increases, the system would tend to “run away” since the source effort exceeds the load in that case. Further,
should the flow variable decrease, then the load effort exceeds the source and again the operation would move away from
the unstable intersection point. Thus, case (b) is an unstable equilibrium. If the characteristics of source S1 and load S2 do
not intersect, then there is no equilibrium. Essentially, the source S1 is unable to “drive” load S2 .
Since many practical engineering systems spend a significant amount of their operational time in steady state, these
are useful concepts for defining and analyzing equilibrium (or steady-state) conditions. As such, they form the basis for

Stable equilibrium
point of operation Unstable equilibrium

Rising load (or demand)

characteristic
Load characteristic
Falling source (or supply)
characteristic Source characteristic

(a) (b)

Figure 4.11 (a) Stable equilibrium as balance of supply (source) and demand (load) curves (b) Unstable equilibrium.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
4.1 Equilibrium Analysis 163

Source Load Source Load

1 arbitrary
assignment
(a) (b)

Figure 4.12 (a) A permanent-magnet dc (pmdc) motor as source coupled to a resistive load. (b) Steady-state bond graph.

analysis and design of source–load systems. Using effort and flow curves, it becomes possible to determine the power flow
between a source and load during operation, since the product of e and f at the intersection defines this value, o = eo ⋅ fo .
It is insightful to superimpose on such effort-flow graphs the loci of constant power, r , which plot as hyperpolas,
e = r ∕f .
Given the established utility of source–load analysis, it can be useful to adopt equilibrium bond graphs to model both
sources and loads in some applications. To demonstrate, consider the permanent magnet dc (pmdc) motor shown in
Figure 4.12(a) which is to be coupled to a resistive load. To build a steady-state model of the motor, we construct an equi-
librium bond graph as in (b). The source causality is applied  1 followed by equilibrium causality on the two I elements.
An arbitrary assignment is required and applied at  4 , having the load dictate the speed 𝜔 to the source. The subsequent
causality assignments  5 –
8 follow. The implied algebraic relations take on a useful causal form. For example, the back
emf, Vm , is dictated by the motor shaft speed and together with the source Vs dictate motor current, im = (Vs − Vm )∕Rm .
These relations help construct the source torque–speed curve, beginning with the output torque as follows:
𝜏 = 𝜏m − 𝜏Bm = rm im − Bm 𝜔,
where the gyrator relation has been used, 𝜏m = rm im . Using the current relation from above and the relation for back-emf,
Vm = rm 𝜔m , the torque–speed source characteristic for the pmdc motor becomes,
[ 2 ]
r r
𝜏 = rm (Vs − Vm )∕Rm − Bm 𝜔 = m Vs − m − Bm 𝜔 = 𝜏s − Rc 𝜔,
Rm Rm
a linear function of speed, 𝜏 = 𝜏(𝜔), where 𝜏s is the stall torque and Rc a characteristic resistance composed of key motor
parameters.
A typical torque–speed curve is plotted against a nonlinear resistive load curve in Figure 4.13. Superimposed on the
source and load curves is a constant maximum power line based on the given torque–speed curve. As the input voltage
level is decreased, the torque–speed curve shifts down, remaining in parallel.
An approach similar to that used to model the basic pmdc motor can be taken to model a wide range of electromechanical
and hydromechanical sources. In most cases, the internal losses which influence the source characteristic may need to be
determined experimentally. The model structure formed can be used to guide such formulations and testing. For many
applications, only steady-state models are needed and insight can be gained into these devices through the source model
development.

Figure 4.13 Typical pmdc motor torque–speed curve as source characteristic Maximum
with nonlinear load line. power line
Nonlinear resistive
load line

Torque-speed
curve
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
164 4 System Model Formulation and Evaluation

Rotational Figure 4.14 A rotational inertia, J, driven at controlled speed, 𝜔(t), through a
inertia, clutch.
Clutch

Rotational
damping,

Load curve due

to bearing friction

Arbitrary
assignment

Effective source curve

from clutch

(a) (b)

Figure 4.15 (a) Bond graph for steady-state analysis. (b) A graphical steady-state solution for the clutch-inertia system.

Example 4.4 Clutch-flywheel system

A rotor with inertia J is mounted on bearings to be driven by a speed-controlled source of speed, 𝜔o , through a clutch to
achieve a steady state, hJo = J𝜔Jo , with stored kinetic energy, h2o ∕(2J) = J𝜔2Jo ∕2. Develop a bond graph and conduct analysis
to assess steady-state performance.
Solution
The steady state can be found directly using a steady-state bond graph as shown in Figure 4.15(a). Causality of the source
at 1 is followed by equilibrium causality on the inertia  2 . An arbitrary assignment is required and made on the bond
connected to the clutch  3 , leading to assignments 
4 and 5 .
At steady state, ḣ J = 0, and we seek 𝜔Bo = 𝜏Bo ∕B, which would dictate 𝜔Jo (per causality). But note 𝜏Bo = 𝜏co = Φc (𝜔co ) =
Φc (𝜔o − 𝜔Bo ). Thus, we have two equations to solve for the unknowns (𝜏co , 𝜔Bo ):
𝜔Bo − 𝜏co ∕B = 0
𝜏co − Φc (𝜔o − 𝜔Bo ) = 0,
which can be solved numerically provided the function Φc () is known. Alternatively, one can plot the induced clutch torque
as a function of 𝜔Bo as a source characteristic and the bearing torque as a load function of 𝜔Bo as well. This is shown in
Figure 4.15(b), with an assumed form for the clutch torque imposed on a linear bearing load. It is implied that the induced
clutch torque is a function of both 𝜔o and 𝜔Bo = 𝜔J .

4.1.4 Summary
This section described how system equilibrium can be defined in terms of the mathematical state-space equations, for
both unforced and forced systems. Equilibrium conditions were also interpreted using a bond graph perspective, useful for
contrasting the concepts of static and stationary equilibrium. This introduction lays a foundation for subsequent discussions
on: (i) how equilibrium conditions can be used to define initial conditions, (ii) how we can build “linearized” systems about
equilibrium states, and (iii) how we study stability about equilibrium states.
This section also introduced equilibrium bond graph analysis and assignment of equilibrium causality. These methods
can be useful for stationary equilibrium modeling of purely resistive systems, stationary equilibrium analysis, development
of steady-state actuator models, and source–load analysis. In constructing bond graphs in these cases, care should be taken
when a loop arises in the bond graph structure. In general, these can be more difficult to analyze as there is no unique
causal form. Repeated causal assignments may be needed to identify the best solution in those cases.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
4.2 Linearization Techniques 165

Linear
Linear
approximation
approximation

‘tangent’ linearization ‘secant’ linearization

(a) (b)

Figure 4.16 Nonlinear curves with (a) tangent and (b) secant linear approximations.

4.2 Linearization Techniques

From a mathematical standpoint, linearization refers to the linear approximation of a function at and about a given value
or values of the function. The selection of that value or values is often based on an expected state of operation. For example,
it can be useful to linearize a function about a system equilibrium for a dynamic system. In such cases, the equilibrium
concepts introduced in Section 4.1 are useful. Indeed, a linearized approximation of a system is often used to study response
and stability behavior of the system for small deviations about an equilibrium. Stability concepts will be introduced in
Section 4.5, and linear response of systems is the topic of Chapter 5. For now, this section reviews some common ways for
developing approximate linear models from nonlinear models.

4.2.1 Tangent Linearization

If we consider a linear approximation to a nonlinear curve, one such approximation is a tangent line at a point on the curve.
In a tangent linearization, we assume that the variables deviate only slightly from some operating point. Consider a system
whose input is x(t) and output is y(t), and the relation between y(t) and x(t) is given by,

y = f (x). (4.1)

If the operating point corresponds to x0 , then, if f (x) is differentiable, equation 4.1 can be expanded into a Taylor series
about this point as follows:

y = f (xo ) + f ′ (xo )Δx + f ′′ (xo )Δx2 + · · · , (4.2)

where Δx = x − x0 , and the following notation is adopted:

𝜕f | 𝜕2 f |
f ′ (xo ) = | , f ′′ (xo ) = | .
𝜕x |x=xo 𝜕x2 |x=xo
If the deviation Δx is small, we may neglect higher-order terms to form a linear approximation from equation 4.2,

y ≈ fo + kt Δx, (4.3)

where, f0 = f (x0 ) and kt = f ′ (x0 ). Equation 4.3 indicates that the output y is linearly related to x.
This linearization technique can be used with nonlinear state and output equations in a straightforward manner. Consider
a set of nonlinear state and output equations of the form,

ẋ = f(x, u) (4.4)

y = h(x, u) (4.5)

If the operating vector points correspond to (x𝟎 , u𝟎 ), then, if f(x, u) and h(x, u) are differentiable, equations 4.4 and 4.5 can
be approximated as follows:
𝚫x = x − x𝟎
𝚫u = u − u𝟎 ,
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
166 4 System Model Formulation and Evaluation

[ ] [ ]
𝜕f || 𝜕f ||
ẋ = f(xo , uo ) + 𝚫x+ 𝚫u
𝜕x ||xo ,uo 𝜕u ||xo ,uo
[ ] [ ]
𝜕h || 𝜕h ||
y = h(xo , uo ) + 𝚫x+ 𝚫u,
𝜕x ||xo ,uo 𝜕u ||xo ,uo
where higher-order terms in 𝚫x and 𝚫u are neglected. If x𝟎 and u𝟎 are chosen as the equilibrium point such that,

ẋ 𝟎 = f(x𝟎 , u𝟎 ) = 𝟎

then a linear approximation to the state equations can be expressed as,

𝚫ẋ = A𝚫x + B𝚫u, (4.6)

𝚫y = C𝚫x + D𝚫u, (4.7)

where 𝚫y = y − h(x𝟎 , u𝟎 )
𝜕f 𝜕f 𝜕f 𝜕f 𝜕f 𝜕f
⎡ 𝜕x1 𝜕x1 … 𝜕x1 ⎤ ⎡ 𝜕u1 𝜕u1 … 𝜕u1 ⎤
⎢ 1 2 n
⎥ ⎢ 1 2 r
⎥
⎢ 𝜕f2 ⋱ ⋮ ⎥ ⎢ 𝜕f2 ⋱ ⋮ ⎥
𝜕f | 𝜕f |
A= |
𝜕x |xo ,uo
= ⎢ 𝜕x1 ⎥ B= |
𝜕u |xo ,uo
= ⎢ 𝜕u1 ⎥ ,
⎢⋮ ⋱ ⋮⎥ ⎢⋮ ⋱ ⋮⎥
⎢ ⎥ ⎢ ⎥
⎢ 𝜕fn … … 𝜕fn ⎥ ⎢ 𝜕fn … … 𝜕fn ⎥
⎣ 𝜕x1 𝜕xn ⎦x ⎣ 𝜕u1 𝜕ur ⎦x
o ,uo o ,uo

𝜕h 𝜕h 𝜕h 𝜕h 𝜕h 𝜕h
⎡ 𝜕x 1 𝜕x 1 … 𝜕x 1 ⎤ ⎡ 𝜕u1 𝜕u1 … 𝜕u1 ⎤
⎢ 1 2 n
⎥ ⎢ 1 2 r
⎥
⎢ 𝜕h2 ⋱ ⋮ ⎥ ⎢ 𝜕h2 ⋱ ⋮ ⎥
𝜕h | 𝜕h |
C= |
𝜕x |xo ,uo
=⎢ 1
𝜕x ⎥ D= |
𝜕u |xo ,uo
=⎢ 1
𝜕u ⎥ ,
⎢ ⋮ ⋱ ⋮ ⎥ ⎢ ⋮ ⋱ ⋮ ⎥
⎢ ⎥ ⎢ ⎥
⎢ 𝜕hm … … 𝜕hm ⎥ ⎢ 𝜕hm … … 𝜕hm ⎥
⎣ 𝜕x1 𝜕xn ⎦x ⎣ 𝜕u1 𝜕ur ⎦x
o ,uo o ,uo

Example 4.5 Nonlinear mass–spring–damper system

For a spring–mass–damper system with a cubic spring, Fx = k(x + 𝛼x3 ), and a square-law damper, FD = 𝛽|𝑣|𝑣, obtain a
linear approximation about an equilibrium position achieved when a constant force F0 is applied to the system. The set of
nonlinear state equations has been derived as,
[ ] [ ]
ẋ p∕m
= ,
ṗ −𝛽|p|p∕m2 − k(x + 𝛼x3 ) + F
where x is the position, p is the mass momentum, F is the force, and m is the mass.
Solution
The equilibrium operating point can be found by setting ẋ = ṗ = 0, or the equations,
0 = p0 ∕m
0 = −𝛽|p0 |p0 ∕m2 − k(x0 + 𝛼x03 ) + F0 ,
are satisfied by the equilibrium points (x0 , p0 ). We can see immediately that p0 = 0 and x0 satisfies,

0 = −k(x0 + 𝛼x03 ) + F0 .

Using equation 4.6, a tangent linear approximation about the operating point is obtained as,
[ ] [ ][ ]
Δẋ 0 1∕m Δx
=
Δṗ −k(1 + 3𝛼x02 ) 0 Δp
[ ]
0
+ ΔF,
1
where Δx = x − x0 , Δp = p − p0 , and ΔF = F − F0 .
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
4.2 Linearization Techniques 167

4.2.2 Secant Linearization

Tangent linearization is strictly a local approximation. If 𝚫x gets large, then the approximation becomes increasingly inaccu-
rate in estimating the response of the nonlinear system. Even for small deviations, tangent linearization can have substantial
error. Consider the result from Example 4.5, which leads to a result where the damping effect in the original model is elimi-
nated. This anomaly is due to the square-law nature of the damping. For very small velocity, the damping force approaches
zero at a faster rate than a straight line. Thus, it would appear for small motions about zero that the system would oscillate
as if there were no damping.
Another problem with tangent linearization is the necessity for differentiability. For example, if the damper were of the
coulomb type with constitutive relation,
{
𝛽 sgn(𝑣) 𝑣 ≠ 0
Fd =
0 𝑣 = 0,
then the tangent linearization of this function is everywhere zero except at 𝑣 = 0, where it is undefined. The function sgn(x),
referred to as the signum function, gives +1 or −1 depending on whether x is positive or negative, respectively.
Secant linearization can alleviate some of the problems associated with tangent linearization. In its simplest implemen-
tation, secant linearization uses a linear approximation that passes through two points on a nonlinear curve. Referring to
Figure 4.16(b), the secant approximation is given as

y ≈ fa + ks Δx, (4.8)

where Δx = x − x0 , and,

fa = fL + ks (x0 − xL ),

with,
fU − fL
ks = (x − xL ),
xU − xL 0
and the upper and lower points xU and xL are a priori estimates of the range of the value of x.

4.2.3 Statistical Linearization

A more rational basis for secant linearization can be obtained if we try to obtain a linear approximation that in some sense
minimizes the error between a nonlinear function and its approximation.2

4.2.3.1 Statistical Linearization Approach

Defining the error, e, as the difference between the nonlinear function and the linear approximation as,

e = f − fa − ks Δx,

the average or mean square error, J, can be obtained as,

J =< e2 >=< (f − fa − ks Δx)2 >,

where < ⋅ > stands for expectation or an averaging operator.

A standard method for defining expectation is in terms of a probability density function, p(x). The probability density
function is nothing more than an (apriori) estimate of the expected distribution of a variable x. For example, if a variable
x is equally likely to be found in any interval between 0 and 1, then the probability density function is a uniform density
function depicted in Figure 4.17(a).
In order to find the probability of finding a variable x in an interval x1 to x2 , the density function is integrated between
these two limits as,
x2
P(x1 < x < x2 ) = p(x)dx.
∫x1

2 See Gelb et al. [52] for more detailed discussion of statistical linearization.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
168 4 System Model Formulation and Evaluation

Uniform Gaussian
distribution distribution

(a) (b)

Figure 4.17 Common probability density functions.

In order to find the expected value or average value of a function f (x), the product of f (x) and the density function integrated
over all possible values of x, or,
∞
< f (x) > = f (x)p(x)dx. (4.9)
∫−∞
Statistical linearization can then be formulated as the minimization of,
∞
J = < e2 > = (f − fa − ks Δx)2 p(x)dx, (4.10)
∫−∞
where p(x) is a given probability density function.
In order to minimize the mean square error, the following necessary conditions must be satisfied:
𝜕J
= −2 < f − fa − ks Δx > = 0,
𝜕fa
𝜕J
= −2 < Δx(f − fa − ks Δx) > = 0,
𝜕ks
For simplicity, assume that < Δx > = 0, which implies that the operating point x0 is equal to the mean value of x. In this
case, we have,
ks = < f Δx > ∕ < Δx2 >, (4.11)

fa = < f > . (4.12)

Example 4.6 Nonlinear mass–spring–damper system

Obtain the statistical linearization of f (x) = x3 about the operating point x0 = 0 with x uniformly distributed between −a
and a.
Solution
Using 4.11 and 4.12 and the uniform probability density function,
{
1∕(2a) −a < x < a
p(x) =
0 otherwise
we obtain
a
1
fa = x3 dx = 0,
∫−a 2a
a 1
∫−a x4 2a dx
ks = a 1
= 3a2 ∕5.
∫−a x2 2a dx
Note that as a increases, the effective linear slope, ks , increases, and that as a → 0, ks → 0, which is identical to the tangent
linear slope at x0 = 0.

4.2.3.2 Formulating Statistically Linearized State Equations

If x0 and u0 are chosen as average equilibrium points such that,
< ẋ > = < f > = 𝟎,
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
4.2 Linearization Techniques 169

and,
< 𝚫x > = < 𝚫u > = 𝟎,
then the minimum mean square approximation of a nonlinear set of state and output equations 4.4 and 4.5 can be expressed
as,
𝚫ẋ =A𝚫x + B𝚫u, (4.13)

𝚫y =C𝚫x + D𝚫u, (4.14)

where,
A = < f𝚫xT > P−1
xx

B = < f𝚫uT > P−1

uu

C = < h𝚫xT > P−1

xx

D = < h𝚫uT > P−1

uu

Pxx = < 𝚫x𝚫xT >

Puu = < 𝚫u𝚫uT >,
and Puu and Puu are covariance matrices. It is assumed here that 𝚫x and 𝚫u are uncorrelated or < 𝚫u𝚫xT >= 0 (see [51]).

Example 4.7 Statistical linearization of nonlinear mass–spring–damper system

For the spring–mass–damper system of Example 4.5 obtain statistically linearized state equations about the mean equilib-
rium point with x, p, and F Gaussian distributed with standard deviations 𝜎x , 𝜎p , 𝜎F , respectively, and < F >= F0 .
Solution
The state equations were given previously as,
[ ] [ ]
ẋ p∕m
= .
ṗ −𝛽|p|p∕m2 − k(x + 𝛼x3 ) + F
At the mean equilibrium point, we have
< ẋ > = < p∕m > = 0,

< ṗ > = < −𝛽|p|p∕m2 − k(x + 𝛼x3 ) + F > = 0,

which implies that
< p > = p0 = 0, (4.15)
−k < x > −k𝛼 < x > +F0 = − kx0 −
3
k𝛼(x03 + 3𝜎x2 x0 ) + F0 = 0, (4.16)
where we have used the Gaussian density functions:
[ ]
1 (x − x0 )2
p(x) = √ exp − , (4.17)
2𝜋 2𝜎x2
[ ]
1 (p − p0 )2
p(p) = √ exp − , (4.18)
2𝜋 2𝜎p2
[ ]
1 (F − F0 )2
p(F) = √ exp − , (4.19)
2𝜋 2𝜎F2
to evaluate the expectations. Equations 4.15 and 4.16 are used to evaluate x0 and p0 .
Using 4.13 and 4.14 with density functions 4.17–4.19 results in the linearization,
[ ] [ ][ ] [ ]
Δẋ 0 1∕m
√ Δx 0
= + ΔF.
Δṗ −k(1 + 3𝛼)(x02 + 𝜎x2 ) −2𝛽 2∕𝜋𝜎p Δp 1
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
170 4 System Model Formulation and Evaluation

√ √
Note that the eigenvalues for this linear system are s = −𝛽 2∕𝜋𝜎p ± k(1 + 3𝛼x0 )∕m − 2𝛽 2 𝜎p2 ∕𝜋, which indicate that
the linear system is damped and damping increases with momentum (velocity) variation as expected.

4.2.4 Summary of Linearization

We introduced three techniques to linearize a set of nonlinear state equations in this section. Tangent linearization is a
local linearization technique which requires differentiability. Secant and statistical linearization are more global and do
not require differentiability of the nonlinear function. Statistical linearization provides a formal basis on which to choose
a secant approximation at the expense of increased calculations. The advantage of statistical linearization over secant
linearization is a more precise estimate of the error associated with the approximation.

4.3 Signals and Block Diagrams with Causal Bond Graphs

A block diagram is a graphical representation of the functional input–output relationships in a system model. The blocks
represent functional relationships between input–output signals. A complete block diagram of a system may represent a
mathematical model; as such, the context here is not of conceptual or schematic diagrams. Rather, we emphasize opera-
tional block diagrams, as the blocks represent mathematical (causal) relations between the input(s) and output(s). However,
a block diagram does not enforce physical constraints inherent in a bond graph, and therefore they can be used to model
nonphysical systems such as computational elements, or even systems that are of an economic or social nature. Block dia-
grams are widely used in control engineering to convey the interaction of systems and their control elements and will be
used extensively in Chapter 7.
A standard form for a “block” in a block diagram is a functional operator, as shown in the first row of Table 4.1. This
operator block is not necessarily linear or time invariant. Inputs are indicated with a signal arrow and outputs with an
output signal arrow. In general, there are an infinite number of allowable functionals, some of which are nonphysical.
For representing physical systems, the functional may be as simple as a linear relation with a constant coefficient or a
generalized functional with multiple inputs and outputs, as shown in Figure 4.18. Part (c) of that figure uses a more compact

Table 4.1 Three basic block diagram elements used in physical and
functional block diagram descriptions.

Type Block Functional representation

Function y y = f (u)
u
f (·)

Summation u1 + y y = u1 − u2

−
u2
t
Integration y y = y(0) + udt
u ∫0
dt

(a) (b) (c)

Figure 4.18 (a) Generalized functional form of physical system model. (b) Block diagram form. (c) Block diagram with vector/matrix
notation.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
4.3 Signals and Block Diagrams with Causal Bond Graphs 171

vector/matrix notation that is common, especially in control system diagrams. From the basic blocks conveyed in Figure 4.1,
a wide range of physical and functional system models can be represented.
Note that summation includes both addition and subtraction, with signals to be subtracted indicated by a minus sign (−).
An integration block is also commonly indicated with the inverse of the derivative operator, s, i.e., 1∕s, a notation derived
from Laplace transform theory. Differentiation can also be included as a block type but it is not needed for causally correct
physical models and therefore is not included here.
A bond graph can include block diagram descriptions, especially since the latter may more effectively describe certain
types of systems of interest. It may also be the case that a block diagram description is already available from a previous
or collaborative study. More fundamentally, block diagrams can graphically convey information flow that is an essential
part of the model formulation. Recall that a causal stroke represents information about the direction of the bilateral signal
flows which can also represent power interactions, as shown in Figure 4.19(a). Also, a signal bond can graphically represent
unilateral (causal) information flow from a bond graph. For example, information about the effort or flow variable at a 0 or
1 junction, respectively, as shown in Figure 4.19(b), can be represented by signals. These signal bonds are inherently causal,
so causality is not assigned on these bonds. Further, they extract no power. The signal takes on the value of the variable as
imposed by the assigned causality.
It is common to use signal bonds in the way shown in Figure 4.19(b) to indicate the placement of ideal sensors that can
detect the relevant effort or flow variable without drawing power. The signal can then flow into a functional block that
models the sensor input–output characteristics. For example, the function may be an empirically derived model (static or
dynamic) for how a variable of interest relates to the sensed effort or flow variable. The implicit assumption that power flow
is zero (or at least negligibly small) for an ideal sensor or measurement can also be represented using ideal source models
as shown in Figure 4.20. An ideal flow source can model sensing effort while an ideal effort source models sensing of flow.
The effort or flow variables at junctions may also be needed for defining information states. Information state equations
are then simply q̇ = f or ṗ = e, where f and e are the flow and effort at a given 1- or 0-junction, respectively. In this case,
the q and p states do not arise from ESEs. It is not uncommon in mechanical systems to define position information states
arising from kinematics of mechanical elements, so displacement states may not always require presence of an elastic energy
storage element. Those states would be distinguished as information states as opposed to energetic states which arise from
energy storage elements.
The prototypical example of how an information state arises is for the position of a translational mass, as depicted
in Figure 4.21. Note that the signal bond has a full arrowhead (→) as opposed to the half-arrowhead of a power bond.
This is a relevant distinction since the signal bond conveys only information. In addition, only power bonds are assigned
causality – causality is not needed on a signal bond since the flow of information is implied by the arrow direction.

(a) (b)

Figure 4.19 (a) Causal power bond conveying bilateral information flow. (b) Extracting effort (e1 ) and flow (f1 ) information from
causal ideal junctions.

(a) (b)

Figure 4.20 An ideal sensor draws no power, es ⋅ fs = 0. (a) An ideal effort sensor measures es with negligible flow, fs , and is thus
represented by an ideal flow source. (b) An ideal flow sensor measures fs with negligible effort, es , which thus requires use of an ideal
effort source.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
172 4 System Model Formulation and Evaluation

(a) (b)

Figure 4.21 (a) Schematic of translational mass with applied force. (b) Bond graph with a signal bond used to show that no power
flows in the conveyance of information from the 1-junction. The integrating block identifies creation of an information state; that is,
one that is not related to an ESE, yet important for the purposes of this model.

The energy content of the system is described by the state equation for p,
ṗ = F, (4.20)
but there are many cases in which the position of the mass is also desired. Since the energy content of a basic translational
mass does not depend on its position, an additional information state x is introduced. The rate equation for this position
state is,
ẋ = 𝑣m = p∕m. (4.21)
The decision to include the information state can be depicted in the bond graph by a signal flowing from the 1-junction,
which has a flow causally indicated as the mass velocity, 𝑣m . An integrator block is used to identify creation of x, as shown
in Figure 4.21(b). Note that solution of the energetic state equation 4.20 does not require the simultaneous solution of 4.21.
This implies that the information state x can be determined after p(t) has already been completely determined. This is a
general feature of a basic information state. The need for information states can arise in different contexts. These states
can help track parts of a system that are not being considered in detail in a preliminary model. For example, consider the
hydraulic ram depicted in Figure 4.22(a). The model of the system is assumed to have linear friction between the piston
and cylinder. It is desired to have the position x of the ram as an output of the model. This state could be useful in tracking
the state of the hydraulic piston, for example, or to track the change in the contained volume. In this model, this variable
is an information state and it is indicated by the integrated velocity signal in Figure 4.22(b).
A set of state equations can be derived using standard techniques yielding,
ṗ = −Fb − Fp = −(b∕m)p − APs , (4.22a)

ẋ = 𝑣m = (1∕m)p, (4.22b)
where the equation for information state x is implied by the integration block in Figure 4.22.
Information states defined by integrating a flow variable result in variables that are physically intuitive. Specifically,
these are mechanical displacements or positions, fluid volumes, and so on. In contrast, integrating effort variables results
in momentum-type states in the domain of interest. Unless it is related to kinetic energy storage, it may not always be clear
that a momentum state is useful.
Signal flow is essential for not only drawing information from a bond graph but also imposing the value of a variable onto
a bond graph element, notably where there is no conveyance of power. This can be essential in representing many types of
physical and functional processes in a model.

Piston area,

(a) (b)

Figure 4.22 (a) Schematic of a hydraulic ram and (b) bond graph with signal bond to indicate information state, x.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
4.3 Signals and Block Diagrams with Causal Bond Graphs 173

This use of signal or information flow also arises when modeling modulation or when integrating a control system
description with a bond graph. Modulation will be discussed in Section 4.7 and integration of control system block diagrams
with bond graphs will be described in Chapter 7.
In the remainder of this section, the emphasis is on converting a causally complete bond graph into block diagram form.
The block diagram can be formulated from the system state equations or directly from a causal bond graph. The equivalent
block diagram is also convenient when integrating with existing functional input–output representations for parts of a
system, as well as when the intent is to analyze a model using one of many available block diagram description languages.
If state equations have already been derived from a bond graph, then converting those equations into block diagram form
requires use of the basic integrator, summing, and functional blocks. The focus here is on converting a causally complete
bond graph directly into block diagram form. In this case, the underlying relationship between basic bond graph elements
and their block diagram equivalents should be understood. A system that has n independent ESEs will have n integrator
blocks, identifying the system states. Additional integrator blocks should be added for information states.
An advantage in going directly from a bond graph to a block diagram is that, while not as compact as a form derived from
reduced state equations, the result can be easy to relate to the original bond graph. Any model updates or refinements can
be readily made, and it can be easier to communicate the model to another user along with the original bond graph. Block
diagrams can be difficult to debug. Further, the close relation of the bond graph can help avoid algebraic loops, which can
easily arise in block diagram model formation.
Causally complete bond graphs can be systematically converted into block diagrams by: (a) identifying the block elements
required, using an integrator for each ESE or information state, and (b) using a summation element to construct the effort
and flow relations at each 1 and 0 junction, respectively. Figure 4.23 illustrates this process for the example of the hydraulic
system previously discussed. For illustration purposes, all the basic elements are shown in isolation in Figure 4.23. These
block diagram elements can then be composed into a complete block diagram as shown in Figure 4.24.

Figure 4.23 Relation between block diagram and bond graph elements.

Figure 4.24 Block diagram directly from bond graph.

 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
174 4 System Model Formulation and Evaluation

Bearing
friction
(a) (b)

Figure 4.25 Permanent-magnet dc motor schematic and bond graph.

Example 4.8 Permanent-magnet dc motor bond graph and block diagram

A permanent-magnet dc (pmdc) motor is shown in Figure 4.25(a) with a corresponding bond graph with assigned causality
in (b). Here an information state is also defined for the angular position of the motor shaft, 𝜃m = ∫ 𝜔m dt, where 𝜔m is the
motor shaft speed. All the model elements are assumed to have linear constitutive behavior, except for the effective bearing
friction. For this case, we assume a nonlinear functional relation, Tf = Φ(𝜔m ). Develop a block diagram description by
converting the bond graph.
Solution
It is not always necessary to lay out the individual block elements required, but for convenience, these are provided in
Figure 4.26. These elements are used to compose the completed block diagram in Figure 4.27.
This block diagram has one nonlinear element representing the relationship between friction torque and the motor
shaft speed. In its current form, the block diagram provides a graphical depiction of a mathematical model for a
system. This form can be used directly in a block diagram computing environment for nonlinear simulation studies.
Alternatively, a linearized form can be generated and used to derive a linear transfer function form as discussed later in
Section 4.4.

It can be convenient to create reusable block diagram models within these modeling environments. A benefit of building
such models from a bond graph is that an emphasis can be placed on ensuring proper causal relations at the ports,
avoiding algebraic loops, and identifying augmentation as needed to avoid issues such as derivative causality and algebraic
loops. In such cases, it is best to re-examine the model assumptions and rebuild a model rather than to make ad hoc
adjustments.

Figure 4.26 Block diagram elements corresponding to elements of the bond graph for a pmdc motor.

Figure 4.27 Completed block diagram from causal bond graph of pmdc motor.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
4.4 Transfer Functions 175

4.4 Transfer Functions

Transfer functions are linear and causal functional relations between a system input, u, and a system output, y, shown in
block diagram form as follows:

u y
G(s)

where G(s) is a transfer function, G(s) = y∕u, being a function of the linear operator, s. Transfer functions are commonly
used in block diagram descriptions of physical system models and control systems. A linear transfer function can be derived
from mathematical models, denoting any derivative operations dn ∕dtn by the linear sn operator and integration by 1∕s. For
example, the second-order differential equation for a basic mass–spring–damper with force input is transformed into an
s-domain transfer function as follows:

mẍ + bẋ + kx = F (t)

F 1 x
(ms2 + bs + k)x(s) = F (s) ⇒
ms2 + bs + k
x =
G(s) = F 1
ms2 + bs + k

Using the s-operator allows for quick conversion of linear time-invariant ODEs from state space or nth-order form into
transfer function form. The resulting transfer functions can be used in block diagram descriptions. Block diagram algebra
can then be used to reduce the models into simpler form for analysis including available software programs. Rules for block
diagram reduction can be found in Ogata [53].

Example 4.9 Block diagram with feedback path

The block diagram shown below is a common form found in many physical and feedback control system models. This
diagram can be reduced into a single transfer function, G, relating the output, y, to the input, u, by using block diagram
algebra.

u + e y u y
G1 Simplify ⇒ G
−

b G2

The output is y = G1 e = G1 (u − b), where b = G2 y. Thus,

y = G1 (u − G2 y) = G1 u − G1 G2 y,

or,

(1 + G1 G2 )y = G1 u,

The transfer function, G, is then defined by,

y G1
=G= .
u 1 + G1 G2
This is a well-known form in block diagram reduction. In feedback control, the G is referred to as a closed-loop transfer
function.

The Laplace transform of a function of time can be used to transform differential equations into algebraic form. For a
[ ]
function f (t), the Laplace transform is  f (t) = f (s) and established theorems provide ways to obtain results similar to
the s-operator approach just described. For the purpose of physical system modeling as presented in this book, it is not
necessary to adopt Laplace transform methods.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
176 4 System Model Formulation and Evaluation

Ideal displacement
sensor
(a) (b)

Figure 4.28 Force measurement system: (a) schematic and (b) bond graph model with block diagram showing ideal measurement.

Physical system models in the s-domain can be readily derived by adopting the linear s-operator approach. As an example,
consider a force measuring system formed by a mass–spring–damper configuration as shown in Figure 4.28(a). Assume
that an ideal displacement measurement yields an output voltage related to mass displacement, xm , or eo = c ⋅ xm . We seek
a transfer function relating the output voltage eo and the applied input force F(t).
The bond graph in Figure 4.28(b) indicates two independent states, p and xk , which have the state equations,
ṗ = −b p∕m − kxk + F(t)
ẋ k = p∕m.
We can transform these equations into the s-domain, thus,
sp = −b p∕m − kxk + F
sxk = p∕m,
to allow algebraic derivation of a transfer function. First, note that p = m𝑣m = msxm , a relation indicated in block diagram
from in Figure 4.28(b). Thus, the first equation gives,
ms2 xm = −bsxm − kxk + F,
and the second, xk = xm , thus,
ms2 xm = −bsxm − kxm + F,
allowing us to solve for xm ∕F,
xm 1
= .
F ms2 + bs + k
The electrical output, eo , can then be found by multiplying this transfer function by c,
eo c
= .
F ms2 + bs + k
For systems of low order, this type of manipulation can usually be accomplished reliably. For more systems with higher
order (n ≳ 3), however, it is preferred to use a matrix approach.
From the linear system state and output equations,
Linear State: ẋ = Ax + Bu
Linear Output: y = Cx + Du,
begin by transforming the state equations into s-domain,
sx = Ax + Bu,
then rearrange, sIx − Ax = Bu, where I is the identity matrix, so [sI − A]x = Bu gives the states x,
x = [sI − A]−1 Bu = Gxu (s)u, (4.23)
where Gxu (s) is an n × r matrix of transfer functions between the n states and r inputs.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
4.4 Transfer Functions 177

If transfer functions are desired between the m outputs, y, and the system inputs u, then a matrix relation can be derived
using the relation for x equation 4.23, thus,
[ ]
y = Cx + Du = C[sI − A]−1 Bu + Du = C[sI − A]−1 B + D u = G(s)u.
where,
G(s) = C[sI − A]−1 B + D, (4.24)
represents an m × r matrix of transfer functions between the r inputs and m outputs. A single element of this transfer
function matrix can be put in the form of a ratio of polynomials in s,
N(s)
G(s) = , (4.25)
D(s)
where N(s) is the numerator polynomial,
N(s) = bk sk + · · · + b0 ,
and D(s) is the denominator polynomial,
D(s) = an sn + · · · + a0 .
The denominator polynomial D(s) is found from the determinant as shown above, and the degree of this polynomial is equal
to the number of states in the systems n. In general, for causally correct models, we also have the degree of the numerator
N(s) less than or equal to the degree of the denominator D(s) or k ≤ n.

Example 4.10 Transfer function of force-measuring system

For the force measuring system, we have previously derived the state equations in the matrix form as,
[ ] [ ][ ] [ ]
ṗ −b∕m −k p 1
= + F,
ẋ 1∕m 0 x 0
with an output equation which can be expressed in the matrix form as,
[ ]
[ ] [ ] p
eo = 0 c + [0] F.
x
Find the transfer function eo ∕F.
Solution
For this system, we first identify the linear state-space matrices:
[ ] [ ]
−b∕m −k 1
A= B=
1∕m 0 0
[ ] [ ]
C= 0 c D= 0 .
Now, apply equation 4.24 to find the implied transfer function,
[ ]−1 [ ]
eo [ ] s + b∕m k 1
= G(s) = 0 c + [0]
F −1∕m s 0
[ ][ ]
[ ] s −k 1
0 c
+1∕m s + b∕m 0
= + [0]
s2 + (b∕m)s + k∕m
c∕m
= 2 .
s + (b∕m)s + k∕m
As to be expected, this transfer function is identical to the one derived previously.

If transfer functions between the states and inputs are desired, then equation 4.23 can be used directly. Further, a specific
transfer function between the ith state and the rth input, ur , can be determined using Cramer’s rule [54] from,
xi |sI − A|(with ith column replaced by br ) Nir (s)
= = , (4.26)
ur |sI − A| D(s)
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
178 4 System Model Formulation and Evaluation

where | ⋅ | indicates a determinant, D(s) is the system characteristic equation, and br is the rth column of the B matrix.
To illustrate, consider the force measurement system example from above, where,
| [ 1 0 ] [ −b∕m −k ]| | s + b∕m +k |
| | | |
D(s) = |sI − A| = |s − |=| | = s(s + b∕m) + k∕m,
| 0 1 1∕m 0 | | −1∕m s |
| | | |
| 1 +k |
| |
Nir (s) = |sI − A|(with ith column replaced by br ) = | | = s.
|0 s |
| |
Thus,
p 𝑣 s∕m
= m = 2 .
F F s + (b∕m)s + k∕m
Dividing both sides by s and multiplying by m gives,
xm 1
= ,
F ms2 + bs + k
as derived before. This approach using Cramer’s rule is conducive to hand-based derivations and can be readily imple-
mented in a symbolic processor as n gets large.
Using s as a derivative operator, the transfer function 4.25 also represents the nth -order differential equation,
an y(n) + an−1 y(n−1) + · · · + a1 ẏ + a0 y = bk u(k) + bk−1 u(k−1) + · · · + b1 u̇ + b0 u.
For most cases, state equations, matrix transfer functions, and a sequence of nth-order differential equations (one for each
output) are functional equivalents as illustrated in Figure 4.29.3

Example 4.11 Transfer function for LRC circuit

Consider the linear LRC circuit shown in Figure 4.30(a) and its associated bond graph. Find transfer functions that relate
the voltages across the inductor, VL = 𝜆,̇ the capacitor VC , to the input, Vs (t).
Solution
The state equations in terms of flux linkage in the coil, 𝜆, and charge on the capacitor, q, are
[ ] [ ][ ] [ ]
𝜆̇ 0 −1∕C 𝜆 1
= + V (t).
q̇ 1∕L −1∕RC q 0 s

Figure 4.29 Functional equivalents for linear systems.

(a) (b)

Figure 4.30 (a) L–R–C circuit and (b) causal bond graph model.

3 Exceptions can occur for uncontrollable or unobservable systems for which there is pole-zero cancelation in the transfer function. In this case,
the state equations will be of higher order than the transfer function. Related discussion can be found in Takahashi et al. [55] and/or Schultz and
Melsa [56].
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
4.4 Transfer Functions 179

Choosing two outputs as the voltage across the coil, VL = 𝜆,̇ and the capacitor voltage, Vc = q∕C, gives,
[ ] [ ][ ] [ ]
VL 0 −1∕C 𝜆 1
= + 𝑣.
VC 0 1∕C q 0
We then have:
[ ] [ ]
0 −1∕C 1
A= B=
1∕L −1∕RC 0
[ ] [ ]
0 −1∕C 1
C= D= .
1∕L −1∕RC 0
and the transfer function matrix is,
[ ][ ]−1 [ ] [ ]
0 −1∕C s 1∕C 1 1
G(s) = +
0 1∕C −1∕L s + 1∕RC 0 0
[ ]
−1∕LC
[ ]
1∕LC 1
= 2 +
s + (1∕RC)s + (1∕LC) 0

[ ] ⎡ −1∕LC ⎤
VL ∕Vs ⎢ s2 + (1∕RC)s + (1∕LC) + 1⎥
G(s) =
VC ∕Vs
=⎢ 1∕LC ⎥.
⎢ 2 ⎥
⎣ s + (1∕RC)s + (1∕LC) ⎦
Note that the transfer function between VL and Vs can also be expressed as
VL s(s + 1∕RC)
= 2 .
Vs s + (1∕RC)s + (1∕LC)
Given this transfer functions, a second-order differential equation in terms of VL can be found as,

V̈ L + (1∕RC)V̇ L + (1∕LC)VL = V̈ s + (1∕RC)V̇ s .

Likewise, the transfer function between VC and Vs can also be used to find an alternative second-order differential
equation in terms of VC ,

V̈ C + (1∕RC)V̇ C + (1∕LC)VC = (1∕LC)Vs .

It is common to simplify block diagrams into reduced forms [53]. This can be convenient when the model is to be reused,
for example, or to derive an overall transfer function of a block diagram description. For example, the block diagram shown
in Figure 4.31(a) is obtained directly from the state equations derived previously for the system of Figure 4.22. The system
can alternatively be represented by the reduced block diagram with a transfer function that relates the input pressure, Ps ,
to output position, x, as shown Figure 4.31(b).
While the condensed form in Figure 4.31(b) is functionally equivalent to Figure 4.31(a), the former conveys more informa-
tion about how the physical model equations are formulated, and could thus be more readily extended. As such, reductions
are not always desirable. On the other hand, the transfer functions for a given linear system reveal the denominator as the
characteristic equation of a system, useful in linear analysis, as will be shown in Chapter 5. The characteristic equation
offers key insights into linear stability analysis as discussed in Section 4.5.

Ps + ṗ
1
p ẋ 1 x
A s 1/m s
−
Ps A/m x
b/m s(s + b/m)

(a) (b)

Figure 4.31 Block diagrams for hydraulic ram: (a) block diagram from state equations and (b) reduced block diagram.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
180 4 System Model Formulation and Evaluation

User-specified User-specified
filter load

Controller Amplifier &

Motor Sensor
Reference
command

Figure 4.32 Schematic of legacy motor-drive system with manufacturer specified models and user-specified requirements.

Example 4.12 Transfer function of legacy motor drive

A legacy motor drive needs to be evaluated to determine if it can be used in a prototype application. Some manufacturers’
information is available in the form of a system schematic shown in Figure 4.32, which includes ideal specifications for
available drive components. The user must specify additional components to complete the system. Here, an analog filter
circuit is required between the Controller and the power Amplifier & Motor. The user can also incorporate a model of a
typical rotational load, and here a basic inertia with linear damping is shown.
a) Develop an integrated block diagram of the system, using bond graph descriptions for the user-specified systems.
b) For each user-specified subsystem, develop a transfer function that can integrated into a final system block diagram.
c) Develop a block diagram and derive a transfer function that relates the shaft position, 𝜃L to the input, V1 . The results
from Example 4.9 can be useful in deriving this result.
Solution (a)
In forming a model to represent a system composed from the components provided, a first key assumption is that the
Controller (K1 ), Amplifier & Motor (K2 ), and Sensor (K3 ) can be represented by the implied ideal relationships. This means,
for example, that the constant gain values given can be used in block diagram descriptions. Note also that the sensor provides
a measure of position so it is implied that an integrating function is required to give 𝜃L from 𝜔L and can be represented here
by 1∕s.
The diagram in Figure 4.33 shows a first step in integrating the block diagram elements with bond graph models of the
user-specified subsystems.
Note how signals are used in two ways. The output from K1 and K2 shown a signal flowing into a 0-junction to specify
the effort at that junction. This conveys the assumption that these ideal elements can deliver these efforts regardless of the
power flow. Those are assumptions that could be revisited in a later model if necessary. The input to K2 , Vb = Vf , is also a
signal, which indicates K2 is ideal amplifier model (negligible current draw). Lastly, the sensor model is assumed to draw
no power from the load.
Solution (b)
Each of the user-specified physical elements can be modeled using the causal bond graphs shown. For the analog filter,
there is a single state, qf , with state equation,
q̇ f = iRf = (Va − Vf )∕Rf .

R : Rf C : Cf I : JL

VRf iRf Vf q̇f ḣL ωL

r + r − V1 Va Va Vf Vb τm τm ωL 1 θL
K1 0 1 0 K2 0 1 s
− iRf
τBL
V1
R : BL

K3

Figure 4.33 Block diagram with bond graphs of physical models.

 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
4.5 Stability of Physical Systems 181

r + r − V1 Va 1 Vb τm 1 ωL 1 θL
K1 Rf Cf s + 1 K2 JL s + BL s
−

V1

K3

Figure 4.34 Final form of block diagram.

Recall Vf = qf ∕Cf so we can rewrite to get the desired output as Vf ,

Rf Cf V̇ f = −Vf + Va ,

Rearranging, then transforming into s-domain, we can get the transfer function desired,
Vf 1
= .
Va Rf C f s + 1
A similar process can be followed with the JL − BL load to find,
𝜔L 1
= .
𝜏m J L s + BL
Solution (c)
The two transfer functions from part (b) can now be integrated into a final system block diagram as shown in Figure 4.34.
The output in this model is 𝜃L and input is r, and the closed-loop transfer function relation found in Example 4.9 can be
used,
𝜃L G1
=G= ,
r 1 + G1 G 2
where here G1 is the product of all the block diagram relations in the forward path and the feedback gain is G2 = K3 . Here,
K1 K2 1 K1 K 2
G1 = ⋅ ⋅ = .
Rf C f s + 1 J L s + B L s s(Rf Cf s + 1)(JL s + BL )
The final closed-loop transfer function is then,
𝜃L G1
=G= ,
r 1 + G1 G 2
or,
𝜃L K1 K2
= .
r s(Rf Cf s + 1)(JL s + BL ) + K1 K2 K3

4.5 Stability of Physical Systems

Stability is an important property of physical systems with practical relevance to design and control. It is also important to be
mindful of how stability characteristics arise in physical system model formulation. Key model features such as parameter
values and functional trends in constitutive relations influence stability properties. It is possible to build an understanding
for how physical stability behavior relates to analytical model formulations for low-order linear systems. These insights can
help avoid errors during the model formulation itself.

4.5.1 Stability Concept

Stability describes the tendency for a system to persist in an equilibrium state so long as it is undisturbed. In this way, a
system is stable if it returns to the equilibrium state after being disturbed slightly and then released. We distinguish between
static and dynamic stabilities.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
182 4 System Model Formulation and Evaluation

Statically unstable
about

Statically unstable Statically stable

Statically stable Neutrally stable
about

(a) (b)

Figure 4.35 (a) Simple pendulum in stable and unstable equilibrium states. (b) Three cases of a marble in neutral, unstable, and
stable equilibrium states.

Consider a simple pendulum which has equilibrium states at 𝜃 = 0 and 𝜃 = 𝜋. As shown in Figure 4.35(a), the pendu-
lum behavior about 𝜃 = 0 represents a statically stable system, while the inverted pendulum is understood to be statically
unstable about the equilibrium at 𝜃o = 𝜋. The inverted pendulum quickly “falls” when given the slightest deviation from its
equilibrium. We can introduce the concept of neutral stability for systems that would neither return to nor move away from
an equilibrium state when disturbed. The physical mechanism at work is the gravitational restoring torque, 𝜏𝜃 = mgl sin 𝜃,
which when linearized (about 𝜃 = 0) takes the form, 𝜏𝜃=0 = mgl𝜃 = k𝜃 𝜃, where k𝜃 = mgl (for equilibrium at 𝜃 = 0) is an
effective stiffness. This is similar to the linear spring (restoring) force, Fk = kx, where x is the displacement from an equi-
librium. The physical effects can be related to the stability concepts commonly illustrated by a marble on the three surfaces
shown in Figure 4.35(b). The force on the marble induces an effective stiffness, not unlike the pendulum effect, that dic-
tates whether the state is stable. If the effective stiffness in any of these systems becomes negative, the system is statically
unstable.
A system that is statically stable is not necessarily dynamically unstable, as it is often the case that different physical
effects influence the two properties. A useful way to think about how static stability is different from dynamic stability is
to consider the basic mass-spring-damper system, m̈xm + bẋ m + kxm = 0. If b > 0, then power dissipated, d = bẋ 2m > 0, so
energy leaves the system when b > 0, and the system is dynamically stable. On the other hand, an apparent b < 0 would
imply that energy is being injected into the system, typically by drawing it from the environment. This would make the
system dynamically unstable.

4.5.2 Stability and Eigenvalues

Physical insight into stability can be developed for low-order linear systems (n < 3). As systems become more complicated,
insights into the influence of system model parameters on stability characteristics must usually be gained from analytical
descriptions. Consider the linear first-order system in standard form, 𝜏t ẋ + x = u(t), where 𝜏t is the time constant.4 This sys-
tem has the characteristic equation 𝜏t s + 1 = 0 which gives a single eigenvalue, s = −1∕𝜏t . Recall the homogeneous solution
is then xh (t) = Aest . So as long as 𝜏t is positive in this case, the system is dynamically stable. Since 𝜏t is related to physical
parameters of a system, it is clear that as long as those parameters are positive then a dynamic instability will not arise in
any first-order system. In addition, one should see how the terms in the first-order equation convey sign convention and
an implicit check on correctness of the derivation. That is, if the equation was derived or presented as 𝜏t ẋ − x = u(t) and if
the system is known to be stable then clearly an error has been made in the model formulation.
Consider now the second-order system in standard from, ẍ + 2𝜁𝜔n ẋ + 𝜔2n x = 𝜔2n u(t), where 𝜁 and 𝜔n are the damping
ratio and undamped natural frequency, respectively, and u(t) is an input. The characteristic equation is now a quadratic
equation,

s2 + 2𝜁𝜔n s + 𝜔2n = 0, (4.27)

4 Analysis of linear systems will be more fully discussed in Chapter 5.

 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
4.5 Stability of Physical Systems 183

so the two eigenvalues can be found as,

√ √
s1,2 = −𝜁𝜔n ± 𝜔n 𝜁 2 − 1 = −𝜁𝜔n ± j𝜔n 1 − 𝜁 2 , (4.28)
where the latter form is convenient for highlighting how 𝜁 < 1 leads to eigenvalues that are complex conjugates. Now the
homogeneous solution is,
xh (t) = A1 es1 t + A2 es2 t ,
and we see here that as long as 𝜁 and 𝜔n are both positive the stability of a second-order system is ensured. Again we see
that the 𝜁 and 𝜔n parameters depend on the physical parameters in a system (e.g., resistance, damping, mass, and stiffness),
which should all be positive for stability. Again for this form of the model a negative coefficient in the characteristic equation
clearly indicates either an unstable system or that an error has been made in the model formulation (e.g., model parameter
definition and sign convention).
It is common to examine these concepts using system transfer functions, which are a more convenient form since the
denominator, D(s), is the system characteristic equation. For example, we can compare a simple (undamped) pendulum
and a mass–spring–damper in transfer function form:

Simple pendulum Mass-spring-damper

𝜃 1 xm 1
= 2 =
𝜏 s + (g∕l) F ms2 + bs + k
D(s) = s2 + (g∕l) D(s) = ms2 + bs + k

where 𝜏 and F are input torque (at pivot of pendulum) and force (on mass), respectively. The system eigenvalues are found
by setting D(s) = 0. In these cases, the two eigenvalues are (in physical parameter and standard form terms):

Simple pendulum Mass-spring-damper

√ √
[ ]2
g b k b
s1,2 = ±j s1,2 =− ±j −
l 2m m 2m
√
s1,2 = ±j𝜔n s1,2 = −𝜁𝜔n ± j𝜔n 1 − 𝜁 2

It is common to plot the eigenvalues, or poles, of the system characteristic equation in the complex or s-plane as shown
in Figure 4.36. For s = 𝜎 + j𝜔, the real part 𝜎 is plotted along the horizontal axis and the imaginary component 𝜔 is plotted
along the vertical complex axis. In this way, it is clear that systems with eigenvalues on the left-hand plane (negative real
part) are stable. This is the case for the undamped stable pendulum, which has complex conjugate eigenvalues for motion
about 𝜃o = 0, as shown in Figure 4.36(a). For the inverted pendulum (𝜃o = 𝜋), there would be two real eigenvalues and one
would be positive as shown in Figure 4.36(a), indicating instability.

Increasing
Statically stable
about
Statically unstable
about

Increasing

(a) (b)

Figure 4.36 (a) Pole locations for the simple pendulum (stable) and inverted simple pendulum (unstable). (b) Poles for a mass–spring
damper showing effect of increasing stiffness k as well as unstable case when k < 0.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
184 4 System Model Formulation and Evaluation

For the mass–spring–damper system, the two eigenvalues will have a negative real part as long as b > 0 and k > 0.
Figure 4.36(b) shows how the eigenvalues (or poles) change as k varies from a value of 0 to large values. The poles would be
equal when k∕m = b∕2m and then diverge into complex conjugate values. Only if k < 0 is the system stable. These trends
in the poles or root loci are commonly used in control system analysis where the effect of changes in control gains and the
influence on stability and performance are of interest.
For the mass–spring–damper, either b < 0 or k < 0 will indicate the system is not stable. The case of an inverted pendu-
lum would lead to an apparent negative stiffness, k𝜃 = −mgl, a statically unstable system. There are many other physical
examples commonly used to describe how a negative stiffness (or compliance) leads to static instability. Later in Section 4.6,
we will argue that a negative stiffness, for example, can be rejected as a viable model on the basis of passivity. Passive sys-
tems, composed of elements with positive parameters and thus passive elements, store or dissipate energy but they don’t
have elements that supply energy that was not previously stored.

4.5.3 Physical Instabilities

Many physical instabilities are often explained by trends in constitutive behavior that indicate apparent negative capaci-
tance or resistance. The inverted pendulum, for example, appears to have a negative stiffness, but fundamentally of course
this is due to a linearization for small motion. More will be said in Section 4.6 about the need in some cases to more fun-
damentally examine the physical behavior to arrive at a proper constitutive model. Nevertheless, it can be informative to
examine how some of these models or effects arise in physical system models.
One example is the friction force induced between surfaces, described by the system in Figure 4.37(a). The bond graph
of Figure 4.37(b) shows how the friction is modeled by a nonlinear R element, with associated slip velocity defined by the
moving belt and the mass, 𝑣f = 𝑣m + 𝑣belt . The induced friction force shows a negative trend versus the slip velocity, 𝑣f ,
as shown in the graph of Figure 4.37(c). This can be contrasted with the typical positive trend for linear friction with slip
velocity. As discussed by Rabinowicz [57], the interaction between mass and belt induces what is referred to as stick-slip
dynamics or “chatter” as a result of the induced trend of the friction force with slip velocity. Similar trends can be seen in
other friction force models, including the friction-slip trends used to model traction forces in tire-ground contact [58].
An apparent negative stiffness is sometimes used to explain some fluid–structure interactions commonly associated with
observed unstable behavior. For example, the mechanical stiffness of a valve spring can combine with a flow-induced pres-
sure around a closing valve to create a net force–displacement characteristic, Fn , on the valve mass that has a negative
stiffness as illustrated in Figure 4.38(a). The bond graph in (b) includes some effective fluid damping as well. While such
a model is common for explaining the source of the instability through the negative slope in Figure 4.38(c), we prefer
constitutive relations that do not combine effects (see Section 4.6).
Note also that in many instances negative capacitance or resistance commonly arise also because of the presence of a
source of energy. As such, a more fundamental model will reveal this effect. For example, the mass–spring system released
on the belt in Figure 4.37 when it is not moving will come to a stop; chatter would only arise if the belt was driven by an
outside source.
In summary, the stability characteristics for low-order systems are readily evident from the characteristic equations,
which should have positive coefficients for a stable system. Otherwise the system is unstable. The foregoing has focused on
low-order systems, for which it is easy to either argue from a physical perspective or by solving for the eigenvalues directly

Linear friction
with positive slope

Positive dry friction

Friction
with negative slope

(a) (b) (c)

Figure 4.37 (a) Dry friction between mass and driven belt. (b) Bond graph showing induced relative friction velocity, 𝑣f . (c) Trend in
friction force versus slip velocity compared to linear trend.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
4.5 Stability of Physical Systems 185

Linear mechanical
Force Spring spring force
force

Net force
with negative slope

Closed position
Fluid pressure
force
(a) (b)
(c)

Figure 4.38 (a) Valve closure under the combined effect of spring and flow-induced forces. (b) Bond graph showing combined effect
of spring and flow forces in effective C, including some effective damping with R. (c) Description of force–displacement trends
emphasizing the negative slope induced in effective stiffness by flow-induced forces.

that a system is stable or not. In order to understand how system characteristics influence stability for higher-order linear
systems (n ≥ 3), we introduce the Routh criterion.

4.5.4 Routh Stability Criterion

E.J. Routh [59] and A. Hurwitz [60] each published in the late 1800s independent methods for investigating the stability
of a system based on the characteristic equation. The Routh–Hurwitz stability criterion can tell us whether or not the
characteristic equation has any roots in the right-hand complex plane. It is a necessary and sufficient condition for a linear
dynamic system to be dynamically stable if all its poles lie in the left half of the s-plane. The method was originally developed
using determinants but is now used primarily in an array formulation. The method calls for ordering the coefficients of the
characteristic equation from the form,

an sn + an−1 sn−1 + · · · + a1 s + a0 = 0, (4.29)

into an array,

sn an an−2 an−4
n−1
s an−1 an−3 an−5
⋮ b1 b2 b3
⋮ c1 c2 c3

where the elements follow the relations given below:

a a − an an−3
b1 = n−1 n−2
an−1
an−1 an−4 − an an−5
b2 =
an−1
b1 an−3 − an−1 b2
c1 =
b1
b1 an−5 − an−1 b3
c2 =
b1
The criterion then states that the roots of the characteristic equation will have negative real parts (and thus stable) if and
only if the elements of the first column of the Routh table have the same sign. The number of sign changes is equal to the
number of roots with positive real parts. This is sufficient for determining absolute stability; that is, is the system stable or
not. Questions about relative stability which address how close a system is to being unstable or about the oscillatory nature
of a system can also be addressed [53]. Relative stability is more commonly examined using root locus techniques, which
will be briefly discussed in Chapter 7.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
186 4 System Model Formulation and Evaluation

Example 4.13 Routh table examples

Use the Routh–Hurwitz criterion to determine the stability of a system with the following characteristic equations.

Example 1 s3 + 4s2 + 8s + 12 = 0.
Solution
The coefficients are computed as follows:
4 ⋅ 8 − 1 ⋅ 12
b1 = =5
4
b2 = 0
5 ⋅ 12 − 1 ⋅ 0
c1 = = 12
5
c2 = 0

The Routh table is:

s3 1 8 0
2
s 4 12 0
s1 5
s0 12

Since there are no sign changes in the first column, all the roots of the characteristic equation have negative real parts
and the system is stable. A numerical root solver confirms all the roots have negative real parts.

Example 2 Determine the range allowed for K to ensure stability for the characteristic equation: s3 + 3s2 + 3s + 1 + K = 0
Solution
The Routh table can be filled in as follows:

s3 1 3 0
s2 3 1+K 0
1
s (8 − K)∕3 0
s0 1+K 0 0

We require that both of the following conditions hold for stability,

8−K >0

1+K >0

therefore,

−1 < K < 8.

This example demonstrates how the Routh–Hurwitz criterion is useful in determining the range of allowable system
parameters to ensure stability.

Special case: a first-column term in any row is zero. If the first term in a row of the Routh table is zero but other terms in
that row are not, or there are no other terms, then the zero term can be replaced by a very small value represented by 𝜖 so
the table can be completed. The stability criterion is then applied taking the limit 𝜖 → 0. For example, for a characteristic
equation,

s3 + 2s2 + s + 2 = 0,
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
4.5 Stability of Physical Systems 187

the Routh table should be formed as:

s3 1 1 0
s2 2 2 0
s1 0≈𝜖
s0 2

In such a case, if the sign of the coefficients in the first column above and below the 𝜖 term are the same this indicates
the characteristic equation has a pair of imaginary roots. The numerical example above has a pair s ± j.
If the sign above and below 𝜖 are opposite, this indicates a sign change, indicating there are two sign changes and thus
two roots in the right-hand plane. The system would be unstable. Another case that can arise is that all the coefficients in a
row of the Routh table are zero. This case indicates that there are roots of equal magnitude that lie radially opposite in the
s-plane. This can reflect the case seen for the inverted pendulum in Figure 4.36(a), with two real roots on the real line or
simply two complex conjugate roots.
The application of the Routh–Hurwitz criterion for some of these special cases can be found in references such as Cannon
[54], Ogata [53], or Dorf and Bishop [61], while good applications to stability of linear mechanical dynamics can be found
in Den Hartog [62]. Application of the Routh criterion to feedback control will be discussed in Chapter 7.

Example 4.14 Stability of legacy motor drive system

Use the Routh criterion to assess the stability of the legacy motor drive system in Example 4.12.
Solution
The transfer function relating the output rotational position, 𝜃L , to the input reference command, r, was found as,
𝜃L K1 K2 K1 K2 ∕(JRC)
= = 3 .
r s(Rf Cf s + 1)(JL s + BL ) + K1 K2 K3 s + (B∕J + 1∕RC)s2 + (B∕JRC)s + K1 K2 K3 ∕(JRC)
Stability can be assessed from the characteristic equation,

s3 + (B∕J + 1∕RC)s2 + (B∕JRC)s + K1 K2 K3 ∕(JRC) = 0.

To simplify, write as,

a3 s3 + a2 s2 + a1 s + a0 = 0,

so the Routh table becomes:

s3 a3 a1 0
s2 a2 a0 0
a2 a1 − a3 a0
s1 0
a2
s0 a0

Therefore, stability requires:

a2 = B∕J + 1∕RC > 0

a2 a1 − a3 a0 (B∕J + 1∕RC) − (B∕JRC)(K1 K2 K3 ∕(JRC))

= >0
a2 B∕J + 1∕RC

a0 = K1 K2 K3 ∕(JRC) > 0.

The first relation is satisfied for positive parameters, and usually gain values are positive so the last will be satisfied by
this constraint, K1 K2 K3 > 0. The second relation imposes the additional constraint for stability:

(B∕J + 1∕RC) > BK1 K2 K3 ∕(JRC)2

 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
188 4 System Model Formulation and Evaluation

The last two relations combine to constrain K1 K2 K3 ,

(JRC)2 ( B 1
)
0 < K 1 K2 K3 < + ,
B J RC
to ensure stability.

4.6 Constitutive Structure

One goal in modeling a physical system is to obtain functional state equations. In this section, we review concepts that
support this goal when using a bond graph approach. We want to ensure that, within a known bound of utility and validity:
(1) all bond graph models should have physical realizations, and (2) all physical models should have bond graph realizations
[63]. In practical terms, item (1) means if a bond graph is drawn, a physical system that behaves in the same fashion as the
model could be constructed, while item (2) implies if a physical device is built, then a bond graph could be developed
that models this device. Since models are always approximations to physical systems, statements (1) and (2) can never be
satisfied exactly, but the error bound due to the approximate nature of the model and the bound of validity of a well-defined
model should be known. The acceptable bounds depend on the intended use for the model.
As we reject the concept of a perfect model or modeling structure, this goal can never be attained, but this does not
diminish the importance of trying to achieve it. Part 1 of this goal implies that additional (model) structure be imposed on a
modeling method in order that no physical principles are violated. However, if the method is overly restrictive, then Part 2
of the stated goal will be impossible to realize. This section discusses restrictions on the constitutive relations for the energy
storing (I and C) and dissipative elements (R) in order to balance the two somewhat opposing statements given above.
These bond graph elements are not completely specified until these constitutive relations are given. A logical question
then arises – Is there any additional structure that can be imposed on the allowable forms for these relations from physical
principles? If the answer to this question is in the affirmative then, in keeping with the stated goal, this additional structure
should be enforced in order to automatically disallow any nonphysical models. The concepts and methods introduced in this
section are particularly helpful in evaluating constitutive behaviors that deviate from well-known and established physical
relations.

4.6.1 Energy-Storing Constitutive Relations

As discussed in Chapter 3, bond graph elements can represent physical elements that have single or multiple ports through
which power can enter and leave. The concepts introduced here assume a general multiport form, however multiport
modeling will be more fully described in Chapter 8. The ideas, however, lend insight into single port elements as well.

4.6.1.1 Energetic Structure

The energy-storing elements I and C both imply the existence of a unique energy state function (x) when x is generalized
displacement q for the C element and generalized momentum p for the I element. A general energy storing element can
be expressed by the equations:
ẋ i = ui (rate relations), (4.30)

∑
n
(x) = yi dxi (energy function), (4.31)
∫ i=1

𝜕
yi = Φsi (x) = (constitutive relations). (4.32)
𝜕xi
These are in integral form when the state rate relations are determined by the causal inputs ui , where i is the port index
i = 1, … n. In order for an energy function (x) to exist, the constitutive relations Φsi () must be single-valued vector func-
tions of the state vector x. A second restriction for the constitutive relation can be found by differentiating the ith output,
yi , with respect to the jth state to obtain,
𝜕yi 𝜕2 
= , (4.33)
𝜕xj 𝜕xj 𝜕xi
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
4.6 Constitutive Structure 189

(a) (b)

Figure 4.39 (a) Cantilever beam loaded by force and torque at the tip. (b) Modeled as a 2-port C with translational and rotational
ports.

and differentiating the jth output with respect to the ith state to obtain,
𝜕yj 𝜕2 
= , (4.34)
𝜕xi 𝜕xi 𝜕xj
and, since the order of differentiation is interchangeable, 4.33 and 4.34 can be equated to yield
𝜕yi 𝜕yj
= . (4.35)
𝜕xj 𝜕xi
Equation 4.35 represents a reciprocity constraint on the constitutive relation often called Maxwell reciprocity after J. C.
Maxwell. Maxwell reciprocity provides a check on proposed constitutive relations, while Equation 4.32 is commonly used
to derive constitutive relations based on a known energy function (x) (as will be shown in Chapters 8 and 9).
Linear C Element: Consider a linear n-port C with a constitutive relation in the form

e = Kq. (4.36)
T
Maxwell reciprocity requires that K = K or K is symmetric. As an example consider the cantilever beam shown in
Figure 4.39.
Assuming elementary linear beam theory and neglecting kinetic energy storage and dissipation, the constitutive relation
between the efforts (force F and torque 𝜏) and the displacements position (x and angle 𝜃) can be expressed as,
[ ] [ ][ ]
F 2EI 6 −3L x
= 3 , (4.37)
𝜏 L −3L 2L2 𝜃
where E is the elastic modulus and I is the area moment of the beam [15]. Notice that the matrix is symmetric. The energy
stored in the beam can be found from this constitutive relation by integration of the force constitutive relation as
6EI 2 6EI
(x, 𝜃) = Fdx + 𝜑(𝜃) = x − 2 x𝜃 + 𝜑(𝜃), (4.38)
∫ ∫ L3 L
where 𝜑(𝜃) is an arbitrary function of 𝜃 which is determined by using the 𝜏 torque constitutive relation as
𝜕 6EI 6EI 4EI
𝜏= = − 2 x + 𝜑′ (𝜃) = − 2 x + 𝜃. (4.39)
𝜕𝜃 L L L
This result implies that 𝜑(𝜃) = 2EI𝜃 2 ∕L and the energy can be expressed to within an arbitrary constant as,
6EI 2 6EI 2EI 2
(x, 𝜃) = x − 2 x𝜃 + 𝜃 . (4.40)
L3 L L
The two-port constitutive relation 4.37 can be used to derive the equivalent spring rate of a cantilevered beam to a variety
of torsional boundary conditions. For example, for the so-called clamped–clamped beam, the angular twist 𝜃 is constrained
to be zero and the force deflection relation becomes
12EI
F = 3 x. (4.41)
L
If the torque is zero at the right-hand side, or a clamped-free beam, then the force deflection can be derived from 4.37 as
3EI
F= x. (4.42)
L3
with the resultant twist at the end given as,
3x
𝜃= . (4.43)
2L
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
190 4 System Model Formulation and Evaluation

Figure 4.40 (a) Constitutive relation with negative compliance relations. (b) C
element with input flow source, f (t).

(a) (b)

4.6.1.2 Passivity Structure

Consider a linear 1-port C element with negative compliance in the constitutive law, i.e.,
e = q∕C, C < 0, (4.44)
as illustrated in Figure 4.40(a). We will presently show that such an element should not be used for a physical model. Such
an element does not violate the energetic constraints for an energy storing element given above. Nevertheless, this element
is capable of delivering an infinite amount of energy for any finite initial displacement q0 .
To show this is the case, consider the energy stored,
(q) = Cq2 ∕2. (4.45)
If a source of positive flow is attached to this C element as shown in Figure 4.40(b), the state equation would be,
q̇ = f (t), (4.46)
If the initial displacement is q0 , then the initial energy stored in the C is 0 = q20 ∕2C.For t > 0, q(t) > q0 and the energy
stored in the C is t = q(t)2 ∕2C < q20 ∕2C = 0 . This implies that the stored energy decreases monotonically in time as the
displacement increases. The decrease in stored energy in the C element represents an energy gain for the flow source. This
process is continued in time as q increases in order to extract an unbounded amount of energy from the C element.
Although a C element with available energy like the one above would be nice to have during an energy crisis, such
physical devices do not exist, at least not for engineering systems. This leads to the following definition of passivity [64].

passivity: An n-port system is termed passive if the energy that can be extracted from the ports is finite. If a system
is not passive, it is active.

In terms of a bond graph model, this implies,

{ }
T∑n
A (x) = sup − y u dt < ∞, (4.47)
∫0 i=1 i i

where the supremum is for T ≥ 0 and all admissible inputs u and initial states x. A is called the available energy of the
multiport.
Assuming that all activation is due to boundary effects in the form of source elements, then a bond graph model with its
source elements deleted must necessarily be passive for a physical model. For a single energy storing element, this implies
that the stored energy  is bounded from below. It also can be shown that if each element of a model is passive then the
entire model is passive. This is a sufficient, but not necessary condition as will be shown by example later. In Figure 4.41,
examples of active and passive energy functions are given. Note that the active energy function is not bounded from below,

Figure 4.41 (a) Active and (b) passive energy functions.

Active

Passive

(a) (b)
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
4.6 Constitutive Structure 191

while the passive energy function is bounded from below. If one were to place a ball on the contour formed by the energy
function of the active energy storing element, this ball could continually fall “downhill.” This will not occur for a passive
energy function.
Passivity is strongly related to stability of a physical system. As an example, consider a spring mass system with a linear
mass and a negative linear spring. A second-order differential equation for the displacement x of the spring mass system
can be expressed as
m̈x − Kx = 0, (4.48)
where m and K are translational mass and spring rate, respectively. The solution to 4.48 is
√ √
x(t) = Ae− K∕mt
+ Be+K∕mt
, (4.49)
which consists of a decaying exponential and a growing exponential. The growing exponential indicates an unstable system.

4.6.1.3 Causal Structure

An energy-storaging element with integral or independent causality functionally requires the existence of a single-valued
function Φs (⋅) in order to have a unique output. On a physical basis, Φs (⋅) must also be unique in order for the stored energy
(x) to exist. Since the basis of bond graphs depends on the existence of an energy state function, constitutive laws with
nonexistent energy functions have been disallowed as previously discussed. This is in spite of the fact that an energy storing
element with derivative or dependent causality only functionally requires the existence of the inverse Φ−1 s (⋅). Constitutive
laws of the form shown in Figure 4.42(a) are rejected on a physical, not a functional or causal basis.
Consider next the constitutive relation in Figure 4.42(b). Although this constitutive relation has a unique energy
function, this relation does not have a unique inverse, so the coenergy does not uniquely exist. Energy storing elements
with nonunique Φ−1 s (⋅) must have integral causality. This restriction can severely impact the flexibility of a model. As an
example, consider two springs in Figure 4.43 connected in series which have constitutive relations like the one shown
in Figure 4.42(b). Due to the structure of the model, causality assignment will lead to one of the two C elements being
dependent. This implies a nonunique solution to the functional model equations due to the constitutive relation and
represents a pathological causal conflict. Models of this type should be rejected. It is preferred that energetic constitutive
relations be basic, as shown in Figure 4.42(c).

Non-unique
energy Composite Basic

(a) (b) (c)

Figure 4.42 Causal classification of constitutive relations: (a) non-unique (b) composite (c) basic.

Figure 4.43 Nonunique spring model.

 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
192 4 System Model Formulation and Evaluation

In order to better understand the origin of elements with nonunique inverse relations, consider a C element with a
Lennard–Jones potential function used to represent intermolecular forces between spherical molecules. This function can
be expressed as
[ ( ) ( )1 ]
q0 6 q0
(q) = 𝜖 −2 + 2 , (4.50)
q q

where 𝜖 and q0 are constants and q is a one-dimensional intermolecular separation distance. The 6th-power term represents
attraction and the 12th-power term is repulsion. The force or effort can be found as
[( ) ( )1 ]
𝜕 12𝜖 q0 6 q0
e= = − 2 . (4.51)
𝜕q q q q

As seen in Figure 4.44, this represents a constitutive relation for a C element for which Φ−1
s is not unique. If one tried to
input an effort e into a C with this constitutive relation, there would not be, in general, a unique displacement. A model
with this restriction is, in fact, not necessary.
Although it is true that the Lennard–Jones potential has a nonunique inverse relation, this is due to the incorporation
of two physical effects in one element. The 6th-power attractive effect is attributed to the perturbations of the electron
cloud of the two molecules. The 12th-power repulsive part of the potential is due to an approximation to the interelectronic
forces governed by the Pauli exclusion principle. Since there are two separate effects, it seems more logical to model the
Lennard–Jones potential with a C-field containing two basic C elements, one for each effect. The bond graph model with
separated attraction and repulsion effects is shown in Figure 4.45.
The constitutive and energy relations for this model are:
( )6
12𝜖 q0
ea = , (4.52)
qa qa
( )6
q0
a = −2𝜖 , (4.53)
qa
( )12
12𝜖 q0
er = − , (4.54)
qr qr
( )12
q0
r = 𝜖 . (4.55)
qa

Figure 4.44 Lennard–Jones potential and force.

C Figure 4.45 Two element field model.

ea q̇a

e er
1 C
q̇r
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
4.6 Constitutive Structure 193

The advantage of this field model is the indifference to the causality applied to the model since the constitutive rela-
tions are bijective (both Φs and Φ−1 s exist for all values of their argument). A seeming disadvantage is the necessity of
two displacement states qa and qr rather than just one, but when derivative causality is applied to a C element with the
constitutive relation 4.51 another state is in fact necessary in order to uniquely determine the displacement (i.e., for the
effort–displacement curve shown in Figure 4.44 a state is needed in order to know which branch of the curve the system is
on when effort is an input to the element).
It should be noted that even though the attractive C element in the field model is active5 (Ea is not bounded from below),
the field model is passive since the total field energy  is bounded from below. It also should be noted that the repulsive C
element is passive even though its entire constitutive relation lies in the fourth quadrant of the e − qr coordinate axes.
In light of the above example, the supposition is made that all physical models should be indifferent to causal orientation
since this orientation is imposed on the system by the environment through boundary conditions which can, in general,
change from time to time. Causal indifference can be achieved for the energy-storing elements if each element models
only one important physical effect such that the constitutive relation Φs is bijective. In fact, the nonuniqueness of the
inverse relation is taken as a sign that important multiple physical effects are modeled in the same element. This is not
to say that one element cannot be used to model the space average of many microphysical effects, as this is true for any
lumped physical model. The argument is only made that effects which cause nonunique inverses would be better modeled
in separate elements. These requirements lead to the following definitions:

Basic Energy-Storing Element: An energy storing element is basic if Φs is bijective. If an energy storing element is
not basic, it is composite.

Energy Storing Field: A bond graph model containing basic energy storing elements and junction structure elements
(0, 1, T, G) where the individual energy elements may be active but the entire model is passive.

Since models with composite elements can have potential causal problems, their use should be avoided. These composite
energy-storing elements can be modeled with an energy storing field to alleviate these problems. The combined C ele-
ments proposed for the behavior in Figure 4.44 suggest a similar approach could be taken for the effective combination of
spring-like forces arising on a valve as shown in Figure 4.38.

4.6.2 Dissipative Constitutive Relations

4.6.2.1 Entropic or Passivity Structure
Modeling with a dissipative element R implies production of entropy. The constitutive relation for the dissipative element
can be expressed as,
fs = e ⋅ f∕T (entropy production), (4.56)

yi = Φli (u, T) (constitutive relation), (4.57)

where fs is the entropy flow or production rate, e and f represent the efforts, T is the absolute temperature, and power has
been assumed positive into the R element.
An R element with multiple ports will have defined causal inputs u and outputs y, with each possibly comprised of
either effort or flow variables. In order to have a unique output y, Φl must be a single valued function of the inputs u and T.
Another restriction on Φl can be obtained by noting that this function must ensure that,
y ⋅ u ≥ 0, (4.58)
since T ≥ 0 and Ṡ ≥ 0. From the expression of available energy in 4.47 and the inequality 4.58, it can easily be determined
that
{ }
T∑n
A (x) = sup − y u dt ≤ 0, (4.59)
∫0 i=1 i i

and therefore a proper dissipative element is always passive.

5 The implied collapse of molecules due to active attraction was one impetus for the development of quantum mechanics and the development
of the exclusion principle.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
194 4 System Model Formulation and Evaluation

y Figure 4.46 Relations for 1-port dissipative elements showing allowed and not allowed
Passive regions.
Not allowed
Active
Allowed
Allowed u

Not allowed

4.6.2.2 1-Port R Element

For a 1-port R, equation 4.58 implies that the constitutive law for a dissipative element must lie entirely in the first and third
quadrants of the effort-flow coordinate axes. These restrictions are illustrated in Figure 4.46, using trends in passive and
active constitutive relations. Note that y and u are either effort-flow or flow-effort conjugate variables. The active relation
has a portion of its graph in the 4th quadrant of the y − u graph.

4.6.2.3 Causal Structure

Besides the necessary entropic or passivity structure of the dissipative element, there are also causal constraints on the
element in order to have a unique solution to the model equations. As an example, assume that the bond graph shown in
Figure 4.47(a) is a model of a physical system. The proposed constitutive relations are also shown with the bond graph. It is
desired to solve for the effort e as a function of the input flow f (t). This system has an algebraic loop, and for the causality
assignment shown,

e = e1 = Φ1 (f1 ) = R2 (f (t) − f2 ), (4.60)

which must be first solved for flow f1 , where the nonlinear function Φ1 () is graphically represented in the e1 − f1 plot of
Figure 4.47(a). Since Φ−11 () is not unique, the solution to this equation can, in general, have multiple solutions. This is
represented by plotting both e = R2 (f (t) − f1 ) and e = Φ1 (f1 ) versus f1 in Figure 4.48. For the particular values of f and R2
given, there are three solutions to the equation.
Since properly structured physical models with multiple solutions are not practically useful, physical system models
proposed with these types of constitutive relations should be rejected.
If the R element on bond 2 is replaced with a linear C element with parameter C2 , as in Figure 4.47(b), then integral
causality on the C element results in an effort input to the R element on bond 1. However, this causality on this R element

(a) (b)

Figure 4.47 Models with composite R: (a) nonunique two R model and (b) nonunique RC model.

Figure 4.48 Multiple solutions of R model.

Multiple
solutions
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
4.7 Modulation Structure 195

is not allowed in general due to the nonuniqueness of the effort-flow constitutive relationship. If the causality is changed
to make the C have derivative causality, then the result leads to equation,
̇
e = Φ1 (f − C2 e), (4.61)
which cannot, in general, be solved uniquely. So, changing the causality does not alleviate the multiple solution conflict.
This shows that there is nothing to be functionally gained by letting an R element force an energy storing element into
derivative causality. A model of this sort would have to be discarded irrespective of the causality assignment.
In order to avoid causal problems such as those indicated in the two previous examples, the constitutive relation Φl () for a
dissipative element should generally be required to be passive and bijective. In this case, multiple solutions like those above
can be avoided. The restriction to a bijective constitutive relation is not as stringent as it might first appear. Constitutive
relations which are not bijective are a sign that a single R element is used to model important multiple physical effects, and
a more detailed model of these effects would eliminate the nonuniqueness. In an analogous fashion to the energy storing
elements, the following definition is made.

Basic Dissipative Element: A dissipative element is basic if Φl is bijective. If a dissipative element is not basic, it is
composite.

Devices modeled with composite dissipative elements are prevalent in electronics. This is not surprising since these
devices exhibit what is called negative resistance (or conductance). The concept of negative resistance is useful in the
design of amplifiers, oscillators, and memory circuits, therefore devices are specifically designed to operate in this fashion.
Composite dissipative elements outside the field of electronics are considerably more scarce. The one notable exception to
this rule is the models sometimes used for stiction and coulomb friction.
If composite dissipative elements are present in a model, they should not be used to dictate derivative causality on an
energy-storing element. In addition, they should not be allowed in an algebraic loop. Algebraic loops are present if, after all
the sources and energy-storing elements have been assigned causality and extended, there are still unassigned R elements.
If any of these remaining R elements are composite, this will be termed a composite dissipative loop. Composite dissipative
loops should not be allowed in physical models in order to avoid multiple solutions.

4.6.3 Summary
As discussed above, the following structure should be imposed on the constitutive relations for bond graph elements.
1) In order for an energy state function to exist, the constitutive relation must: (a) be a single-valued function of state and
(b) satisfy Maxwell reciprocity.
2) For passive energy storing elements, the stored energy must be bounded from below.
3) A constitutive relation with a nonunique inverse is a sign of important multiple physical effects. Composite elements
which model these multiple effects should be modeled as fields in order to separate these effects, increase the flexibility
of the model, and eliminate pathological causal conflicts due to these elements.
4) Dissipative elements must have passive constitutive relations.
5) Dissipative constitutive relations should be basic when at all possible.
6) Composite dissipative loops should not be allowed in physical systems.

4.7 Modulation Structure

Up to this point, our main emphasis has been on how we model energetic interactions (i.e., those associated with power
flow) among the elements comprising physical systems of interest. There is also the possibility that signal or informa-
tion flow among the elements can directly influence how power flows in a system. The term modulation is adopted to
describe how information conveyed in a signal can influence physical processes. Since it is possible to introduce effects that
might violate conservation principles used to build physical system models, it is useful to review some ways to make sure
modulation is used properly.
A common example of modulation in a mechanical system arises in the case of a rigid, massless link constrained by
frictionless pins to slide on the x − y plane, as shown in Figure 4.49.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
196 4 System Model Formulation and Evaluation

Figure 4.49 (a) Rigid massless link constrained by frictionless pins to

slide on the x − y plane. (b) Bond graph model represented with
holonomic constraint.

(a) (b)

Due to the rigid nature of the link, there is a geometric constraint between y and 𝜃. This constraint can be expressed as,
y = L sin 𝜃, (4.62)
and differentiated to yield,
𝑣 = [L cos 𝜃] 𝜔. (4.63)
where 𝑣 is the velocity at y and 𝜔 is the angular velocity of the link. Since this is a nonenergic (no energy storage) and
lossless system, the following power balance must hold,
𝜏𝜔 = F𝑣, (4.64)
where 𝜏 is the torque applied at x and F is the force applied at y. Combining 4.63 and 4.64 gives
𝜏 = [L cos 𝜃] F, (4.65)
which is nothing more than the summation of torques about the pivot on the x axis. This power conserving transforma-
tion between the kinematic variables 𝑣 and 𝜔 and the related dynamic variables F and 𝜏 is shown in the bond graph of
Figure 4.49(b). The T element represents the power constraint and the signal bond (→) indicates that the modulus of this
transformer is a function of 𝜃, n(𝜃) = L cos 𝜃. Transformers of this type are often called modulated transformers and are
sometimes given a special symbol.6 In this text, modulation will be indicated by the inclusion of a modulated modulus
(e.g., n or r, as commonly used in transformers and gyrators, respectively) and/or by a signal bond (full arrow) explicitly
showing the source of modulation.
Modulation in bond graphs refers to a signal that reflects changes in the power flow in a system or subsystem but does
not contribute directly into the net sum of the power conveyed or coming into a system. There is no power flow conveyed
by a signal bond. For example, in Figure 4.49, we have power 𝜏𝜔 entering positively on the bond on the left port of the T
and power F𝑣 leaving positively on the bond on the right, but there is no power flow associated with the information about
𝜃 conveyed by the signal bond. The value of 𝜃 contains necessary topological information about the system but does not
contribute directly to the power flow through the transformer.

4.7.1 Functional Description of Physical Systems with Modulation

Consider a general physical system with n power ports and r signal ports. After causality is applied to the bond graph
element, it can be depicted in causal form as shown in Figure 4.50(b).
In the following discussion, the most general form that needs to be considered for modulation of bond graph elements
can be represented causally in vector form as,
ẋ = u, (4.66)
ẋ m = v, (4.67)
y = g(x, xm , u, v), (4.68)
where u is the input effort or flow on the power bonds, v is the modulation effort or flow on the signal bonds, and y is the
output effort or flow on the power bonds, and x and xm are energy state and modulation states respectively. In the previous
example, the geometric variable 𝜃 is needed to know the configuration of the link. This configuration or modulation state

6 Indeed, it is fairly common to find TF used for transformers and MTF for modulated transformers. We adopt Paynter’s preference that
single-character symbols be used for bond graph elements.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
4.7 Modulation Structure 197

(a) (b)

Figure 4.50 (a) Acausal multiport bond graph element with n ports and r modulation inputs. (b) Causal block diagram with
equivalent inputs and outputs.

Length, L

(a) (b)

Figure 4.51 (a) Cart with linkage mechanism. (b) Bond graph with modulation.

variable represents a necessary state to define the physical system in addition to the energy states of the independent energy
storage elements of the model.

Example 4.15 Link-cart system model

For the simple mechanism shown in Figure 4.51(a), assume the link shown is rigid, has negligible mass, and has a pinned
end free to slide along the y axis. The other end is pinned to a block that is attached to ground with a linear damper with
damping b. Assume a torque, 𝜏(t), is applied at the sliding pinned end. Build a bond graph model and derive state equations.
Solution
From the model in Figure 4.49, a bond graph model for this system can be developed as shown in Figure 4.51(b). Causality
indicates there is one energetic state, p, and one modulation state, 𝜃.
The state equation for p is, ṗ = FL − Fb , where FL = 𝜏L ∕(L cos 𝜃) where 𝜏L = 𝜏(t). The damping force is Fb = b𝑣b = bp∕m,
thus, ṗ = 𝜏(t)∕(L cos 𝜃) − bp∕m.
The modulation state equation is, 𝜃̇ = 𝜔L , where 𝜔L = 𝑣L ∕(L cos 𝜃). However, 𝑣L = 𝑣m , thus, 𝜃̇ = 𝑣L ∕(L cos 𝜃) =
𝑣m ∕(L cos 𝜃) = p∕(mL cos 𝜃). In summary,
ṗ = 𝜏(t)∕(L cos 𝜃) − bp∕m,
𝜃̇ = p∕(mL cos 𝜃).
Note that the rate of the modulation state 𝜃̇ is determined in a causal sense from 𝜔L as indicated by the causal strokes on
the 1 junction which gives the signal 𝜃.̇ Modulation of the transformer is by the integrated variable, 𝜃.

Other examples of modulation states will be given later in this and subsequent chapters, but, in general, it will be shown
that just knowing the displacements of the independent C elements and the momentum of the independent I elements is
not sufficient to determine the state of a physical system when there is modulation present in the model.

4.7.2 Energy Storing Elements Should Not Be Modulated

Assume, for the sake of discussion, that the constitutive relation for an energy storing element (I or C) can be expressed in
terms of both energy state variables and modulation state variables as
y = Φs (x, xm ). (4.69)
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
198 4 System Model Formulation and Evaluation

(a) (b) (c)

Figure 4.52 Modulated coupling elements with modulation indicated by signal bond for variable 𝜈: (a) acausal transformer, (b)
acausal gyrator, and (c) input–output representation.

This implies that the energy can be expressed as

= Φs (x, xm ) ⋅ dx = (x, xm ). (4.70)

∫
The rate of change of energy in the element can then be expressed as
𝜕 𝜕
 = ̇ = ⋅ ẋ + ⋅ ẋ . (4.71)
𝜕x 𝜕xm m
The last term on the right-hand side of this equation must be identically zero since the power flow associated with modu-
lation is defined to be zero. This implies that neither stored energy  nor Φs can be functions of xm (any variable of time
can be disallowed by the same argument) or energy storing elements cannot be modulated.
If an energy storing element is allowed modulation in spite of this result, the energetic foundation of the bond graph
structure can be undermined.7 As an example, consider a linear translational spring with a stiffness that changes over
time, k(t). The energy of the spring can be expressed as,

= kxdx = kx2 ∕2, (4.72)

∫
and the time derivative of the energy is,
̇ 2 ∕2 = F𝑣 + Kx
̇ = kxẋ + kx ̇ 2 ∕2, (4.73)
which does not equal the input power F𝑣 for k̇ ≠ 0. This analysis identifies a power flow port could have otherwise been
ignored. The power interaction at that port on the system under study needs to be determined. The variation may be a rela-
tively “slow” process compared to the rest of the system, allowing an approximation to be made in the model. An analogous
example is a simple tunable electrical capacitor where capacitance is changing by the rotation of a shaft. If the dynamics of
the rotating shaft interact with those of the electrical aspects then a multiport model can be adopted, as will be discussed in
Chapter 8. On the other hand, it may be more practical in some cases to formulate a model with time-varying parameters,
particularly in certain application areas that have established approaches for analyzing the dynamics and stability of such
systems [65–67].

4.7.3 Modulated Coupling Elements

Unlike the energy storing elements, the power conserving coupling elements (T and G) are allowed modulation as seen by
the bond graph in Figure 4.49. Once causality has been applied to either a two port T or G element, it is possible to obtain
a functional input–output form for these elements as shown in Figure 4.52(c).
As an example, consider the T element of Figure 4.52(a) with effort imposed on bond 1. With this causality, the functional
input–output form for the T can be expressed as,
f1 = n(𝜈)f2 , (4.74)
e2 = n(𝜈)e1 , (4.75)
where n is the transformer modulus which now is a function of the variable 𝜈, the input variables are e1 and f2 , and the
output variables are e2 and f1 . This can be seen to be in the same causal form as depicted in Figure 4.52(c) with u1 = e1 ,
u2 = f2 , y1 = f1 , y2 = e2 , and h = n.

7 This argument does not disallow modulation of elements which do not store energy such as coupling elements.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
4.7 Modulation Structure 199

The power conserving constraint has been expressed previously as,

y ⋅ u = 0, (4.76)

which is satisfied for modulation of the coupling modulus h by any variable, say 𝜈. In this case, h is restricted to be a function
of just the input variables u and v (as indicated by the bonds, causality and signals) and their integrals x and xm . A bond
graph can faithfully adhere to the structure of a physical system by explicitly depicting any modulation on the graph using
signal bonds.
Although requiring that all modulation be shown graphically by signal bonds adds some needed structure to the graph
by indicating the source of the modulation variables, this does not restrict the use of modulation since signal bonds could
be placed on the graph in a somewhat arbitrary fashion. Any further restrictions on the allowed use of modulation for
the coupling elements require a detailed study of the physical causes of modulation of these elements. Several cases are
reviewed in the following to illustrate how modulation can arise in practical physical modeling.

4.7.3.1 Modulation in Analytical Mechanics

One domain in which there has been an extensive study of modulation in the guise of work-less or power conserving
constraints is analytical mechanics [68]. Constraints are classified in analytical mechanics into two types: holonomic and
nonholonomic. Although n port constraints can generally be considered, only two port elements are discussed in this
section.

Two-Port Holonomic Constraint An example of a holonomic constraint was given in Figure 4.49. Basically, a holonomic con-
straint, in terms of bond graph variables, can be represented as an algebraic relation between the displacements on the two
bonds labeled 1 and 2 in Figure 4.53. This relation can be expressed as,

g(q1 , q2 ) = 0. (4.77)

This constraint can be differentiated to yield,

𝜕g 𝜕g
q̇ + q̇ = 0, (4.78)
𝜕q1 1 𝜕q2 2
and then used to solve for the ratio of the flows at the two ports as
( ) ( )
f1 𝜕g 𝜕g 𝜕q
=− ∕ = − 1 = n(q). (4.79)
f2 𝜕q2 𝜕q1 𝜕q2
The variable n is the modulus for the transformer used to enforce the algebraic constraint 4.77. Since q1 and q2 are not inde-
pendent due to this constraint, only one independent displacement, q, is needed to describe the modulus. The independent
displacement can be taken as q1 , q2 , or an algebraic combination of q1 and q2 . This is an arbitrary modeling decision. The
bond graph representation of a holonomic constraint shown in Figure 4.53 assumes that q1 is the independent displace-
ment. This is indicated by a signal emanating from the 1-junction which has flow q̇ 1 , where there is an implied integration
in time that provides q1 . As will be discussed in the following, this can require defining some modulation variables as states
of a system.
Assigning an allowed causality to the graph in Figure 4.53, the functional description of the constraint can be given as,

ẋ m = 𝑣, (4.80)
y1 = h(xm )u2 , (4.81)
y2 = h(xm )u1 , (4.82)

where 𝑣 = f1 and xm = q1 .

Figure 4.53 Bond graph of general two-port holonomic constraint.

 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
200 4 System Model Formulation and Evaluation

Figure 4.54 (a) Knife moving on x − y plane. (b) Nonholonomic constraint

representation in modulated transformer.

(a) (b)

Two-Port Nonholonomic Constraint As mentioned above, only one independent displacement or coordinate is needed for
a holonomic constraint. The existence of kinematic constraints that impose no restrictions on the allowable displace-
ments (q1 , q2 ) was not recognized until Hertz (1894) introduced the distinction between holonomic and nonholonomic
constraints.8 The only type of nonholonomic constraint of present interest9 can be expressed as,
f1 ∕f2 = n(q), (4.83)
where q is a general displacement not directly related to q1 or q2 .
As an example of a nonholonomic constraint, consider a knife edge traveling on ice shown in Figure 4.54. There is a
kinematic constraint between the velocities ẋ and ẏ given as,
ẏ = (tan 𝜃)x,
̇ (4.84)
but there is no constraint between the displacement x and y. Physically, this means that the entire x − y plane can be reached
by the knife.
The knife edge constraint can be modeled by a modulated T element where the modulation variable 𝜃 must be supplied
by an angular rotation submodel. A general bond graph of a nonholonomic constraint is shown in Figure 4.54(b). This
bond graph is almost identical to the holonomic case. The only difference is the arbitrary nature of the modulation
displacement 𝜃. Functionally, the nonholonomic constraint has the same description as holonomic constraint given in
equations (4.80–4.82), except 𝑣 is some flow other than f1 or f2 .

4.7.3.2 Rigid Body Dynamics and Gyroscopic Modulation

The six degrees of freedom of a rigid body with body-fixed (1-2-3) axes can be modeled by the Euler equations, [12], for
translation,
F1 = ṗ 1 + 𝜔2 p3 − 𝜔3 p2 , (4.85)
F2 = ṗ 2 + 𝜔3 p1 − 𝜔1 p3 , (4.86)
F3 = ṗ 3 + 𝜔1 p2 − 𝜔2 p1 , (4.87)
and rotation,
𝜏1 = ḣ 1 + 𝜔2 h3 − 𝜔3 h2 , (4.88)
𝜏2 = ḣ 2 + 𝜔3 h1 − 𝜔1 h3 , (4.89)
𝜏3 = ḣ 3 + 𝜔1 h2 − 𝜔2 h1 , (4.90)
where the left-hand sides represent net applied external forces and torques relative to the 1-2-3 body-fixed axes, hav-
ing translational momenta, (p1 , p2 , p3 ), angular momenta, (h1 , h2 , h3 ). Six-state equations can be expressed by rearranging
equations 4.85 to 4.90 with state derivatives dependent on forces and induced “gyroscopic” effort terms. These gyroscopic
forces and torques are due to the “cross-terms” in the Euler equations, such as torques generated by a change in direction
of the angular momentum given by,
𝜏 = 𝜔 × h, (4.91)
where 𝜏 is the torque, 𝜔 is the angular velocity, and h is the angular momentum. When viewed in this manner, forces
[ ]
such as 𝜔2 p3 = 𝜔2 (m𝑣3 ) in Equation 4.85 take the form, F = m𝜔2 𝑣3 , indicating a modulated gyrator structure. These
gyroscopic forces and torques represent bidirectional coupling of the states by power conserving (gyrator) constraints since

8 A detailed discussion of holonomic and nonholonomic constraints can be found in [69].

9 Other nonholonomic constraints include inequality constraints.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
4.7 Modulation Structure 201

(a) (b)

Figure 4.55 Euler junction structure for (a) translational dynamics and (b) rotational dynamics of a body of mass m and body-fixed
principal axes 123. Gyroscopic forces and torques coupling degrees of freedom are represented using gyrator (G) elements modulated
by translational and rotational states. External forces and torques are indicated at each degree of freedom with open bonds.

they are perpendicular to the respective motional state (and induced by the cross-product). This modulation structure
enables construction of an Euler junction structure for representing these equations in bond graph form [38, 70] as shown
in Figure 4.55. This model assumes that the 123 coordinates are principal axes with corresponding moments of inertia (e.g.,
I11 about axis 1).
While the Euler equations convey the inherent coupling between degrees of freedom, the Euler junction serves as a
convenient graphical representation that can ensure key effects are included. For example, the special case where the axis
of rotation of a body always passes through a fixed point is of special interest, and relates to the theory of the gyroscope.10
A prototype example of exhibiting this form of gyroscopic modulation is a symmetric spinning disc with controlled spin
velocity, 𝜔1 , constrained by a pivot at one end as shown in Figure 4.56(a). In this figure, the 1-2-3 coordinate frame rotates
with the rigid body. There is no need here to consider the translational motions and assume the system is at equilibrium
spinning about axis 1.
The torques induced and coupling axes 2 and 3 are,
𝜏13 = h1 𝜔3 , (4.92)

𝜏12 = h1 𝜔2 , (4.93)
where h1 is the modulus of a gyrator that models this constraint. These expressions should be compared to corresponding
torques in equations 4.89 and 4.90.
It should be emphasized that the modulation by h1 = I11 𝜔1 implies stored kinetic energy accompanied by a power flow
(e.g., the input source spinning the disk). Without the kinetic energy, there would be no gyroscopic coupling. The single G
element in this model is in reality a model with multiple physical effects. The signal bond for h1 represents coupling to the
I element for the 1 axis and if convenient could be drawn to emphasize this relationship.
Aside from helping to explain the modulation structure inherent to rigid body dynamics, the Euler junction models
provide a practical way to ensure key degrees of freedom are considered. The system modeler that only encounters such

(a)
(b)

Figure 4.56 (a) Prototypical gyroscopic system. (b) Bond graph using Euler junction to model gyroscopic coupling, with dashed
region indicating relation between axes 2 and 3.

10 Leon Foucault first presented and named the gyroscope in 1852 to prove rotation of the earth. The word gyroscope was derived from the
Greek for gyros “a circle” and skopos “watcher.”
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
202 4 System Model Formulation and Evaluation

Spin axis
Case
Case

Spring restraining
Gimbal Gimbal bearing
gimbal rotation
in z to case fixed to case
damping

Gimbal
Input axis Output axis

(a) (b)

Figure 4.57 (a) Gimbaled gyro. (b) A classical gyro-based sensor.

effects occasionally can confidently use this structure to graphically “prune” away any effects not relevant to a given problem
prior to integrating with other subsystems. As shown in the basic example of Figure 4.56, causality assignment also helps
interpret physical phenomena, which may not be intuitive in some cases.

Example 4.16 Single-axis rate gyro

A classical application for gyros in sensing angular motion takes advantage of gyroscopic coupling.11 A typical gimbaled
gyro is shown in Figure 4.57(a), where the spin momentum of the flywheel is controlled at hx (motor not shown). In (b), this
configuration is attached to a torsional spring with stiffness k, which represents a compliant sensing element (e.g., strain
gauge). In this device, an input motion to be sensed is 𝜔s . The gimbaled device is attached to a case but can rotate with 𝜔z .
Model this system and write state equations that show how the spring deflection can be related to 𝜔s .
Solution
The bond graph in Figure 4.58(a) shows how the dynamic response along the z-axis depends on the gyroscopically induced
torques that depend on the spin velocity, 𝜔s , and the input velocity to be detected, 𝜔i (t).
The only states in this system as indicated by causality are hz = Izz 𝜔z and 𝜃k , the latter being the spring deflection. From
the bond graph,
ḣ z = 𝜏z − 𝜔x hy + 𝜔y hx ,
where the x and y components are derived from the geometry in Figure 4.58(b), thus,
ḣ z = Izz 𝜔̇ z = 𝜏z − 𝜔x ⋅ Iyy 𝜔s sin 𝜃 + 𝜔s cos 𝜃 ⋅ hs ,
∑
where 𝜏z = 𝜏 = −b𝜔z − k𝜃z . The final nonlinear state equations are,
Izz 𝜔̇ z = −b𝜔z − k𝜃 − hs (Iyy ∕Ixx )𝜔i sin 𝜃 + hs 𝜔i cos 𝜃, (4.94)

𝜃̇ k = 𝜔z , (4.95)

Spin

Input motion
(a) (b)

Figure 4.58 (a) Bond graph employing Euler junction structure. (b) Components of input angular velocity.

11 Modern sensors of this type take advantage of micro-electromechanical technology.

 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
4.7 Modulation Structure 203

where hs = Ixx 𝜔s = constant. In the second-order form for 𝜃k ,

[ ]
Izz 𝜃̈ k + b𝜃̇ k + k𝜃k = hs cos 𝜃 − (Iyy ∕Ixx ) sin 𝜃 𝜔i (t),
For Ixx ≫ Iyy and small 𝜃k this leads to a standard linear second-order form,
𝜃̈ k + 2𝜁𝜔n 𝜃̇ k + 𝜔2n 𝜃k = (hs ∕Izz )𝜔i (t) = Kg 𝜔2n 𝜔i (t),
where 2𝜁𝜔n = b∕Izz , 𝜔2n = k∕Izz , and Kg = hs ∕k, which is a sensor gain (or sensitivity) that depends on spin momentum,
hs . While modern angular rate sensors increasingly use microelectromechanical and other technologies, mechanical gyro-
scopes have better long-term stability and are favored in some applications [71].

Modeling systems that have coupled translational–rotational dynamics can be much more tractable using the Euler junc-
tion structure which captures essential modulation structure. When dealing with multibody systems with constraints,
however, adopting a Lagrangian approach can be shown to be more efficient, as will be discussed in Chapter 8. For even
more complex cases, computational approaches employed within commercially available computational software offer
more viable solutions. Haug [72] describes the basis for such methodologies. The same type of modulated gyrator structure
that arises in rigid body dynamics is found in Lorentz magnetic forces as discussed in the following.

4.7.3.3 Electromechanical Modulation

The source of modulation in electromechanics is due to the form of the force exerted by the magnetic field. This force can
be expressed as
Fm = qv × B, (4.96)
where Fm is the magnetic force, q is the charge, v is the velocity, and B is the magnetic flux density. Note that this Lorentz
force constraint represents a work-less or power conserving constraint, just as those discussed in analytical mechanics,
since the force is perpendicular to the velocity of the charged particle. The forces of rigid body constraints discussed above
are also perpendicular to the local velocity of a body particle.
In Figure 4.59, we consider a single charge moving q in a plane with cartesian axes labeled 1 and 2 and a magnetic field
perpendicular to that plane. The applied force on the particles, F = −Fm , is
−F1 = qB𝑣2 , (4.97)

F2 = qB𝑣1 , (4.98)
where F1 and 𝑣1 are the force and velocity in 1 direction and F2 and 𝑣2 are in the two directions.
The net power into the charged particle system is net = F1 𝑣1 + F2 𝑣2 = 0 or the system is power conserving. A bond
graph model for this system consists of a modulated gyrator. The modulus of the gyrator is r = qB which relates effort
(F) to flow (𝑣). This modulus can be varied by changing the strength of magnetic flux density. This is true modulation
since changing this field does not contribute directly to the power flow of the element. It should be noted that B cannot be
changed without a power flow, but this power flow does not directly change the power flow in the G element used to model
the electromechanical coupling.
In the next example, a conducting wire is in horizontal motion through a spatially constrained magnetic field as shown
in Figure 4.60. The Lorentz force F required to keep the wire in equilibrium when a current i flows through the wire can
be calculated from equation 4.96 as,
F = Bli, (4.99)

Figure 4.59 (a) Charge moving in 1–2 plane of a magnetic field. (b) Bond
graph with charge and magnetic field modulation.

(a) (b)
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
204 4 System Model Formulation and Evaluation

Figure 4.60 (a) Current-carrying conductor moving in a magnetic field.

(b) Bond graph with B-field modulation.

(a) (b)

where l is the width of the constrained field. Since the voltage or electromotive force, V, is equivalent to a force through a
distance per charge (energy/charge), the voltage generated across the width of the field can also be found from 4.96 as,
V = (Fe l)∕q = Bl𝑣, (4.100)
where Fe is the force generated on the charges in the conductor due to the velocity 𝑣. This result could also be obtained by
using Faraday’s induction law as V = Φ̇ = d(BA)∕dt = Bl𝑣. Equations 4.99 and 4.100 represent the functional relationship
of a gyrator element which could be modulated by B as seen in Figure 4.60(b).
Up to this point, no mention has been made of the origin of the magnetic field. This field could be generated by a perma-
nent magnet (in which case it will be relatively constant), or by a current. The magnetic field generated by a current can be
found by either the Biot–Savart law or Ampere’s law.
A prototype example of a system in which a magnetic field is generated by a current is displayed in Figure 4.61. A simple
bond graph model which is sometimes given for systems of this type is shown in Figure 4.61(b). The field current, if , is
indicated as a modulation signal input from the environment. Although this model can be useful for specific circumstances
(static field current or negligible power flow in the field circuit), the model could lead to incorrect conclusions since if
cannot be changed without the power required at the indicated terminal,  = Vf if , and therefore does not represent a
true modulation signal. For example, changing if will generate a magnetomotive force proportional to the turns in the
wire around the core. This implies a change in magnetic field. Due to Faraday induction, a changing magnetic field will

(a) (b)

(c)

Figure 4.61 (a) Prototype electromechanical system. (b) Composite modulation model. (c) Basic modulation model.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
4.7 Modulation Structure 205

generate a voltage Vf also proportional to the number of turns. Any attempt to ignore the power flow due to rapidly changing
field current, for example, in an attempt to impose a step change in field current on the system, would lead to incorrect
conclusions.
The inadequacy of this model for time varying field current can be traced to the fact there are important multiple phys-
ical effects being modeled by just one element. In Figure 4.61(c), a more detailed lossless12 model which separates out
the physical effects is given. The upper gyrator of this model accounts for Ampere’s law relating the field current to the
magnetomotive force, M.13 From Ampere’s law,
M = Ni, (4.101)
where N is the number of turns in the wire. The back reaction of this element is due to Faraday induction which can be
expressed as,
Vf = N 𝜑,
̇ (4.102)
where Vf is the field voltage and Φ is the magnetic flux which is equal to BA (A being the cross-sectional area of the gap).
The C element represents the energy storage with constitutive function
[ ][ ]
d 1
M= 𝜑, (4.103)
A 𝜇0
where 𝜇0 is the magnetic permeability, d is the distance across the gap, the field is assumed uniform in the gap, and
energy stored in the core material and fringing field is neglected. The signal bond represents a true modulation of the
electromechanical coupling as given in the previous two examples. It should be noted that the modulation variable 𝜑
cannot be discontinuous in time without an infinite power flow. This also implies that the gyrator modulus cannot change
instantaneously.
A more practical example of electromechanical coupling is that found in a separately excited dc machine. A simplified
schematic is shown in Figure 4.62. In the bond graph shown in Figure 4.62, the R elements model coil resistance, the G
elements model Ampere’s and Faraday’s laws for the two coils, the C elements represent the magnetic field and armature
energy storage, the modulated G is the electromechanical coupling and the remaining T represents a transformation from
linear to rotary motion.

4.7.3.4 Theoretical Structure of Modulated Coupling Elements

As discussed in Section 3.2.2, power conserving elements can be represented in input–output form with a skew-symmetric
matrix. When there is modulation present, it is possible to represent the output y as a function of input u as
y = H(x, xm , u, v)u, (4.104)

(a) (b)

Figure 4.62 (a) Separately-excited DC machine. (b) Bond graph of the field magnetic circuit and DC machine with field flux, 𝜑, shown
explicitly as a modulating variable.

12 Losses are certainly necessary in order to obtain a realistic model of this system, but their inclusion is not relevant to the present discussion
and are therefore neglected.
13 Magnetomotive force is analogous to electric potential, being defined by a line integral between two points in a magnetic circuit, as will be
discussed further in Chapter 8.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
206 4 System Model Formulation and Evaluation

where H is an n × n skew symmetric matrix (H = −HT ) function of energy and modulation states x and xm , and modulation
signal inputs v and power inputs u. This implies that two-port-modulated coupling elements can be represented as
[ ] [ ][ ]
y1 0 h u1
= where: h = h(x, xm , u, v), (4.105)
y2 h 0 u2
for power positive into port 1 and out of port 2.
If the results of the previous study of modulated coupling elements are examined, it can be seen that in every case the
functional form for the modulation of a two port coupling element (T or G) can be represented with the h solely a function
of modulations state xm . In light of this, we make the following definition:

Basic modulated coupling element: A coupling element (T or G) has basic modulation if modulation is solely a
function of modulation state xm , where xm is the integral of an effort or flow indicated by a signal bond attached to
the coupling element. If the modulation is not basic, it is composite.

Although composite modulation can be a good approximation to a physical model, its use should be used with care in
order to maintain a well-structured model.
Other interesting results which can be seen from the previous examples is the absence of material constants in the cou-
pling element moduli for basic modulation. Since material is associated with energy storage, this is to be expected since
there is no change of energy storage with coupling elements. In general, the energy states p’s and q’s on independent energy
storing elements can be thought of as quantifying the energy of the system while the modulation states give the configura-
tion of the system including the necessary geometry. Another interesting result is the continuity of the modulus with basic
modulation. Since basic modulation involves time integration, it cannot change instantaneously.
As a simple example of a system with a modulation state that might be treated with a composite energy storing element
consider the compliant system shown in Figure 4.63(a). Assume we can neglect the mass and the losses in the system.
A composite model of this system is shown in Figure 4.63(b). Due to the composite nature of this representation it could
not be used with derivative causality in order to have unique solutions. Derivative causality might occur for example if two
of these systems were attached together or if a force is impressed on the system. A model with this restriction is not necessary
if both the energy state (compression of the spring) and the modulation state (which can be taken as the displacement x) are
indicated and utilized. A basic model which incorporates both of these potential states is shown in Figure 4.63(c). Although
this example is not particularly practical, similar techniques could be used to develop causally indifferent models (until the
environment is imposed) for components such as latches and bevel springs.

4.7.4 Modulated Dissipative Element

If one includes the temperature–entropy port of an R element, it is a power-conserving element like the coupling elements,
and therefore it should not be surprising to have modulation of these elements. Consider a two-port R element with one of
the ports the temperature–entropy port, as shown in Figure 4.64.

(a) (b) (c)

Figure 4.63 (a) Compliant mechanical system comprise of rigid link with negligible mass. (b) Composite model for compliant
mechanical system. (c) Bond graph of a basic model of the compliant system.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
4.7 Modulation Structure 207

Resistance causality Conductance causality

(a) (b)

Figure 4.64 Modulated dissipative element with (a) resistance causality (b) conductance causality.

Since the R element with a temperature–entropy port is power conserving, it satisfies relation 3.16 in Chapter 3. The
output for resistance causality can then be expressed as
e = h(f , xm )T, (4.106)

fs = Ṡ = h(f , xm )f . (4.107)
For conductance causality, the outputs can be expressed as
f = g(e, xm )T, (4.108)

fs = Ṡ = g(e, xm )e. (4.109)

Besides being power conserving, a dissipative element must also produce entropy. This requires fs = Ṡ ≥ 0. From 4.106
and 4.107, it can be seen that this will only happen in general if h is an odd function of f for resistance causality or h is an
odd function of e for conductance causality. If a dissipative element is thought of as an entropic coupling element, bond
graph decompositions which explicitly indicate the entropic functional relationships may be formulated as in Figure 4.64.14
We do not espouse that dissipative elements always be decomposed in this fashion, but this form does reveal a motivation
for the restricted form of modulation proposed in Section 4.7.3 for the coupling elements. In order to generate entropy,
a coupling element must have internal modulation by power variables (e and f ), where internal refers to modulation by
variables on the power bonds attached to the element. It is comforting to note that modulation by xm does not satisfy the
condition to generate positive entropy flow, and coupling elements with basic modulation will be isentropic as desired.
Besides the purely formal internal variable modulation, there can also be external modulation of a dissipative element.
This modulation requires the addition of a signal bond conveying the variable information to the dissipative element. In
general, this modulation could be due to both the signal itself and its integral. The element’s constitutive relation must
adhere to positive entropy production with the inclusion of the modulation variables.
A common situation where need for a modulated dissipative element arises is with mechanical friction between two
bodies, where the normal force influences the induced frictional force. Consider the case in Figure 4.65(a) where force,
Fn , is applied normal to the motion of the mass, m. The free-body diagram in Figure 4.65(b), shows how the friction force
Ff = 𝜇Fn sgn(𝑣x ), is now modulated by Fn . The force Fn is here indicated as derived from the effort on a 0-junction, presumed
to be causally specified by one of the connected bonds. The nature of this modulating force is that it does not directly interact
with the x-directional dynamics of the mass. The normal force only influences the frictional force.
Another example of a modulated dissipative element is the hydraulic valve shown in Figure 4.66 in which the flow area
can be changed by valve position, x. The pressure drop across such a valve, ΔP = P1 − P2 , is a function of the flow rate and
̇
valve position, commonly of the form, ΔP = K𝑣 (x)Q2 . Note that position is an integrated flow variable, that is, x = ∫ xdt. In
this example, the valve motion is normal to the flow through the valve, so any interaction is negligible. Thus flow through the
valve is assumed to not influence the valve motion, and valve position is considered to only influence the valve coefficient.
This is in contrast to many valve scenarios where there can be some interaction between valve motion and the fluid flow,
such as in Figure 4.38 and practical fluid power scenarios [17, 75]. Section 4.7.5 examines how such scenarios might be
more directly represented in a bond graph.

14 See Paynter [73] or Breedveld [74] for a more detailed discussion of this process.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
208 4 System Model Formulation and Evaluation

(a) (b) (c)

Figure 4.65 (a) Mass sliding on surface with dynamic normal force. (b) Free-body diagram indicating sliding friction, Ff . (c) Model of
friction as modulated mechanical dissipative element, R, with implied dependence on Fn .

(a) (b)

Figure 4.66 (a) Hydraulic control valve with controllable flow area. (b) Model of hydraulic valve as modulated hydraulic dissipative
element, R, with flow coefficient K𝑣 (x).

4.7.5 Modulation in Fluid–Structure Interaction

Forces are induced when fluids flow around and through bodies. One way to account for effects in some system modeling
contexts is to adopt the control volume approach introduced in many introductory fluid mechanics textbooks [14, 76]. Force
and moment relations can be formulated that can be used within a bond graph framework, and a modulation structure is
shown to arise in some cases. Refer to textbooks such as Fox and McDonald [14] and Daugherty et al. [76] which introduce
application of the following formulations to fundamental problems and configurations.

4.7.5.1 Inertial Control Volume

Consider first a control volume (CV) formulation of Newton’s second law for a nonaccelerating (fixed) control volume. The
integral form over a CV accounts for changes in momentum due to the time rate of change of fluid within the CV and to
flow through the control surface (CS). Newton’s law relates these momentum changes to surface (S) and body (B) force
effects by [14],
𝜕 ⃗ = F⃗ = F⃗ S + F⃗ B ,
𝑣𝜌d–
⃗ V+ 𝑣𝜌
⃗ 𝑣⃗ ⋅ dA (4.110)
𝜕t ∫CV ∫CS
where ∫CV () accounts for fluid in the CV and ∫CS () accounts for the fluid flow that crosses the control surface. The fluid
velocity vector is denoted 𝑣,
⃗ and the surface area vector through which flows cross the CS is A. ⃗ Consider, for example, a
steady flow,15 entering a fixed control volume at 1 and leaving at areas 2 and 3 as shown in Figure 4.67(a). Surface forces
due to pressure are assumed to cancel so only a (surface) reaction force at the wall relates to the force induced along x-axis.
Using 4.110 gives,
Fx = 𝜌A1 𝑣21 . (4.111)

15 Steady flow implies that fluid properties in a flow field do not change over time.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
4.7 Modulation Structure 209

Control Fixed
volume wall

Fixed
wall
Incompressible
fluid
(a) (b)

Figure 4.67 (a) Fluid-flow entering a control volume along x, striking a fixed wall and exiting along y at 2 and 3. (b) Bond graph of
fluid reaction force along x modeled with a modulated gyrator.

[ ]
When expressed in terms of fluid flow rate, Q = A1 𝑣1 , Fx = FQ = 𝜌(Q∕A1 ) Q = r(Q) ⋅ Q, suggesting the force could be
modeled by a modulated gyrator, as shown in Figure 4.67(b). In this case of a fluid jet impinging a fixed wall with 𝑣x = 0,
this would suggest ΔP = 0. In actuality, the “back pressure” is assumed to not propagate back to the source in this case, so
a signal bond is used to indicate Q is imposed at the 1-junction. It is also possible to use a modulated flow source in these
situations. Modulated sources are commonly used in many areas, especially in electronic systems as described by Chua
et al. [37], and will be discussed further in Section 4.7.6.
For fluid flowing in a conduit or pipe, the interaction between the fluid and mechanical dynamics can be more explic-
itly represented. For fluid flowing in a 90∘ pipe bend, for example, forces are induced on the pipe structural system.
Figure 4.68(a) shows a control volume and free-body diagram around a bend. There are two reaction forces along the x
and y axes, designated Fx and Fy , respectively. Consider gage pressures and effects of atmospheric pressure cancel out.
Applying equation 4.110 along the x-axis gives,

Fx + P1 A1 = (Q∕A1 )𝜌(Q∕A1 î) ⋅ (−A1 î) = −𝜌(Q∕A1 )Q,

∫CS
so the reaction force is opposite in direction to that indicated,

Fx = −P1 A1 − 𝜌(Q∕A1 )Q = −P1 A1 − r(Q)Q. (4.112)

The bond graph in Figure 4.68(b) shows how this relation can be represented. Note that the contribution of P2 is zero along
the x-axis. From this bond graph fragment, the flow rate is causally determined as by the resistive element. Assume for
simplicity that PR = R ⋅ Q so,

Q = PR ∕R = (P1 − P2 − ΔP)∕R = (P1 − P2 )∕R,

since ΔP = r(Q)𝑣x = 0 for a fixed pipe bend. A similar result as for Fx can be found for Fy .

Control volume

Incompressible
fluid

(a) (b)

Figure 4.68 (a) Fluid-flow entering a control volume along x, striking a fixed wall and exiting along y at 2 and 3. (b) Bond graph of
fluid reaction force along x modeled with a modulated gyrator.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
210 4 System Model Formulation and Evaluation

If the velocity vector 𝑣⃗ in equation 4.110 is now taken as relative to the CV, 𝑣⃗r , then forces induced bodies within a moving
control volume can be treated, with extensions to those with arbitrary acceleration. This more general case of equation 4.110
takes the form [14],
𝜕 ⃗ = F⃗ = F⃗ S + F⃗ B − d𝑣⃗
𝑣𝜌d–
⃗ V+ 𝑣𝜌
⃗ 𝑣⃗ ⋅ dA 𝜌d–
V. (4.113)
𝜕t ∫CV ∫CS ∫CV dt
This latter formulation can be used to derive dynamic models of bodies propelled by impinging flows or for bodies propelled
by expelled mass (e.g., classic rocket equation). These surface forces are also associated with convective terms that can be
used to account for variable-mass effects [63, 74, 77] (introduced in Section 4.7.6 but more fully discussed in Chapter 8).
Finally, extension to the moment of momentum equation leads to a related and fundamental equation for the torque in an
Eulerian turbomachine (ETM). Considering a control volume about a rotor or impeller, for example, as a first approximation
the shaft torque induced under steady flow is,

𝜏⃗ = ⃗
⃗r r × 𝑣⃗r 𝜌𝑣⃗r ⋅ dA, (4.114)
∫CS
which can the form in practical cases [14, 76],
𝜏 = 𝜌Q(r2 𝑣t2 − r1 𝑣t1 ), (4.115)
The tangential velocities 𝑣t2 and 𝑣t1 are related to the specific impeller characteristics and to the impeller shaft speed, 𝜔. In
this way, it is can be found that Eulerian (or dynamic) turbomachines [78, 79] have gyrator relations of the form,
𝜏 = r(𝜔, Q)Q, (4.116)
ΔP = r(𝜔, Q)𝜔, (4.117)
where the modulus r(𝜔, Q) is “doubly modulated” and implicitly a function of the machine geometric properties. This form
of ETM model is contrasted with a common positive-displacement machine (PDM) in Figure 4.69. The PDM class of fluid
machines make use of mechanisms relate the power variables as,
Q = d(q)f , (4.118)
e = d(q)ΔP, (4.119)
where e can be either force or torque and f either velocity or angular velocity. Here d(q) represents the PDM displacement
(typically the volume displaced per revolution), here indicated as a quantity that can be modulated by a signal q. Thus
arising the distinction between fixed-displacement and variable-displacement PDM technology [17, 80].
The ideal transduction in both cases are shown here with loss effects such as friction and leakage. Real machine mod-
els would also incorporate inertia in rotating parts and compliance on the fluid side, notably for high-performance fluid
systems. These basic models can be used to practically represent a very wide class of fluid machinery used as both motors
and pumps, thus some power of the power flow arrows are not in indicated in Figure 4.69. For example, a hydraulic PDM
rotational motor with variable displacement would apply the model in 4.69(b) with power flow positive from the right.

4.7.6 Dependent and Modulated Sources

Dependent or controlled sources refer to a model abstraction in which effort or flow sources are functionally dependent
on effort or flow variables from within a system to be modeled. These types of source model elements were first explicitly

(a) (b)

Figure 4.69 (a) An Eulerian turbomachine (ETM) as modulated gyrator with losses. (b) Typical positive-displacement machine (PDM)
model with a transformer modulated by x and leakage loss on fluid side, and here either translational or rotational power may be used
on mechanical side.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
4.7 Modulation Structure 211

Electrical ideal controlled sources Generalized ideal controlled sources

(a) (b)

(c) (d)

(e) (f)

(g) (h)

Figure 4.70 Summary of ideal electrical and generalized controlled sources (a) CCVS, (b) FCES, (c) CCCS, (d) FCFS, (e) VCVS, (f) ECES,
(g) VCCS, and (h) ECFS, where V = voltage, C = current, E = effort, and F = flow.

defined for and have been applied to electrical and electronic systems since the early 1950s [37, 81, 82] and continue to be
used extensively in electrical and electronic systems models. Dependent voltage sources or current sources are functionally
dependent on a voltage or current variable from within a network of interest. Four commonly used linear dependent sources
are shown in Figure 4.70. These model elements can be generalized in bond graph form as shown in that table. The use
of controlled sources can be used in physical system models when it is not practical or necessary to incorporate certain
underlying physics. Such use should be carefully rationalized to avoid violating relevant physical laws, notably energy
conservation. This was done in Section 4.7.2, where it was shown that modulation of energy storage elements generally
requires a power interaction.
When a controlled source is modulated from across a system by a variable related to a physical device or process the
stability characteristics should be carefully considered.

4.7.6.1 Controlled Source Modeling of Ideal Operational Amplifiers

Controlled sources are useful in modeling the behavior of amplifiers at an abstraction level suitable for certain modeling
applications. For example, circuit modeling of an operational amplifier (op amp) shown in Figure 4.71(a) as an ideal dif-
ference amplifier makes use of a voltage controlled voltage source (VCVS), as shown in Figure 4.71(b). In this model, the
differential input voltage across input terminals defines the output voltage as,

Vout = GOL (V(+) − V(−) ),

where V(+) and V(−) are the non-inverting and inverting input voltages, and GOL is the open-loop gain. The input and output
resistance values are also indicated in Figure 4.71(b).
The most common configuration for an ideal op amp is an inverting amplifier, where the non-inverting terminal (+)
is grounded and the terminals are connected to an input and output using resistors, R1 and R2 , the latter in a feedback
connection from the output to the inverting input as shown in Figure 4.71(c). The output voltage in the inverting amplifier
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
212 4 System Model Formulation and Evaluation

Positive
power supply
Inverting
input

Output
Noninverting
input
Negative
power supply

(a) (b) (c)

Figure 4.71 (a) Symbol for operational amplifier (op-amp) showing key connections, (b) op-amp model using a VCVS, and (c) typical
use of op-amp as inverting amplifier.

(a) (b)

Figure 4.72 (a) Op-amp integrating amplifier. (b) Bond graph based on a VCVS model for amplifier.

model is Vout = −(R2 ∕R1 )Vin , showing that the actual gain, G = R2 ∕R1 , depends only on the resistance values of the input
and feedback resistors. The derivation of this gain relation directly from a bond graph based on Figure 4.71(b) and (c) can
be found in the Solved Problems for this chapter.
Consider now use of a VCVS model for an ideal amplifier at the center of the op-amp integrating circuit shown in
Figure 4.72(a). Using a VCVS model enables construction of the bond graph in (b), and application of causality identi-
fies a state variable for the charge in the capacitor. The causality assignment does indicate an algebraic loop, since for the
controlled source, assume Vout = −GV, where V is the voltage at the inverting input and V = VC + Vout .
The bond graph can now be used to derive the well-known relation between Vout and Vin , which is Vout ≈ −(1∕RC) ∫ Vs dt,
or Vout ≈ −1∕(RCs). The state equation is q̇ C = iC = iR , but note that iR = (Vin − V)∕R and V = VC + Vout = VC − GV
(algebraic loop). Thus, (1 + G)V = VC and V = VC ∕(1 + G), where VC = qC ∕C. The state equation for qC is thus,
1 1
q̇ C = − q + V .
R(1 + G) C R in
Alternatively, we can write
1 1
V̇ C = − V + V ,
RC(1 + G) C RC in
and derive a transfer function in the form,
VC 1+G
= ,
Vin 𝜏t s + 1
where 𝜏t = RC(1 + G). But since Vout = −GV = −G(VC + Vout ) and thus Vout = −GVC ∕(1 + G), this allows us to write,
[ ][ ]
Vout Vout VC −G −1
= = = . (4.120)
Vin VC Vin RC(1 + G)s + 1 RC(1 + 1∕G)s + 1
Note how the gain, G, influences the time constant. But also in the limit as G ≫ 1 this gives the ideal integrating relation
expected. Further, the use of a controlled source can introduce a stability condition. In this case, stability is assured if the
[ ]
pole from the characteristic equation in 4.120, s = −1∕ 𝜏t (1 + 1∕G) < 0, or G > −1.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
4.7 Modulation Structure 213

More discussion about how dependent and controlled sources can be used to model basic electronic devices can be found
in Rosenberg and Karnopp [39]. This includes descriptions of device models (diodes and transistors), which are instructive
but not necessarily practical for most system modeling applications. Nevertheless, these model representations, commonly
introduced in basic electronics for ideal amplifier devices, can be adopted for use in other modeling applications.

4.7.6.2 Controlled Source in Variable-Mass Systems

Mechanical systems of practical interest may have inertia parameters that vary in some way. As previously discussed,
modulating an energy storing element may not properly account for power required to make such a change. These types
of physical effects can be modeled using Lagrangian and open system thermodynamic methods, as will be discussed in
Chapters 8 and 9, respectively. In some cases, however, a simpler approach using a controlled source model can also pro-
vide a way to integrate these effects into a bond graph model. This approach to modeling variable-mass effects can be
demonstrated for the classic rocket model, which is a common way to model mechanical systems that have mass varying
by a specified function of time, say m(t). Examples include mechanical elements that take on mass during motion, balloons
that throw away ballast, or vehicles propelled by expelling a jet, as illustrated in Figure 4.73.
Figure 4.74(a) models a system with total mass m that has mass accumulated at rate m. ̇ During an interval dt, the momen-
tum of the mass changes by m ⋅ d𝑣. During the same interval of time, an amount of mass ṁ ⋅ dt is acquired. The added mass
has velocity 𝑣o ≠ 𝑣. At the instant of contact, the momentum of the added mass changes by (𝑣 − 𝑣o ) ⋅ ṁ ⋅ dt. Therefore,
the total change in momentum of the system along the x-direction in dt is, Δp = m ⋅ d𝑣 + m(𝑣 ̇ − 𝑣o )dt, thus we can write
Newton’s law as,
dp d𝑣
=m ̇ − 𝑣o ) = F,
+ m(𝑣
dt dt
where F represents the net external applied forces. We can rewrite this equation similar to the form given by Tiersten [83],

ṗ = F − 𝜙, (4.121)
̇
where 𝜙 = mu, with u = 𝑣 − 𝑣o , is a form of momentum flux [83, 84]. These expressions are also commonly derived in
dynamics textbooks (e.g., [12, 85]) and enable construction of a bond graph representation, introducing forces induced by

(a) (b) (c)

Figure 4.73 Examples of variable mass systems: (a) a Pelton wheel that captures water, (b) a balloon expelling mass, and (c) basic
rocket with jet.

Friction

(a) (b)

Figure 4.74 (a) Mass with external forces, F(t), and friction, accumulating mass at rate, ṁ (b) Bond graph representation using a
controlled effort source model.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
214 4 System Model Formulation and Evaluation

Heavy chain
mass/length

Friction

(a) (b)

Figure 4.75 (a) A body originally with mass mo slides down an incline while attached to a heavy chain from point S. (b) Bond graph
representation using a controlled effort source model.

the change in momentum due to added mass. The bond graph in Figure 4.74(b) uses common elements for the external
forces, including a friction force. The system mass is represented by an I element that has a total mass, m = mo + mt. ̇
This method can be used to model the case of a mass sliding down an incline while attached to a heavy chain, as shown
in Figure 4.75. As the mass moves down the incline, the apparent mass increases by a rate ṁ = 𝜌𝑣, where 𝜌 is the mass per
unit length of the chain. From the bond graph, there are actually two states in this system: velocity, 𝑣, and x, the modulation
state that represents the length of the chain. The two state equations are:

m𝑣̇ = −𝜌𝑣2 − b𝑣 + mg sin 𝛼, (4.122)

ẋ = 𝑣, (4.123)

where here the friction force Ff between the mass and incline is modeled as linear with damping, b. Note here that,
m = mo + 𝜌x.
For the special case of b = 0, the equation 4.122 can be written,
d
̇ =
m𝑣̇ + m𝑣 (m𝑣) = (mo + 𝜌x)g sin 𝛼,
dt
or,

m𝑣 ⋅ d(m𝑣) = (mo + 𝜌x)g sin 𝛼 ⋅ dt.

This ideal result can be integrated analytically, or a simulation can be used to solve the original form directly.
Many practical problems can be solved using this approach. For example, machines that have varying mass as they lose
matter during vibration can be represented with this approach. These methods are suitable in some applications. In some
cases, there can be additional effects or features in a system that need to be accounted for, such as if the mass changes occur
due to a flow of matter. In some cases, the matter flow across a surface and from a contained volume should be tracked as
discussed in Section 4.7.5. The effects of variable mass and inertia in these contexts are taken up in Chapter 8 using energy
and Lagrangian methods. The Lagrangian approach can be more suitable when there is more information about how mass
or inertia parameters are influenced by a system under study. Additional discussion on open systems with thermodynamic
effects is taken up in Chapter 9.

4.7.6.3 Controlled Source in Fluid Filling of a Tank

Hydraulic systems can have fluid supplied into tanks or vessels in various ways. One way is when fluid flows directly into
the contained fluid, as represented by the flow Q1 in Figure 4.76(a). In this interaction, the flowrate will be influenced
by the back pressure, here equal to the tank base pressure, PC . So in this case, Q1 = Φ1 (Pa (t) − PC ), where Φ−11 () is the
inverse of a hydraulic resistive constitutive relation for the inlet. In contrast, when fluid is supplied in an “uncoupled”
mode indicated by Q𝑣 , which is controlled by the flowrate through the control valve. By using a flow-controlled-flow-source
(item (d) in Figure 4.70), PC is now “uncoupled” and does not influence Q1 . Despite this difference, both source flows
contribute identically to the state equation for V
– C,

–̇ C = Q1 + Q2 − Ql ,
V
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
4.8 Summary 215

Valve

“Uncouple”
“Uncouple”

Constant tank
area,

Directly
coupled
Resistive losses
Directly
coupled

(a) (b)

Figure 4.76 (a) Tank filled in two ways: coupled and uncoupled. (b) Bond graph representation using a controlled flow source to
represent the uncoupled flow into the tank.

however, only Q1 and Ql , in this case, would depend on dynamic effects in the tank. Both types of flow sources, coupled and
uncoupled, are very common in these types of applications and the use of a controlled source provides a way to distinguish
the differences which can impact the overall system dynamics.

4.7.7 Summary of Modulation Structure

This extended section has reviewed restrictions on the use of modulation in physical system modeling. It was shown that
modulated energy storage should not be allowed, but that modulation of coupling elements and dissipative elements can
be used to model a wide range of physical system effects. Modulation plays a role, for example, in kinematic constraints
of rigid bodies, electromechanical and gyroscopic coupling, as well as in modeling some fluid–structure interactions. The
concept of composite modulation of coupling elements should be avoided in well-structured models. The role of modulation
was also shown to play a role in modeling with controlled or dependent sources. While the latter can be controversial,
their use in modeling electronic devices is long standing. Some examples of how controlled sources can be used in other
energy domains was also shown. It is important to make sure that fundamental laws are not violated when applying these
methods.

4.8 Summary

A key step in formulation of physical system models is studying equilibrium conditions, which help identify operating
conditions and initial conditions, both essential for applying system linearization methods. Linearized models are used in
system analysis, design, and in evaluating stability, concepts that will be discussed in Chapters 5 and 6. Model formulation
in terms of linear transfer functions has been reviewed, as well as allowing integration of bond graphs with signal-based
block diagram descriptions. We showed that a causally complete bond graph provides a reliable basis for generating a proper
block diagram description of a physical system.
This chapter also introduced methods that can help ensure models will reflect reality by reviewing certain restrictions on
constitutive relations. These restrictions provide insight into model development and especially into computational model
formulation. Since it can be necessary to use modulation to represent a wider range of physical effects, proper ways to use
and restrict modulation were described. We found that modulation and information states can be introduced and handled
in an analogous manner to energy-based or Hamiltonian (i.e., p and q) state variables.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
216 4 System Model Formulation and Evaluation

4.9 Problems

Equilibrium analysis

Problem B-4.1 Two mass vibration absorber Figure B-4.1 shows mass m2 and spring k12 acting as a vibration absorber,
designed to minimize the effect of unwanted forces F(t) applied to mass m1 .

Figure B-4.1 A mechanical system in

equilibrium.

a) Develop a bond graph, assuming that gravity acts in the downward direction (along 𝑣1 and 𝑣2 ).
b) For the case where F(t) = 0, solve for the equilibrium values of the spring deflections.

Problem B-4.2 Mechanical elastic system equilibrium Consider the mass–spring system shown in Figure B-4.2. The
two masses, A and B, are interconnected by three elastic spring elements labeled 1, 2, and 3.
a) Develop a model by building a bond graph and derive the state equations assuming the springs are generally nonlin-
ear with constitutive relations of the form Φi (xi ), where i is the number of the spring and xi is the associated spring
state. Each cart has a mass, mA and mB . Forces are applied on each mass as shown.

Figure B-4.2 A mechanical system in equilibrium.

b) Assume the springs are linear, Φi (xi ) = ki xi , and k1 = 10, k2 = 20, and k3 = 15. Solve for the equilibrium spring
displacements, xi , for the case where the FA = 100, and FB = 150. All units are SI.
c) Consider now springs that are nonlinear and of the form Φi (xi ) = ki0 + ki3 xi3 , with the same loads and with the
spring coefficients: k10 = 5, k13 = 2, k20 = 3, k23 = 4, k30 = 6, k33 = 5. Solve for the equilibrium states as in (a).

Problem B-4.3 Two-dimensional mass–spring system: finding equilibrium For the mass–spring system shown in
Figure B-4.3, assume that the springs have linear constitutive laws and free lengths of L1 and L2 . The system is
confined to move in a plane, with the mass, m, exposed to gravity, g, in the z-direction.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
4.9 Problems 217

Figure B-4.3 Mass–spring balance.

a) Use Newton’s law to find equations for determining the static equilibrium of this system. Write these equations in
vector–matrix form.
b) Show that the potential energy stored in this system can be expressed as,
[ ]2 [ ]2
1 √ 1 √
U(x, y) = k1 x2 + y2 − L1 + k2 (x − D)2 + y2 − L2 − mgy.
2 2
The equations for static equilibrium can be derived by using the energy expressions. At static equilibrium, the energy
will be a minimum, so ∇U = 0.
c) Contrast the two results. Are they the same? How are they different, if at all?
d) Use these results to find the static equilibrium when k1 = 10 N/m, k2 = 20 N/m, L1 = L2 = 0.1 m, D = 0.1 m,
m = 0.1 kg, and g = 9.81 m/sec2 .

Problem B-4.4 Equilibrium of coupled oscillators For the system shown in Figure B-4.4, develop state equations that
can be used to find an expression for the static equilibrium (in terms of x1 , x2 , and x3 ).

Figure B-4.4 Coupled spring–mass elements in equilibrium.

Write the expression for static equilibrium in the form A x = b. For the case where all springs have stiffness values
equal to 1, and with L equal to 4, determine the values of x1 and x2 when the system is in static equilibrium. Assume
the springs have unstretched lengths of 2, 2, 1, and 1 (from left to right).

Problem B-4.5 Wheatstone bridge for null measurement A Wheatstone bridge circuit configured for a null measure-
ment of an unknown resistance is shown in Figure B-4.5. In a null measurement, the resistor R2 is adjusted until there
is no current detected by the null detector (an ammeter), which has a specified resolution (in amps).

n Null
ow detector
n kn
U

Figure B-4.5 Wheatstone bridge

with null detector.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
218 4 System Model Formulation and Evaluation

a) Show how the null measurement allows us to ideally find R4 through the relation: R4 = R2 R3 ∕R1 .
b) Develop a bond graph model of the entire circuit, including a model for the ammeter. The ammeter sensing can
be modeled ideally (i.e., by an effort source), but it should include an internal resistance of Rm .
c) Apply causality, and count the number of algebraic equations required to determine all unknowns in this system.
Derive the (linear) algebraic system in terms of the currents in resistors 1, 2, and 3. Write the equations in matrix
form.
d) We want to show that you cannot choose the values of R1 = R3 arbitrarily. This can be done by solving for the
current through the ammeter as a function of R2 (adjustable) for the values of R1 = R3 = 1, 100, or 100,000 ohms.
Let Vs (t) = 6 volts and assume the unknown resistance is R4 = 1 ohm. The smallest current the ammeter can detect
is 10−4 amps and assume Rm = 10−3 ohm.

Problem B-4.6 Bond graph equilibrium analysis – 1 A bond graph is given in Figure B-4.6 for a system that requires
solving for equilibrium states of q4o and p7o given input value of e1o .

R : R2 C : C4 R : R6

e2 f2 e4 q̇4 e6 f6

e1 (t) e3 e5 ṗ7
Se 1 0 1 I : I7
f1 f3 f5 f7

Figure B-4.6 Bond graph for equilibrium analysis.

a) Assign causality and derive the state equations.

b) Determine the equilibrium equations by setting ẋ = 𝟎.
c) Use the equations in (b) to derive relations for q4o and p7o in terms of e1o and system parameters.
d) What is the equilibrium power dissipated in each resistor?
e) Is the equilibrium stationary or static?

Problem B-4.7 Bond graph equilibrium analysis – 2 A bond graph is given in Figure B-4.7 for a system that requires
solving for equilibrium states of p6o and q7o given input value of e1o .

R : R2 R : R4 I : I6

e2 f2 e4 f4 ṗ6 f6
e1 (t) e3 e5 e7
Se 1 0 1 C : C7
f1 f3 f5 q̇7

Figure B-4.7 Bond graph for equilibrium analysis.

a) Show that causality assignment indicates there is an algebraic loop.

b) Identify states and derive the state equations.
c) Determine the equilibrium equations by setting ẋ = 𝟎, this will include an equilibrium relation for the loop variable.
d) For values of R2 = 1, R4 = 4, C7 = 1, I6 = 10, and e1o = 10, solve for q7o and f4o .
e) Compute the equilibrium power dissipated in R4 for the numerical values given in (c).

Problem B-4.8 Bond graph equilibrium analysis – 3 Consider the purely resistive network bond graph given in
Figure B-4.8.
R : R2 R : R4 R : R6 R : R8

e2 f2 e4 f4 e6 f6 e8 f8

e1 (t) e3 e5 e7 e9
Se 1 0 1 0 R : R9
f1 f3 f5 f7 f9

Figure B-4.8 Bond graph for equilibrium analysis.

 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
4.9 Problems 219

a) Use causality assignment to identify the algebraic loop(s) and variable(s).

b) Determine the equilibrium equations.
c) For values of R2 = 1, R4 = 4, R6 = 2, R8 = 6, and R9 = 2, compute all equilibrium effort and flow variables and the
equilibrium power dissipated in R9 .

Problem B-4.9 Pressure pump controlling two tanks Your colleague has sketched an electrical analog of the hydraulic
system shown in Figure B-4.9(a), which you are both tasked with analyzing. Develop a bond graph and contrast with
the electrical circuit analog shown in (b). Use equilibrium causality and the resulting relations to find a relationship
between the initial pressure in each tank, PAo and PBo , and the pressure from the constant pressure pump, Po .

Tank A Tank B

Pressure pump,
(a) (b)

Figure B-4.9 (a) Constant pressure pump controlling two tank volumes.
(b) Electrical analog of hydraulic system.

Problem B-4.10 Circulation pumping system in equilibrium The system in Figure B-4.10 has a circulation pump with
controlled flowrate, Q(t) at point C, and can drive flow between tanks A and B. Assume that there is fluid inheritance
and resistance in pipes 1 and 2 as indicated.

Circulation
pump,

Figure B-4.10 Hydraulic fluid circulation

system.

a) Develop a bond graph.

b) Apply equilibrium causality.
c) Derive equations that can be used to solve for the equilibrium states of this system, assuming Q(t) = Qo .

Problem B-4.11 Fluid resistive network You’re asked to analyze a fluid distribution network sketched in the form shown
in Figure B-4.11. The pressure conditions are assumed known at the inlet (A) and outlets (B, C, and D). Each of the
connecting fluid lines (labeled a to f ) have known geometric properties, and the fluid is assumed to be incompressible.

Figure B-4.11 Electrical circuit in equilibrium.

a) Develop a bond graph model of this system, assuming linear resistive elements can represent the connecting fluid
lines (no fluid inertia; assume flow will be laminar).
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
220 4 System Model Formulation and Evaluation

b) Assign causality to reveal the algebraic loop(s). Study the consequence of making “nonoptimal” choices in arbitrary
assignment for algebraic loops by deriving equations in multiple cases and showing whether any problems arise,
simplifications are made, etc.
c) Summarize your findings.

Problem B-4.12 Pushrod-lifter-valve simulation A pushrod-lifter-valve system was studied in Problem B-3-29. Develop
a simulation model to study floating of the valve which can occur for a given design (mass, stiffness, etc.) and for
certain values of the cam rotational velocity, 𝜔, which shows up as the input velocity forcing frequency; that is, 𝑣cam (t) =
𝑣o sin(𝜔t).
a) Assume that the pushrod initially has a preload force of Frod = 1 N. Determine the values of the other states under
this condition.
b) Develop a simulation of the system and determine if floating occurs for the given conditions and parameters given
below. Generate a figure with plots of: (1) valve displacement and cam displacement (in mm), and (2) pushrod force
(in N). For this system, you should show whether floating occurs, which is when Frod < 0.
c) If floating does occur, do two things: (i) adjust one or more system parameters to eliminate float, and (ii) adjust 𝜔
to a value where floating will not occur.
For the initial study: 𝑣o = 2.4 m/s, and 𝜔 = 240 rad/s, n = 1, m𝑣 = 0.01 kg, k𝑣 = 10 N/m, and let b𝑣 = 0.3 N s/m.

Problem B-4.13 Mass–spring on a rail A mass that is constrained to move along a rail is shown in Figure B-4.13

L L

θ z
k k
M

Figure B-4.13 Mass constrained on rail

with spring restraints.

a) Formulate a constitutive relation F(z) that models the total force due to both springs which is applied in the z
direction on the mass as it moves along the rail. Assume each spring itself is linear and account for the angle 𝜃 and
that the spring is unstressed when z = 0 and 𝜃 = 0.
b) Develop a bond graph and develop the state equations, making sure to account for the effect of gravity, and add a
linear damping force along the rail with damping constant b.
c) Develop a simulation to study vibration of the mass about a static equilibrium. Let m = 1 kg, k = 100, b = 0.5 N s/m,
and L = 1 m, and find the static equilibrium position, ze . Study the unforced response when releasing the mass from
rest and displaced from ze by an amount 2ze .
d) If this system was used to model the support of a package so that the ground supports had prescribed motion (e.g.,
from vertical motion of a ground vehicle), show how the bond graph would change. Assume only motion in the
vertical direction. The problem can be extended to include use of this model to refine a package support design.

Problem B-4.14 Maximum power transfer in source–load problems This problem considers a few cases of power trans-
fer from a source to a dissipative load.
a) A general steady-state source model can be represented by an ideal effort source, es , connected to a resistive element
with common flow, leading to a Thevenin equivalent model, or an ideal flow source, fs , connected to a resistive
element with common effort, called a Norton equivalent model. Sketch an electric circuit realization of each of
these models and the associated bond graph. Label the resistive element as the source resistance, Rs , in each case.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
4.9 Problems 221

b) Figure B-4.14 illustrates an electrical Thevenin source model driving an electrical resistive load, RL . Show that
maximum power is achieved in this case when Rs = RL .

Figure B-4.14 Electrical Thevenin source model driving a load

resistance.

c) An AC power adapter for an electronic load provides DC power rated at 9 V DC for 300 mA. The “open” circuit
voltage on the power adapter reads 14.4 V DC, and the “short” circuit current is 1700 mA. Estimate the power
dissipated in the “load” as well as how much is dissipated in the adapter? Sketch the operating conditions on
voltage–current plots. Is maximum power transfer achieved? Is the operating condition desirable? Explain why or
why not.

Problem B-4.15 Real source–load analysis A general source model with the data provided in Figure B-4.15(a) is mod-
eled using a Thevenin source, represented by the Se − R within the dashed box in the bond graph shown in (b). The real
source is characterized by the data: eo = [5, 4.1, 3.500, 2.600, 1.000], and fo = [0, 0.005, 0.010, 0.015, 0.020]. This system
drives a linear RC load.

6
5
4 R R:R
eo

3 eR fR
2 eo eC
Se 1 1 C:C
1 fo q̇C
0
0 0.5 1 1.5 2
fo .10–2
(a) (b)

Figure B-4.15 (a) Source characteristics and (b) bond graph model with source represented by
Thevenin form and source driving linear RC load.

a) Apply causality to show that the system has an algebraic loop, which arises because of the structure of the real-source
model (an implied resistive element).
b) Derive state equations and write an expression for the loop flow variable, fR , assuming a given function for the
source, eo = eo (fo ).
c) Develop a simulation that compares the case where the source is ideal (not dependent on fo ) with eo,ideal = 5 and
with R = 200 and C = 0.01. Compare the results from the ideal source case with those from the real source case.
Generate three plots for comparing: (a) the input effort, eo , (b) the flow, fR , and (c) the effort, eC .
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
222 4 System Model Formulation and Evaluation

Problem B-4.16 Motor fan loading A motor and fan are to be connected as shown in Figure B-4.16. The torque–speed
characteristics of the motor and fan are plotted on the same graph in Figure B-4.16(b).

Fan 12 Fan load

Belt
Torque
8
Motor (in-lbf)
Pulley 4 Motor
torque-speed
0
0 1000 2000
Speed (rpm)
(a) (b)

Figure B-4.16 (a) Motor-fan system (b) Motor and fan characteristics.

a) Draw a bond graph model of this system, neglecting any storage of energy (i.e., include only sources, loads, and
ideal power conversion).
b) Determine the speed of the motor that maximizes its allowable power, and find this power.
c) Determine the speed and torque of the fan for maximum power transfer as well as the pulley ratio that achieves
them. Neglect belt losses (and stretching), so power is conserved.

Problem B-4.17 Equilibrium of closed cylinder The cylinder shown in Figure B-4.17 has a circular piston of mass
mp placed into it, trapping the contained volume of air, which is initially at standard temperature and pressure
(Patm ,Tatm ).16 There is negligible leakage, but the piston (diameter dp ) can move within the cylinder (diameter dc ) and
settles at a point where the weight of the mass balances the pressure of the contained air, Pc .

Negligible friction
and damping

Air at STP

Figure B-4.17 Closed piston-cylinder in

equilibrium.

a) Assume there is no leakage and that there is linear friction with constant damping b, and the piston can move
within the cylinder. Gravity acts on the mass in the vertical direction. Develop a bond graph model of this system.
Represent the contained air using a two-port C but indicate by causality that the temperature of the contained air
Tc is constant.
b) Identify state variables for this system and write the state equations.
c) Write the constitutive equations for the two-port C model for the contained air relating the pressure and temperature
of the air contained in the cylinder, Pc and Tc , to volume and entropy states. Assume the amount of air remains
constant (no leakage).
d) Derive relations that describe the equilibrium condition(s).

16 Standard Temperature and Pressure (STP) is defined by 273.15 K (0 ∘ C, 32 ∘ F) and an absolute pressure of exactly 105 Pa (100 kPa, 1 bar).
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
4.9 Problems 223

Problem B-4.18 Steady-state pmdc motor modeling The torque–speed curve for a permanent-magnet dc (pmdc) motor
relates the torque available from the motor as a function of the shaft angular speed at steady state. The electrical circuit
diagram for a pmdc motor in steady state is shown in Figure B-4.18(a). In steady-state (or equilibrium) operation,
motor inductive and inertial effects are negligible, so 𝜆̇ and ḣ are indicated as equal to zero. These effects are neglected,
so the pmdc torque–speed curve only depends on three parameters: motor constant, rm , electrical resistance, Rm , and
rotational damping, Bm (only to account for linear losses).

LOAD

LOAD

(a) (b)

Figure B-4.18 (a) pmdc schematic indicating steady-state operation. (b) Bond graph of the pmdc motor
with equilibrium causality.

a) Begin by applying equilibrium causality on the bond graph provided in Figure B-4.18(b), assuming input voltage,
Vs is known. Specify load velocity, 𝜔o , on the output shaft as an input (indicating one algebraic loop). Complete the
equilibrium causality.
b) Use the equilibrium bond graph in (a) to guide derivation of the torque-speed relation. Show that the torque deliv-
ered to a load attached to the output shaft as a function of the shaft speed at steady state, 𝜔o , is 𝜏o = 𝜏o (𝜔o ) =
𝜏s − Be 𝜔o , where 𝜏s = rm Vs ∕Rm (stall torque), and Bo = Bm + rm
2 ∕R . This is the ideal torque–speed curve.
m
c) Derive an expression for the output power o as a function of 𝜔o
d) Select parameters for a pmdc motor that has parameters specified by a motor manufacturer. Select the motor rate
input voltage and summarize the key parameters. Plot the power curve on the same graph as an efficiency curve
(scale as necessary or use 2 independent vertical scales), with efficiency defined as, 𝜂 = out ∕in .
e) For the pmdc motor studied in (d), determine a value Bo,max for a purely linear rotational damping load that will
result in maximum power being delivered to the pmdc motor. What is the value Bo,eff for maximum efficiency?

Problem B-4.19 Thermal model of a pmdc motor The steady-state motor model of Problem B-4-18 shows that the
torque–speed curve depends only on the motor internal electrical (Rm ) and mechanical (Bm ) loss effects. Account-
ing for thermal effects (e.g., Problem B-2-34) allows one to quantify limits on the effective rotor/winding temperature,
Tr , relative to the ambient environment at temperature Ta .
a) Extend the equilibrium bond graph model to include how the key heat transfer effects from Rm and Bm influence
the stored internal energy in an effective rotor/winding thermal capacitance, Cr .
b) Write the relations for heat transfer due to electrical, Q̇ e , and mechanical, Q̇ m , effects, and develop expressions for
the corresponding entropy flow rates, fse , and fme , respectively.
c) Write the state equation for the rotor/windings entropy, Sr , which gives temperature Tr = Sr ∕Cr .
d) Impose an equilibrium thermal condition, Ṡ r , and use the resulting relation to find a relation between the motor
current, im and the differential temperature, ΔT = Tr − Ta .

Linearization

Problem B-4.20 Simple pendulum interacting with U-magnets. A pendulum with a metal bob hangs between two
U-shaped magnets as shown in Figure B-4.20. The force one magnet exerts on the bob is given by Fm = c∕x2 , where c
is a constant.
a) Develop a bond graph model as in Problem A-4-13.
b) Derive the state equations.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
224 4 System Model Formulation and Evaluation

l
θ

m
x
h

Figure B-4.20 Pendulum between two

U-shaped magnets.

c) Linearize the state equations about 𝜃 = 0 for small motions (l𝜃 ≪ h − x).
d) Determine the characteristic equation and use it to assess the effect of h and c on stability.

Problem B-4.21 Analysis of linearized mass–spring–damper Revisit Example 4.5 on equilibrium and linearization of
a mass–spring–damper with nonlinear spring and damper. This system has a spring force, F(x) = k(x + 𝛼x3 ), and
square-law damper, FD = 𝛽|V|V, with the state equations,
[ ] [ ]
ẋ p∕m
= .
ṗ −𝛽|p|p∕m2 − k(x + 𝛼x3 ) + F

a) Solve for the equilibrium relations given F = Fo is a constant.

b) For parameter values, m = 1 kg, k = 0.1 N/m, 𝛼 = 1e10 m−2 , and 𝛽 = 1000 kg/m, numerically determine the equi-
librium position, xo , for values of Fo = −10, −5, −2.5, 0, 2.5, 5, 10 ×109 N.
c) At each value of xo , plot the linearized spring force against the plot of F(x) (the nonlinear spring force).
d) Assume an initial applied force of Fo = 1e9 N sets the system at a corresponding equilibrium xo (mass velocity zero).
Beginning at time t = 0, simulate the response of the system given a “small” step change ΔF from the equilibrium
force to another value, that is, F = Fo + ΔF. Compare to a simulation of the response of the linearized system model.
Determine how large a change be made ΔF before the linear model becomes invalid?
e) For this system, tangential linearization leads to a model that has no damping for xo = 0. Use secant linearization
of the damping term to improve the linearized model. Show that the secant-linearized model better predicts the
damped response about xo = 0.

Problem B-4.22 Linearization of a nonlinear hydraulic damper A proposed damper design requires a model of a device
that will damp the motion of a large mass. The system proposed is shown in schematic form in Figure B-4.22.

Rigid rod Orifices

Seals

Piston with area Orifice areas,

rod area,

(a) (b)

Figure B-4.22 (a) Mass coupled to nonlinear damper. (b) Bond graph of mass-rod-damper system.

a) A simplified bond graph is given in Figure B-4.22(b) to guide a first analysis. In this first model, consider the rod
to be rigid (no compliance). Assume the flow between chambers 1 and 2 through any orifice follows the relation,
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
4.9 Problems 225

√ √
2
Q = Co ⋅ Ao ⋅ 𝜌
∣ P1 − P2 ∣, where Co is a constant (that depends on Reynolds number, etc.), Ao is the orifice
√
area, and 𝜌 is the fluid density. Simplify this notation by letting, Ko = Co ⋅ Ao ⋅ 𝜌2 , and let ΔP ≡∣ P1 − P2 ∣. The
√
relation above becomes either: Q = Ko ΔP, or ΔP = Ko−1 Q ∣ Q ∣. Consider the two possible causal assignments you
can have on the R-element (which represents this nonlinear damper): either Q in or ΔP in. For each case, linearize
the constitutive relation about a general equilibrium point (either Qe or ΔPe ). Are there any possible difficulties that
may arise in using the linear approximations.
b) Assume now that you must size the rod, so consider its compliance in this model. Let the rod stiffness be kr .
Draw a new bond graph, develop the nonlinear state equations, and linearize again (choosing proper equilibrium
conditions).

Problem B-4.23 Air-cushion vehicle: vertical equilibrium A plenum chamber air-cushion vehicle is shown schemati-
cally in Figure B-4.23. Assume the platform is essentially rectangular, and that there is an effective cushion area, Ac ,
that can be approximated by the known length and width of the cushion. A fan with known source characteristic
(Pf − Qf ) provides air flow into the plenum chamber of rectangular shape (length L, width 𝑤). The pressure (gage)
in the plenum chamber, Pcu , supports the total mass, m, and flow is discharged through the clearance height, hc . The
cushion wall angle with the horizontal is 𝜃c (e.g., 45∘ ).

Ideal fan
curve
Typical fan
curve

(a) (b)

Figure B-4.23 (a) Plenum chamber in an air cushion vehicle. (b) Ideal and real fan
characteristics.

a) Assuming that air in the chamber is essentially at rest, use Bernoulli’s equation to show that the flow exiting the
chamber can be approximated by the relation,
√
Qc = hc lcu Dc 2Pcu ∕𝜌,
where lcu is the cushion perimeter, 𝜌 is the air density, and Dc is a discharge coefficient. The value of Dc depends on
the wall angle, 𝜃c , as shown in the table below [58].

𝜃c 0∘ 45∘ 90∘ 135∘ 180∘

Dc 0.50 0.537 0.611 0.746 1.000

b) Use a linearization procedure to find a relation between Pcu and V– cu , where V – cu represents the change in volume,
not the total volume in the chamber. Assume that the air in the chamber can be compressed adiabatically. Show
that you can define a compliance of the air in the chamber by, Ccu = – Vo ∕𝛾Po , where 𝛾 is the ratio of specific heats,
and V– o and Po define the operating equilibrium point.17

c) Build a bond graph model for the vertical motion of this vehicle. Your model should be such that the fan flowrate
is specified into the chamber. Note, however, that this is Qf = Qf (Pf ) (i.e., it is not an ideal flow source). Apply
causality and derive state equations.

17 While not a closed system, assume there are no significant changes in contained mass of air.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
226 4 System Model Formulation and Evaluation

d) Assume that a source characteristic for the fan is given between the pressure and flow in the form (as adapted
from [7]),
Qf = Qo − a1 Pf + a2 Pf2 − a4 Pf4 ,

where Qo = 180 m3 ∕s, a1 = 0.08 m5 ∕N s, a2 = 2.0 × 10−5 m7 ∕N2 s, a4 = 1.482 × 10−12 m9 ∕N3 s, and Pf is the associ-
ated fan pressure (in Pa). For air, let 𝜌 = 1.23 kg∕m3 . For each case in the table below, determine the elevation, hc at
equilibrium.

Case Width (m) Length (m) Height (m) Total mass (kg)

Case 1 [58] 6.09 12.19 1.0 8164

Case 2 [7] 4.00 10.00 0.8 4000

Problem B-4.24 Material temperature with embedded heating element A system is designed to maintain a block of
material (e.g., aluminum) at a specified temperature using an embedded heater that has resistance, Rh . When current
flows through the resistive heater, it generates heat, Q̇ h = V ⋅ i = Rh i2 .
a) Explain how the material can be modeled as a lumped linear thermal capacitance with constitutive relation for the
temperature Tm as a function of entropy Sm , Tm = Sm ∕Cm .
b) Develop a bond graph that models all the significant power flows influencing the rate of change of internal energy
in the block. Include heat transfer to the ambient. Apply causality to the bond graph and show that the entropy
state for the thermal capacitance of the material, Sm , is a state of the system.
c) Derive a state equation first in terms of the entropy of the block material, Sm , and then show that you can convert
this equation into a state equation in terms of the temperature, Tm . You should find that this equation is linear in
Tm , but the current is a nonlinear term.
d) The heat generating function is Q̇ h = Rh i2 . Let the equilibrium current value, io , be the value required to maintain
the material at Tmo . Find a relation for io . Assume that the following parameters are provided for this prob-
lem: Rh = 2Ω, Ka = 1.46 × 10−6 W/K (thermal conductivity), Cm = 0.00146 J/K (K = Kelvin) (effective thermal
capacitance)
e) What is io if Tmo = 100 ∘ C and Ta = 50 ∘ C?
f) To obtain a linear model, linearize the nonlinear current relation in the temperature ODE of part (c) about the
quiescent (equilibrium) point (Tmo , io ). One form of the resulting linear equation is,
Km Ṫ m + Ka Tm = qho + Kh Δi + Ka Ta ,
where Δi = i − io and qho = Rh i2o (the equilibrium heat required to maintain at Tmo ).

Signals, block diagrams, and transfer functions

Problem B-4.25 Transfer function from block diagram Consider the nonlinear block diagram in Solved Problem A-4-7.
Make the same linear approximations adopted in that problem and redraw the linearized block diagram of the system.
Use block diagram reductions to derive the transfer function, G(s) = 𝑣m ∕Ps .

Problem B-4.26 Vibration absorber Develop a transfer function that relates the velocity of mass m1 in the system of
Problem 4.1 to the applied force F(t). Assume that the system is at equilibrium with gravity and that motions are all
relative to the equilibrium condition.

Problem B-4.27 Rotational drive subsystem A segment of a drive train under study is shown in Figure B-4.27(a), with
a known controlled torque 𝜏c (t) applied onto J2 and a disturbance torque 𝜏d on J1 . Verify the bond graph provided in
Figure B-4.27(b) and then apply causality, identify states, and derive state equations. Use the state equations to derive
a transfer function 𝜔1 ∕𝜏c .
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
4.9 Problems 227

I : J1 C : 1/K I : J2

ḣ1 ω1 τk θ̇k ḣ2 ω2

τd τk τk τc
Se 1 0 1 Se
ω1 ω2

(a) (b)

Figure B-4.27 (a) Rotational drive subsystem and (b) bond graph.

Problem B-4.28 Electrical circuit transfer function from state space The electrical circuit in Figure B-4.28(a) is powered
by a current source and an electrical source as shown.

R:R C R:R

iR q̇C
Vs (t)
Sf 1 0 1 Se
is (t)

λ̇L

I:L
(a) (b)

Figure B-4.28 (a) Circuit with multiple electrical sources and (b) proposed bond
graph.

a) Confirm the bond graph proposed in Figure B-4.28(b).

b) Apply causality and derive state equations.
[ ]T
c) Write an output equation for y = V1 V2 .
d) Derive the transfer function matrix (a 2 × 2 result).

Stability

Problem B-4.29 Chamber-valve stability Figure B-4.29 shows air entering a chamber at port A that can exit through a
valve of mass m. At equilibrium, the flowrate QA is fixed and equal to flowrate exiting through the valve, Qo . Pressure
is constant under these conditions.
The area of the valve piston is A𝑣 , and xm is the valve position relative to the chamber wall. Assume the flowrate of
air through the valve is proportional to the valve position and to the pressure drop, in a generally nonlinear form,
Qo = f (xm , Pc ).

(a) (b)

Figure B-4.29 (a) Air chamber with spring-loaded valve

(b) Valve-spring geometry.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
228 4 System Model Formulation and Evaluation

√
a) Develop a bond graph model of this system, assuming a nonlinear form for flow through the valve, Qo = K(xm ) Pc ,
where Pc is the gage pressure. This chamber pressure can be modeled as a function of the steady-state pressure
and volume of air in the chamber, (Pco , –Vco ), and assume Pc –
Vc = constant. A linearized form of the system can be
simplified to the form,
m̈xm + kxm = A𝑣 Pc ,
Ṗ c = Gx,

where G is a constant that incorporates a linear trend in the function, K(xm ) = K𝑣 xm . Note that this trend
can take a different form based on whether the valve has the configuration in Figure B-4.29(a) or the one in
Figure B-4.29(b).
b) After showing the system model can be expressed in the simplified form given in (a), explain how the sign of G
depends on the slope sign of K𝑣 .
c) Discuss the stability of the system when the sign in G takes different signs as determined in part (b).
d) If a linear damper b is added in mechanical parallel to the spring, how will this influence the results found for
part (c)?

Problem B-4.30 Valve stability with controlled inflow The spring-loaded valve used to regulate flowrate Qin may exhibit
stability problems. The valve piston of mass m and area A is restrained by a spring k within a chamber as shown
in Figure B-4.30. Assume that the air is incompressible. For a fixed pressure source of air, it is found that the air
entering the perforated section has flowrate, Qin = c1 z, where c1 is a constant. Further, the leaking flowrate can be
modeled by Ql = c2 Pl , where c2 is a constant and Pl is the pressure drop to the exterior. Gage pressures are assumed in
this model.

Air flow in

Figure B-4.30 Modulated valve flow

into chamber.

a) It is convenient to model the inflow Qin as a modulated source. Develop a bond graph that includes the spring, mass,
and leakage resistive elements. Assign causality and show that the mass velocity 𝑣z , and spring deflection, zk , are
energetic states, and include a modulation state z for mass position.
b) Use the model to determine condition(s) for system stability.
c) At equilibrium, the mass is not moving, 𝑣zo = 0, the spring has preload, zko , and Qlo = Qino . Assuming the mass
equilibrium position, zo , is known, find an expression for equilibrium pressure in the chamber, Pco .
d) It is suggested that a more complete model of this system would represent the perforated section as a modulated
resistive element. To show how this would change the system model, assume a known source pressure Pin and
model the flow through the perforated section by the relation, Qin = c3 z(Pin − Pc ), where c3 is a constant. It may be
necessary to deal with an algebraic loop and additional linearization of the resulting equations in order to determine
if there are any condition(s) for stability.

Problem B-4.31 Mass suspended by electromagnet The schematic in Figure B-4.31 depicts an electromagnetic coil
fixed to ground, which may also have velocity, 𝑣i (t). It is desired to control the gap, xg , between the coil and the per-
manent magnet (PM). Magnetic levitation of a nonmagnetic mass relies purely on the attractive force induced by
variable-reluctance effects.18 Assume here that the total electromagnetic force can be modeled by the combined effect

18 Chapter 8 introduces modeling of electromechanical devices of this type.

 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
4.9 Problems 229

of a current-induced force and a force due to the position of the permanent magnet. The latter will be present even
when there is no current. Short of developing electromagnetic Finite Element Method (FEM) models to predict how
these forces depend on key variables, it is possible to estimate them experimentally.

due to if is
current

Air gap
Suspended
mass
Magnetic
flux lines

Figure B-4.31 Mass suspended by electromagnet, with insert showing electromagnetic

force due to current and effect of permanent magnet.

a) Assume the forces on the mass are given as Fem = Gi2 ∕xg2 + Fpm (xg ), as indicated in Figure B-4.31, and derive the
state equations, assuming the current, i, is specified as an input.
b) Linearize the state equations about an equilibrium gap, xgo , and assumed equilibrium current, io .
c) Show that the transfer function relating the gap, xg , to the drive current, i, is given by,
xg Ko
= ,
−i s2 − a 2
where Ko and a are defined for equilibrium values of xg and i.
d) Comment on the stability of the system about a given equilibrium point.

Constitutive structure

Problem B-4.32 C-element with exponential energy function A single-port C-element has an energy function given by
Uq = eq . Comment on the validity of this element.

Problem B-4.33 Composite C-element A single-port C-element has the composite constitutive relation shown in
Figure B-4.33. Decompose this relation to form basic C-elements.

1 2 3

Figure B-4.33 Composite, nonlinear C-element constitutive

function.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
230 4 System Model Formulation and Evaluation

Problem B-4.34 Cantilever beam model as two-port C Consider the cantilever beam system shown in Figure B-4.34(a).

(a) (b)

Figure B-4.34 Composite, nonlinear C-element constitutive function.

a) Develop a bond graph for this system (neglecting gravity).

b) If the constitutive relation for the beam is given as: F1 = x1 + x2 and F2 = 2x1 + x2 , calculate the work required to
traverse path  1 to point (1,1). Repeat for path 
2 .
c) Perform a digital simulation of the system with m1 = m2 = 1 and initial conditions p1 (0) = 0, x1 (0) = 1. p2 (0) = 0,
and x2 (0) = 1.
d) Comment on the results and validity of the model.

Problem B-4.35 Hysteresis modeling Hysteresis behavior arises in many types of physicals systems. In mechanical sys-
tems, this behavior can arise due to combined energy storing and loss effects due to elastoplastic behavior of materials.
This leads to a limiting effect in the forces (the effort variable). In magnetic systems, magnetic flux (the displacement
variable) is limited by saturation. These two cases result in the qualitative differences seen in typical “hysteresis loops”
as in Figure B-4.35(a). The area enclosed by a hysteresis loop is related to energy lost as the system cycles the variables.
Hysteresis effects can be modeled using combinations of RC elements, such as in Figure B-4.35(b). Such a concept
has been investigated computationally in different ways, including the works by Caughey [86], Paynter [87, 88], and
Karnopp [89].

q̇c
eh
0
fh

R
Mechanical Magnetic
(a) (b)

Figure B-4.35 (a) Mechanical and magnetic hysteresis trends and (b) hysteresis model bond graph.

8 1
qh eh vs qh
6 fh 0.3

4 0.5 0.2

2 0.1

0 0 0
–2 –0.1
–4 –0.5 –0.2
qh
–6 qc –0.3
–1 –0.5 0 0.5 1
–8 –1
0 5 10 0 5 10
Time, sec Time, sec
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
4.9 Problems 231

Study these ideas/concepts to formulate a simple simulation that demonstrates hysteresis loop construction. Present
the results as three plots: (a) input flow, fh , (and qh = ∫ fh dt) versus time, (b) state qch versus time, and (c) effort, eh ,
versus qh . Typical results are shown below using the form of the model in Figure B-4.35(b), with a specified flow input,
fh (t). Note that the variables with the “h” subscript represent those at the input to the complete hysteresis model.

Modulation structure

Problem B-4.36 Slider-crank as modulated transformer Prototypical piston and slider-crank assembly schematics are
shown in Figures B-4.36(a) and (b).

Piston
mp, bp
Motor shaft velocity, vp Jc rc
speed, ωm
θc φ

ωc Pp (t)
Crank radius, rc τc (t)
Connecting rod length, lc xp

(a) (b)
Figure B-4.36 (a) Slider crank from Problem B-3-1. (b) Slider crank for analysis.

a) Derive the explicit relationships for representing the slider-crank mechanism as a modulated transformer. Deter-
mine the relation for the modulus, n(𝜃c ), where,
𝑣p = n(𝜃c )𝜔c .
𝜏c = n(𝜃c )Fp ,
and 𝜃c is a modulation state with 𝜃̇ c = 𝜔c
b) Develop a complete bond graph that represents a model for the slider-crank connected to a piston mass, mp , and
damping, bp . Assume an input crank torque, 𝜏c , is applied onto inertia Jc , and that there is also a known pressure
load, Pp (t) on the piston. Assume power flows from crank to piston.
c) Study causality on the bond graph of (b). Note that there will be dependent I element and there may be a preferred
causality.

Problem B-4.37 Link-cart system model and simulation Figure B-4.37 (a) shows a schematic of a cart of mass m that
moves along the x -axis. The cart is connected via a link of length L to a block that is constrained to move along the
vertical y -axis. A preliminary bond graph is given in Figure B-4.37(b). As in Example A-4-15, this bond graph suggests

Block

Se Se
Negligible mass in
block and link
F (t)
Fo (t)
Link n(x) ṗx
1 vy T̈ vx 1 vx I:m

Fk ẋk

C : 1/k

(a) (b)

Figure B-4.37 (a) Schematic of a link-cart system model and (b) bond graph with modulation of transformer
T by position of cart, x = ∫ 𝑣x dt, where 𝑣x = 𝑣m .
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
232 4 System Model Formulation and Evaluation

a modulation related to the position of the cart x or to the spring deflection, which can be taken as a modulation state. The
force Fo (t) is applied onto the block, which has a velocity 𝑣y represented by the adjoined 1-junction (which can be removed
in a simplification). Both the block and link are assumed to have negligible mass. The force F(t) is applied on the cart which
is also restrained by a spring k. Some damping may be introduced in the cart motion.
a) Study the system and bond graph provided. Apply causality to the bond graph and identify energetic states as mass
momentum velocity, 𝑣x = px ∕m, and spring deflection, xk . Then form a complete system state vector x by appending
the mass position, x, as a modulation state.
b) Determine a suitable modulation function, n(x) or n(xk ).
c) Derive state equations.
d) Write output equations for the block positions, x and y.

Problem B-4.38 Simulation study of link-cart model system Complete the link-cart system model in the previous prob-
lem and conduct an equilibrium and dynamic simulation study in the following steps.
a) Solve for an equilibrium initial condition assuming a known Fo . Assume 𝑣xo = 0 (no motion) and determine an
equilibrium value of Faxo (at time t = 0) that will set xko = 0 (unstressed spring in equilibrium state). Make sure to
determine the initial configuration states xo and yo . Equilibrium relations should be found in terms of Fo and system
parameters. Note that here gravity is not relevant.
b) Set up a simulation of this system assuming Fo remains at its equilibrium value and for F(t) = Fax sin(2𝜋fa t), where
fa is the frequency in Hz. Use the following parameters: m = 10 kg, L = 1 m, k = 1000 N/m, Fo = 10 N. Include a
damping force in the x direction on the cart with b = 10 N s/m. Assume the spring has a free length (unstressed)
Lo = 0.5 m. For the forcing function: Fax = 100 N and fa = 1 Hz. Simulate the system from the equilibrium state at
t = 0 assuming F(t) is Faxo until t = 1 second. From t = 1, let F(t) = Faxo + Fax sin(2𝜋fa t). Simulate as described until
t = 10 seconds and plot the following over time: (1) x and y, (2) 𝑣m , (3) n(x).

Problem B-4.39 Rocking on the mound Formulate a bond graph model for the system shown in Figure B-4.39 which
has a uniform bar of mass, m, rocking on a rigid, semicircle mound.

Figure B-4.39 Rocking on the mound.

Problem B-4.40 Modulation structure in a flyball governor A classical flyball governor is shown in Figure B-4.40(a),
composed of two flyballs that rotate about the z-axis with angular velocity, 𝜔 (known input). The pendula are attached
to one end of rigid L-shaped rod (of negligible mass), and the other end is attached to the spring-loaded sleeve of mass
ms . Each L-shaped rod pivots at the point indicated in the diagram, forming angle 𝜃 relative to the z-axis. The bond
graph proposed in Figure B-4.40(b) indicates three different modulated coupling elements to represent the nonlinear
mechanical dynamics in this system. Gravitational effects are ignored in this model study.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
4.9 Problems 233

Pivot
Sleeve
Mass,

(a) (b)
Figure B-4.40 (a) Spring-loaded governor schematic. (b) Bond graph with explicit modulation
structure.

a) Explain how the modulated gyrator and transformer elements are used to represent the nonlinear mechanical
dynamics in this system and confirm the indicated moduli.
b) Causality indicates the momentum of the sleeve mass, ps (or velocity 𝑣s ), and the spring deflection, zk , are state
variables. Derive the nonlinear state equations, and note that momentum h is a dependent state.
c) Assume the input velocity 𝜔 is at a reference (or equilibrium) value 𝜔o . Determine the equilibrium value of spring
deflection, zko , for this case.
d) For the case in (c), formulate the system as a linearized second order differential equation.
e) Determine stability condition(s) for the model equation in (d).
f) Derive a linear transfer function that relates zs to input 𝜔.

Problem B-4.41 Flat-follower drive It is desired to evaluate part of a machine that uses a flat-follower mechanism to
generate a reciprocating drive motion, as shown in Figure B-4.41. A circular cam of radius rc with eccentricity e is
chosen to achieve a desired motion z.

Effective mass of
follower and attached load/parts
Bearing
Connecting rod

Flat-follower

Cam

Camshaft

Figure B-4.41 Flat-follower drive mechanism.

 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
234 4 System Model Formulation and Evaluation

a) Study the motion of the cam and the relationship between 𝜃c and the position of the flat-follower, zc . Make note
of the cam radius, rc , and the eccentricity, e. Show that a modulated transformer with modulus, n(𝜃c ), can be used
to model the coupling between the camshaft which has angular velocity, 𝜔c = 𝜃̇ c , and the flat-follower which has
velocity, 𝑣c = żc . Identify the associated torque-force relationship, 𝜏 = n(𝜃c )F, for the modulated transformer as well.
b) Consider the contact point between the cam and the flat-follower face, designated by C in the diagram. Formulate
a model for the torque imposed on the cam due to the sliding friction force induced at C.
c) Integrate the models from (a) and (b) into a complete model of the cam/flat-follower system. Assume the connecting
rod has stiffness, kr . The total load mass is m and assume there is an effective damping b associated with the mass
̇ Assume negligible mass at the lower region of the flat-follower, which is contact with the cam. The
velocity, 𝑣 = z.
cam/camshaft has total rotational inertia Jc . Develop a bond graph that assumes the input is a known velocity, 𝜔c ,
and clearly identify all modulation effects with signal bonds. Apply causality and identify state variables. Since the
velocity 𝜔c is specified, the Jc will be in derivative causality for the purposes of this model.
d) Write the system state equations, which for the given assumptions should be for the momentum or velocity state of
the load mass, 𝑣, the connection rod displacement, xr , and a modulation state, 𝜃c .
e) Write an output equation for the torque, 𝜏c , associated with the input flow source for 𝜔c . It should be clear from the
bond graph that this torque is comprised of components due to the flat-follower, the sliding friction, and the inertial
torque from the cam/camshaft.

Problem B-4.42 Field-excited generator and motor load As shown in Figure B-4.42, a generator is driven by a motor
at constant speed, 𝜔g (t) (not shown), while its field circuit is excited by voltage Vf . The generator circuit meanwhile
excites the motor field, while the motor armature circuit is driven by a constant current, ia . Build a bond graph, identify
states, and derive state equations. Include an information state for the load position, 𝜃L .

constant

Gear ratio,

Load torque

load inertia

Figure B-4.42 dc generator/dc motor load.

Problem B-4.43 Hydrostatic transmission system A simplified schematic for a hydrostatic transmission system
is shown in Figure B-4.43(a), composed of a variable-displacement hydraulic pump powered by a prime mover
(e.g., an engine) with shaft angular velocity, 𝜔p (t). The pump transmits power to a fixed-displacement hydraulic
motor via fluid lines (with known inertia and compliance), and the motor drives a rotational inertia and damping
load (J, B). The variable-displacement hydraulic pump has a displacement, dp , and the ideal power transforma-
tion can be modeled by the bond graph shown in Figure B-4.43(b), where the modulus is d = dp (𝜙), and 𝜙 is a
controllable modulation variable. The fixed-displacement positive-displacement motor (PDM) can be modeled by
a similar bond graph form with a constant transformer modulus, dm . While the fluid recirculates between the
pump and motor, leakage can occur between the high-pressure side to the low-pressure side (internal resistance)
as well as to the external environment (external resistance). A replenishing system compensates for external loss
of fluid.
Assuming 𝜔p = constant, we seek a model to study the dynamic response of key system variables due to variations in
the value of 𝜙. The dynamic response can be influenced by the effective compliance in the fluid lines, Cf , as well as any
significant inertial or loss effects in the fluid lines. Internal and external leakage for each machine can be accounted for
by resistances Rli (internal) and Rle (external), which assume linear flow-pressure relations, Pli = Rli ⋅ Qli and Ple = Rle ⋅
Qle , respectively.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
4.9 Problems 235

Check valve d
τ ·· ΔP P2
Hydraulic Fluid replenishing Hydraulic
ω T 1
pump system
motor Q Q
Sump
Check valve
P1 Q
(a)

(a) (b)

Figure B-4.43 (a) Hydrostatic transmission system. (b) Bond graph of ideal positive-displacement machine,
Q = d ⋅ 𝜔, 𝜏 = d ⋅ ΔP, d = displacement (units of length3 ).

a) Develop a model of the system that incorporates a total effective line compliance on the high-pressure side equal to
Cf . Neglect any inertial or resistance effects in the fluid lines but incorporate a total effective leakage coefficient Rl .
Assume the pump displacement is given by dp = dpo 𝜙, where 𝜙 is the angle of the pump control lever in radians.
Neglect any rotational inertia or damping on the pump side, and assume J and B of the load incorporate the motor
rotational effects. Build the bond graph and assign causality to show that the incremental volume Δ– Vf = Cf Pf of the
effective fluid line compliance and the rotational momentum, hm of the motor load are the system state variables
and 𝜙 is a known (modulation) input.
b) Formulate the system state equations.
c) For a steady-state value of value of 𝜔po = 180 rad/sec, and 𝜙o = 0.1 rad, solve the equilibrium values of hmo and Δ– Vfo .
d) Simulate the system by initializing the system at the equilibrium conditions solved in (c). Let the system remain
at equilibrium until t = 0.025 seconds. At t = 0.025 seconds, increase 𝜙 from 𝜙o (initial value) to 1.5 𝜙o . Plot 𝜔m , Pf
and the value of 𝜙 from t = 0 to t = 0.15 seconds.

Problem B-4.44 Hydrostatic drive Recall the hydrostatic drive system used in the differential track system of Problem
B-3-4. A word bond graph was proposed in that problem to understand the system. In the differential drive, the “tilting”
swashplates are controlled by a “T-bar” control lever, and this mechanism controls the variable-displacement pump
system, as in Problem B-4.43.
a) For a purely performance model of the vehicle, integrate a model of the hydrostatic transmission system developed
in Problem B-4.43 with a simple model of the sprocket creating traction on a hard terrain. That is, do not consider any
steering functionality, only a drive mode is considered (forward/reverse) where the swashplates would be identically
driven. Develop a complete bond graph and identify the need for constitutive relations without specifying details
forms (functional relations are sufficient). Apply causality on the final bond graph to assess the completeness of the
form and identify state and modulation variables that would be needed. Derive state equations in complete a form
as possible. It is possible to focus only on 1/2 of the drive (e.g., the right side) for drive mode. This model should
have the same states as in Problem B-4.43 plus one additional state for the translational velocity (or momentum) of
the vehicle.
b) A more complete model would examine the steering performance, which would require also a two-dimensional
rigid-body model of the vehicle to track forward, lateral, and yaw dynamics. Such a model study would be more
appropriate as an extended case study.

Problem B-4.45 Half-car vehicle suspension model A half-car model with passive suspension elements is shown in
Figure B-4.45, and a bond graph is provided further below. It is assumed that the pitch angle, 𝜃, remains relatively
small.
Complete the following:
a) Verify the bond graph and complete the state equations initiated below, then write them in linear state-space form
b) Set up an equilibrium problem and solve for the initial deflections of the front and rear suspension springs and the
front and rear tire spring elements assuming gravity acts in the downward direction. The suspension elements are
linear, and parameters are given further below. Assume a static equilibrium with all mass velocities zero, and that
60% of the chassis mass is at the front axle in order to determine L1 and L2 .
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
236 4 System Model Formulation and Evaluation

g
vcm
ωcm
vr
L2 L1 vf vx
mv, Jcm
ksr bsr ksf bsf

mtr mtf
ktr, btr ktf, btf
vgr vgf

Figure B-4.45 Half-car suspension model.

I : mv I : Jcm

ṗcm vcm ḣcm ωcm

L2
mv g ·· L2 Fr
Se 1 T 1
ωcm
L1 Ff

Fr T : L1
Ff

0 0

C Fr vr Ff vf C

ẋsr Fr Ff ẋsf
1 0 0 1

R Fr vtr Ff vtf R
ṗtr mtr g mtf g ṗtf
I 1 Se Se 1 I
vtr vtf

C Ftr Ftf C

ẋtr Ftr Ftf ẋtf

1 0 0 1

R Ftr żr Ftf żf R

Sf Sf

c) Simulation study can be conducted by defining each of the ground inputs at the front and rear contact points as
known functions of time. Alternatively, ground profile, zg (x), can be specified so that each of the ground inputs can
be determined by,

dzg (x) dzg (x) dx dzg

𝑣g (t) = = ⋅ = ⋅ 𝑣x .
dt dx dt dx

where the vehicle forward velocity is 𝑣x . Consider the road ground profile given below. For 𝑣x = 40 km/hr, simulate
the vehicle response transient response simulations.
xg = [0, 5, 6, 10, 11, 15] m
zg = [0, 0, 0.1, 0.1, 0, 0] m.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
4.9 Problems 237

System equations:
ṗ cm = m𝑣 𝑣̇ cm = Fr + Ff − m𝑣 g 𝑣cm = pcm ∕m𝑣
ḣ cm = Jcm 𝜔̇ cm = +L2 Fr − L1 Ff 𝜔cm = hcm ∕Jcm
ẋ sr = 𝑣tr − 𝑣r 𝑣r = 𝑣cm + L2 𝜔cm
ẋ sf = 𝑣tf − 𝑣f 𝑣f = −𝑣cm + L1 𝜔cm
ṗ tr = Ftr − Fr − mtr g 𝑣tr = ptr ∕mtr
ṗ tf = Ftf − Ff − mtf g 𝑣tf = ptf ∕mtf
ẋ tr = 𝑣gr (t) − 𝑣tr Fr = Fsr + Fsbr = ksr xsr + bsr (𝑣tr − 𝑣r )
ẋ tf = 𝑣gf (t) − 𝑣tf Ff = Fsf + Fsbf = ksf xsf + bsf (𝑣tf − 𝑣f )
Ftr = Ftsr + Ftbr = ktr xtr + btr (𝑣gr (t) − 𝑣tr )
Ftf = Ftsf + Ftbf = ktf xtf + btf (𝑣gf (t) − 𝑣tf )
System parameters:

Vehicle body/chassis Front suspension and tires Rear suspension and tires

L = 2.7 m ksf = 30 N/mm ksr = 20 N/mm

h = 0.55 m bsf = 0.75 N-s/mm bsr = 0.75 N-s/mm
m𝑣 = 1700 kg mtf = 100 kg mtr = 80 kg
Ixx = 400 kg-m2 ktf = 200 N/mm ktr = 200 N/mm
Iyy = 2704 kg-m2
Izz = 3136 kg-m2
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
239

Linear System Modeling and Analysis

Designing or modifying a system to meet specifications requires understanding the physical characteristics, modeling, use
of functional system properties, and analysis methods for design and control. Linear system analysis is often made possible
only after linear approximations (or linearization) of a system model, as was discussed in Chapter 4. Linear methods of
analysis have always played a useful role in engineering, often making it possible to use analytical solutions of the dynamic
response. These methods were essential in the last century prior to the advent of digital computational tools, and they
remain very useful despite the wide availability of contemporary computational software. Very often, insight into relevant
behavior of a system becomes more evident only after such analysis, and application of linear analysis in system design and
control remains popular and essential.
A necessary precursor to using linear analysis methods is a linear model assumption or application of linearization
techniques as in Chapter 4. This chapter assumes such a model exists, and the simplest types of dynamic systems (first-
and second-order systems) are presented first along with methods for analyzing these systems. A good understanding of
these first- and second-order system results can be very helpful in a wide range of practical engineering problems. Often
it is the low-order models (n = 1, 2, 3) that can be effectively used to answer questions about a system. From such prelim-
inary models, we develop ways to understand and analyze higher-order systems, adopting use of transfer functions and
free and forced response of state equations. Forced oscillatory response (frequency response) will be taken up in Chapter 6.
Throughout the discussion in this chapter, we demonstrate that knowledge of the analytical results can reinforce practical
use of available computational tools.

5.1 Analysis of First-Order System Models

The simplest type of system that exhibits dynamic characteristics is a first-order system. These systems typically have one
independent energy storing element and therefore one energy state. A single modulation state would also suffice to form a
first-order system. These types of systems are modeled by a first-order differential equation, which we recognize can arise
in many different physical forms. For the purposes of reviewing first-order analysis methods, a simple hydraulic tank being
filled by a known source of flow, Qs (t), is used in this section. The tank leaks fluid via a linear resistive valve as shown in
Figure 5.1(a), and the bond graph causality in (b) identifies volume –V as a single energetic state.
The constitutive relations for the valve and fluid tank are given as,
Ql = P∕R (resistive leakage) (5.1)
P = –V ∕C (capacitive tank) (5.2)
where –V is the fluid volume, C = A∕pg, and R is the leakage resistance. From the structure of the model in Figure 5.1(b)
and the constitutive relations for the elements, a single state equation can be derived as,
–V̇ = −–V ∕RC + Qs ,
or,
–V̇ = −–V ∕𝜏t + Qs . (5.3)
where 𝜏t = RC is the system time constant (units of seconds).

Modeling of Physical Systems: Simulation and Control, First Edition. Raul G. Longoria and Joseph J. Beaman.
© 2025 John Wiley & Sons Ltd. Published 2025 by John Wiley & Sons Ltd.
Companion Website: www.wiley.com/go/Modeling_of_Physical_Systems/1e
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
240 5 Linear System Modeling and Analysis

Constant tank
area, A
g C:C

P V
–˙
V h
Qs (t) Ql Pl
Sf 0 R:R
Qs (t) Ql

Linear resistance, R
(a) (b)

Figure 5.1 Example first-order system: (a) tank with leakage and (b) first-order model in bond graph form.

From the solution of this equation, we might want to answer the following engineering questions:
1) If the input flow is turned on at t = 0 and remains constant, how long does it take for the tank to reach a steady-state
(SS) maximum height?
2) What is the maximum height attained?
3) What is the flowrate lost to leakage?
4) What is the net flowrate into the tank?
Once we have a solution for –V from 5.3, we can obtain a solution for almost any other output variable of interest by direct
substitution. For example, the tank height h is related to the tank volume V
– by h = –V ∕A, the leakage flow is,
Ql = P∕R = –V ∕𝜏t ,
and the tank inflow QT can be found as,
QT = Qs − Ql ,
We also recognize that –V̇ = QT . Now, rather than first solving for V – , relations for any output system variables can be
substituted directly into state equation 5.3 to yield differential equations in terms of those variables. For example, for the
height, h,
Aḣ = –V̇ = Qs − –V ∕RC = −(A∕RC)h + Qs ⇒ ḣ = −h∕𝜏t + Qs ∕A.
Or, if an equation is sought for the leakage flow, Ql = –
V ∕𝜏t , then,
𝜏t Q̇ l = Qs − Ql . (5.4)
For the tank inflow, QT = –V̇ = −Ql + Qs , so
Q̇ T = −Q̇ l + Q̇ s ,
where Q̇ l = –V̇ ∕𝜏t = QT ∕𝜏t , thus,
Q̇ T = −QT ∕𝜏t + Q̇ s .
All of these differential equations for the different variables of interest describe the same physical system. If we write
these equations as a group, they can be expressed as:
𝜏t –V̇ + –V = 𝜏t Qs (5.5a)

𝜏t ḣ + h = 𝜏t Qs ∕A (5.5b)

𝜏t Q̇ l + Ql = Qs (5.5c)

𝜏t Q̇ T + QT = 𝜏t Q̇ s . (5.5d)
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
5.1 Analysis of First-Order System Models 241

By expressing the equations in this form, we can immediately see that it is possible to obtain an almost identical equation
for each system variable, the only difference being the right hand side of the equation. We can then propose a standard
first-order form,
𝜏t ẏ + y = u(t), (5.6)
where y is the variable of interest, u(t) is the forcing function of time, with 𝜏t the time constant. The time constant has
dimensions of time and for the present example hydraulic system is,
[ ][ 3 ]
[ ] FL2 L
𝜏t = [R] [C] = 3 −1 = [t] = unit of time.
Lt FL2
Although we used a particular example to propose the form 5.6, any linear first-order differential equation model with
constant parameters can be put in this general form. Note also that the u(t) will change in each case here. For example, for
equation 5.5b,
u(t) = 𝜏t Qs ∕A,
which has the same units as h, as required in any of these forms.
Linear differential equations have an extremely important property, namely, superposition. For example, consider a lin-
ear first-order system with two different inputs ua (t) and ub (t) applied at different times with the same initial conditions.
Indicating the solution to input ua as ya and to ub as yb , we obtain the respective equations
𝜏t ẏ a + ya = ua . (5.7)

𝜏t ẏ b + yb = ub . (5.8)
Now assuming the solutions ya (t) and yb (t) are known, we can find a solution to the system equation when ua and ub are
both applied with ua scaled by a factor a and ub scaled by a factor b. Mathematically, we desire a solution to the differential
equation,
𝜏t ẏ c + yc = aua + bub ,
where yc (t) is the solution. Multiplying 5.7 by a and 5.8 by b and then adding the results yields,
𝜏t (aẏ a + bẏ b ) + (aya + byb ) = aua + bub ,
which implies that yc can be found by simply adding and scaling ya and yb in the same manner as for inputs ua and ub , with
the result,
yc = aya + byb .
This is an example of the superposition property, which holds for all linear systems with constant parameters, as we next
demonstrate.
A formal mathematical statement of the superposition property can be stated as,
(aua + bub ) = a(ua ) + b(ub ),
where  now stands for any linear operator and not just for a first-order linear differential equation.
As a counter example of a first-order system that is not linear for which superposition does not hold, consider the system,
𝜏t ẏ + y3 = u.
For two different inputs ua and ub we may find the solutions to the equations
𝜏t ẏ a + y3a = ua , (5.9)

𝜏t ẏ b + y3b = ub , (5.10)
and we now seek the solution when both inputs are applied,
𝜏t ẏ c + y3c = u1 + u2 . (5.11)
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
242 5 Linear System Modeling and Analysis

However, if we add 5.9 and 5.10, we obtain as before

𝜏t (ẏ a + ẏ b ) + y3a + y3b = ua + ub ,
and we note that in this case, yc ≠ ya + yb , since it will not produce 5.11 upon substitution.

5.1.1 First-Order Linear Time Invariant Solution

The general solution to the linear first-order system uses superposition directly. The first step is to solve for the general form
of the homogeneous equation, yh ,
𝜏t ẏ h + yh = 0. (5.12)
The next step is to find any particular solution yp for a given input, u(t), to the equation,
𝜏t ẏ p + yp = u(t). (5.13)
The total solution y is then found by simply adding,
y = yh + yp .
Assuming a constant value for 𝜏t , the general solution to the homogeneous or unforced equation can always be
expressed as,
yh = Aest ,
where A and s are constants to be determined. Substituting the exponential form into the homogeneous equation 5.12
yields,
𝜏t Asest + Aest = 0 ⇒ Aest (𝜏t s + 1) = 0,
which implies either A = 0, which is the trivial solution, or else,
𝜏t s + 1 = 0.
Thus,
s = −1∕𝜏t . (5.14)
The homogeneous or unforced solution can then be expressed as,
yh = Ae−t∕𝜏t , (5.15)
where A is an arbitrary constant for the moment. The particular or forced equation can only be solved after a particular
input is specified. For a step input, we have
{
0, t < 0,
u(t) =
u0 , t ≥ 0,
u(t) = u0 Us (t), (5.16)
where Us is the unit step function which is 0 for t < 0 and 1 for t ≥ 0. Since a step is constant after t = 0, the constant
solution,
yp = u0 (5.17)
satisfies 5.13 for t ≥ 0.
The total solution is then given by,
y = Ae−t∕𝜏t + u0 . (5.18)
In order to determine the constant A, an initial condition y(0) = y0 is needed. Using this initial condition in 5.18 results
in y0 A + u0 or,
A = y0 − u 0 ,
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
5.1 Analysis of First-Order System Models 243

so that the total response to a step input can be expressed,

y(t) = y0 e−t∕𝜏t + u0 (1 − e−t∕𝜏t ). (5.19)
This result provides insight into how y0 and u0 each influence the response.1

5.1.1.1 First-Order Response Solutions and Trends

The exponential trend can be emphasized by rearranging equation 5.19 into a nondimensional form,
y(t) − u0
= e−t∕𝜏t ,
y0 − u0
which is plotted in Figure 5.2. The exponential trend in Figure 5.2 dictates that a steady-state condition is only asymptot-
ically reached with the decay rate determined entirely by the system time constant, 𝜏t . We can thus readily identify how
much the response has decayed relative to the value of 𝜏t . Notably, after t = 1𝜏t the response is 63% decayed, 2𝜏t = 86%
decayed, 3𝜏t = 95% decayed, 4𝜏t = 98% decayed, and t = 5𝜏t = 99% decayed. These results lead to the definition of standard
values for 5% settling time as 3 time constants, and a 2% settling time as 4 time constants, commonly denoted t95 and t98 ,
respectively.
For our tank example, where 𝜏t = RC, we can then state that the tank reaches 95% of its steady-state height in t95 = 3𝜏t
and reaches 98% of this height in t98 = 4𝜏t . Although theoretically, the tank never reaches a steady-state height, quite often
either the 5% or 2% settling times are sufficient for engineering analysis. Other factors such as nonlinear friction provide
for a finite time to steady state in the real system. Conversely, one might say that observing a definitive “stop” in motions
or flows, for example, indicates the presence of nonlinear effects.
The maximum height for the system occurs when t → ∞ and, if we note the differential equation for the height 5.5b and
the general step response for a first-order system 5.19, we can find a steady-state height by,
hss = lim h(t) = u0 𝜏t Q0 ∕A, (5.20)
t→∞

where Q0 is the constant flow applied at t = 0. This equation results from the fact that the maximum height occurs at steady
state when all the input flow is leaking and none remains to fill the tank volume.
An expression for leakage flow can be found by using the general step response equation 5.19 with u0 = Q0 and the initial
condition Ql (0) = 0 to obtain,
Ql (t) = Q0 (1 − e−t∕𝜏t ), 𝜏t = RC. (5.21)

1
y0 = initial condition
0.9 u0 = step magnitude
0.8
y(t) – u0
y0 – u0 = e
0.7 –t/τt

0.6
0.5

0.4 0.32
0.3
0.2
0.14
0.1 0.05
0
0 0.5 1 1.5 2 2.5 3 3.5 4
t/τt

Figure 5.2 First-order response to both input step of magnitude u0 and initial condition, y0 , with some key decay values indicated for
values of t∕𝜏t = 1, 2, and 3, 𝜏t = time constant.

1 Solved Problem A-5-3 shows how this result can be used to find the response to a unit ramp input, u(t) = t.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
244 5 Linear System Modeling and Analysis

1
QT
0.8 Qo

0.6 V
τQo

0.4

0.2

0
0 0.5 1 1.5 2 2.5 3 3.5 4
t/τt

Figure 5.3 Tank volume –V a continuous function versus QT discontinuous net inflow function.

For the tank volume, the differential equation can be expressed,

𝜏t –V̇ + –V = 𝜏t Qs . (5.22)
Then, for a step change in flow, Qs (t) = Q0 Us (t), with Us (t) the Heaviside step function,2 shown in Figure 5.4(a), and with
–V (0) = 0, the solution to this equation is,
–V = 𝜏t Q0 (1 − e−t∕𝜏t ). (5.23)
The tank net inflow rate, QT , can be obtained from 5.23 by noting that QT is the time derivative of the tank volume, or
QT = –V̇ = Q0 e−t∕𝜏t . (5.24)

5.1.1.2 Practical Consideration of Model Input Types

Energy and power considerations can be used to assess the practical nature of models used for inputs. For example, tank
volume, –V , directly quantifies the potential energy stored in the example system under study. Practically, this stored energy
function must be a continuous function of time. A discontinuity such as a step change in volume V – would imply – V̇ is
infinite. This would require infinite power:  = ̇ = P ⋅ – V̇ → ∞. In contrast, the tank net inflow QT , which is not directly
related to energy content, can and does change discontinuously. These trends are illustrated in Figure 5.3. Recognizing that
certain trends might be possible for particular system variables and not in others is important when building a model, and
may guide the selection of variables to be used.
For example, consider the differential equation that was derived in terms of the net inflow, QT ,
𝜏t Q̇ T + QT = 𝜏t Q̇ s . (5.25)
If the input flow Qs is a step input then the time derivative Q̇ s is not defined in the usual mathematical sense. The time
derivative of a step function is called an impulse. A unit impulse 𝛿(t), applied at t = t0 , can formally be defined as,
∞

𝛿(t0 )dt = 1 (5.26)

∫
−∞
𝛿(t0 ) = 1, t ≠ t0 .
Conceptually, an impulse has an infinite magnitude over an infinitely short-time duration, as shown in Figure 5.4(b).
The time integral of the function yields a step.
Although the impulse function can be a useful mathematical concept for the analysis of physical systems, the unbounded
nature of this input is not physically possible and can lead to analysis and modeling errors. Impulse inputs arise in models
for primarily three reasons: (i) the choice of variable, (ii) the choice of model, and (iii) input approximation. In the above
example, the tank net inflow QT is equal to the derivative of the energy state V – . Describing the system performance in
terms of this variable requires an impulse input when there is a step change in source flow Qs . If the system analyst does

2 The Heaviside step function, Us (t) = 0 for t ≤ 0 and 1 for t > 0.

 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
5.1 Analysis of First-Order System Models 245

Δt → 0
1
1 Δt Δt

t t
(a) (b)

Figure 5.4 (a) Unit step, U(to ), and (b) impulse input, 𝛿(to ).

Ideal velocity source Sf

v(t)
1 v(t) 2 p1 p2
m1:I 0 I:m2
v1 v2

(a) (b)

Figure 5.5 Model with derivative input: (a) two masses with relative velocity source, V (t), and (b) bond graph model of system in (a).

not wish to deal with an impulse, another variable such as the tank volume V – may be preferred. The variable QT can then
be determined in terms of V –.
For a large number of models, a simple change of variable, like the one from tank net inflow to tank volume, will suffice
to eliminate the need for an impulse, but there are models for which this is not the case. This occurs whenever an energy
storing element depends directly on the input. As an example, consider two translational masses connected by a device that
controls the relative velocity between them as shown in Figure 5.5(a).
A state equation for this system follows from the bond graph in Figure 5.5(b), identifying there can only be one indepen-
dent I element, the other is dependent. Causally, ṗ 1 = ṗ 2 = m2 𝑣̇ 2 = m2 dtd (𝑣(t) − 𝑣1 ) = −m2 𝑣̇ 1 + m2 𝑣(t),
̇ or,
m2
𝑣̇ 1 = 𝑣(t).
̇ (5.27)
m1 + m2
A step change in velocity 𝑣(t) would imply an impulse input to the system, which is not physically realistic since in this
case it also implies an infinite power transfer. This is shown from the power at the input to mass 1 which is, 1 = ṗ 1 ⋅ 𝑣1 , and
ṗ 1 = m2 ⋅ ∞. Since models with infinite power transfer are not realistic, either step velocity inputs should not be allowed
or the model should be changed.
Although not entirely realistic, an impulse can sometimes be used as a good approximation to an actual physical input.
An example is the extremely high forces of short duration that can occur during mechanical impact. Testing and analysis
with impact forces modeled as impulse inputs can be quite effective in experimental identification of physical systems [90].
Another common input used in model analysis is the unit ramp function, r(t) = t, which like the step and impulse is
useful in representing many practical physical signals. The unit ramp is the time integral of the unit step function, and if
another integral is taken the unit parabola, t2 , is obtained. The ramp and parabola functions are commonly used as test
signals and to model some ideal processes. Combinations of the step, ramp, and parabola are usually sufficient to create
models of more complex signals. By understanding the response of basic first- and second-order systems to these test inputs,
it becomes possible to compose analytical approximations of practical systems to different types of inputs of interest. As we
will see in this chapter, a known response of a linear system to one input can be used to determine the response to an input
that may be the derivative or integral of the other.

5.1.2 Summary
The simplest type of dynamic system, the first-order linear time invariant system, can always be cast in the form
𝜏t ẏ + y = u(t).
The solution to this equation for u(t) a step input is,
y(t) = yo e−t∕𝜏t + yo (1 − e−t∕𝜏t ),
where yo is an initial condition y(0) = yo and uo is the magnitude of the step input.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
246 5 Linear System Modeling and Analysis

An impulse is formally defined as the time derivative of a step. An impulse response to a first-order linear system can be
found by taking the derivative of the step response. Throughout this section, the basic hydraulic system of Figure 5.1 was
used to describe how these results can be used in first-order analysis. A similar exercise can be used for any system that can
be reduced to standard first-order form.

5.2 Analysis of Second-Order System Models

After the first-order system, a second-order system, which is characterized by two independent states, is the next in order
of complexity. Many engineering devices and systems are designed to exhibit linear second-order characteristics, so under-
standing this class of system has practical value. A standard linear second-order form can be expressed as,

ÿ + 2𝜁𝜔n ẏ + 𝜔2n y = 𝜔2n u(t), (5.28)

where y is the state or variable of interest, u(t) is the forcing function of time, 𝜔n is the undamped natural frequency, and
𝜁 is the damping ratio. While the natural frequency has units of 1/time, it is typically expressed as radians/sec (or rad/sec)
where the radian is a dimensionless measure of angle (𝜋 radians = 180 ∘ ). The damping ratio is dimensionless.
It can be very useful to convert a linear second-order state-space model directly into this standard form, since the
undamped natural frequency and damping ratio can be quickly identified in terms of system parameters. As an example,
consider the force measuring system studied in Chapter 4 and shown again in Figure 5.6.
Two coupled first-order differential equations can be derived directly from the system bond graph in Figure 5.6(b),
ṗ = −(b∕m)p − kxk + F(t) (5.29a)
ẋ k = (1∕m)p. (5.29b)

Since p = m𝑣m = mẋ m , in this case, equation 5.29a can be used to find an equation in terms of xm directly,

ṗ = m̈xm = −(b∕m)(mẋ m ) − kxm + F(t),

where we also see that xk = xm , thus resulting in,

ẍ m + (b∕m)ẋ m + (k∕m)xm = F(t)∕m. (5.30)

Note that xk is the spring deflection and ẋ k = 𝑣m = ẋ m . The undamped natural frequency and damping ratio can now be
determined as,
√
√ [ ] F L
𝜔n = k∕m → 𝜔n = = 1∕t, (5.31)
L Ft2
(where we’ve used “[ ]” to indicate “units of”) and,
√
𝜁 = b2 ∕4km. (5.32)

F(t)

I:m

m p vm
vm = xm vm xm eo
F(t) 1 c
Se 1 s
Fb xk
vb F k
b k
eo R:b C:1/k
Ideal displacement
sensor
(a) (b)

Figure 5.6 (a) Force measurement system schematic and (b) bond graph model.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
5.2 Analysis of Second-Order System Models 247

Conversion of the linear state space to second-order form requires selecting the variable of interest, y. Recall this
discussion for the standard first-order form (Section 5.1). In some cases, this can be done with a direct variable substitu-
tion. For example, if it is desired for the force measuring system above to have an equation in terms of the output voltage,
eo = c ⋅ xm , this leads directly to,
ë o + (b∕m)ė o + (k∕m)eo = cF(t)∕m. (5.33)
Alternatively, a second-order differential equation in terms of momentum p can be derived by differentiating
equation 5.29a once to find,
̇
p̈ + (b∕m)ṗ + (k∕m)p = F(t). (5.34)
Notice that the input for this model is the derivative of the input force, F(t). Note that having a derivative of an input as in
this case may not be desirable in some cases, as discussed previously for first-order systems. In addition, as for first-order
systems, any of the differential equations with alternative variables model the same physical system. As such, they all have
the same undamped natural frequency and damping ratio.
In most cases, converting a second-order linear state-space model into second-order ordinary differential equation (ODE)
form can be completed by direct substitution. In some systems, it may not be clear how to complete this conversion
algebraically. In those cases, it can be helpful to first convert the state-space form into a transfer function form, y∕u = G(s),
using the methods in Section 4.4. The transfer function form can often more easily be converted to the second-order ODE
form by the substitution s → d∕dt.

5.2.1 Second-Order Linear Time Invariant Solution

In a similar fashion to the first-order solution, the homogeneous form for the second-order system in equation 5.28,
ÿ h + 2𝜁𝜔n ẏ h + 𝜔2n yh = 0, (5.35)
is solved by first assuming an exponential solution,
yh = Aest . (5.36)
Substituting this exponential form into the homogeneous equation gives,
(s2 + 2𝜁𝜔n s + 𝜔2n )Aest = 0, (5.37)
which implies that either A = 0 or,
s2 + 2𝜁𝜔n s + 𝜔2n = 0, (5.38)
which is identical that found by applying Cramer’s rule to a state-space form of the model equations. In that case, the s
variable is considered an exponent of an exponential function rather than a derivative operator. This correspondence is due
to an identity for the exponential function,
d(est )∕dt = sest . (5.39)
These two interpretations for s (s as derivative operator and s as exponent of an assumed exponential form) are used
extensively in linear system analysis.
There are two solutions for s from the quadratic equation 5.38. These two solutions can be expressed as,
√
s1,2 = −𝜁𝜔n ± 𝜔n 𝜁 2 − 1. (5.40)
Since this is a linear system, the sum of the two possible solutions is also a solution. The general homogeneous solution
or unforced solution is then,
xh = A1 es1 t + A2 es2 t , (5.41)
where,
√
s1 = −𝜁𝜔n + 𝜔n 𝜁 2 − 1, (5.42a)
√
s2 = −𝜁𝜔n − 𝜔n 𝜁 2 − 1, (5.42b)
and A1 , A2 are arbitrary constants.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
248 5 Linear System Modeling and Analysis

The total solution for the second-order system is then formed by,
y = yh + yp , (5.43)
where yp is any solution to the particular equation,

ÿ p + 2𝜁𝜔n ẏ p + 𝜔2n yp = 𝜔2n u(t). (5.44)

The particular equation can only be solved after a particular input is specified. For example, for a step input u(t) = u0 Us (t),
where Us (t) is the Heaviside step function,3 the constant solution,
yp = u0 (5.45)
satisfies 5.44 for t ≥ 0, and the total solution is then given as
y = yh + yp = A1 es1 t + A2 es2 t + u0 . (5.46)
̇
In order to determine the constants A1 and A2 , the initial conditions y(0) = y0 and y(0) = ẏ 0 are needed. Using these initial
conditions in 5.46 results in,
y0 = A1 + A2 + u0 , (5.47)

ẏ 0 = A1 s1 + A2 s2 , (5.48)
which can be solved for A1 and A2 as,
s2 ( y0 − u0 ) − ẏ 0 ẏ 0 − s1 ( y0 − u0 )
A1 = and A2 = . (5.49)
s 2 − s1 s 2 − s1
̇ and a step of magnitude u0 can now be seen to be composed of two parts,
The total response of y to initial conditions y0 , y,
y = ys + yI , (5.50)
where,
s2 ( y0 − u0 ) s t s1 ( y0 − u0 ) s t
ys = e1 − e 2 + u0 ⇒ Response due to step and initial y0 . (5.51)
s 2 − s1 s 2 − s1
ẏ 0 s t ẏ 0 s t
yI = − e1 + e2 ⇒ Response due to initial ẏ 0 . (5.52)
s 2 − s1 s 2 − s1
These results can be used to derive specific response functions for different combinations of the initial conditions with
a step input for second-order systems: (a) ys : y0 with step input, ẏ 0 = 0, (b) yI : initial ẏ 0 , with y0 = u0 = 0, (c) y𝛿 : impulse
response, with y0 = 0. For each of these cases, the value of 𝜁 will dictate a different form of the response function.

5.2.2 Response to Initial Condition and Step Input

With ẏ 0 = 0, the total response of y is due just to y0 and the step input which is result ys in equation 5.51. The solution ys
can be further classified in terms of the damping ratio 𝜁. For the time being, we restrict ourselves to positive, real values of
𝜁 and 𝜔n , as these correspond to stable second-order systems.

5.2.2.1 Overdamped, 𝜻 > 1

If the damping ratio is greater than one, the two solutions for s are both real and negative and the system is classified
as overdamped. It is common in these cases to think about the response comprised of two first-order response functions,
determined by slow and fast time constants defined, respectively, as follows:
1
Slow time constant: 𝜏s = −1∕s1 = √ , (5.53)
(𝜁 − 𝜁 2 − 1)𝜔n
1
Fast time constant: 𝜏f = −1∕s2 = √ , (5.54)
(𝜁 + 𝜁 2 − 1)𝜔n

3 The Heaviside step function, Us (t) = 0 for t ≤ 0 and 1 for t > 0.

 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
5.2 Analysis of Second-Order System Models 249

1.5 Slow response

0.5
Total response
0

–0.5
Fast response
–1
0 0.2 0.4 0.6 0.8 1
Time, t

Figure 5.7 Qualitative picture of overdamped system as the sum of fast and slow exponential responses.

where 𝜏s > 𝜏f . A step response, ys , can then be expressed in scaled form as a sum of fast and slow exponential components,
ys − u 0 𝜏s 𝜏f
= e−t∕𝜏s − e−t∕𝜏f = Cs (t) + Cf (t). (5.55)
y0 − u0 𝜏s − 𝜏f 𝜏s − 𝜏f
Alternatively the slow and fast components can be expressed in a form similar to a first-order system response,
[ ]
ys = u0 + ( y0 − u0 ) Cs (t) + Cf (t) . (5.56)

The fast time constant exponential will decay at a faster rate than the slower time constant exponential. This implies that
for t ≫ 3𝜏f the solution can be approximated by,

ys ≈ u0 + ( y0 − u0 )Cs (t), t ≫ 3𝜏f . (5.57)

For this reason, the slow time constant, 𝜏s , is sometimes referred to as the dominant time constant, as it will influence
the response over a longer period. A graphical description of an overdamped response with slow and fast components is
shown in Figure 5.7. For our force measuring system, an overdamped response would occur for relatively large amounts of
damping.

5.2.2.2 Critically Damped, 𝜻 = 1

If the damping ratio is identically equal to one, then the two solutions for s are identical and the system is termed critically
damped. In this case, there is no dominant time constant. The critically damped response for y can be found by defining,

𝜏 t = 𝜏f , (5.58)

𝜏s = 𝜏t + 𝜖, (5.59)

and letting 𝜖 go to zero in the solution 5.55 or,

[ ]
ys − u 0 𝜏 + 𝜖 −t∕(𝜏 +𝜖) 𝜏t −t∕𝜏
= lim t e t − e t
y0 − u0 𝜖→0 𝜖 𝜖
[ ]
(𝜏t + 𝜖)e−t∕(𝜏t +𝜖) − 𝜏t e−t∕𝜏t
= lim
𝜖→0 𝜖
d(𝜏t e−t∕𝜏t )
=
d𝜏t
ys − u0
= e−t∕𝜏t + (t∕𝜏t )e−t∕𝜏t . (5.60)
y0 − u0
Critically damped systems exhibit the fastest decay to steady state for a step input with zero initial conditions, and have
no overshoot of the steady-state value. For these reasons, it can be desirable to design some systems, such as the force
measuring system, with critical damping.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
250 5 Linear System Modeling and Analysis

5.2.2.3 Underdamped, 0 < 𝜻 < 1

If the damping ratio is between 0 and 1, then the two solutions have both real and imaginary components and the system
is called underdamped. The solutions can be expressed as,
( √ )
s1 = −𝜁 + j 1 − 𝜁 2 𝜔n = −𝜎 + j𝜔d , (5.61)
( √ )
s2 = −𝜁 − j 1 − 𝜁 2 𝜔n = −𝜎 − j𝜔d , (5.62)
√ √
where 𝜎 = 𝜁𝜔n , 𝜔d = 𝜔n 1 − 𝜁 2 , and j = −1 is the imaginary number. The quantity 𝜔d is called the damped natural
frequency.
It will be convenient to express the complex numbers s1 and s2 in polar form as a magnitude and an angle in the complex
plane as shown in Figure 5.8. Each complex root is indicated by a “×” symbol.
Mathematically, the relation between the Cartesian real-imaginary coordinates and the polar magnitude-angle coordinate
can be expressed as,
√
s1 , s2 = −𝜎 ± j𝜔d = (cos 𝜙 + j sin 𝜙)𝜑 = tan−1 (𝜔d ∕(−𝜎)), (5.63)
but, using the identity,
𝜎 2 + 𝜔2d = (𝜁𝜔n )2 + (1 − 𝜁 2 ) 𝜔2n = 𝜔2n (5.64)
and Euler’s formula,
e jq = cos q + j sin q, (5.65)
the two roots can also be expressed as,
s1 = 𝜔n e j𝜙 . (5.66)
s2 = 𝜔n e −j𝜙
. (5.67)
Note that s1 and s2 are complex conjugates and therefore have equal real and opposite imaginary parts. In polar notation,
this relates to equal magnitude and opposite angle. Using both the Cartesian and polar representations for the complex
conjugate roots s1 and s2 in the general solution for ys given in equation 5.51 results in
ys − u 0 𝜔 e−j𝜙 (−𝜎+j𝜔 )t 𝜔n e−j𝜙 (−𝜎−j𝜔 )t
= n e d − e d
x0 − u0 −2j𝜔d −2j𝜔d
[ (−𝜎+j𝜔 )t ]
𝜔 e d − e(−𝜎−j𝜔d )t
= n e−𝜎t
𝜔d 2j
𝜔n −𝜎t
= − e sin(𝜔d t − 𝜙), (5.68)
𝜔d
where we have used the identity,
(e j𝜃 − e−j𝜃 )∕2j = sin 𝜃.
√
Since 𝜎 = 𝜁𝜔n , and the damped natural frequency 𝜔d = 𝜔n 1 − 𝜁 2 , and 𝜙 = tan−1 (𝜔d ∕ − 𝜎), 5.68 can also be
expressed as,
( √ (√ ))
ys − u0 e−𝜁 𝜔n t
= √ sin 𝜔n 1 − 𝜁 2 t + tan−1 1 − 𝜁 2 ∕𝜁 = D1 (t), (5.69)
y0 − u0 1 − 𝜁2

s-plane Im(s) Figure 5.8 Complex roots represented in polar form on s-plane.

s1 +jωd
ωn
+φ

–σ Re(s)
–φ
ωn
s2 –jωd
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
5.2 Analysis of Second-Order System Models 251

or alternatively,
( )
( √ ) 𝜁 ( √ )
ys − u0 e−𝜁 𝜔n t
= √ cos 𝜔n 1 − 𝜁 2 t + √ sin 𝜔n 1 − 𝜁 2 t = D2 (t). (5.70)
y0 − u0 1 − 𝜁2 1 − 𝜁2
The response of an underdamped second-order system can be viewed as sinusoid with an exponential decay envelope as
seen in Figure 5.9.
The exponential decay envelope has a time constant of 𝜏e = 1∕𝜎 = 1∕𝜁𝜔n . This time constant can be used to estimate the
settling time for system as t95 ∼ 3𝜏e , t98 ∼ 4𝜏e , and t99 ∼ 5𝜏e for the 95%, 98%, and 99% settling times, respectively. These are
only estimates due to the sinusoidal nature of the response. A more accurate value for these settling times can be obtained
directly from Figure 5.10(a) and (b), which are plots of the initial condition, y0 , and step response for a second-order system,
respectively, for selected values of damping ratio ranging over undamped to overdamped√cases.
Besides critical damping, another value for damping that is often used for design is 𝜁 = 2∕2 = 0.707. This value provides
a good trade-off between low overshoot and fast settling time, where overshoot refers to the amount that the response rises
above its steady state value (above 0 in Figure 5.10).
Returning to our force measuring system we can now do a preliminary design of the system. Assume that a measurement
response time of .01 second is desired with small overshoot, and using 98% settling time as a measure of response time we
then have
t98 ≈ 4𝜏e = 4𝜁𝜔n = 0.01 second,

Exponential envelope
0.5

0
Total
response
–0.5

–1
0 0.5 1 1.5 2 2.5 3 3.5 4
Time, t

Figure 5.9 Qualitative picture of underdamped system as a sinusoid with exponential decay.

1 2
ζ = 1.4 ζ = 0.0
u0 = 0 x0 = 0
ζ = 0.2
0.5 ζ = 1.0 1.5 ζ = 0.4
ζ = 0.707 ζ = 0.707

ys 0 ys 1

ζ = 0.4
–0.5 0.5 ζ = 1.0
ζ = 0.2

ζ = 0.0 ζ = 1.4
–1 0
0 1 2 3 4 0 1 2 3 4
Dimensionless time, ωnt/π Dimensionless time, ωnt/π
(a) (b)

Figure 5.10 (a) Initial condition, y0 , response for second-order system ÿ + 2𝜁𝜔n ẏ + 𝜔2n y = 𝜔2n u, u0 = 0. (b) Step response with y0 = 0
(ẏ 0 = 0 in both cases). Note the time scaling, 𝜔n t∕𝜋.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
252 5 Linear System Modeling and Analysis

and therefore,
2𝜁𝜔n = b∕m = 800 1∕sec.
Additionally choosing a damping ratio of 𝜁 = 0.707 to provide a small overshoot for step changes in force levels gives,
𝜔n = (400∕0.707) 1∕sec
𝜔2n = k∕m = 32 × 104 1∕sec2 .
Assuming the force plate has a mass of m = 100 gm yields,
k = 0.1 kg ⋅ 32 × 104 1∕sec2 = 3.2 × 104 N∕m = 320 N∕m = 320 N∕cm
b = 0.1 kg ⋅ 800 1∕sec = 80 Nsec∕m = 0.8 Nsec∕cm.
This completes a preliminary design for the force measuring system.

5.2.3 Initial Derivative Condition Response

Another practical situation that can arise for a second-order system is that there is an initial value for y, ̇ with y0 = 0 and
no applied input u(t) = 0. Some examples of this situation include the mass with initial velocity coming into contact with
a spring–damper system, or when an LR circuit with initial current gets switched into capacitor and load, as in a classical
ignition circuit.
The total response for y in this case is given by yI in equation 5.52. It is also convenient to classify yI in terms of damping
ratio as done in the preceding discussion for the ys .

5.2.3.1 Overdamped, 𝜻 > 1

For an overdamped system, s1 and s2 are both real and negative and we can define slow and fast time constants, 𝜏s and 𝜏f ,
as before. The ẏ 0 response for the overdamped case can be expressed as,
ẏ 0 𝜏s 𝜏f [ ]
yI = e−t∕𝜏s − e−t∕𝜏f , (5.71)
𝜏s − 𝜏f
which is the sum of two exponential functions.

5.2.3.2 Critically Damped, 𝜻 = 1

For the critically damped case, the roots are real and negative and s1 = s2 . The solution for yI can be obtained by a limiting
process as
[ ]
(𝜏t + 𝜖)(e−t∕(𝜏t +𝜖) − e−t∕𝜏t )
yI = ẏ 0 𝜏t lim
𝜖→0 𝜖
d(e −t∕𝜏t )
= ẏ 0 𝜏t2
d𝜏t
yI = ẏ 0 te−t∕𝜏t . (5.72)

5.2.3.3 Underdamped, 0 < 𝜻 < 1

For the underdamped case, s1 and s2 are complex conjugates and yI can be expressed as,
ẏ 0 [ (−𝜎+j𝜔 )t ]
yI = − e d − e−(−𝜎−j𝜔d )t
−2j𝜔d
[ ]
ẏ 0 −𝜎t e j𝜔d t − e−j𝜔d t
= e
𝜔d 2j
ẏ 0 −𝜎t
= e sin 𝜔d t
𝜔d
which can also be expressed as
ẏ 0 −𝜎t ( )
yI = e sin 𝜔d t . (5.73)
𝜔d
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
5.2 Analysis of Second-Order System Models 253

1
ζ = 0.0
ζ = 0.2
0.5 ζ = 0.4
ζ = 0.707
yI ωn
y0 0 ζ = 1.0
ζ = 1.4

–0.5

–1
0 0.5 1 1.5 2 2.5 3 3.5 4
Dimensionless time, ωnt/π

Figure 5.11 Initial condition (ẏ 0 ) response for second-order system, ÿ + 2𝜁𝜔n ẏ + 𝜔2n y = 𝜔2n u, equivalent to impulse response. Note
that y0 = u0 = 0 in these plots.

Once again the underdamped solution can be represented as a sinusoid with an exponential envelope. A plot of the yI
response, scaled by 𝜔n ∕ẏ 0 , for a range of values of damping ratio 𝜁 is given in Figure 5.11. The total response for both initial
conditions y0 and ẏ 0 and a step input can be found by adding ys and yI .

5.2.4 Impulse Response

Recall that the derivative of a step function is an impulse function. As such, it is common that a step input may be
differentiated when a set of model equations are expressed in the second-order form. This was the case, for example,
when the force measuring system model equations were expressed in the second-order form in terms of momentum (or
̇
velocity), particularly equation 5.34. In that case, a derivative of the applied force, F(t), appears as the input, which is then
an impulse function. The impulse response function itself, g(t), is a useful linear system characterization. Given any input,
u(t), g(t) can be used to find the response of a system using the a convolution integral by,
t
y(t) = g(t − 𝜏)u(𝜏)d𝜏. (5.74)
∫0
References [53, 91, 92] provide additional discussion.
One way to determine the impulse response is to take advantage of the fact that the derivative is a linear operator. Thus,
given the response of a linear system to a known input, the derivative of the known response is the response of the system
to the derivative of the known input. Thus:
d [ ]
step response of linear second-order ODE → impulse response of linear second-order ODE.
dt
Thus, we can find the impulse response of the linear second-order system by taking the derivative of ys , which is the step
response solution given previously as equation 5.51.
Further, the impulse response with zero initial conditions is identical to the ẏ 0 response given in the preceding discussion.
In general, then, a step response is equivalent to changing y0 and an impulse response is equivalent to changing ẏ 0 or,

step ⇔ y0
impulse ⇔ ẏ 0 .

In this sense, the scaled plot of the ẏ 0 response in Figure 5.11 can also be interpreted as the impulse response of a
second-order system with zero initial conditions.
To show this explicitly, take y0 = 0, so ys from equation 5.51 becomes,
+u0 [ ]
ys = −s2 es1 t + s1 es2 t . (5.75)
s 2 − s1
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
254 5 Linear System Modeling and Analysis

Differentiating this expression yields the impulse response,

ss [ ]
y𝛿 = ẏ s = +u0 1 2 es2 t − es1 t , (5.76)
s 2 − s1
which is identical to the ẏ 0 response given in equation 5.52 with,
ẏ 0 = u0 s1 s2 = u0 𝜔2n ,
since s1 s2 = 𝜔2n . Thus, the unit impulse response functions for the three 𝜁 cases of interest are summarized as follows:
𝜔 ( )
g(t) = √ n e−𝜎t sin 𝜔d t for 0 < 𝜁 < 1 (5.77)
1−𝜁 2

= 𝜔2n te−t∕𝜏t for 𝜁 = 1 (5.78)

𝜏s 𝜏f [ −t∕𝜏 ]
= 𝜔2n e s − e−t∕𝜏f for 𝜁 > 1, (5.79)
𝜏s − 𝜏f
where u0 → 1. These are identical to the results that can be derived using Laplace transforms (e.g., see Chapter 2 of Ogata
[53]).

5.2.5 Summary
A linear time invariant second-order system can always be put in the standard form for a variable of interest, y,
ÿ + 2𝜁𝜔n ẏ + 𝜔2n y = 𝜔2n u(t)
The solution to this equation is highly dependent on the damping ratio 𝜁. If 𝜁 > 1, then the response will be exponential
in character similar to a first-order system. But, if 0 < 𝜁 < 1, then the response will be both exponential and oscillatory in
nature. If 𝜁 = 0, the response will be entirely oscillatory. The responses for various values of 𝜁 are depicted in Figures 4.19
and 4.20. The student would be well advised to commit the essence of these figures to memory.
It is also worthwhile to remember the corresponding transfer function form of the standard second-order form,
y 𝜔2n
= G(s) = .
u s2 + 2𝜁𝜔n s + 𝜔2n
Both the time-domain and s-domain (transfer function) form of the second-order equation arise frequently in system
modeling and design.
̇
Response relations for y(0) = y0 , y(0) = 0, and step input, u0 :
1) Overdamped, 𝜁 > 1
1 1
𝜏s = √ (slow), 𝜏f = √ (fast).
(𝜁 − 𝜁 2 − 1)𝜔n (𝜁 + 𝜁 2 − 1)𝜔n
[ ]
( ) 𝜏s 𝜏f
ys (t) = u0 + y0 − u0 e−t∕𝜏s
− e−t∕𝜏f
.
𝜏s − 𝜏f 𝜏s − 𝜏f

2) Critically damped, 𝜁 > 1

( )[ ]
ys (t) = u0 + y0 − u0 e−t∕𝜏t + (t∕𝜏t )e−t∕𝜏t .
3) Underdamped, 𝜁 < 1
( ) e−𝜁 𝜔n t [ √ (√ )]
ys (t) = u0 + y0 − u0 √ sin 𝜔n 1 − 𝜁 2 t + tan−1 1 − 𝜁 2 ∕𝜁 .
1 − 𝜁2

Example 5.1 Model of a field-excited dc motor

A field-excited dc motor has a motor constant, rf , that depends on the state of a magnetic field inductor, Lf , that is powered
by source Vf (t). Consider the field-excited dc motor shown in Figure 5.12(a) which has rf = Kf if , where Kf is a constant and
if is the field circuit current. The motor torque is thus, 𝜏m = Kf if io , where io is a constant armature drive current. The bond
graph in Figure 5.12(b) shows that the field flux linkage, 𝜆f , and the motor speed, 𝜔m , can be taken as states. Derive state
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
5.2 Analysis of Second-Order System Models 255

I : Lf

λ̇m if

Vf (t) Vs(t) VRf

Rf Se 1 R : Rf
Field circuit if
if
Lf J rf (if)
·· τm ḣm
io(t) Sf G 1 I : Jm
Load io(t) ωm ωm
ωm
τBm
Armature circuit B

R : Bm
(a) (b)

Figure 5.12 (a) Field-excited schematic. (b) Bond graph model.

equations and convert these to a single second-order ODE in standard form for the motor shaft speed, 𝜔m . Identify the
system damping ratio, 𝜁, and undamped natural frequency, 𝜔n , in terms of system parameters.
Solution
From the bond graph, the state equation are given as follows. For the flux linkage,
𝜆̇ f = −VRf + Vs (t) = −Rf 𝜆f ∕Lf + Vs (t).
For the motor momentum,
ḣ m = Jm 𝜔̇ m = 𝜏m − 𝜏Bm = Kf if io − Bm 𝜔m = (Kf io ∕Lf )𝜆f − Bm 𝜔m .
Differentiate this equation,
Jm 𝜔̈ m = (Kf io ∕Lf )𝜆̇ f − Bm 𝜔̇ m ,
and thus,
[ ]
Jm 𝜔̈ m = (Kf io ∕Lf ) −Rf 𝜆f ∕Lf + Vs (t) − Bm 𝜔̇ m ,
To eliminate 𝜆f , us the original 𝜔̇ m equation,
Jm B
𝜆f ∕Lf = 𝜔̇ + m 𝜔 .
Kf io m Kf io m
Now substitute into the above equation for 𝜔m ,
[ ]
[ ] −Rf Kf io Jm Bm Kf io
Jm 𝜔̈ m = (Kf io ∕Lf ) −Rf 𝜆f ∕Lf + Vs (t) − Bm 𝜔̇ m = 𝜔̇ + 𝜔 + V (t) − Bm 𝜔̇ m ,
Lf Kf io m Kf io m Lf s
which simplifies to,
[ ] [ ] [ ]
Rf Bm Rf Bm Kf io Rf Bm Kf io
𝜔̈ m + + 𝜔̇ m + 𝜔m = V (t) = V (t).
Lf Jm Lf Jm Lf Jm s Lf Jm Rf Bm s
Thus, in the standard form,
𝜔̈ m + 2𝜁𝜔n 𝜔̇ m + 𝜔2n 𝜔m = 𝜔2n u(t),
where u(t) = (Kf io )∕(Rf Bm ) ⋅ Vs (t),
Rf Bm Rf B m
2𝜁𝜔n = + , and, 𝜔2n = .
Lf Jm Lf Jm
The standard second-order model response relations can now be studied using the values of 𝜁 and 𝜔n based on given
parameters.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
256 5 Linear System Modeling and Analysis

5.3 Application of Model Response

The analytical solutions for linear first- and second- order systems provide insight into experiment design and system
design. Many systems or part of systems can be well represented by linear first- or second- order systems. This allows
for their isolated testing and/or design, informed by methods that take advantage of well-known linear model response
relations.

5.3.1 Experiment Design and Parameter Determination

Linear second-order system response relations provide a basis for designing experiments from which data can be gath-
ered for parameter determination. First- and second-order system models are used to guide the design of experiments that
take advantage of the analytical response relations, which depend on 𝜏t , 𝜁, and 𝜔n in known ways. Often the parameters
values for I and C elements may be given or known, or can be inferred from known material and geometric parameters or
manufacturer data. It is most often the case that determining parameter data for dissipative elements requires experimental
measurement.

5.3.1.1 Undamped or “Lightly Damped” Systems

Some second-order systems with very low 𝜁 can often be set up to oscillate harmonically. If the damping is so low that
observed oscillations appear to not decay, it is possible to estimate the undamped natural frequency from Tn ≈ 2𝜋∕𝜔n .
By setting 𝜁 = 0 in equations 5.40 and plugging into equations 5.49, the unforced response of a standard second-order
system is given by,
ẏ
y(t) = y0 cos(𝜔n t) + 0 sin(𝜔n t), (5.80)
𝜔n
where y0 and ẏ 0 are the initial conditions. Thus, an experiment should be set up with solely an initial condition y0 with no
forcing, while forcing ẏ 0 to zero as well. The ideal y0 response is shown in Figure 5.10(a) for 𝜁 = 0. Since all real systems
have some damping, any real measurement of an oscillation period will actually be a damped period, Td > Tn . If losses
are more significant, then alternative methods are needed. If a resistive parameter cannot be measured directly (e.g., with
an ohmmeter and force gauge), then a common approach for underdamped systems is to infer system damping using the
logarithmic decrement. This approach takes advantage of the underdamped response solution, as described in the following.

5.3.1.2 Logarithmic Decrement for Underdamped Systems

A system that can be modeled as a linear underdamped system will undergo damped oscillations, when released from y0 ,
with ẏ 0 and no input, u(t) = 0. The response can be approximated by ys with u0 = 0, from equation 5.69,
e−𝜎t ( ) ( )
ys = y0 √ sin 𝜔d t + 𝜙 = Ao e−𝜎t sin 𝜔d t + 𝜙 , (5.81)
1 − 𝜁2
(√ )
where 𝜙d = tan−1 1 − 𝜁 2 ∕𝜁 .
From this form of the response, we derive a commonly used technique for experimentally identifying the damping
ratio under the specified conditions. The amplitude decay envelope, shown in Figure 5.13(a), is √ solely determined by
A(t) = Ao e−𝜁 𝜔n t . The period of oscillation is the damped natural period, Td , which is 2𝜋∕𝜔d , 𝜔d = 𝜔n 1 − 𝜁 2 .
The peaks and envelope for a typical linear underdamped second-order system response is shown in Figure 5.13(a).
For such a measured decayed oscillation, the relation between successive oscillations is,
Ai [ ]
= exp 𝜁𝜔n (ti+1 − ti ) ,
Ai+1
where i is the cycle number. The period between any two peaks is: Td = ti+1 − ti = 2𝜋∕𝜔d , so,
[ ]
Ai 2𝜋𝜁
= exp √ .
Ai+1 1 − 𝜁2
Now, focusing on the peaks and the oscillation number, i, let Ao be any first oscillation peak and Ai the ith peak. Then,
[ ]
Ao 2𝜋i𝜁
= exp √ ,
Ai 1 − 𝜁2
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
5.3 Application of Model Response 257

1.00 A0 1.0
A1
0.75
Ai 0.8
0.50
0.25 0.6
A0
y(t) In
0.00 Ai
0.4
–0.25
2πζ
–0.50 0.2 1 – ζ2
–0.75
Exponential decay
envelope, A(t) 0.0
–1.00
0 20 40 60 0 2 4 6 8
Time, sec Cycle number, i
(a) (b)

Figure 5.13 (a) Decayed oscillation for a second-order system with linear damping. (b) Log of amplitude ratios versus cycle number
indicating linear slope.

from which we can write,

[ ]
A 2𝜋𝜁
ln o = √ ⋅ i. (5.82)
Ai 1 − 𝜁2
For
[ ]any two successive peaks, the logarithm of the amplitudes is referred to as the logarithmic decrement. If the plot of
A
ln Ao versus i follows a straight line, as in Figure 5.13(b), the assumption that the system is linear and second order is
i
reasonable. In this case, the slope of the line described by equation 5.82 then provides an estimate of the system damping
ratio by,
𝛽n
𝜁= √ , (5.83)
4𝜋 2 + 𝛽n2
where,
[ ]
1 Ao 2𝜋𝜁
𝛽n = ln = √ .
n An 1 − 𝜁2
A deviation of the data from a straight line in the log decrement plot is typically a good indication that the linear
second-order system model assumption is not valid. This type of data reduction is illustrated in Figure 5.13, which shows
measured angular position for an oscillating pendulum that has nonlinear pivot friction. In this case, the linear decay
envelope as shown in Figure 5.14 for the measured position would likely signal that the damping is dominated by dry
friction.
As a point of reference, some typical values for damping ratio for several types of systems are listed below.
metals (elastic range) 𝜁 < 0.01
auto shock absorbers 𝜁 ∼ 0.3
rubber 𝜁 ∼ 0.05
building during earthquake 0.01 < 𝜁 < 0.05
recoil mechanism 𝜁 ∼ 1.0.

The logarithmic decrement method is a classical method that can be applied in many areas for system testing and charac-
terization. It is best when a sufficient number of oscillations can be measured so that estimates of the amplitude ratios can
be made. If the system is not underdamped, or there are not enough peaks to get a good estimate, then alternative methods
that employ either first- or second-order models can be applied to estimate the system parameters. The following examples
illustrate how this can be done for a permanent-magnet DC (PMDC) motor.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
258 5 Linear System Modeling and Analysis

1.5
3.0
1.0
2.5
0.5
2.0
A0 Nonlinear trend
y(t) 0.0 In
Ai 1.5 in log decrement

–0.5 1.0

–1.0 Linear decay 0.5

envelope
–1.5 0.0
0 5 10 15 20 0 5 10
Time, sec Cycle number, i
(a) (b)

Figure 5.14 Reduction of decayed oscillation test data to evaluate type of damping. (a) Data shows a linear decay envelope. (b) Log
decrement with a nonlinear trend shows linear damping not a suitable model.

Example 5.2 Model-based testing of PMDC motor

A bond graph model of the PMDC motor shown in Figure 5.15(a) has been developed as in Figure 5.15(b). The model will
be used to interpret results from two tests used to determine the motor inductance Lm , resistance Rm , motor constant rm ,
rotational inertia Jm , and linear bearing resistance Bm . The two tests are as follows:
Test 1 (blocked-rotor test). With the motor blocked from rotating, apply a step dc voltage and measure the resultant current
trace.
Test 2 (unblocked test). With the motor unblocked, apply a step dc voltage and measure the resultant angular velocity trace
of the motor shaft.
Results from these two tests on a particular motor using a step voltage of 4 volts are presented in Figure 5.16.

I : Lm I : Jm

Lm Rm
λ̇m im ḣm ωm
Jm
rm
Vs(t) Vm ·· τm τ=0
Se 1 G 1 Se
im ωm
Vs(t) im Vm
VRm im τBm
τm ωm

τm = rmim Bm R : Rm R : Bm
(a) (b)

Figure 5.15 (a) PMDC schematic. (b) Bond graph model.

Solution
State equations are derived from the bond graph model given in Figure 5.15(b) for the rotor blocked case (𝜔 = 0):
𝜆̇ m = Vs − Rm 𝜆∕Lm ,
or,
Lm dim 1
+ im = V,
Rm dt Rm s
which is a first-order differential equation with time constant, 𝜏t = Lm ∕Rm . The solution for the motor current can be found
using equation 5.19,
im = iss (1 − e−t∕𝜏t ),
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
5.3 Application of Model Response 259

iss = 0.11 A
1
100 Blocked 0.8 ωss = 0.65 rad/s
80 0.63iss = 0.7 A 0.6
Current, mA

Speed, rad/s
Unblocked
60 0.4

40 0.2

20 Unblocked 0

0 –0.2
0 2 4 6 8 0 2 4 6 8
τt ≈ 0.85 sec Time, msec Tp ≈ 1.0 sec Time, msec
(a) (b)

Figure 5.16 (a) Motor current for blocked and unblocked motor with a Vs = 4 volt step input. (b) Speed for the unblocked case, Vs = 4
volt step input.

where steady state current is iss = Vs ∕Rm . From the measurement in Figure 5.16(a), we see that iss = 0.11 A, and therefore
the motor resistance can be found as,
Vs 4V
Rm = = = 36.4 Ω.
iss 0.11 A
At t = 0, we have,
dim (0) iss
= ,
dt 𝜏t
which says that the initial slope can be used to determine the time constant as indicated in Figure 5.16(a) as,

𝜏t = iss ∕(dim (0)∕dt) = 8.5 × 10−4 sec .

An alternative is to find 𝜏t from the time it takes to reach 0.63iss . This result can then be used to find inductance,

Lm = Rm 𝜏t = 0.031 H.

With the motor unblocked, we have the state equations,

𝜆̇ m = −Rm 𝜆m ∕Lm − rm hm ∕Jm + Vs ,

ḣ m = rm 𝜆m ∕Lm − Bm hm ∕Jm ,

and output equation,

𝜔m = hm ∕Jm .

This test has the motor operating with second-order dynamics. A second-order differential equation for 𝜔 can be derived
in two ways. One way is to convert the two first-order state equations directly, as illustrated with the force measuring system
at the beginning of Section 5.2. Start by differentiating the ḣ equation and substituting the 𝜆̇ equation.
Another approach is to use the state-space equations to derive a transfer function (see Section 4.4), which allows manipu-
lating the equations algebraically before transforming back into a simplified second-order ODE form. First, transform into
s-domain by substituting any d()∕dt with the s operator. For the state-space form of the equations,
[ ] [ ][ ] [ ]
𝜆 −Rm ∕Lm −rm ∕Jm 𝜆m 1
s = + Vs
h rm ∕Lm −Bm ∕Jm hm 0
[ ][ ] [ ]
(s + Rm ∕Lm ) rm ∕Jm 𝜆m Vs
= .
−rm ∕Lm s + Bm ∕Jm hm 0
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
260 5 Linear System Modeling and Analysis

Note that these equations are now in the s-domain. Since we want an equation for 𝜔 = h∕J, we can first find h by applying
Cramer’s rule,
| s + R ∕L V |
| s|
| m m
|
| −rm ∕Lm 0 | rm Vs ∕Lm
hm = | | = .
| s + R ∕L | s 2 + (B ∕L + B ∕J )s + (r 2 + R B )∕L J
| rm ∕Jm | m m m m m m m m m
| m m
|
| −rm ∕Lm s + Bm ∕Jm |
| |
Then,
Rm Vs ∕Lm Jm
𝜔m = hm ∕Jm = 2
.
s2 + (Rm ∕Lm + Bm ∕Jm )s + (rm + Rm Bm )∕Lm Jm
Now this s-domain form can be easily transformed back into time domain by using the s ↔ d()∕dt relationship:
( )
Rm B m r 2 + Rm Bm r
𝜔̈ m + + 𝜔̇ m + m 𝜔m = m Vs .
Lm Jm Lm Jm Lm Jm
Noting Figure 5.10, we see that the per cent overshoot on Figure 5.16(b) matches 𝜁 ≈ 0.3 and from the time to peak
overshoot (𝜔n tp ≈ 3.14 on Figure 5.10) that,
𝜔n (0.001 sec)
= 1 ⇒ 𝜔n ≈ 3141 rad∕sec.
𝜋 rad
Also note that the steady-state angular velocity is 𝜔ss = 0.65 rad/sec. If we cast our second-order differential equation in
standard form 5.28, we have
2 +R B
rm m m
𝜔2n = = 9.87 × 106 ,
Lm Jm
Rm B m
2𝜁𝜔n = + = 1885,
Lm Jm
and
2
(rm + Rm Bm )𝜔ss = rm Vs .

Since we have previously determined Rm = 36.4 Ω and Lm = 0.031 H, these three equations can be used to solve for rm ,
Bm , and Jm as:
rm = 5.6 N-m/A
Jm = 1.14 × 10−4 kg-m2
Bm = 8.05 × 10−2 N-m-sec.
From this example, we see that our knowledge of responses of first- and second-order systems, besides its obvious analysis
purpose, is also useful in system identification. Once identified, these parameters can be used with the model to predict
system response to arbitrary inputs.

Example 5.3 Model-based testing of PMDC motor

Predict the response of the electromechanical (EM) system of Example 5.2 when the rotor is initially blocked and a voltage
of 4V is applied to the terminals of the motor. At a time long compared to the system time constant, the block on the rotor
is suddenly released and the motor begins to spin. Develop an equation which describes the angular velocity of the motor.
Solution
Using the numerical parameters identified in Example 5.2, a second-order differential equation for motor shaft speed, 𝜔m ,
can be obtained as

𝜔̈ m + 1885𝜔̇ m + 9.87 × 106 𝜔m = 1.6 × 106 Vs ,

which can be expressed in standard form 5.28 as

𝜔̈ m + 1885𝜔̇ m + 9.87 × 106 𝜔m = 1.6 × 106 Vs

 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
5.3 Application of Model Response 261

where,
𝜁 = 0.3, 𝜔n = 3141.
To use the standard step response results, determine the step input magnitude, u0 , by comparing to the standard
second-order form, that is, rm ∕(Lm Jm )⋅s = 𝜔2n u0 , so
[ ]
u0 = rm ∕(Lm Jm ) Vs ∕𝜔2n = 0.65 sec−3 .
The initial conditions when the rotor is blocked are,
𝜔m (0) = 0 (blocked rotor),
but now there is an initial current induced in the coil so there is an initial motor torque,
𝜏m (0) = rm im (0) = J 𝜔̇ m (0).
This results in an initial acceleration in the motor, or an initial condition,
𝜔̇ m (0) = rm im (0)∕J = rm Vs ∕(Rm Jm ) = 5453 rad∕sec2 .
The total response, 𝜔m (t), can now be obtained as,
𝜔 = 𝜔s + 𝜔I ,
where
( √ (√ ))
e−𝜁 𝜔n t
𝜔s = u0 − (u0 − 𝜔m (0)) √ sin 𝜔n 1 − 𝜁 2 t + tan−1 1 − 𝜁 2 ∕𝜁 .
1 − 𝜁2
𝜔̇ m (0) ( √ )
𝜔I = √ e−𝜁 𝜔n t sin 𝜔n 1 − 𝜁 2 t .
𝜔n 1 − 𝜁 2
A plot of this response is depicted in Figure 5.17 where it is labeled “Blocked step response.” The “normal step response”
which occurs when the motor is given a step input without blocking the rotor is also plotted in this figure. Note that the
blocked response has a higher overshoot before returning to the steady-state speed. This is due to the initial torque that
results from the initial current, im (0), induced in the coil while the rotor is blocked. This additional overshoot would also
result if there were an impulsive torque on the rotor.

2
Blocked step response
Speed, rad/sec

1.5

1 Normal step response

0.5

0
0 1 2 3 4 5 6 7
Time, msec

Figure 5.17 Response of blocked and unblocked rotor to step response.

5.3.2 Time-Domain System Specifications

The characteristic response features of both first- and second- order systems are often used to define dynamic performance
of systems. Many systems are designed to achieve a response rate, for example, that relates to a system time constant. Or the
overshoot in a system during a transient response may need to be less than a certain value to avoid collisions. Defined
specifications can also be referred to when testing to determine whether a system has met requirements.
First-order response. The unforced response or step response of a first-order system both provide a basis for defining a
system time constant, 𝜏t , as a response specification. This parameter can be shown to be related to the underlying parameters
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
262 5 Linear System Modeling and Analysis

Allowable
1.25 Mp tolerance
Unit step
response 1
Tolerance of
0.75 td 0.05 or 0.02

0.5

0.25

0
tr Time, t

tp
ts

Figure 5.18 Time-domain specifications based on second-order unit step response.

that quantify the response rate. As will be shown in Chapter 6, the time constant is also very useful in defining frequency
response specifications.
Second-order response. The unit step response of a second-order system is one of the most commonly used as a test
basis for defining specifications of systems, even those that are said to have dominant second-order behavior. This means
that they may be higher order but their performance is for all intents and purposes second-order in nature. The following
common specifications are illustrated in Figure 5.18 and can be referred to in testing and design of such systems:

● Delay time, td , is the time to reach 1/2 the final value for the first time
● Rise time, tr , defined for different systems in different ways; for second-order, underdamped, use the 0 to 100% rise time
while for overdamped you use the 10 to 90% time
● Peak time, tp , is time to first peak
● Maximum overshoot, Mp , peak relative to SS value in %
● Settling time, ts , again different values, 2%, 5%, etc.

Rise time, tr , for an underdamped system is the first time at which the steady-state value is reached. So, in this case,
when D1 (t) = 1. This means, sin(𝜔d tr + 𝛽) = 𝜋, from which we can find,
(𝜔 ) 𝜋 −𝛽
1 d
tr = tan−1 = , (5.84)
𝜔d −𝜎 𝜔d
√
where 𝜎 = 𝜁𝜔n and 𝛽 = tan−1 ( 1 − 𝜁 2 ∕𝜁).
Time to first peak, tp , can be found by taking dD1 ∕dt (in equation 5.69), setting it equal to zero solving for the time.
Thus,
[ ]
( √ (√ ))
dD1 d e−𝜁 𝜔n t
= √ sin 𝜔n 1 − 𝜁 t + tan
2 −1
1 − 𝜁 ∕𝜁
2
dt dt 1 − 𝜁2
( √ (√ ))
e−𝜁 𝜔n t
= −𝜁𝜔n D1 (t) + 𝜔d √ cos 𝜔n 1 − 𝜁 2 t + tan−1 1 − 𝜁 2 ∕𝜁 .
1 − 𝜁2
(√ )
Setting dD1 ∕dt = 0 leads to, tan(𝜔d t + 𝛽) = tan 𝛽, where, 𝛽 = tan−1 1 − 𝜁 2 ∕𝜁 . Thus, peaks occur as expected at,
𝜔d t = 0, 𝜋, 2𝜋, etc. The first peak is at,

tp = 𝜋∕𝜔d . (5.85)

This peak time corresponds to one half cycle of the frequency of damped oscillation, which is based on the damped period,
Td = 2𝜋∕𝜔d . So tp could also be inferred directly from, 𝜔d tp = 𝜋, so tp = 𝜋∕𝜔d .
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
5.4 Analysis Using System Model Transfer Functions 263

Maximum overshoot, Mp , occurs at the peak time or at t = tp = 𝜋∕𝜔d . So, you can solve from the nondimensional
response equation 5.69 to get, Mp = D1 (tp ) − 1, giving,
√
1−𝜁 2 )𝜋
Mp = e−(𝜁 ∕ , (5.86)
which is often reported as a percentage. Note that overshoot depends only on 𝜁, so given Mp , it is possible to solve for the 𝜁
needed to achieve a certain Mp . You can show that,
ln Mp [ ]−1∕2
𝜁= √ = 1 + (𝜋∕ ln Mp )2 . (5.87)
[ ]2
𝜋 + ln Mp
2

√
Settling times for underdamped second-order systems are based on the envelope curves, 1 ± exp [−𝜎t] ∕ 1 − 𝜁 2 , which
have a time constant of 1∕𝜎, 𝜎 = 𝜁𝜔n . For convenience in comparing the responses of systems, we commonly define settling
times by the percentage, e−𝜎t . Use ln Δ, where Δ is the settled level (0.01 for 1%, 0.02, etc.). Some common approximations
for settling times are:
4.6 4.6
t99 = 4.6𝜏e = = 1% criterion,
𝜎 𝜁𝜔n
4 4
t98 = 4𝜏e = = 2% criterion,
𝜎 𝜁𝜔n
3 3
t95 = 3𝜏e = = 5% criterion,
𝜎 𝜁𝜔n
where we have used a common designation by percentage in reaching a value.
The time-domain specifications reviewed here can be used in various ways to design passive systems, as measures during
qualification testing, or, as will be shown in Chapter 6, to set controller gains.

Example 5.4 Response from transfer functions and time-domain specifications

Consider the second-order system,
y 10
= .
u s2 + 2s + 9
Determine the rise time, tr , time to first peak, tp , percent overshoot, Mp , and 2% settling time.
Solution √ √
From the characteristic equation, 2𝜁𝜔n = 2 and 𝜔2n = 9, giving 𝜔n = 3, 𝜁 = 1∕3, and thus, 𝜔d = 𝜔n 1 − 𝜁 2 = 8.
The specifications can then be determined as follows:
𝜋−𝛽
tr = = 0.457,
𝜔d
√
where 𝛽 = tan−1 ( 1 − 𝜁 2 ∕𝜁). Time to first peak,
tp = 𝜋∕𝜔d = 1.11 sec.
Percent overshoot is,
√
1−𝜁 2 )𝜋
Mp = e−(𝜁 ∕ = 0.329,

and the approximate 2% settling time is, t98 = 4∕(𝜁𝜔n ) = 4 sec. The step response computed in MATLAB® enables these
characteristics to be extracted either directly from the figure plot or from a built-in function, stepinfo().

5.4 Analysis Using System Model Transfer Functions

We recognize that there are equivalent forms for a linear system: linear state and output equations, an nth-order differential
equation, transfer function(s). Analysis techniques for determining the response of the system given certain inputs are avail-
able for all of these formulations, but some may be more convenient or insightful for a particular application. This section
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
264 5 Linear System Modeling and Analysis

Table 5.1 Standard first- and second-order forms

with transfer functions.

First-order form Second-order form

𝜏t ẏ + y = u(t) ÿ + 2𝜁𝜔n ẏ + 𝜔2n y = 𝜔2n u(t)

y 1 y 𝜔2n
u = 𝜏t s + 1 u = s2 + 2𝜁𝜔 s + 𝜔2
n n

examines some analysis methods for transfer function forms. A transfer function4 is convenient for determining certain
properties of a system, and response characteristics such as final and values can be determined as well. Analytical expres-
sions for the time-domain response to test inputs can be derived using inverse Laplace transforms and partial fraction
expansions, as described by Ogata [53]. The discussion that follows shows how some of these techniques can be used to
arrive at time-domain response relations from transfer function representations.

5.4.1 Standard Transfer Function Forms

The standard forms of first- and second-order system ODEs are summarized in Table 5.1, along with corresponding trans-
fer functions relating a system variable y to the input u. It is common to capitalize s-domain variables, such as Y (s) for
the Laplace transform of y(t), in transfer function relations. Here lowercase variables are retained, such as y and u, when
referring to either time or s-domain variables, since the context is evident from the presence of s terms in the expressions.
First- and second-order systems expressed in standard form facilitate the use of known solutions, such as those derived
in Sections 5.1 and 5.2. Solutions for systems of lower order, say n ≤ 3, can also be found using tabulated Laplace transform
pairs [53]. Recall that the order of a system is equal to the order of the denominator, D(s), in the transfer function, G(s) =
N(s)∕D(s).
In order to take advantage of these known solutions, transfer functions with n > 2 can be expressed as products of first
and second systems. That is, a system transfer function relating y to an input u might be expressed in the general form,
∑ ∑
y N(s) K k (s + zk ) l (s2 + 2𝜁l 𝜔nl s + 𝜔nl )
= G(s) = = ∑r ∑I 2 (5.88)
u D(s)
i (s + 𝜆i ) i (s + 2𝜁i 𝜔ni s + 𝜔ni )
2

with K an overall gain constant. A partial fraction expansion can then be used to decompose such a transfer function into
first- and second-order forms.
It is helpful to first illustrate how partial fractions relate to individual factors in a linear system. For example, consider a
second-order system in state-space form,
ẋ 1 = −2x1 − u. (5.89a)
ẋ 2 = −x2 + u. (5.89b)

The solution of this particular system is easy since the state equations are decoupled. That is, the rate (state) equation for
x1 is only a function of the input u and x1 , and the rate equation for x2 is solely a function of the input u and x2 . This implies
that x1 and x2 can be solved for independently. Now, if the system output y is taken as the sum of these two states,

y = x1 + x2 , (5.90)

then y can also be found by simple addition of the two solutions for x1 and x2 .
In transfer function notation, the system can be expressed by,
x1 −1
= , (5.91a)
u s+2
x2 1
= , (5.91b)
u s+1

4 Ways for determining transfer functions are introduced in Chapter 4 and impedance-based methods are discussed in Chapter 6.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
5.4 Analysis Using System Model Transfer Functions 265

and y is given by,

x x
y = 1 u + 2 u,
u u
or,
y x x −1 1
= 1 + 2 = + . (5.92)
u u u s+2 s+1
In the combined form,
y 1 1
= = , (5.93)
u (s + 2)(s + 1) s2 + 3s + 2
which represents the second-order differential equation,
ÿ + 3ẏ + 2y = u. (5.94)
The analyst then has the option of solving either the second-order differential 5.94 or the two first-order differential
equations 5.89 with output 5.90. This example illustrates how a partial fraction expansion relates or reduces a high-order dif-
ferential equation or transfer function to a sequence of low-order differential equations (in standard form). In this example,
the second-order transfer function 5.93 can be reduced to the sum of two first-order transfer functions 5.92.
It should be made clear that the partial fraction expansion is made here to decompose the higher-order transfer function
into the lower-order standard form. This is in contrast to using a partial fraction expansion in a traditional way to directly
solve for the actual response relations. This subtle difference will be made more clear in the following discussion.

5.4.2 System Decomposition Using Partial Fractions

Typically, a general system transfer function,
y N(s)
= G(s) = ,
u D(s)
might have an nth -order denominator polynomial D(s) and kth- order numerator polynomial N(s). The first step in a partial
fraction analysis of a transfer function in this form is to find the solutions to the equation,
D(s) = an sn + · · · + a0 = 0. (5.95)
It is a well-known mathematical theorem that there are n solutions to a polynomial equation of this form, and for real
coefficients an , … , a0 , these solutions are either real or complex conjugates. Designating the real solutions as −𝜆1 , … , −𝜆r
and the complex conjugate solutions as −𝜎1 ± j𝜔d1 , … , −𝜎I ± j𝜔dI where there are r real solutions and 2I complex conjugate
solutions. The values of s that satisfy 5.95 are called characteristic values or eigenvalues, and equation 5.95 is called the
characteristic equation. Using the eigenvalues, the characteristic equation can be factored as,
[ ][ ]
D(s) = (s + 𝜆1 ) · · · (s + 𝜆r ) (s2 + 2𝜎1 s + 𝜎12 + 𝜔2d1 ) · · · (s2 + 2𝜎I s + 𝜎I2 + 𝜔2dI ) = 0. (5.96)
r real eigenvalues 2I complex eigenvalues

Note that 𝜎 = 𝜁𝜔n and + 𝜎2= 𝜔2d 𝜔2n .

In our previous introductory example, the transfer function 5.93 has the characteristic equation,
s2 + 3s + 2 = 0,
with two real eigenvalues 𝜆1 = −2 and 𝜆2 = −1. The transfer function can be expanded in terms of these eigenvalues as,
y 1 −1 1
= = + .
u s2 + 3s + 2 s + 2 s + 1
A logical question then arises: Is it always possible to simply expand a transfer function in terms of its eigenvalues?
As long as the eigenvalues are distinct5 (not equal) then the answer to this question is yes with the general partial fraction
expansion for a transfer function expressed as,
y k1 kr 𝑤 1 s + z1 𝑤I s + zI
= G(s) = +···+ + +···+ + d. (5.97)
u s + 𝜆1 s + 𝜆r s2 + 2𝜎1 s + 𝜎12 + 𝜔2 s 2 + 2𝜎 s + 𝜎 2 + 𝜔2
I I
d1 dI

5 For equal eigenvalues, a limiting argument can be used to give the correct expansion, but this will not be developed in this text. See Ogata [53].
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
266 5 Linear System Modeling and Analysis

The transfer function must also be a proper fraction; that is, the order of D(s) must be greater than that of N(s). If not,
then synthetic division must first be applied to separate the fractions.
Now define the first-order transfer functions,
y11 k1
= , (5.98a)
u s + 𝜆1
⋮
y1r kr
= , (5.98b)
u s + 𝜆r
and the second-order transfer functions,
y21 𝑤1 s + z I
= . (5.99a)
u s2 + 2𝜎1 s + 𝜎12 + 𝜔2 d1
⋮
y2I 𝑤I s + z1
= . (5.99b)
u s + 2𝜎I s + 𝜎I2 + 𝜔2dI
2

Now, continuing with the previous example, the output y can be found by,
r [
y1i ]
I [
∑ ∑ y2i ] ∑ ∑
r I
y= ⋅u+ ⋅u= y1i + y2i , (5.100)
i=1
u i=1
u i=1 i=1

where y1i is found by independently solving,

ẏ 1i + 𝜆i y1i = ki u, i = 1, … , r, (5.101)
which can be put in the standard first-order form,
𝜏i ẏ 1i + y1i = u, i = 1, … , r (5.102)
with,
𝜏i = 1∕𝜆i , (5.103)

ui = (ki ∕𝜏i )u, (5.104)

and y2i is found by independently solving
ÿ 2i + 2𝜎i ẏ 2i + (𝜎i2 + 𝜔2di )y2i = 𝑤i u̇ i + zi u, i = 1, … , I.
This second-order differential equation can also be put in the standard second-order form,
ÿ 2i + 2𝜁i 𝜔ni ẏ 2i + 𝜔2ni y2i = 𝜔2ni ui , i = 1, … , I, (5.105)
where,
√
𝜔ni = 𝜎i2 + 𝜔2di . (5.106)
√
𝜁i = 𝜎i ∕ 𝜎i2 + 𝜔2di . (5.107)

ui = (𝑤i u̇ i + zi u)∕𝜔2ni . (5.108)

Note how the individual (standard) factors are dealt with to compose a total solution. Familiarity with solutions of the
standard first- and second-order systems is essential but also sufficient for composing quick solutions of the response of
systems from higher-order transfer functions.

Example 5.5 Response from transfer functions

For the a third-order system with the transfer function,
y s2 + 2s + 2
= G(s) = 3 ,
u s + 2s2 + 2s + 1
solve for the response, y(t).
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
5.4 Analysis Using System Model Transfer Functions 267

Solution
The given transfer function is equivalent to the differential equation,
...
y + 2̈y + 2ẏ + y = u̇ + 2u̇ + 2u.

The characteristic equation is,

s3 + 2s2 + 2s + 1 = 0,
√
which has a real and a complex conjugate pair of eigenvalues with 𝜆 = 1 for the real root and 𝜎 = 1∕2 and 𝜔d = 3∕2 for
the complex roots. The transfer function can then be expanded as,
s2 + 2s + 2 k 𝑤s + z
G(s) = = + .
s3 + 2s2 + 2s + 1 s + 1 s2 + s + 1
The parameters k, 𝑤, and z can be determined by requiring the expansion to have the same numerator polynomial as the
original transfer function when collapsed back to a single fraction as,
k 𝑤s + z k(s2 + s + 1) + (𝑤s + z)(s + 1) s2 + 2s + 2
+ 2 = = .
s+1 s +s+1 (s + 1)(s2 + s + 1) s3 + 2s2 + 2s + 1
This implies:
k(s2 + s + 1) + (𝑤s + z)(s + 1) = s2 + 2s + 2.
(k + 𝑤)s2 + (k + 𝑤 + z)s + k + z = s2 + 2s + 2.

Equating like powers of s results in the three linear equations:

s2 ∶ k + 𝑤 = 1,
s1 ∶ k + 𝑤 + z = 2,
s0 ∶ k + z = 2,

which have the solution: k = 1, 𝑤 = 0, and z = 1. The partial fraction expansion of the transfer function can then be
expressed,
y 1 1
= + .
u s + 1 s2 + s + 1
Defining the transfer functions,
y1 1
= ,
u s+1
y2 1
= 2 ,
u s +s+1
the output y can be found as,

y = y1 + y2 ,

where y1 is found by solving,

ẏ 1 + y1 = u,

and y2 is found by solving,

ÿ 2 + ẏ 2 + y2 = u.

If u is a unit step input with u0 = 1 and all initial conditions are zero, then from 5.102 and 5.19, we have,

y1 (t) = 1 − e−t ,

and from 5.105 and 5.70, we have,

( )
√ 1 √
y2 (t) = 1 − e−t∕2
cos( 3t∕2) + √ sin( 3t∕2) ,
3
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
268 5 Linear System Modeling and Analysis

and the output y is then,

( )
√ 1 √
−t
y(t) = y1 (t) + y2 (t) = 2 − e − e −t∕2
cos( 3t∕2) + √ sin( 3t∕2) .
3

In the above example, we took a linear third-order system and used a partial fraction expansion to express this system as
the sum of decoupled first- and second-order systems. This is the principal concept we wish to associate with the method
of partial fraction expansions. The response of all linear system are simply the sum of a sequence of first- and second-order
systems. This implies, at least for linear systems, that a thorough understanding of just first- and second-order systems
is sufficient for understanding the response of higher-order systems. Before the wide accessibility of digital computers,
partial fraction expansion methods along with Laplace transform techniques6 provided one of the most effective ways for
obtaining the exact time response of higher-order linear systems. Obtaining the total analytical solution for high-order
systems is rarely done in present times, but this does not at all dilute the importance of the general concept. Since first-
and second-order systems can be characterized by time constants, damping ratios, and natural frequencies, the response of
high-order linear systems can also be characterized by these same parameters. This is an important concept for analysis and
design of systems which can be used before any computer simulations are used to obtain the total response. Understanding
these concepts helps build intuition into how these systems can respond under different forced and unforced conditions.

5.5 Poles and Zeros from Transfer Function Models

A transfer function can be described in terms of its poles and zeros. A zero of a transfer function is a value of s that makes
N(s) = 0 and therefore G(s) = 0. A pole of a transfer function is a value of s that makes D(s) = 0 and therefore G(s) = ∞.
The poles of a transfer function are the same as the eigenvalues of the system. A knowledge of the poles and zeros is
sufficient to completely describe the response of a linear system (within a constant factor) to any specified input function.
In addition, quick insights can be gained particularly about the steady-state response using the Final Value Theorem.

5.5.1 First- and Second-Order Systems

In general, the poles and zeros may be either real or complex conjugates and some may be either real or complex conjugates
and some may be repeated. The poles and zeros are frequently shown in the complex s-plane, with poles indicated by × and
zeros by ⚬.
For a first-order system, the transfer function is
1
G(s) = ,
𝜏t s + 1
which has a single pole at −1∕𝜏t . This is plotted in Figure 5.19.
As this pole is placed further to the left, the time constant 𝜏t becomes smaller and the system responds at a faster rate.
Conversely, as the pole is placed closer to the origin, the time constant increases and the system responds slower. If the pole
crosses the origin and becomes positive, then the response will be described by a growing exponential of the form

y = Aet∕𝜏t ,

jω Figure 5.19 s-plane plot of first-order pole.

Faster Slower Unstable

–1/τ σ

6 See Dorf and Bishop [61] or Ogata [53] for more complete discussions of the use of Laplace transforms.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
5.5 Poles and Zeros from Transfer Function Models 269

Faster jω jω
oscillation ζ=0
ζ decreasing
+μ
Faster ωn
decay Unstable
ζ increasing
θ θ
–ν σ σ
Unstable ζ=1
ωn
–μ

(a) (b)

Figure 5.20 (a) s-plane plot of second-order poles. (b) Locus of roots as 𝜁 varies.

and the system is called unstable. Energetically passive systems are always stable, but if we provide an external power
source, as is often the case in control systems, then systems can be unstable.
For a second-order system with no input derivatives (no s in the numerator), the transfer function can be expressed
𝜔2n 𝜔2n
G(s) = = ,
(s + 𝜎 + j𝜔d )(s + 𝜎 − j𝜔d ) s2 + 2𝜁𝜔n s + 𝜔2n )
which has a pair of complex conjugate poles at −𝜎 + j𝜔d and −𝜎 − j𝜔d . Figure 5.20 shows a plot of these poles in the s-plane.
As the poles are placed farther to the left of the imaginary axis, then 𝜎 increases and the system decays at a faster rate
with an exponential decay envelope time constant of 𝜏e = 1∕𝜎. If the poles are placed farther from the real axis, then the
frequency 𝜔d increases and the system oscillates at a faster rate. The cosine of the angle 𝜃 in Figure 5.20 is

cos 𝜃 = 𝜎∕𝜔n = 𝜁𝜔n ∕𝜔n = 𝜁

or

𝜃 = cos−1 (𝜁). (5.109)

Equation 5.109 shows that poles with the same angle 𝜃 have the same damping ratio and therefore the same overshoot.
For example, a pole with 𝜃 = 45o has a damping ratio of 0.707. Note that if the system is overdamped (𝜁 > 1), the two poles
are real and can be treated as two first-order systems.

5.5.2 The Influence of Zeros

In order to understand the significance of the zeros of the transfer function, consider G(s) written in the following factored
form:
N(s) K(s + z1 )(s + z2 ) … (s + zk )
G(s) = = , (5.110)
D(s) (s + p1 )(s + p2 ) … (s + pn )
where p1 , p2 , …, pn and z1 , z2 , …, zk are either real or complex conjugate pairs with k ≤ n. For distinct non-repeated poles,
−pi , the G(s) can always be expanded into a sum of simple fractions as,
N(s) k1 k2 kn
G(s) = = + +···+ + d, (5.111)
D(s) s + p 1 s + p2 s + pn
where ki are constants. Here ki is called the residue at the pole s = −pi . The value of ki can be found by multiplying both
sides of equation 5.111 by (s + pi ) and letting s = −pi , giving
[ ] [ n ]
N(s) ∑ kj
(s + pi ) = (s + pi ) + d(s + pi )
D(s) s=−pi j=1
s + pj
s=−pi

= ki .
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
270 5 Linear System Modeling and Analysis

Figure 5.21 Pole-zero plot with vectors for k1 residue calculation.

We see that all the expanded terms drop out, with the exception of kj=i , thus, the residue ki is found from,
[ ]
N(s)
ki = (s + pi ) . (5.112)
D(s) s=−pi

Note that if p1 and p2 are complex conjugates, then the residues k1 and k2 are also complex conjugates. A direct coupling
factor d can be found letting s → ∞ in 5.111. Note that d = 0 if k < n.
Consider a system with two poles and two zeros as
k(s + z1 )(s + z2 )
G(s) = .
(s + p1 )(s + p2 )
The function G(s) has two poles at s = −p1 and s = −p2 and two zeros at s = −z1 and s = −z2 . The partial fraction expan-
sion of G(s) is written as
k1 k2
G(s) = + + d.
s + p 1 s + p2
The residues can be calculated from equation 5.112 as follows:
[ ]
k(s + z1 )(s + z2 ) k(z1 − p1 )(z2 − p1 )
k1 = = ,
s + p2 s=−p1 p2 − p1
[ ]
k(s + z1 )(s + z2 ) k(z1 − p2 )(z2 − p2 )
k2 = = ,
s + p1 s=−p2 p1 − p2
and d can be found as
d = lim G(s) = k.
s→∞

The value of the residues can be interpreted graphically by referring to Figure 5.21. The poles have been plotted as complex
conjugates and the zeros as real just for the sake of illustration. The quantity z1 − p1 may be considered as a vector in the
complex plane drawn from −z1 to −p1 which can be expressed in polar notation as
z1 − p1 = |z1 − p1 |e j𝜙1 ,
similarly,
z2 − p1 = |z2 − p1 |e j𝜙2
and
p2 − p1 = |p2 − p1 |e j𝜃1 .
Then, the residue k1 is found to be
k|z1 − p1 ||z2 − p1 | j(𝜙 +𝜙 −𝜃)
k1 = e 1 2 . (5.113)
|p2 − p1 |
From 5.113, we see that the magnitude of the residue for pole −p1 is proportional to the distance between this pole and
the zero −z1 and likewise it is proportional to the distance to the zero at −z2 .
This observation is a general result: poles that are closer to zeros will have smaller residues than those at greater distances,
and these poles will have proportionally less effect on the forced response. For example the pole at −p1 will have a larger
residue than the pole at −p2 due to the zero at −z shown in Figure 5.22.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
5.5 Poles and Zeros from Transfer Function Models 271

Figure 5.22 Pole-zero location for a system with two real poles and a zero.

Note that the residue only scales the forced part of the response and the exponential response associated with this pole
may still be present in the total response due to initial conditions, even if the residue at a pole is zero. Also note that if the
number of finite zeros is equal to the number of poles k = n and d ≠ 0. This implies a portion of the output y is directly
proportional to the input in the form y∕u = d.

Example 5.6 Pole-zero response analysis

Use pole and zero analysis to estimate the unit step response of the two transfer functions
y 1
G1 (s) = =
u (s + 1)(s2 + 8s + 32)
and
y s + 0.99
G2 (s) = = .
u (s + 1)(s2 + 8s + 32)
Solution
The poles of G1 are a complex conjugate pair, p1,2 = −4 ± j4, and p3 = −1, and these are plotted in Figure 5.23.

4 4

2 2

−6 −4 −2 2 −6 −4 −2 2

−2 −2

−4 −4
(a) (b)

Figure 5.23 (a) Pole-zero locations for G1 and (b) G2 .

The pole closest to the imaginary axis is sometimes called the dominant pole since its exponential decay is slower than the
other poles. For G1 (s), the complex pair of poles are four times farther from the imaginary axis, and, therefore, the response
due to these poles decays four times faster than the dominant real pole at −1. The system can then be approximated by a
first-order form and the response is,
1
y≈ (1 − e−t ),
32
for time greater then about 0.25 sec (the time constant of the fast poles). The value 1/32 in this expression is the steady state
output of G1 (s) (G1 (0)) for unit step input.
For G2 (s), the poles are identical, but there is a zero at z = 0.99. Now the response will more closely resemble that of a
damped second-order system with poles at p1,2 = −4 ± j4 because the zero implies that the residue associated with the dom-
inant pole at p3 = −1 will be small compared to the complex conjugate pair. In this case, we sometimes state that the effect
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
272 5 Linear System Modeling and Analysis

of a pole has been effectively “canceled” by a zero. The response with zero initial conditions can then be approximated as,
0.99 [ ]
y≈ 1 − e−4t (cos 4t + sin 4t) .
32

5.5.3 Summary
Poles and zeros in the complex domain can be used to give a qualitative picture of the response of linear time invariant
systems. The methods reviewed here can be extended to n > 2. While contemporary computer-aided tools are available to
compute response of systems based on the transfer function forms, insights gained from both partial fraction and pole-zero
analysis remain useful, especially in control system analysis [53, 61, 93].

5.6 Response of nth-Order Linear Systems

As discussed in Sections 5.4 and 5.5, a partial fraction expansion can be used to decompose a high-order system transfer
function into a sequence of first- and second-order systems. This general result will be further exploited in this section in
which we analyze high-order systems directly in the time domain.

5.6.1 Free Response of nth-Order Systems

Consider first the case of an unforced n-dimensional linear system with state equations in the form,
ẋ = Ax, (5.114)
with a set of given initial conditions,
x(𝟎) = x𝟎 .
The solution to this unforced system can be found by assuming a general exponential solution of the form,
x(t) = Vest ,
where V is an n-dimensional constant vector. Substituting this exponential form into 5.114 yields,
sVest = AVest

sV − AV = 𝟎
so,
[sI − A] V = 𝟎, (5.115)
where sI is the identity matrix. The solution to equation 5.115 is a classic eigenvalue–eigenvector solution in which either,
V = 𝟎,
which gives a trivial solution to 5.114, or,
|sI − A| = 0.
This expression is the same as the characteristic equation discussed in Section 5.2 (and earlier in Chapter 4). There are
n solutions s1 , …, sn to this equation for an nth-order system and these solutions are called the eigenvalues of the system.
Along with each eigenvalue sj there is an associated eigenvector Vj found from,
[ ]
sj I − A Vj = 𝟎. (5.116)
These eigenvectors are only determined up to a multiplicative constant, so the total unforced response is expressed,
x(t) = 𝛼1 V1 es1 t + · · · + 𝛼n Vn esn t . (5.117)
The 𝛼’s are multiplicative constants that must be determined by initial conditions using the relation,
x(𝟎) = x𝟎 = 𝛼1 V1 + · · · + 𝛼n Vn . (5.118)
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
5.6 Response of nth-Order Linear Systems 273

This expression shows that if x(𝟎) is set commensurate to a specific eigenvector Vi then 𝛼i = 𝟏n and 𝛼j = 𝟎n for j ≠ i. In
this case, the unforced response of the system will then become,
x(t) = Vi esi t . (5.119)
So in this case, every system state responds with a function defined by the specified eigenvalue, si .

Example 5.7 Response of mass–spring–damper system: real eigenvalues

For a linear spring–mass–damper system with mass m = 1, spring-rate k = 2, and damping coefficient b = 3, we have the
state equations,
[ ] [ ][ ]
ẋ 1 0 1 x1
= ,
ẋ 2 −2 −3 x2
where x1 is the spring extension and x2 is the mass momentum. Assume initial conditions are given as x1 (0) = 1 and
x2 (0) = 0. Solve for the eigenvalues and eigenvectors and for the response to initial conditions provided.
Solution
Assume an exponential form for the solution,
[ ] [ ] [ ]
x1 V1 st V1 est
= e = ,
x2 V2 V2 est
and substitute this result into the state equation to yield,
[ ] [ ][ ]
V1 st 0 1 V1 st
s e = e
V2 −2 −3 V2
[ ] [ ][ ] [ ]
V1 0 1 V1 0
s − =
V2 −2 −3 V2 0
[ ][ ] [ ]
s −1 V1 0
= ,
2 s+3 V2 0
which has the characteristic equation,
| s −1 |
| |
| | = s2 + 3s + 2 = 0.
|2 s+3|
| |
Since this is a second-order system, there are two eigenvalues: s1 = −2 and s2 = −1.
For s1 = −2, we have the eigenvector equation,
[ ][ ] [ ]
s1 −1 V11 0
= ,
2 s1 + 3 V21 0
which represents the two equations,
−2V11 − V21 = 0,
2V11 + V21 = 0,
where the first subscript on V11 stands for first state and second subscript stands for first eigenvalue, and the subscripts on
V21 stand for second state, first eigenvalue. Note that these two equations are not independent and we can only solve for
the relative ratio between V11 and V21 as,
V21 = −2V11 .
We can then express the eigenvector related to the first eigenvalue as,
[ ]
1
V1 = ,
−2
which is only determined up to a multiplicative constant. The second eigenvalue is s2 = −1 and the eigenvector
equations are,
−V12 − V22 = 0,
2V12 + V22 = 0,
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
274 5 Linear System Modeling and Analysis

from which the second eigenvector can be derived up to a multiplicative constant,

[ ]
1
V2 = .
−1
The total unforced response now takes the form,
[ ] [ ]
1 −2t 1 −t
x(t) = 𝛼1 e + 𝛼2 e ,
−2 −1
when the 𝛼’s are determined from the initial conditions,
[ ] [ ][ ]
1 1 1 𝛼1
x(0) = =
0 −2 −1 𝛼2
giving: 𝛼1 = −1 and 𝛼2 = 2.
The eigenvalues and eigenvectors can also be used to qualitatively determine the response of the system before initial
conditions are applied. For this example, we have a fast mode associated with the eigenvalue s1 = −2 which has an expo-
nential time constant of 𝜏t1 = 0.5, and a slow mode associated with the eigenvalue s2 = −1 with a time constant of 𝜏t2 = 1.
The eigenvector for the fast mode V1 , as derived previously, is,
[ ]
1
,
−2
which indicates that x1 , the position state variable, is half the magnitude of x2 , the momentum state variable, and opposite
in sign if the system is responding entirely in the fast mode. The eigenvector gives the relative shape of the fast mode or its
mode shape. The eigenvector for the slow mode is,
[ ]
1
,
−1
which indicates that the mode shape for this eigenvalue is both states having the same magnitude with opposite signs. Note
that it is possible, depending on the initial conditions, to have an exclusively fast or slow response.

Example 5.8 Response of mass–spring–damper system: complex eigenvalues

In the previous example, the eigenvalues and associated eigenvectors were real numbers. As we have previously seen, eigen-
values can also be complex. In this case, the eigenvectors will be complex. As an example consider a spring–mass–damper
system with m = 1, k = 1, and b = 1. The state equations are
[ ] [ ][ ]
ẋ 1 0 1 x1
= .
ẋ 2 −1 −1 x2
Assuming an exponential form for the solution gives
[ ]
V1 st
x(t) = e .
V2
[ ][ ] [ ]
s −1 V1 0
= .
1 s+1 V2 0
The characteristic equation for this system is
| s −1 |
| |
| | = s2 + s + 1 = 0,
|1 s+1|
| |
√
which has solutions of the form s1,2 = −𝜎 ± j𝜔d with 𝜎 = 0.5 and 𝜔d = 3∕2 = 0.866. The eigenvector associated with the
eigenvalue s1 = −𝜎 + j𝜔d is found as
[ ]
1
V1 = ,
−𝜎 + j𝜔d
which is only determined up to a (complex) multiplicative constant. The second eigenvector is
[ ]
1
V2 = .
−𝜎 − j𝜔d
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
5.6 Response of nth-Order Linear Systems 275

Note that the eigenvectors for the two complex conjugate eigenvalues can always be scaled as complex conjugates. The
total unforced response can then be expressed as
[ ] [ ]
1 (−𝜎+j𝜔d )t 1 (−𝜎−j𝜔d )t
x(t) = 𝛼1 e + 𝛼2 e , (5.120)
−𝜎 + j𝜔d −𝜎 − j𝜔d
where the 𝛼’s are determined by initial conditions. In order for there to be a real response, x(t), the two complex terms
which are combined in 5.120 must be complex conjugates, that is,
x = (a + jb) + (a − jb) = 2a.
Due to this fact, we can analyze the response of x due to just the first term and take twice the real part in order to obtain
the total response. The response can then be expressed as
[ ]
1 [ ]
x(t) = 𝛼1 e(−𝜎+j𝜔d )t + complex conjugate
−𝜎 + j𝜔d
{ [ ] }
1
= 2Re 𝛼1 e(−𝜎+j𝜔d )t . (5.121)
−𝜎 + j𝜔d
Now 𝛼1 and −𝜎 + j𝜔d can also be expressed in the polar form as
𝛼1 = A1 e j𝜙1 ,
[𝜔 ] √
d
−𝜎 + j𝜔d = 𝜔n e j𝜙 ; 𝜙 = tan−1 , 𝜔n =𝜎 2 + 𝜔2d ,
−𝜎
where A1 and 𝜙1 are the amplitude and phase of the complex number 𝛼1 . Equation 5.121 can then be written as
{[ ] }
1 −𝜎t j(𝜔d t+𝜙1 )
x(t) = 2Re A e e
𝜔n e𝜙 1
[ ]
2A1 e−𝜎t cos(𝜔d t + 𝜙1 )
= .
2𝜔n A1 e−𝜎t cos(𝜔d t + 𝜙1 + 𝜙)
The eigenvector, which can also be written in the form,
[ ]
1
V1 = ,
𝜔n e j𝜙
is seen to scale both the magnitude and the phase shift between the two states. The second state will be scaled √ by an
amplitude of 𝜔n relative to the first state and shifted in phase an amount 𝜙. For our particular example, 𝜎 = 1∕2, 𝜔d = 3∕2,
𝜔n = 1, and 𝜙 = 2𝜋∕3, and choosing an initial condition such that 2A1 = 1 and 𝜙1 = 0 gives
√
x1 (t) = e−t∕2 cos( 3t∕2),
√
x2 (t) = e−t∕2 cos( 3t∕2 + 2𝜋∕3),
which says that state 2 (the momentum) has the same magnitude as state 1 (the spring extension) but it leads in phase by
2𝜋∕3 radians.

Deriving the unforced response of an nth- order linear system as described can be desirable in some cases, and the use
of contemporary symbolic processors makes the process much more amenable and less error-prone. Another way to derive
the unforced response is described in Section 5.6.2.

5.6.2 Unforced Response Using Matrix Exponential

Besides eigenvalues and eigenvectors, the unforced response of a system,
x = Ax,
can also be obtained by assuming an infinite series solution of the form,
( )
x = S(t)x(0) = I + C1 + C2 t2 + · · · + Cn tn + … x(0),
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
276 5 Linear System Modeling and Analysis

where S(t) is the solution matrix and x(0) is an initial condition vector. In order for the solution matrix to give the correct
response for arbitrary initial conditions, the solution matrix must satisfy,
Ṡ = AS. (5.122)
To determine the coefficients of the solution matrix, we substitute the series expansion for S(t) into 5.122 which gives
( )
C1 + 2C2 t + 3C3 t2 + · · · + nCn tn+1 + · · · = A I + C1 t + C2 t2 + · · · + Cn tn + … .
Equating like powers of t yields
C1 = A
1 1
C2 = AC1 = A2
2! 2
1 1 3
C3 = AC2 = A
3! 2⋅3
⋮
1 1
Cn = ACn−1 = An ,
n! n!
and therefore
S(t) = I + At + A2 t2 + · · · + An tn + · · · .
This series is defined to be the matrix exponential
eAt = I + At + A2 t2 + · · · + An tn + · · · , (5.123)
which is a general solution for the unforced linear time-invariant system.
There are several methods for calculating the matrix exponential, such as truncation of the series, Cayley–Hamilton
methods, and Laplace transform techniques.7 We will introduce a technique closely related to our previous derivation for
the unforced response using eigenvalues and eigenvectors. The simplest case for the calculation of the matrix exponential
is for a diagonal matrix Λ,
⎡ s1 0⎤
Λ = ⎢ 0 ⋱ ⎥,
⎢ ⎥
⎣0 sn ⎦
which has the property that
⎡ sn1 0⎤
⎢ ⎥
Λ = ⎢ 0 ⋱ ⎥,
n

⎢ ⎥
⎣0 snn ⎦
and therefore
⎡ es1 t 0 ⎤
⎢ ⎥
eΛt =⎢ 0 ⋱ ⎥.
⎢ s t ⎥
⎣ 0 e ⎦
n

This form should not be surprising since the state equations

ẋ = Λx
are a set of n decoupled first-order differential equations. Each of which has an exponential solution.
For the more general case,
ẋ = Ax,

7 Useful references include Brogan [94], Takahashi et al. [55], and Schultz and Melsa [56].
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
5.6 Response of nth-Order Linear Systems 277

we seek a transformation T and a new set of states x∗

x = Tx∗ , x∗ = T−1 x,
Tẋ ∗ = ATx∗ ,
x∗ = T−1 ATx∗ ,
such that
⎡ s1 0⎤
⎢ ⎥
T AT = Λ = ⎢ 0 ⋱ ⎥
−1
(5.124)
⎢ ⎥
⎣0 sn ⎦
In this case, we have diagonalized the system equations by a new choice of states. Note that this new diagonal state
equation has the identical eigenvalues of the original system. This can be shown by the derivation
|sI − Λ| = |sI − T−1 AT| = |T−1 ||sI − A||T| = |sI − A| = 𝟎.
From equation 5.124, we have
AT = TΛ, (5.125)
and if we partition the n × n transformation matrix T into n n-dimensional vectors V as
[ ]
T = V1 , V2 , … , Vn ,
equation 5.125 can also be expressed as a sequence of n equations as,
AV1 = V1 s1
⋮
AVi = Vi si
⋮
AVn = Vn sn .
If we take the ith equation in this sequence and rewrite it in the form
[ ]
si I − A Vi = 𝟎, (5.126)
we see that it is identical to the eigenvalue and eigenvector problem discussed previously. This implies that the columns
of the transformation matrix T are identical to the eigenvectors found in equation 5.116. In order to compute the matrix
exponential of At, we use the fact that
A = TΛT−1
A2 = TΛT−1 TΛT−1 = TΛ2 T−1
⋮
An = TΛn T−1 .
which can be used to demonstrate that,
eAt = TeΛt T−1 . (5.127)

Example 5.9 Matrix exponential for mass–spring–damper system

Consider again the spring-mass-damper system of Example 5.7 which has an A matrix of
[ ]
0 1
A= ,
−2 −3
and a characteristic equation
| s −1 |
| |
|sI − A| = | | = s2 + 3s + 2 = 0,
|2 s+3|
| |
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
278 5 Linear System Modeling and Analysis

with eigenvalues of s1 = −2 and s2 = −1 and eigenvectors of

[ ] [ ]
1 1
V1 = , V2 = .
−2 −1
The transformation matrix T which diagonalizes the A matrix can then be represented as
[ ]
1 1
T=
−2 −1
and
[ ]
−1 −2 0
T AT = = Λ.
0 −1
The matrix exponential is then found as
[ ] [ −2t ][ ] [ −2t ]
1 1 e 0 −1 −1 −e + 2e−t −e−2t + e−t
eAt = T−1 eΛt T = = .
−2 −1 0 e−t 2 1 2e−2t − 2e−t 2e−2t − e−t
The solution for x(t) given x(0) is then
[ −2t ]
−e + 2e−t −e−2t + e−t
x(t) = x(0).
2e−2t − 2e−t 2e−2t − e−t

5.7 Forced Response of nth-Order Systems

The general response for an nth-order linear time invariant system can be constructed in a similar manner to a first-order
system discussed in Section 5.1.1. For the nth-order system,
ẋ = Ax + Bu (5.128)
with a set of given initial conditions,
x(0) = x0 .
The total solution is found by adding the unforced response and the forced responses,
x = xh + xp ,
where the unforced solution was found in Section 5.6 as,
ẋ h = Axh , xh (0) = x0 .
xh = eAt x0 .
The forced part of the response (particular solution) is found from,
ẋ p = Axp + Bu, xp (0) = 𝟎. (5.129)
The solution to 5.129 can be developed by using a vector–matrix form of variation of parameters [55]. Suppose we assume
that the solution to 5.129 can be generated from the unforced solution as,
xp = eAt p(t), (5.130)
where p(t) is yet to be determined. Substituting 5.130 into 5.129 yields,
AeAt p + eAt ṗ = AeAt p + Bu,

eAt ṗ = Bu,
ṗ = e−At Bu, (5.131)
where we have used the fact that,
[ At ]−1
e = e−At .
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
5.7 Forced Response of nth-Order Systems 279

Equation 5.131 can be integrated to solve for p(t) using the equation,
t
p(t) = e−A𝜏t Bu(𝜏)d𝜏. (5.132)
∫0
Note that p(0) = 𝟎 which insures xp (0) = 𝟎. This result can be combined with 5.130 to obtain the forced response,
t
xp = eAt e−A𝜏 Bu(𝜏)d𝜏
∫0
t
= e−A(t−𝜏) Bu(𝜏)d𝜏. (5.133)
∫0
The integral form in 5.133 is sometimes termed a matrix convolution integral. The total solution can then be formed by
adding the unforced and forced solutions,
t
x = eAt x𝟎 + e−A(t−𝜏) Bu(𝜏)d𝜏. (5.134)
∫0
The methods described in these sections enable semi-analytical solutions for nth-order solutions (i.e., formalized analyt-
ically but more practically implemented numerically), which can be needed in some applications (e.g., see Brogan [94]).
However, for most intents and purposes in modeling of physical systems, computer-based implementations of these meth-
ods provide a more practical and effective way to analyze unforced and forced linear system response. Solved Problem A-5-12
demonstrates both unforced and forced responses of a two-story system using the methods describe in this section.

5.7.1 General Step Response

Before the convolution integral in 5.134 can be evaluated, the input vector function of time u(t) must be specified. Consider
first a step response, extended to the n-order state equations as illustrate in Figure 5.24, with each Ui the magnitude of the
step inputs.
The general solution for a step input can then be found from 5.134 by first pulling out the constant Bu,
[ t ]
At −A(t−𝜏)
x = e x𝟎 + e d𝜏 Bu. (5.135)
∫0
The integral in the brackets can be evaluated in series form,
t 0
e−A(t−𝜏) d𝜏 = − e−Az dz, z = t − 𝜏
∫0 ∫t
[ 0 ]
1
I + Az + · · · + An zn + … dz
=−
∫t n!
1 2 1 n−1 n
It + At + · · · + A t + … .
2 n!
If A is nonsingular such that A−1 exists, this integral can also be expressed as,
t [ ]
eA(t−𝜏) d𝜏 = A−1 eAt − I . (5.136)
∫0
Using 5.136 the general step response can be expressed,
[ ]
x(t) = eAt x0 + A−1 eAt − I Bu. (5.137)
Assuming a stable system such that,
lim eAt → 𝟎,
t→∞

the steady-state solution for a step response can be expressed as,

xss = −A−1 Bu. (5.138)

Figure 5.24 n-order step input.

 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
280 5 Linear System Modeling and Analysis

5.7.2 Difference Equation

Using the step response solution 5.137, an exact difference equation can be generated for the propagation of x(t) at discrete
time intervals,
tm = mT, m = 0, 1, 2, … ,
where T is a fixed time interval. The discrete equation can be expressed as,
[ ]
x(mT) = eAT x((m − 1)T) + A−1 eAT − I .
Defining the matrizant M as,
M = eAT ,
and the Paynter matrix P as,
[ ]
P = A−1 eAt − I ,
the exact discrete equation can then be found from,
xm = Mxm−1 + PBu, (5.139)
where xm = x(mT).

Example 5.10 Difference equation response

Consider a scalar system with,
a = −1, b = 1, U = 1, xo = 0, and T = 0.5.
This represents the differential equation,
ẋ = −x + us (t),
to be solved at time increments of T = 0.5 with a zero initial condition. The matrizant for this system is,
M = e−aT = e−0.5 = 0.607,
and the Paynter term is,
P = −1[0.607 − 1] = 0.393.
The solution can then be propagated by iterating the difference equation,
xm = 0.607xm−1 + 0.393.
x0 = 0.
The table below shows the results, plotted to the right.

m tm xm xm
1
0 0 0.0
1 0.5 0.393
2 1.0 0.632 0.5
3 1.5 0.776
4 2.0 0.864 tm
0.5 1 1.5 2

As should be expected, the continuous and discrete solutions are identical at the discrete time intervals.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
5.8 Forced Response from Transfer Functions 281

Figure 5.25 Arbitrary input approximated as a sequence of steps.

u(t)

5.7.3 Approximate Response to Arbitrary Input

An approximate method for the response of systems to arbitrary inputs can be obtained by dividing the input function of
time into a sequence of step inputs, as shown in Figure 5.25.
Defining either,

ū m = u((m − 1)T), (m − 1)T ≤ t < mT,

or with the midpoint formula,

ū m = (u(mT) + u(m − 1)T)∕2, (m − 1)T ≤ t < mT,

then an approximate solution for the response is,

xm = Mxm−1 + PBūm . (5.140)

̇ is large, or a higher-order approximation
The value of T could be varied to better approximate u(t) in regions where u(t)
such as a point-slope approximation could be used.

5.8 Forced Response from Transfer Functions

In Section 5.4, we studied the relation between transfer functions, partial fraction expansions, and time responses. For a
linear output equation,

y = Cx + Du,

the transfer function for a single input–single output system can be expressed as,
y
= G(s) = C[sI − A]−1 B + d.
u
If we assume, for simplicity, that x(0) = 0 and d = 0, then the output as a function of time can be expressed in convolution
form,
t
y(t) = CeA(t−𝜏) Bu(𝜏)d𝜏. (5.141)
∫0
For a unit step input, the output ys reduces to,
{ [ ]
CA−1 eAt − B , t ≥ 0,
ys =
0, t < 0.
As discussed in Section 5.7, the unit impulse response g(t) can be formally defined as the time derivative of the unit step
response. The unit impulse response can then be expressed as,
{
CeA B, t ≥ 0,
g = ẏ s =
0, t < 0.
Letting g(t) be the unit impulse response of the transfer function G(s), we can then express the output response function as
t
y(t) = g(t − 𝜏)u(𝜏)d𝜏, (5.142)
∫0
where the integral is a (scalar) convolution integral.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
282 5 Linear System Modeling and Analysis

5.8.1 Geometric Interpretation of Convolution

Given the impulse response g(t) to a unit impulse excitation, the response of the system by an arbitrary input u(t) can
be obtained by convolution as shown above. This result can also be understood graphically. For this development, we
consider the arbitrary input to consist of a series of pulses as shown in Figure 5.26. If we examine one of the pulses (shown
crosshatched) at time t = 𝜏, its area is,

û = u(𝜏)Δ𝜏.

Approximating this pulse as an impulse of strength u, ̂ the contribution to the output response at time t is dependent on
the elapsed time t − 𝜏. The response at t due to a pulse at 𝜏, yI (t, 𝜏), can be expressed,

yI (t, 𝜏) = g(t − 𝜏)û = g(t − 𝜏)u(𝜏)Δ𝜏.

Since the system we are considering is linear, the principle of superposition holds. Thus, by combining all such contri-
butions, we have,
∑
n
y(t) = g(t − kΔ𝜏)u(kΔ𝜏)Δ𝜏, t = kΔ𝜏.
k=0

Taking the limit as Δ𝜏 → 0, y(t) can be represented by the convolution integral,

t
y(t) = g(t − 𝜏)u(𝜏)d𝜏.
∫0

Example 5.11 RC filter with square pulse input

Suppose that a square pulse is the input voltage Vi (t) to a low-pass RC-filter, as shown in Figure 5.27.
The transfer function between the output voltage Vo and Vi is
Vo 1
= .
Vi RCs + 1
The unit impulse response is given by

g(t) = e−t∕𝜏t ∕𝜏t , t ≥ 0,

where 𝜏t = RC is the time constant of the system. Now to find eo (t), we use
t
Vo (t) = g(t − 𝜏)u(𝜏)d𝜏.
∫0
Figure 5.28 shows the result of convolution for various values of t, and the response Vo (t) (reference [95], pp. 73–75).

Figure 5.26 Geometric interpretation of convolution.

Vi(t) R
1
Vi(t) C Vo(t)

1 t
(a) (b)

Figure 5.27 (a) Square pulse input voltage, Vi (t). (b) RC filter circuit.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
5.10 Problems 283

(a)

(b) (d)

(c)

Figure 5.28 Results of convolution for an RC filter. (a) t < 0, (b) 0 < t < 1, (c) t > 1, and (d) output voltage response.

5.9 Chapter Summary

This chapter reviews methods of analysis for linear (or linearized) time invariant physical systems. An emphasis is given
to reviewing traditional analysis of first and order systems, given their importance in practical model analysis as well as
in engineering design. The use of standard nth-order forms promotes the relation to well-established definitions of system
time constants, damping ratios, and undamped natural frequency. These system characteristic parameters play a role in
modeling, experiment design, and control system analysis and design.
For systems of order three and higher, the use of linear state-space methods and eigenvalue/eigenvector analysis becomes
essential. This chapter has reviewed how to determine the unforced response from the eigenvalue-eigenvector problem.
The matrix exponential is also reviewed and its use in the unforced and forced response of a system is demonstrated.
How these methods lead to approximations of solutions used in software programs and by difference equations is also
discussed. Lastly, a brief review of the role played by the convolution integral in forced response of a system is given.

5.10 Problems
First-order systems

Problem B-5.1 Tank with two valves The tank shown in Figure B-5.1(a) is filled by water from a source at a constant
rate, Qo , through valve 1. When the level in the tank reaches a desired height, a controller shuts off valve 1 and opens
valve 2 as in Figure B-5.1(b). Valve 2 is never open when valve 1 is open. Assume the tank has a constant area A
with capacitance C and that the flow through valve 2 can be modeled with a linear resistive element with hydraulic
resistance R2 .

Fluid Fluid
source source

Valve open Valve closed Valve closed Valve closed

(a) (b)

Figure B-5.1 (a) Tank in Mode 1 (filling): valve 1 open, valve 2 closed. (b) Tank in Mode 2
(emptying): valve 1 closed, valve 2 open.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
284 5 Linear System Modeling and Analysis

a) First-order differential equations are suggested as below to model the dynamic volume of fluid, V – , in the tank for
each case:
Mode 1 (valve 1 ON): –V̇ = Qo
Mode 2 (valve 2 ON): 𝜏 –V̇ + –V = 0
Explain how these equations are derived.
b) In terms of the given parameters, what is the system time constant, 𝜏t , when the tank has valve 2 open?
c) Assume that valve 1 has been open for some time and the tank is at level h0 . At this time t = 0, valve 1 is turned off
and valve 2 is opened. Develop an expression for the time, Δt, that it takes for the fluid level fluid level in the tank,
h, to drop from h0 to h1 .
d) Once the level reaches h1 , valve 2 is closed and valve 1 is opened again to raise the level in the tank back to h0 .
This filling/emptying process is repeated for several cycles by switching the valves on and off. Neatly sketch the
response of the tank level, h, versus time. Label the graph axes clearly and indicate any level amplitudes and/or
times of interest. Explain trends in the response as needed.

Problem B-5.2 Spring–damper design Study the two spring–damper configurations shown in Figure B-5.2, which
have linear spring and damper components and applied force, F(t)

(a) (b)

Figure B-5.2 (a) Configuration 1. (b) Configuration 2.

a) Show that the two systems have the same bond graph.
b) Apply causality to show that only one spring can be assigned integral causality. Choose xk1 as the system state.
c) Determine the system time constant.
d) One purpose for adding another spring is to increase the amount of energy that can be stored within this system.
Compare the power dissipated in the damper when k1 < k2 .

Problem B-5.3 Hydrostatic drive The hydrostatic drive shown in Figure B-5.3(a) is driven by a torque source, 𝜏(t).
The hydraulic pump and motor are assumed to be positive-displacement (PDM) hydraulic machines with displace-
ments of dp and dm , respectively, where pump flow, Qp , is related to pump shaft speed, 𝜔p , by the relation, Qp = dp 𝜔p ,
and similarly for the motor Qm = dm 𝜔m . The PDM model in Figure B-5.3(b) can be used to represent both devices.

Hydraulic Hydraulic
pump motor

Sump B

(a) (b)

Figure B-5.3 Hydraulic drive system.

a) What are the SI units for the ‘displacement’ (the transformer modulus)? Show units consistency for both transformer
relations.
b) Develop a bond graph of the system in Figure B-5.3(a) that incorporates the PDM models and other basic elements to
form a final model of the first-order form. Assumptions need to be made to simplify the system. In particular, show
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
5.10 Problems 285

how losses can combine so that total (or net) losses in the hydraulic components (mechanical and hydraulic) can
be lumped into a single hydraulic resistive element as shown in (b) using a linear pressure-flow relation, Pl = Rl Ql .
On the mechanical load, consider only the rotational elements with rotational inertia, J, and damping, B. Discuss
other assumptions implied.
c) Derive a first-order form in standard form and identify the time constant 𝜏 in terms of given system parameters.
Show unit consistency; that is, show that the time constant has units of seconds by considering the proper units of
the parameters defining 𝜏t .

Problem B-5.4 Thermocouple between pipe and insulation A thermocouple is mounted as shown on the outside of a
metal steam pipe but inside the insulation as shown in Figure B-5.4(a). The pipe is in a room with a temperature, Ta ,
and the steam temperature varies in a known manner, Ts .
a) Develop a bond graph model of this system assuming that the metal should be modeled with a RC T-model as shown
in Figure B-5.4(b) while the insulation can be treated as a pure thermal resistor. Indicate all ideal source elements.
Assume the thermocouple is an ideal temperature sensor (i.e., no dynamics).
b) Assign causality, identify state variables, and derive state equations from the bond graph developed in part (a) above,
then write an output equation for temperature at the thermocouple location.
c) Assume a mild steep pipe with a wall thickness of 0.5 inches and a polystyrene insulation is 4 inch thick. Find
physical data and estimate values for all model parameters.
d) If the steam is initially at 80 ∘ F and undergoes a step change to 280 ∘ F. Determine the response at the thermocouple
location.
e) Assume all that is known is that the thermocouple reaches a final (steady-state) value of 230 ∘ F after a step input of
280 ∘ F. Is it possible to formulate a model and solve for the transient time response of the thermocouple.

Thermocouple

Steam
Ts = 280 F Metal
C
Insulation Steam
Ta = 80 F

(a) (b)

Figure B-5.4 (a) Model of heated steam pipe and insulation with thermocouple
measurement. (b) T-junction model.

Second-order systems

Problem B-5.5 Elevator lift revisited Consider the elevator lift scenario in Solved Problem A-5-7.
The fully loaded elevator is now at rest at the first floor and the cable is given a step input at the upper end of 𝑣c .
Determine an expression for the maximum extension (or stretch) that the cable will experience if the value of 𝑣c is that
found in Solved Problem A-5-7.

Problem B-5.6 Torque meter A torque meter is made from a circular shaft that is rigidly mounted at one end and has
strain gauges attached to it as in Figure B-5.6(a). A static calibration test with a known constant torque 𝜏1 provides an
estimate for a spring constant of K of this shaft.
The intent is to use this torque sensor to measure an oscillating, time-varying torque, 𝜏(t). To damp or “filter” out some
high-frequency components in the torque, a change to the system has been proposed as shown in Figure B-5.6(b). This
system design needs to be evaluated so you will develop two models.
a) First, determine an expression for K in terms of shaft material and geometry properties.
b) For low-frequency (quasi-static) operation, the rotational inertia of the gears and shafts can be neglected. Build a
bond graph model for the system in (b) that only includes B and K. Rather than using unity gear ratios, assume
that you have the option to select different gear ratios for coupling to the damper (nB ) and the torque shaft (nK )
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
286 5 Linear System Modeling and Analysis

components. Derive a mathematical model for this system in the form of a single standard first-order differential
equation and show how the system time constant, 𝜏t , is related to B, K, and the gear ratios nB and nK .
c) To assess the system at higher frequencies, the rotational inertia of the gears and shafts needs to be included.
Begin by composing an effective rotational inertia for all rotating components (input shaft, gears), and call this Jeff .
Then develop a bond graph model for this system. Derive a mathematical model for this system, first by deriving
the state-space form and then by converting this to a single standard second-order differential equation in terms of
the shaft angular displacement, 𝜃K . From the second-order form, derive relations for the system damping ratio 𝜁,
and undamped natural frequency, 𝜔n .
d) For the two models developed in (b) and (c), derive transfer functions between the angular twist of the sensing shaft,
𝜃k , and the input torque 𝜏(t).

B
Unknown torque

Geartrain with
Strain gages 1:1 gear ratios, K
negligible inertia
(a) (b)

Figure B-5.6 (a) Torsion rod with strain gauges and (b)
proposed sensing configuration with damper.

Problem B-5.7 Beam-mass vibration A mass m is attached to the end of a cantilever beam as shown in Figure B-5.7.
The beam is to be modeled as a simple spring. If it takes about ts seconds for the beam to stop vibrating after it is slightly
displaced at the tip and released from rest, how would you estimate the damping coefficient if given the effective beam
stiffness, k. Briefly provide the reasoning for estimating damping ratio, 𝜁. For values of m = 10 kg and k = 1000 N/m,
determine a value for 𝜁.

Figure B-5.7 Mass on tip of a beam.

Problem B-5.8 Weighted spar in the ocean A weighted spar design as shown in Figure B-5.8(a) is a common base
for offshore drilling platforms. To study this concept and to guide development of a dynamic model, a scaled physical
model has been built in the lab.

Platform 2
Displacement, m

1.5
Spar Spar Water
Force, N

20 1

10 0.5
Weighted
end 0
1 2 0 4 8 12
Displacement, m Time, sec
Spar platform Model
concept
(a) (b)

Figure B-5.8 (a) Weighted spar. (b) Test results.

 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
5.10 Problems 287

The lab spar has been made and two tests conducted. About 1/10 of its length is above the water level. The spar will
be used in the first test, and the spar was pushed down from its free-floating (equilibrium) position. The force needed
to do this was measured as a function of the position. In a second test, the spar was pushed down to a known position
and then released and allowed to vibrate freely in the vertical direction. The measured test results are summarized in
Figure B-5.8(b).
a) Develop a set of state equations to model the vertical motion of the spar.
b) Use the tests and the resulting measurements to estimate any model parameters.
c) The oscillatory response indicates the system is underdamped. Use your model to determine the system poles and
plot them on the real-imaginary axis.
d) Estimate the damping ratio 𝜁 and undamped natural frequency 𝜔n .

Problem B-5.9 Read/write head positioning The system below is a simplified model of a positioning mechanism for
a read/write head on a disk drive. Assume only those elements for which parameters are indicated are to be included
in the model and that all constitutive relations are linear. Assume effective rotational inertia of the link and r/w head
is J and the effective stiffness about the pivot is K. Assume the bearing resistance at the pivot is also linear with a
parameter, B.
1) Develop the second-order differential equation that governs the position, 𝜃. You may use a bond graph or any other
method to get the result. You must clearly show the steps to its development.
2) What is the expression for the natural frequency, 𝜔n ?
3) What is the expression for the damping coefficient, 𝜁?
4) In your equation, what is the standardized input u(t) and what are its units?
5) Briefly give two factors that would influence your choice for a design 𝜁.
6) What value of 𝜁 would you choose to achieve minimum overshoot with the fastest response time?
7) For an underdamped system, choose a k and b for a 𝜁 = 0.9 given that J = 10 g cm2 , and if we want a settling time
of t95 = 3𝜏e = 3∕(𝜁𝜔n ) = 0.01 sec.
8) For the underdamped case (0 < 𝜁 < 1), derive the response of a standard second-order system to a step input of
magnitude us .

Problem B-5.10 Motor-belt-fan dynamics A fan is driven by a constant velocity motor through a rubber belt as illus-
trated in Figure B-5.10(a). The fan blades, rotor, and shaft can be modeled as a lumped inertia, J, mounted on rotational
bearings. Assume that total losses of the fan (bearings and fluid damping) are to be modeled with linear damping B,
and the belt is elastic with total effective stiffness K. The system can be modeled by the bond in Figure B-5.10(b).

C : 1/K I:J

τK θ̇
K
ḣJ ωJ
Fan
Belt τ
Sf 0 1
ωm(t)
τB ω B
Pulley
Motor
R:B
(a) (b)

Figure B-5.10 (a) Motor-belt-fan system and (b) bond graph.

We want to study the dynamic behavior of the system when it is initially at rest and the motor is turned on. Assume
the nominal parameter values are: J = 0.39 kg-m2 , K = 3.86 N-m/rad, and B = 0.245 N-m-s/rad
a) For a step input in motor speed, 𝜔m = 3.5 rad/sec, find the maximum value of the speed of the fan, 𝜔J .
b) At what time after the step is applied will a peak speed value occur?
c) Find the steady-state value of the torque in the spring
d) It is desired that the maximum value of 𝜔J not exceed the motor speed by more than 50%. Find the smallest value
of B for which this is the case.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
288 5 Linear System Modeling and Analysis

e) When the fan has reached its steady-state speed, the motor is turned off and the bond graph of the system simply
becomes (assume no losses in the motor shaft):

ḣJ τB
J :I ωJ 1 ωB R:B

Using the same parameter values as given above, estimate how long it will take for the fan to come approximately to
rest?

Problem B-5.11 Field-controlled dc motor with load The armature circuit of the dc motor shown in Figure B-5.11 is
driven by a known current source, is (t). The motor operation is controlled via a separately-excited configuration using
the field circuit which uses an input voltage, Vf (t), to induced field current, if . The field circuit coils have effective
inductance Lf and resistance Rf . Assume that the induced motor torque is then 𝜏m = r(if )im , where r(if ) indicates that
the motor torque constant is a function of the field current.
a) Develop a bond graph of a model for this system that reflects the assumptions given. Be sure to indicate the modu-
lation of the motor torque constant by if .
b) Apply causality to justify selection of the state variables as the field current and the load inertia angular speed 𝜔.
Derive the state equations.
c) Convert the state equations into a single second-order ODE in terms of 𝜔.
d) Determine relations for the overall system undamped natural frequency and the damping ratio.

Excitation on
field circuit

B
Armature
circuit

Figure B-5.11 Field-excited DC motor with load.

Problem B-5.12 Switched circuit load response The electrical circuit in Figure B-5.12 has switch configuration where
SW 1 is initially closed with SW 2 open to create a steady-state current in the LR circuit. At time t = 0, SW1 is opened
and SW 2 is closed.
SW 1 SW 2

Vb
C

Figure B-5.12 Switched LR circuit with

CR load.

a) Write an expression for the time constant of the LR circuit.

b) Write an expression for the steady-state current in the LR circuit assuming that SW 1 has been closed for a long
time?
c) Model the circuit that is formed when SW 1 opens and SW 2 closes by formulating a bond graph, applying causality
and showing that there are two states: 𝜆L of the inductor and qC of the capacitor.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
5.10 Problems 289

d) Write the state equations for the two states.

e) Write a second-order linear ODE in terms of the flux linkage, 𝜆L .
f) Write a second-order linear ODE in terms of the voltage across the capacitor, 𝑣C = qC ∕C.
g) Confirm that the expressions for the damping ratio, 𝜁 and the undamped natural frequency, 𝜔n , for the equations
in (e) and (f) are the same.
h) For the equation from (f), determine the two initial conditions: 𝑣C (0) and 𝑣̇ C (0).
i) Write an expression for the solution of 𝑣C (t) given the initial conditions in (g).
j) Assume Vb = 12 volts and L = 1H. Choose an R value so the steady-state current is 1 amp.
k) For a value C = 0.1 𝜇F, determine a RL so that 𝜁 = 0.707.
l) Compute the undamped natural frequency, 𝜔n , given values in (i) and (j).
m) For parameters values from (i), (j), and (k), plot the response 𝑣C (t) over three damped periods.
n) Plot the current in the load resistor, iRL ?
o) What is the value of the peak power dissipation in RL ?

Problem B-5.13 CRL circuit model, poles, and response The circuit shown in Figure B-5.13 has a current source, is (t),
with a shunt capacitor C and an RL load.
C : 1/K I:J
R
τK θ̇
K
ḣJ ωJ

τ
Sf 0 1
L ωm(t)
C
τB ω B

R:B
(a) (b)

Figure B-5.13 (a) Current driven CRL circuit and (b) bond graph.

a) Use the bond graph model in Figure B-5.13(b) to show that the charge qC on the capacitor and inductor current iL
(= 𝜆∕L) are states for this system, and write state equations.
b) Convert the state equations into standard (linear) second-order form in terms of iL and identify the system
undamped natural frequency and damping ratio in terms of system parameters.
c) If the system is given a step input, for what range of R would the system oscillate?
d) Find the transfer function iL ∕is
e) Two different sets of (C, R, L) parameters define Systems 1 and 2 and lead to the complex conjugate poles indicated
below.
jω
System 2
3
2
System 1
1

σ
−6 −5 −4 −3 −2 −1 1 2 3 4
−1
System 1
−2
−3
System 2

In response to a step input, which system: (i) oscillates more rapidly?, (ii) has a larger overshoot, and (iii) damps out
more quickly?
(f) Draw a detailed block diagram of this system showing iL as an output and is as an input.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
290 5 Linear System Modeling and Analysis

Problem B-5.14 Analog voltage meter modeling and response characteristics Refer to the modeling of the analog meter
in Problem B-3-32 which calls for developing a bond graph model and derivation of state equation for an analog voltage
meter.
a) In contrast to Problem B-3-32, neglect the inductance, Lm , develop a new bond graph, and derive two state equations
in ABCD for needle angle 𝜃 and angular velocity, 𝜔.
b) Derive a transfer function, 𝜃∕Vin .
c) Determine the damping 𝜁 and undamped natural frequency 𝜔n for the following parameter values: Rm =
15 085 ohm, rm = 0.03 N m/A, Jm = 2 × 10−7 kg m2 , Km = 200 × 10−6 N m/rad, and Bm = 12 × 10−7 N m s/rad.
d) Compute and plot the step response for an input voltage of 2.5 volts and determine the steady-state angular
deflection.
e) How well does Equation 5.86 predict the overshoot observed in computed step response?
f) What is the maximum step input (voltage) that will achieve a peak response of 90 ∘ , at what time after the step does
this peak occur (Tp ), and what is the steady-state value of the meter needle for this case?

Problem B-5.15 Pneumatic joint actuator A pneumatic design is proposed to actuate a joint, and a test configuration
will be built as shown in Figure B-5.15(a). An ideal source of air flow, Qs (t), is controlled into one of the bellows. Assume
the air is incompressible and that a change in the contained volume in each bellows, V – B , can be related to the pressure
by, PB = –VB ∕CB , where CB is a fluid capacitance. Each bellows also has some mechanical stiffness, kB , as indicated in
Figure B-5.15(b). The area of each bellows is AB , and the distance from the rotational pivot to each bellows centerline
is a. Assume a mechanical effector with rotational inertia J about the fixed pivot, and some linear rotational damping
B. Assume that the load at the tip of the effector is zero, FL (t) = 0, for the purposes of this problem.

θ
FL(t)

J, B
Bellows fluid Bellows mechanical
Bellows area, AB
capacitance, CB stiffness, kB

Qs(t) Closed
Rh a a a a
valve
(a) (b)

Figure B-5.15 (a) A pneumatic bellows actuator test setup. (b) Bellows actuator with angular motion.

a) Develop a bond graph that accounts for energy storage in each bellows (fluid and mechanical) as the rotational
inertia of the link, J. Consider one bellows is closed (as shown), but air can be pumped into the other. Apply causality
and verify that there are two states: volume change of the bellows, V – B , and momentum (or angular velocity) of the
“link” that has inertia, J. Assume small angular motion of the link by angle 𝜃.
b) Use your bond graph to develop state equations.
c) Derive a single second-order ODE in standard form in terms of the angular velocity of the link about the joint
pivot, 𝜔.
d) Write expressions for the damping ratio, 𝜁, the undamped natural frequency, 𝜔n , and for the standardized input,
u(t).
e) Design case study. It is desired to move 𝜃 from a zero position to 10∘ in the shortest time possible. Assume a prototype
end effector is to be made with L = 2 inches, a = 1 inch, using stock aluminum rod with a square cross-section of
1/4 inch. Using the ideal (linear) model relations, size a bellows so you can achieve this motion with a damping
ratio 𝜁 = 0.707 and an undamped natural frequency of 2𝜋(1) Hz. Rationalize initial estimates for any unspecified
parameters needed (e.g., bellows mechanical stiffness, damping, and flow resistances), or neglect if reasonable.
Sizing of the bellows will be related to estimates of CB .
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
5.10 Problems 291

Problem B-5.16 Bucking bronco Two dc machines are connected as shown in Figure B-5.16. Machine 1 is a series dc
generator (meaning it has a field that is in series with the electromechanical conversion as shown). This generator is
driven by a constant speed source, 𝜔g . Machine 2 is a dc motor that has a fixed field powered by a constant current If ,
so that motor 2 has a constant torque constant, rm . Demonstration systems used to be built in this way to show how an
oscillator could be built.

Field
circuit

Generator

1 2
constant Motor
shaft

Figure B-5.16 A “bucking bronco” circuit.

Assume the generator has inductance Lg and resistance Rg , and the generator voltage is given by Vg = rg 𝜔g . The motor
shaft/rotor has inertia, Jm , and damping constant is Bm , but there is no other load. The motor shaft rotates at angular
velocity 𝜔m .
a) Develop a bond graph model, apply causality, and show that you have two states: 𝜆g (generator flux linkage) and
hm = Jm 𝜔m (motor angular momentum, or 𝜔m ).
b) Derive state equations for states 𝜆g and 𝜔m .
c) Derive a second-order ODE in terms of the angular velocity of the motor shaft, 𝜔m .
d) From the result of (c), identify the undamped natural frequency and the damping ratio in terms of the given system
parameters
e) This system was used in the mid-1900s as a demonstration of an oscillator. The motor shaft could be “perturbed”
(given an initial velocity) and then it would oscillate indefinitely, earning the name “bucking bronco.” Explain how
this can be done by making rg = 𝛼 ⋅ ig , where 𝛼 is a constant and ig is the current in the generator, ig = 𝜆g ∕Lg .
f) Assume we built a bucking bronco with Rg = 0.1 Ω; L= 0.01 H, B = 1 N-m-s/rad, J = 10 kg-m2 , rm = 1 N-m/A, and
𝜔g = 10 rad/s. Using your result from (e), what would you predict for the amplitude (in rad/s) and frequency of
oscillation (in Hz) of the response of 𝜔m , assuming it was given an initial velocity of 𝜔mo ?

Transfer functions, nth-order systems, and system response

Problem B-5.17 State space, transfer function, and step response A linear motor-driven rotational system results in
the state-space system,
[ ] [ ][ ] [ ]
𝜆̇ −1 −1 𝜆 1
= + Vs ,
ḣ 1 0 h 0
where 𝜆 is the motor inductance flux linkage, h is the motor rotor angular momentum, and Vs (t) is the input voltage.
Assume the motor rotor rotational speed is 𝜔 = h∕J and consider J = 1.
a) Determine a transfer function, G(s) = 𝜔∕Vs .
b) Sketch the response of 𝜔 for unit step input in Vs (t). Compare to the results from using a linear system simulation
of the transfer function form (e.g., step() in MATLAB or Python SciPy).

Problem B-5.18 Transfer function and step response A transfer function is given as,
s2 + 2s + 6
G(s) = .
(s + 2)(s2 + s + 4)

a) Obtain a partial fraction expansion.

b) Sketch the response y(t) if u(t) is a unit step input.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
292 5 Linear System Modeling and Analysis

Problem B-5.19 Response from poles Consider the two systems represented by the poles in Figure B-5.19. Sketch and/or
plot the response of each system to a unit step input.

jω
2

System 1
1

System 2
−4 −3 −2 −1 1 2σ

−1
System1

−2

Figure B-5.19 Two systems on imaginary plane.

Problem B-5.20 Fluid clutch response Consider the simplified model of a fluid clutch shown in Figure B-5.20, which
has a drive torque 𝜏d (t) applied onto inertia J1 and a known load torque 𝜏L (t) on inertia J2 . The two inertias interact via
shear forces induced in the coupling with an effective rotational damping coefficient B.
a) Develop a bond graph, apply causality, and indicate the state variable(s).
b) Derive state equations for the state variable(s).
c) Determine the transfer function 𝜔2 ∕𝜏d , assume 𝜏L = 0.
d) For the case where J1 = J2 = 1 kg m2 , and B = 1 N s/rad, obtain a partial fraction expansion of 𝜔2 ∕𝜏d .
e) Sketch or plot the response of 𝜔2 when 𝜏d is a unit step.

Figure B-5.20 Fluid clutch

formed by two rotational inertias
coupled by linear damping.

Problem B-5.21 Two-car train For the two-car train system in Figure B-5.21, complete the following:
a) Develop a bond graph and derive state equations.
b) Write the equations in linear state-space form.
c) Develop a transfer function relating 𝑣1 to input force Fc , 𝑣1 ∕Fc .

Figure B-5.21 A two-car train

system with driving force, Fc (t),
and load force, Fd (t).
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
5.10 Problems 293

Problem B-5.22 Two-disk spring coupled For the two-disk, spring-coupled system in Figure B-5.22, complete the
following:
a) Develop a bond graph and derive state equations.
b) Write the equations in linear state-space form.
c) Develop a transfer function relating 𝜔1 to input torque 𝜏c , 𝜔1 ∕𝜏c .

Figure B-5.22 A two-

disk, spring-coupled
system with driving
torque, 𝜏c (t), and load
torque, 𝜏d (t).

Problem B-5.23 Two-tank with pipe For the two-tank system in Figure B-5.23, complete the following:
a) Develop a bond graph and derive state equations.
b) Write the equations in linear state-space form.
V1 ∕Qc .
– 1 to pump flow Qc , –
c) Develop a transfer function relating V

Pump
Long pipe

Figure B-5.23 A two-tank system with controlled pumping,

Qc (t), and disturbance inflow, Qd (t).

Problem B-5.24 PMDC motor with inertial load The pmdc motor in Figure B-5.24 has ideal EM torque 𝜏m = rm im , where
rm is the motor torque constant. The motor has parameters Rm , Lm , Bm , and Jm . The motor is directly coupled via gears
to an attached rotational inertia load, J and B.
a) Develop a bond graph model and identify states.
b) Derive state equations.
c) Develop a transfer function that relates the angular velocity, 𝜔2 , to the input voltage, Vs (t).
d) From the characteristic equation of the transfer function, identify the system damping ratio and undamped natural
frequency.

Gear 1,

Gear 2,

Figure B-5.24 PMDC motor with inertial load.

 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
294 5 Linear System Modeling and Analysis

Problem B-5.25 Rocket pressure detection A rocket propulsion experimentalist wants to convert dynamic pressure
from a rocket motor combustion chamber directly into an analog voltage. To do so, the scheme illustrated in
Figure B-5.25 is proposed where a pressure tap on the combustion chamber wall connects to a fluid-filled pressure
line. The fluid is essentially incompressible and has density 𝜌 and dynamic viscosity 𝜇. The long fluid line, meant to
isolate a pressure detection mechanism from the high temperatures of the combustion chamber. The fluid line feeds
into a bellows which moves a mass that is connected to the slider on variable-resistor. The mass/slider position, xm ,
modulates the variable sensing resistance, R(xm ). The variable resistor (linear potentiometer) is part of a voltage-divider
circuit that outputs analog voltage, VP = 𝛽xm , where 𝛽 is a constant related to the potentiometer characteristics.

Rocket motor
combustion chamber
High T
pressure, Pr
Area, Ap vm, xm
Fluid density, ρ VP
viscosity, μ
Fluid
Long fluid line Lb
length L, diameter D R Potentiometric
circuit
Bellows spring Plat
constant, kb mass, mp
Vb Battery

Figure B-5.25 Scheme for converting dynamic combustion pressure, Pr , into analog voltage, VP .

a) Develop a model of this system considering the chamber pressure as an input to fluid line. Account for all
energy-storing and dissipative effects in the system starting with the fluid line. Assume negligible damping in the
slider contact. Build a bond graph and assign causality.
b) Derive the system state equations, including an output equation for the voltage y = VP .
c) For the elements included in the model, determine the parameter values considering the following geometric and
material properties:
Fluid: 𝜌 = 800 kg/m3 , 𝜇 = 0.1172 Pa-sec
Fluid line: L = 0.75 m, D = 2.54 mm
Bellows: kb = 875.6 N/m, Lb = 15.2 mm
Plate: mp = 0.0454 kg, Ap = 12.9 cm2
Sensing constant: 𝛽 = 4 V/mm
d) Assume the system has been at a steady state with Pr = 3.5 MPa and the pressure suddenly increases by 50%.
Solve for the response of the output voltage VP (t), including for additional insight the response of all key state
variables.
e) Find the transfer function, VP ∕Pr , in terms of the system parameters (i.e., do not plug in numerical values).
f) Use the results from studying this problem to comment on the design of the system, and especially on the role
played by the fluid inertia and resistance in the fluid line and the mechanical properties of the bellows (stiffness)
and plate. Use either the state-space model or the transfer function and any associated analysis to determine whether
the design can be improved.

Problem B-5.26 Modeling and response of ship powertrain A simplified model for an engine-driven ship power trans-
mission system is shown in Figure B-5.26.
a) Assume the engine is an ideal source of rotational speed, 𝜔e (t). Develop a bond graph model and show that the
system states can be taken as the propeller speed, 𝜔J , and the shaft angular displacement, 𝜃K . The input speed,
𝜔e (t), and the load torque, 𝜏L , are inputs to the system.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
5.10 Problems 295

Propeller
inertia
Diesel
engine
Fluid Drive shaft
coupling stiffness Friction
drag

Figure B-5.26 Ship powertrain schematic.

b) Derive the system state equations.

c) It is found that when the system is running at steady state (a stationary equilibrium), the following conditions hold:
𝜔Jo = 0.95𝜔eo , the load torque is 𝜏Lo = BL 𝜔Jo , and B2 = B1 ∕20. Use these conditions to solve for B1 and B2 in terms
of BL .
d) After the engine has been running steadily under the conditions in (c), the speed is suddenly dropped at t = 0 to
one half its initial value. Find the response of 𝜔J to this change in the input 𝜔e (t), assuming the shaft stiffness K is
such that the system damping is 0.5. For this case, let the load torque be modeled as an additional linear damping
element, 𝜏L = BL 𝜔J during this transient. Simulate this system using either an ode solver or lsim() routine.

Problem B-5.27 1/4-car suspension modes A commonly used model for studying vibration of ground vehicle suspen-
sions is the 1/4-car model shown in Figure B-5.27. The base-excitation is modeled as an input velocity, 𝑣g (t), which
depends on the ground profile, zg (x), and vehicle forward speed, 𝑣x . For a given zg (x), one can determine 𝑣g (t) =
(dzg (x)∕dx) ⋅ 𝑣x , where 𝑣x = dx∕dt is the vehicle forward speed.
a) Develop a bond graph model assuming a known 𝑣g (t) input and show that there are four energetic states: zt (tire
deflection), 𝑣us = pus ∕mus (unsprung mass velocity), zs (suspension deflection), and 𝑣s = ps ∕ms (sprung mass veloc-
ity). Assume the model is already at equilibrium with gravity, so the deflections are relative to equilibrium values.
Thus, no effect of gravity need be considered in deriving the linear state equations.
b) Derive the A and B matrices.
c) Neglect damping in the system and derive expressions for the system undamped natural frequencies.
d) It is common that passenger vehicles have a sprung mass, ms , that is, 10 times larger than the unsprung mass, mus ,
while the suspension stiffness, ks , is 10 times lower than the tire stiffness,
√ ks . Show that for this case you can define
an undamped natural frequency for the sprung mass of 𝜔n−s = 2𝜋 kRR ∕ms , where kRR = ks kt ∕(ks + kt ) is defined
√
as the ride rate, and also an undamped natural frequency for the unsprung mass, 𝜔n−us = 2𝜋 (ks + kt )∕mus .
e) Determine 𝜔n−s and 𝜔n−us when ms = 1814 kg, mus = 181 kg, ks = 88 kN/m, and kt = 704 kN/m.
f) Compare the approximate values of the undamped natural frequencies from (e) to those for the same parameter
values as determine by solving the eigenvalue problem with the A matrix from (b) (neglecting damping parameters).

Figure B-5.27 Quarter-car

suspension model schematic.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
296 5 Linear System Modeling and Analysis

Problem B-5.28 Three tank system modes Each of the three tanks in Figure B-5.28 receives flow at specified rates, and
the contained volumes induce flow through the interconnecting pipes and leakages.
a) Develop a bond graph and show that the three (linear) state equations for tank volume deviations about their equi-
librium (or reference) values, (–V1 , –V2 , –V3 ), are given by:

̃̇ ⎤ ⎡ a a a ⎤ ⎡ –V
⎡ –V ̃ ⎤ ⎡ Q (t) ⎤
⎢ ̇ 1 ⎥ ⎢ 11 12 13 ⎥ ⎢ 1 ⎥ ⎢ 1 ⎥
̃ 2 ⎥ = a21 a22 a23
⎢ –V ̃
– 2 + Q2 (t) ,
V
⎢ –V
̃ ̃ 3 ⎥⎦ ⎢⎣ Q3 (t) ⎥⎦
̇ ⎥ ⎢⎣ a31 a32 a33 ⎥⎦ ⎢⎣ –V
⎣ 3⎦
where a11 = −(1∕R12 + 1∕R1 )∕C1 , a12 = 1∕(R12 C2 ), a13 = 0, a21 = 1∕(R12 C1 )4, a22 = −(1∕R12 + 1∕R2 + 1∕R23 )∕C2 ,
a23 = 1∕(R23 C3 ),a31 = 0, a32 = 1∕(R23 C3 ), a3 = −(1∕R23 + 1∕R3 )∕C3
b) Compute the eigenvalues and eigenvectors for this system with the following parameters:
tank areas: A1 = 0.029 m3 , A2 = 0.116 m3 , A3 = 0.058 m3 ,
flow resistances: R1 = R2 = 1.0132 × 107 Pa/(m3 /s), R3 = R1 ∕2 Pa/(m3 /s), R12 = R1 , R23 = R3 .
c) Demonstrate how a specific mode response can be excited by giving this system initial conditions prescribed by the
system eigenvectors, as suggested by equation 5.119.

Reference
level

Figure B-5.28 Three tanks with inlet flows and leakage flows.

Problem B-5.29 Two-story mode response Study Solved Problem A-5-12 and show how the system can be made to
vibrate in one of the two modes by prescribing initial conditions on the states specified by the system eigenvectors.
This method is suggested by equation 5.119.

Problem B-5.30 Three reservoirs in a delta connection Figure B-5.30 shows three reservoirs in a fluid distribution sys-
tem connected in a delta form. Reservoir 1 has input flow that can supply or remove fluid at a controlled flowrate,
Q1 (t).
a) Develop a bond graph for this system.
b) Derive the state equations and assume linear resistance to the flow between each tank, R, and all tanks have capac-
itance C.
c) Find the eigenvalues and eigenvectors for this system assuming C = 1 and R = 1.
d) At time t = 0, a step flow of 1 m3 /s is introduced at Q1 (t). Solve for the response of three volumes. Assume the
volumes are initially at reference (or equilibrium values).

Figure B-5.30 Three fluid reservoirs in

Δ-connection. The connecting pipes are
labeled A, B, and C, with a flow rate
Q1 (t) specified into tank 1.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
5.10 Problems 297

Problem B-5.31 Forced shaker table dynamics The electromechanical shaker system shown in Figure B-5.31 is an
extended version of that in worked Problem A-5-11.
In this case, the coil mass and table mass interaction dynamics are to be considered, and as shown there is also damping
in the moving elements. It is desired to evaluate the response dynamics of this system, in particular to assess how the
table responds given a triangular input voltage pulse.
a) Build a bond graph of this system and use causality to identify the five states as: 𝜆, for inductor flux linkage, 𝑣c , the
velocity of the coil mass, xtc , the deflection of the coil-table coupling spring, 𝑣t the table velocity, and xs , the flexure
spring/stiffness deflection.
b) Derive the five state equations and write in linear ABCD state-space form. Write an output equation with three
outputs: table deflection, xt , table velocity, 𝑣t , and table acceleration, at .
c) Solve for the output response assuming a triangular pulse voltage input, Vs (t), that has the following characteristics:
t = 0, 0.02 sec, Vpeak = 1 V at 0.01 seconds
d) Change the voltage input to be a sinusoidal voltage. Build a table using your model with the following columns:
input voltage amplitude, forcing frequency, peak table amplitude, peak table velocity, and peak table acceleration.
Fill in the table response values when the input voltage has the following amplitude/frequency pairs: (1.2 V, 50 Hz),
(2.4 V, 50 Hz), (1.2 V, 100 Hz), (2.4 V, 100 Hz).

Ground

Permanent-magnet

Ground

Moving coil

Figure B-5.31 Electromechanical shaker with coil-table interaction dynamics.

For the system analysis, use the following parameter values:

Rc = 3 Ω, Lc = 0.0012 H, rc = 190 N m/A (electromechanical constant), mc = 1.815 kg, ktc = 8.16e8 N/m, btc = 3850
N s/m, mt = 6.12 kg, ks = 6.3e5 N/m bs = 1120 N s/m.

Problem B-5.32 1/2-car suspension modes Study the 1/2-car vehicle suspension model posed in Problem B-4-45.
a) Consider the unforced case (i.e., vehicle is sitting at equilibrium) and present the full linear state-space model.
b) For the parameters given in Problem B-4-45, solve for the eigenvalues and eigenvectors.
c) The vehicle body has two key vibrational “modes” of interest: heave, which corresponds to the vertical motion (z-axis
direction) of the center of gravity (CG), and pitch, which corresponds to the angular motion about the CG (about
y-axis). Conduct two separate simulations in which each of these modes is excited by imposing an appropriate initial
condition. Plot all the system states for each case.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
299

Frequency Response and Impedance-Based Modeling

The focus of this chapter is on expanding methods that can be used for developing transfer function models of physical
systems and on reviewing how they provide a way to analyze systems in the frequency domain. Chapters 4 and 5 provide
an introduction to the concept of system transfer functions, which relate system outputs to inputs. It was shown how these
can be derived from system models in the form of ODEs. This chapter first reviews frequency response functions, which take
advantage of s-domain representations. By letting s = j𝜔, in a transfer function, a frequency response function as may be
required in many practical problems can be derived. We also introduce linear impedance relations for bond graph elements.
Impedance functions are defined by the ratio of effort to flow at a system port, Z(s) = e∕f , thus naturally fitting within a
bond graph context. Since impedance relations are widely used in many fields, they allow bond graph methods to be applied
to systems and problems in a compatible fashion.
Finally, the use of impedance relations and related forms provide a basis for using two-port model representations, which
can be very helpful in some system modeling applications. We introduce transmission matrices and show how they can be
used within bond graph models. Two-port modeling provides an efficient means for modeling distributed-parameter effects
in systems using impedance-based formulations.

6.1 Frequency Response

An understanding of oscillatory or frequency response, both forced and unforced, is of great use in system analysis as well
as in system identification.1 Oscillatory or frequency response techniques are used extensively in many areas, including
vibration control, signal conditioning, and control system design. Frequency response refers to the characterization of a
system’s response in the frequency domain, such as to predict response to basic harmonic or sinusoidal input, as well
as more general types of inputs. Systems that oscillate naturally (or unforced) due to initial conditions are discussed in
Chapter 5.

6.1.1 Frequency Response

The frequency response of a system can be obtained in many ways. Consider the transfer function form of a system,
y b sm + · · · + b 0
= G(s) = m n ,
u an s + · · · + a0
which also represents the nth -order differential equation,
an y(n) + · · · + a0 y = bm u(m) + · · · + b0 u.
The frequency response of this system refers to the steady-state solution2 to the differential equation if u(t) is a sinusoid of
given amplitude and frequency. A sinusoidal input can be expressed in a variety of forms. It is convenient to express the
sinusoidal input as a special case of an exponential input,
[ ]
u(t) = ℜ Ue st s=j𝜔 , (6.1)

1 As will be discussed in Section 6.4.3, system identification refers to the process of determining a suitable model for a system based on measured
input and output response data.
2 Assuming an asymptotically stable system, that is, all eigenvalues have negative real parts.

Modeling of Physical Systems: Simulation and Control, First Edition. Raul G. Longoria and Joseph J. Beaman.
© 2025 John Wiley & Sons Ltd. Published 2025 by John Wiley & Sons Ltd.
Companion Website: www.wiley.com/go/Modeling_of_Physical_Systems/1e
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
300 6 Frequency Response and Impedance-Based Modeling

[ ]
where ℜ() implies taking the real part of the complex expression Ue st s=j𝜔 , and, in general, U is a complex number that
explicitly contains both magnitude and phase information when expressed in the polar form as,
U = |U|e j𝜃 .
The connection between the exponential form and a standard sinusoidal form can then be seen by the following derivation:
[ ]
u(t) = ℜ Ue st s=j𝜔
[ ]
= ℜ Ue j𝜔t
[ ]
= ℜ |U|e j𝜃 e j𝜔t
[ ]
= ℜ |U|e j(𝜔t𝜃)
[ ]
= ℜ |U|(cos( j𝜔t + 𝜃) + j sin( j𝜔t + 𝜃))
= |U| cos(𝜔t + 𝜃). (6.2)
Assuming an exponential form for both the input u and the output y,
u = Ue st ,
y = Y e st ,
where it is implied that s = j𝜔 and the real part is to be taken, then the output amplitude Y = Y ( j𝜔) can be found by
substituting into the differential equation which gives
(an sn + · · · + a0 )Y e st = (bm sm + · · · + b0 )Ue st .
Canceling the exponential yields,
Y b sm + · · · + b 0
= m n = G(s),
U an s + · · · + a0
from which it can be seen that the ratio of the exponential amplitudes is exactly the same as the previously given transfer
function. This gives us yet another interpretation of a transfer function. The transfer function alternatively can represent:
(i) a differential equation with s the derivative operator, or (ii) the ratio of the amplitudes of assumed exponential input and
output variables with s the coefficient in exp(st).
Using this last interpretation with s = j𝜔, we can solve for complex amplitude Y as,
Y = G( j𝜔)U,
and the frequency response output can be expressed as
[ ] [ ]
y(t) = ℜ Y e j𝜔t = ℜ G( j𝜔)Ue j𝜔t (6.3)
The complex function G( j𝜔) can be expressed in the polar form as,
G( j𝜔) = |G( j𝜔)|e j𝜙 .
Using the polar form of G( j𝜔) in equation 6.3, we can see that the output y(t) is a sinusoid of magnitude,
|Y | = |G( j𝜔)||U|,
and phase,
∠Y = ∠G + ∠U = 𝜙 + 𝜃,
which are functions of 𝜔. This implies that the ratio of input and output real magnitudes is,
|Y |
|G| =
|U|
and the phase shift between input and output is,
𝜙 = ∠G = ∠Y − ∠U.
The polar and time representations of the input and output are illustrated in Figure 6.1.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
6.1 Frequency Response 301

Y = GU
Im
Y
U
U

Re
(a) (b)

Figure 6.1 (a) Polar and (b) time representations of input, u(t), and output, y(t).

The magnitude and phase functions, |G(𝜔)| and 𝜙(𝜔), are also commonly referred to as frequency response functions
(FRFs). Derivation of a particular FRF for a system requires specifying the output variable, y, of interest in relation to the
input, u, with a transfer function. However, a given application and system may call for finding multiple transfer functions.
Consider first the basic case of a first-order system.

6.1.2 First-Order System

A simple electrical system is shown in Figure 6.2(a), a classical design for an analog “low pass” filter. From the bond graph
in Figure 6.2(b), a single state equation is, q̇ C = iR = (Vin (t) − VC )∕R and Vout = VC . Now, express the ODE in terms of VC ,

𝜏t V̇ C + VC = Vin .

since we will seek a transfer function between Vout and Vin . Substitute Vout for VC and a transfer function can then be
derived as,
Vout 1
= ,
Vin 𝜏t s + 1
which is a first-order transfer function with time constant 𝜏t = RC. The bond graph causality should be used to identify
relations between inputs and outputs.
In order to find the sinusoidal response of the first-order system, we evaluate this transfer function for s = j𝜔,
1
G( j𝜔) = .
j𝜏t 𝜔 + 1
√
Since the amplitude of the complex number j𝜏t 𝜔 + 1 is (𝜏t 𝜔)2 + 1, the amplitude of G( j𝜔) is,
1
|G( j𝜔)| = √ ,
(𝜏t 𝜔)2 + 1
and its phase is, 𝜙(𝜔) = −tan−1 (𝜏t 𝜔). The complex function G( j𝜔) can then be expressed in the polar form as,

G( j𝜔) = |G|e j𝜙 .

The input u(t) can be expressed as,

[ ] [ ]
u(t) = ℜ Ue j𝜔t = ℜ |U|e j(𝜔t+𝜃) .

R C

VR iR VC q̇C
Vin Vout Vin(t) VC Vout
Se 1 0 Sf
i=0
(a) (b)

Figure 6.2 (a) First-order filter with sinusoidal input and (b) bond graph.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
302 6 Frequency Response and Impedance-Based Modeling

Figure 6.3 Polar plot of a first order system, with the√ response indicated at the
break, or cutoff, frequency, 𝜔 = 1∕𝜏t , where |G| = 1∕ 2 and 𝜙 = 𝜋∕4.

Then from equation 6.3, the sinusoidal response for this system can be obtained as,
[ ] [ ]
̇ j𝜔t = ℜ |G||U|e j(𝜔t+𝜙+𝜃) .
y(t) = ℜ G( j𝜔)Ue
If u(t) = sin 𝜔t, as might occur in a filter system, then |U| = 1 and 𝜃 = 𝜋∕2, and the steady-state output y(t) is
[ ] 1
y(t) = Re |G|e j(𝜔t+𝜙+𝜋∕2) = √ sin(𝜔t + 𝜙), (6.4)
(𝜏t 𝜔)2 + 1
√
where 𝜙 = −tan−1 (𝜏t 𝜔). This means that the output is attenuated relative to the input by a factor√1∕ (𝜏t 𝜔)2 + 1 and shifted
in phase by an amount −tan−1 (𝜏t 𝜔). For example, for 𝜔 = 1∕𝜏t , the output is decreased by 1∕ 2 and shifted in phase by
−𝜋∕4.
Note that magnitude function, |G(𝜔)|, and phase shift, 𝜙(𝜔), are both functions of frequency 𝜔. The function G( j𝜔) can
also be displayed on the real-imaginary plane, as shown in Figure 6.3. In this type of polar plot, the frequency is an implicit
parameter along the curve and varies from 0 to ∞. This form of plot is commonly used in control system design.

6.1.3 Formulating Frequency Response Functions

For low-order systems, such as first- and second-order systems, frequency response functions can be quickly derived by
converting to s-domain and with algebraic manipulation to find G(s) = y∕u, as discussed in Section 4.4. Deriving transfer
functions from a linear state-space model was also discussed in Chapter 4, and given the standard set of linear or linearized
equations,
ẋ = Ax + Bu,
y = Cx + Du
the transfer function matrix is given by,
G(s) = C[sI − A]−1 B + D, (6.5)
where each Gjr (s) relates the jth output, yj , to the rth input, ur . The output equation needs to be defined to specify the
transfer function(s) desired. Certain outputs are chosen because they represent variables of interest, they may be variables
that are being directly measured by sensors, or will be used in control design. If the output(s) of interest are system states,
then Cramer’s rule (equation 4.26) can be used to derive the transfer functions that relate state variables to inputs.

Example 6.1 Electromechanical shaker with voltage input

A schematic for a small laboratory electromechanical shaker is shown in Figure 6.4(a), along with a bond graph for
model studies. The intent is to understand the performance characteristics of this shaker, which is designed to provide
controlled vibrational testing of objects attached to the table.3 The input voltage, Vs (t), is supplied by a voltage amplifier
(not shown), and the induced armature (coil) current, ic , generates a force on the moving coil (and attached table),
Fem = rc ic , where rc is the shaker force constant. Develop a state-space model from which a transfer function can be
developed relating the table acceleration to the input voltage. Derive the frequency response function for this transfer
function.

3 Vibration shaker models are also reviewed in Solved Problems A-5-11, B-5-31, and B-6.22.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
6.1 Frequency Response 303

I : Lc I : mt

Package under
test λ̇c ic ṗt vt
rc
Ground Vs(t) Vb ·· Fem Fks
Se 1 G 1 C : 1/ks
Permanent-magnet
ic vt ẋks
Ground
VRc Fbs vbs
Moving coil

R : Rc R : bs
(a) (b)

Figure 6.4 (a) Schematic of electromechanical shaker with coil rigidly attached to table, table leaf spring suspension ks and damping
bs , and voltage input Vs (t) and (b) bond graph with voltage source indicated at input.

Solution
[ ]T
From the bond graph, three independent states are indicated: x = 𝜆c , pt , xks . The state equations are,
⎡ 𝜆̇ c ⎤ ⎡ −Rc ∕Lc −rc ∕mt 0 ⎤ ⎡ 𝜆c ⎤ ⎡ 1 ⎤
⎢ ṗ ⎥ = ⎢ r ∕L −bs ∕mt −ks ⎥ ⎢ pt ⎥ + ⎢ 0 ⎥ Vs (t)
⎢ t ⎥ ⎢ c c ⎥⎢ ⎥ ⎢ ⎥
⎣ ẋ ks ⎦ ⎣ 0 1∕mt 0 ⎦ ⎣ xks ⎦ ⎣ 0 ⎦
and for table acceleration as the output, y = at = ṗ t ∕mt , so,
𝜆
[ ]⎡ c ⎤
y = rc ∕(Lc mt ) −bs ∕m2t −ks ∕mt ⎢ pt ⎥ + [0] Vs (t).
⎢ ⎥
⎣ xks ⎦
The implied ABCD matrices can now provide the desired transfer function from,
G(s) = C[sI − A]−1 B + D
which gives,
−1
] ⎡⎡
s 0 0 ⎤ ⎡ −Rc ∕Lc −rc ∕mt 0 ⎤⎤ ⎡1⎤
at [
= G(s) = rc ∕(Lc mt ) −bs ∕m2t −ks ∕mt ⎢⎢ 0 s 0 ⎥ − ⎢ rc ∕Lc −bs ∕mt −ks ⎥⎥ ⎢ 0 ⎥ + 0.
Vin ⎢⎢ ⎥ ⎢ ⎥⎥ ⎢ ⎥
⎣⎣ 0 0 s ⎦ ⎣ 0 1∕mt 0 ⎦⎦ ⎣0⎦
The transfer function between the table acceleration and the input voltage is,
at r c s2
= ,
Vin a 3 s + a 2 s2 + a 1 s + a 0
3

where a3 = mt Lc , a2 = (bs Lc + mt Rc ), a1 = (rc2 + bs Rc + ks Lc ), and a0 = ks Rc .

The frequency response function is found by setting s = j𝜔,
−rc 𝜔2
G( j𝜔) = [ ] [ ].
a0 − a2 𝜔2 + j𝜔 a1 − a3 𝜔2
So the magnitude and phase function are,
rc 𝜔2
|G(𝜔)| = √
[ ]2 [ ]2
a0 − a2 𝜔2 + 𝜔2 a1 − a3 𝜔2
and,
[ ]
𝜔(a1 − a3 𝜔2 )
𝜙(𝜔) = ∠G(𝜔)| = 𝜋 − tan−1 .
a0 − a2 𝜔2
Given these functions, the peak acceleration of the table for a given voltage peak, Vs,peak , and frequency, 𝜔, can be computed
from,
at,peak
= |G(𝜔)| ⋅ Vs,peak (in units of g).
g
Such a calculation is useful in prediction and design. Alternatively, plots of |G| and 𝜙 over frequency in units of decibels
and degrees, respectively, are commonly used to convey a system’s overall frequency response.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
304 6 Frequency Response and Impedance-Based Modeling

6.1.4 Logarithmic Plots of Frequency Response

The complex transfer function may also be represented in two separate plots, one for the magnitude versus frequency and
one for the angle versus frequency. It is useful to plot the logarithm of |G( j𝜔)| and the phase (in degrees) versus the frequency
(in logarithmic scale). Historically, the logarithmic magnitude of G( j𝜔) is taken as 20 log |G( j𝜔)|, where the basis of the
logarithm is 10. The unit in this representation is called a decibel, usually abbreviated dB. A “bel” is a unit in acoustics for
expressing, in logarithms, the ratios of power; that is, bel = log(∕0 ). The prefix deci- refers to 10 times this quantity or
decibel = 10 log(∕0 ). Since sound power is proportional to the square of pressure, the decibel was expressed as,
decibel = dB = 10 log(∕0 )2 = 20 log(∕0 ).
This is the historical basis for the unit. In actuality, any logarithmic scale would do and the factor of 20 is superfluous for
our application. However, since it is such a common logarithmic unit we retain its use. In these “Bode plots,” the curves
are typically drawn on semilog paper, using the log scale for frequency and the linear scale for either magnitude (in dB) or
phase angle (in degrees).
Contemporary computer-aided software tools for linear system analysis have made it relatively easy to generate the mag-
nitude and phase (or Bode) plots directly from a transfer function description. Such tools in the form of function calls are
helpful when numerical system parameters values are available, and especially when there is a need to conduct studies on
how variations in parameters influence a system’s frequency response functions. It is assumed that the reader has or can
gain sufficient familiarity with either open-source or commercial software tools for applying these software tools. Examples
throughout this chapter and in the rest of the book demonstrate their use.
The classical Bode plot of a transfer function is an estimated sketch of the magnitude and phase plots. These sketches
take advantage of being able to break down any linear transfer function into basic factors. Each of the basic factors can
then be individually shown to contribute to the overall Bode plots. The methodology for making these sketches is described
in the following. While making these sketches with great accuracy can be tedious, there is significant value gained from
understanding how individual factors contribute to overall trends in the Bode plots.
As discussed in Chapter 5, a transfer function can be expressed in factored form, such as,
K(s + z1 ) · · · (s + zk )
G(s) = . (6.6)
(s + p1 ) · · · (s + pn )
Evaluating this form with s = j𝜔 and taking the magnitude we have,
K|( j𝜔 + z1 )| … |( j𝜔 + zk )|
|G( j𝜔)| = . (6.7)
|( j𝜔 + p1 )| … |( j𝜔 + pn )|
The main advantage of using the logarithmic plot is that multiplication and division in equation 6.7 are converted into
addition and subtraction as,
log |G( j𝜔)| = log K + log |( j𝜔 + z1 )| + · · · + |( j𝜔 + zk )| − |( j𝜔 + p1 )| − · · · − |( j𝜔 + pn )|.
Furthermore, a simple method for sketching the individual factors and graphically adding and subtracting them is made
possible. Additionally, the logarithmic frequency representation is useful in showing both the high- and low-frequency
characteristics or trends in one plot.
For the electrical low-pass filter of Figure 6.2, for example, the log magnitude can be expressed in decibels as,
√
20 log |G| = −20 log (𝜏t 𝜔)2 + 1.
It is convenient to develop low- and high-frequency approximations to this magnitude. For low frequencies 𝜔 ≪ 1∕𝜏t ,
we have
20 log |G| ≈ −20 log(1) = 0 𝜔 ≪ 1∕𝜏t .
Thus, the log-magnitude curve for low frequencies is a constant at 0 dB. For high frequencies, 𝜔 ≫ 1∕𝜏t ,
20 log |G| ≈ −20 log(𝜏t 𝜔).
This is an approximate expression for the high-frequency range. At 𝜔 = 1∕𝜏t , the approximate log-magnitude equals 0 dB;
at 𝜔 = 10∕𝜏t , the log-magnitude is −20 dB. Thus, the value of −20 log(𝜏t 𝜔) decreases by 20 dB for every decade (i.e., 10-fold
increase in frequency) in 𝜔. For 𝜔 ≫ 1∕𝜏t , the log-magnitude curve is a straight line with a slope of −20 dB/decade.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
6.2 Basic Factors of Frequency Response Functions 305

–10
Exact curve
|G|dB –20 20 dB

–30

–40
10–2 10–1 100 101 102

φ Exact curve
–45

–90
10–2 10–1 100 101 102

Figure 6.5 Bode diagram of 1∕(𝜏t s + 1).

This analysis shows that the logarithmic representation of the frequency response of 1∕( j𝜏t 𝜔 + 1) can be approximated
by two straight-line segments. The exact curve and the asymptotic approximation are shown in Figure 6.5.
The frequency 𝜔 = 1∕𝜏t is called the break frequency. The maximum error in the asymptotic plot occurs at the break
frequency. This error can be calculated as,
√
−20 log 1 + 1 = −3.03 dB,
or approximately 3 dB error. At low frequency, the phase angle is 0∘ . At the break frequency, the phase angle is
𝜙 = −tan−1 (1∕1) = −45∘ . At high frequency, the phase becomes −90∘ .

6.2 Basic Factors of Frequency Response Functions

The factored form of the transfer function in equation 6.6 is extended to incorporate the standard second-order factors as
well. In this way, basic factors can be defined that comprise Bode diagrams of the transfer function for a finite-order system.4
These basic factors are:
1) Gain factor, K
2) First-order factors, (𝜏t s + 1)±1
[ ]±1
3) Second-order factors, (s∕𝜔n )2 + 2𝜁s∕𝜔 + 1
4) Integral and derivative factors, s±1
Note that integral and derivative factors are just special cases of first-order factors. These basic factors can be graphically
added and subtracted and are sufficient for sketching finite-order system Bode diagrams. The process can be further sim-
plified by using asymptotic approximations as previously described.
The Gain, K. The gain K is simply a constant versus frequency with a constant phase of 0∘ for positive gain and −180∘ for
negative gain.
First-Order Factors. We have previously obtained the Bode diagram for (𝜏t s + 1)−1 . The Bode diagram for (𝜏t s + 1)+1 is
similar. The only difference is the log magnitude plot “breaks up” rather than down at a final slope of 20 dB/dec, and the
phase increases positively from 0∘ to 90∘ . The asymptotic and exact values are shown in Figure 6.6.

4 Finite-order polynomial functions typically result from lumped-parameter models. When distributed-parameter elements are included in a
system, as will be discussed later in Section 6.6.3, alternative methods are required or approximations must be made.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
306 6 Frequency Response and Impedance-Based Modeling

40

30
|G|dB + 20 dB
20 Exact curve

10

0
10–2 10–1 100 101 102

90

φ Exact curve
45

0
10–2 10–1 100 101 102

Figure 6.6 Bode diagram for (𝜏t s + 1).

Second-Order Factors. For the second-order factor,

1
G(s) = ,
(s∕𝜔n )2 + 2𝜁s∕𝜔n + 1
the frequency response is found from,
1
G( j𝜔) = .
1 − ((𝜔∕𝜔n ))2 + 2j𝜁(𝜔∕𝜔n )
The amplitude of this factor is,
1
|G( j𝜔)| = √ ,
[ ]2 [ ]2
2
1 − ((𝜔∕𝜔n )) + 2𝜁𝜔∕𝜔n
and the phase is,
[ ]
2𝜁𝜔∕𝜔n
𝜙 = −tan −1
.
1 − ((𝜔∕𝜔n ))2
For low frequency, we have 20 log |G| = 0 and 𝜙 = 0o for 𝜔 ≪ 𝜔n , and for high frequency, 20 log |G| ≈ 40 log(𝜔∕𝜔n ) and
𝜙 = −180o for 𝜔 ≫ 𝜔n . A bode plot of this second-order factor for a range of damping 0 < 𝜁 < 1 is shown in Figure 6.7.
The break frequency for a quadratic term is 𝜔 = 𝜔n and the slope at high frequency is√−40 dB/dec.
It can be shown that√the peak amplitude occurs at a resonant frequency, 𝜔r = 𝜔n 1 − 2𝜁 2 . The amplitude at this fre-
quency is |G|r = 1∕2𝜁 1 − 𝜁 2 . Note that at resonance, the amplitude is solely a function of the damping ratio. For this
reason, the amplitude at resonance is said to be damping controlled. For damping ratios 𝜁 > 0.707, there is no resonant
peak. As damping approaches zero, |G|r approaches infinity. This means that if the undamped system is excited at its
natural frequency, the steady magnitude of the response approaches infinity.
The Bode plot of the quadratic factor for a zero (i.e., in numerator),
G(s) = (s∕𝜔n )2 + 2𝜁s∕𝜔n + 1,
is similar to that for a pole and is shown in Figure 6.8. Trends in magnitude and phase are similar to those discussed earlier
hold in this case as well, with the implied changes.
Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
307
6.2 Basic Factors of Frequency Response Functions

102
102

102

102
40 dB/dec

101
101

101

101
Increasing ζ

Bode diagrams for (s∕𝜔n )2 + 2𝜁s∕𝜔n + 1 .

]−1

Bode diagrams for (s∕𝜔n )2 + 2𝜁s∕𝜔n + 1 .

]
100
100

100

100
10–1
10–1

[
10–1

10–1
10–2
10–2

10–2

10–2
20
0
–20
–40
–60
–80
–100

–45

–90

–135

–180

100
80
60
40
20
0
–20

180

135

90

45

0
Figure 6.7

Figure 6.8
|G|dB

|G|dB
φ

φ
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
308 6 Frequency Response and Impedance-Based Modeling

20

10

–10

–20
10–1 100 101

100

50

–50

–100
10–1 100 101

Figure 6.9 Bode diagrams for s±1 .

Integral and Derivative Factors. For the factor s±1 , the log magnitude is

20 log |G| = ±20 log | j𝜔| = ± log 𝜔.

As illustrated in Figure 6.9, a derivative term, s±1 , has a log magnitude curve versus log frequency of a straight line with
slope of +20 dB/dec, while the integral term 1∕s is a straight line with slope −20 dB/dec. As shown, the phase for the s factor
(in the numerator) is 𝜙 = +90∘ , while for s−1 = 1∕s, it is −90∘ .

Example 6.2 Example approximations of Bode plots

For the transfer function
y s2 + 2s + 2
= G(s) =
u 4(s3 + 2s2 + 2s + 1)
present the amplitude of the frequency response as a Bode diagram.
Solution
The first essential step in obtaining the Bode diagram is to factor the transfer function into basic factors. The given transfer
function can be factored as,
[ √ ]
2 (1∕2) (s∕ 2)2 + s + 1
s + 2s + 2
= = G1 (s) ⋅ G2 (s) ⋅ G3 (s) ⋅ G4 (s).
4(s3 + 2s2 + 2s + 1) (s + 1)(s2 + s + 1)
This transfer function has a real pole, a complex conjugate pair of poles, and a complex conjugate pair of zeros. The factors
are identified in tabular form below, along with key parameter values that aid in sketching the diagrams. It can be helpful
to tabulate the factor data as follows:
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
6.2 Basic Factors of Frequency Response Functions 309

Factor Gain Second order First order Second order

[ √ ]
G1 (s) = K G2 (s) = (s∕ 2)2 + s + 1 G3 (s) = (s + 1)−1 G4 (s) = (s2 + s + 1)−1
√
K = 1∕2 𝜔n = 2 𝜏t = 1 𝜔n = 1
𝜁 = 0.707 𝜔b = 1 𝜁 = 1∕2

To sketch the Bode diagram, the separate asymptotic curves for each factor are first placed on the plot. The composite
curve is obtained by graphically adding the individual curves (recall, this is made possible by using the dB scale). The factors
and composite curves for the magnitude function are illustrated in Figure 6.10.
Adjustments can be made at the break frequencies to increase the accuracy of the sketch. Note that the final slope of the
Bode plot is −20 dB/dec. The final slope of a rational5 transfer function will always be equal to −20(n − m), where n is the
highest denominator power of s and m is the highest numerator power. This result is useful when experimentally measured
magnitude plots are used for system identification (see Section 6.4.3).

Sketching Bode plots of complex transfer functions may seem unnecessary given that contemporary software tools can
produce exact plots of the magnitude and phase functions so easily. There is value, however, in understanding how a
system transfer function is composed of basic factors that can be more easily related to a physical system under study.
Even more insight can be gained about how such functions relate to possible physical realizations of the system by employ-
ing impedance forms directly on a bond graph, as discussed in the next section.

80

60

40

20

–20
Exact
composite, G(s)
–40

–60

–80
10–1 100 101 102
, rad/sec

Figure 6.10 Basic factors shown as asymptotes (solid) and exact (dashed) curves along with composite curve for
2
G(s) = s + 2s + 2 .
4(s3 + 2s2 + 2s + 1)

5 Note, a proper or rational transfer function has a numerator with order m less than the degree of its denominator. Otherwise, the transfer
function is referred to as improper.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
310 6 Frequency Response and Impedance-Based Modeling

6.3 Impedance Methods Using Bond Graphs

Transfer functions can be derived in many different ways, as first shown in Chapter 4 and again in Chapter 5. Yet another
way to formulate a transfer function between an output of interest and an input is by using impedance relationships.
An impedance in this text is defined as a transfer function between the effort and flow on a power bond. For a general
system, the corresponding linear impedance relation is as follows:

e ⇒ e
SYSTEM Z(s) ≡
f f

Although some of the concepts of impedance can be generalized to nonlinear systems [96], we restrict ourselves here
to linear impedance. Using impedance directly on a bond graph provides a more direct and often quicker way to derive a
desired transfer function. As will be shown, no differential equations are needed in this approach. Further, impedance study
with a bond graph can provide insight into a system that is not gained in other ways. It is useful to refer to the impedance
seen at the ports of a system, for example, by their input or output impedance relations.

6.3.1 Basic Impedance Elements

For a linear one-port R element, e = R ⋅ f and the impedance is,
e
ZR (s) ≡ = R. (6.8)
f
For a linear C element, e = q∕C = (∫ fdt)∕C = f ∕Cs and the impedance is

Z0 (s) ≡ 1∕Cs. (6.9)

For an I element, f = p∕I = (∫ edt)∕I = e∕Is and the impedance is

ZI (s) ≡ Is. (6.10)

It should now be understood that these impedance relations can be defined for the linear modeling elements in all of our
energy domains of interest.
Based on this fundamental definition of impedance as well as admittance, Y , which is the inverse of Z, that is, Y = 1∕Z,
it is possible to identify some useful relationships for basic bond graph modeling elements. Consider when two impedance
elements are connected at a 1 junction as in Figure 6.11(a).
Since Z1 = e1 ∕f1 and Z2 = e2 ∕f2 , then by using the effort relation, e = e1 + e2 , we can write, e = Z1 f + Z2 f , which leads to,
e
= Z1 + Z2 ⇒ Impedances add at a 1-junction.
f
Note that these relations depend on the power sign definitions. A similar relation can be found for a 0 junction, shown
in Figure 6.11(b). In this case, f = f1 + f2 = (1∕Z1 )e + (1∕Z2 )e, so that f ∕e = 1∕Z1 + 1∕Z2 . Or,
e 1 Z1 Z2
= = . (6.11)
f 1∕Z1 + 1∕Z2 Z1 + Z2

Z1 Z1

e1 f1
e e2
1 Z2 0 Z2
f f f2
e
(a) (b)

Figure 6.11 (a) Impedances at a 1-junction add, Z = Z1 + Z2 . (b) Admittances at a 0-junction add, Y = Y1 + Y2 .
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
6.3 Impedance Methods Using Bond Graphs 311

Recall admittance Y is the inverse of the impedance, Y = 1∕Z, so the result in 6.11 can also be expressed as,
f
= Y = Y1 + Y2 ⇒ Admittances add at a 0-junction,
e
where Y1 = 1∕Z1 , Y2 = 1∕Z2 .
Two other useful impedance relations to remember are the bond graph equivalents of the voltage and current dividers.
The effort divider relation gives the transfer function of input and output efforts across a 1-junction in terms of impedances.
This relation can be derived as follows. Say it was desired to find the effort on Z2 in Figure 6.11(a). First write,
Z2
e2 = Z2 f = e.
Z1 + Z2
Thus, the output e2 can be related to the input e by,
e2 Z2
= .
e Z1 + Z2
Similarly, a “flow divider” relation relates input and output flows for the case in Figure 6.11(b). Here you can show that,
( )
1 1 Z1 Z2
f2 = e= f.
Z2 Z2 Z1 + Z2
Thus,
f2 Z1
= . (6.12)
f Z1 + Z2
In terms of admittances, equation 6.12 can also be expressed as,
f2 Y2
= . (6.13)
f Y1 + Y2
Using the bond graph structure makes it convenient to use impedance relations to derive transfer functions. As an example,
consider the electric circuit shown in Figure 6.12(a).
Impedance techniques are commonly used to derive transfer function relations for electric circuit systems. Circuit analogs
for acoustical and electromechanical systems are similarly used for this purpose. These methods can be further generalized
by using impedance forms of bond graphs to derive system transfer functions. For example, consider the case where it is
desired to find Vout ∕Vin directly from the bond graph in Figure 6.12(b). For relatively simple systems, a quick approach is
to replace the (linear) bond graph elements with their impedance equivalents and then to use the effort and flow divider
templates (derived earlier) to reduce to a final form. An example of this approach is illustrated in Figure 6.13. The reduced
form in Figure 6.13(c) shows the result of replacing the parallel RC with an equivalent impedance. It is then possible to
take advantage of a basic “voltage-divider” type relation from Figure 6.13(c): 𝑣out = Z∕(ZL + Z)𝑣in .
This relation leads to,
Vout R∕(RCs + 1) R 1∕LC
= = 2
= 2 .
Vin R∕(RCs + 1) + Ls LRCs + Ls + R s + s∕RC + 1∕LC
In general, this approach is appropriate when the impedance templates defined above can be easily identified and used
in a system to reduce the bond graph. For relatively simple cases, there is no faster method by which to derive a transfer
function for linear system models.

I R

Vin Vout
Vin Vout
Se 1 0 C
(a) (b)

Figure 6.12 (a) LRC circuit and (b) bond graph.

 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
312 6 Frequency Response and Impedance-Based Modeling

sL R sL R sL
Y = YR + YC
Y = 1/R + sC Z= R
RCs + 1
Z = 1/Y
Vin Vout Vin Vout Vin Vout
1 0 1/sC 1 0 1/sC 1 Z

(a) (b) (c)

Figure 6.13 (a) LRC circuit, (b) bond graph, and (c) reduced impedance form.

6.3.2 Impedance-Based Transfer Functions

To apply impedance techniques to more complex systems, it is preferred to formulate desired transfer function relations
by strategically using the bond graph structure. Applying these techniques in a multistep fashion considerably widens the
effective use of impedance. The methods described in the following are especially useful for systems composed of tree-like
structures. There is a large class of practical engineering systems that fall under this category. Deriving impedance relations
for systems that have loops in the bond graph must be dealt with using alternative methods (e.g., see Brown [97]).
Consider the impedance form of a bond graph shown in Figure 6.14. A transfer function that relates, say, the flow, f5 ,
to effort e1 can be found by a product of impedance and admittance relations that can be derived from the bond graph
and chosen by examining the effort and flow junction relations. For example, consider the desired transfer function can be
found by forming a product of impedance or admittance relations; that is,
[ ] [ ] [ ]
f5 f e f
= 5 ⋅ 3 ⋅ 1 .
e1 e5 f3 e1
First, note that the leftmost term on the right-hand side has the output variable in the numerator and the rightmost term
has the input term of the desired transfer function in the denominator. Further, intermediate variables in each of the
impedance/admittance relations are equal to variables in a neighboring relation according to the bond graph structure.
For example, e5 = e3 because of the common effort junction shared by bonds 3 and 5. Similarly, f3 = f1 by the shared com-
mon flow junction. These relations are necessary so that the product of these terms yields the desired transfer function
on the left-hand side. To be clear: we do not literally cancel these variables, as doing so would not give the result desired.
Rather, each of the bracketed impedance/admittance relations is derived from the bond graph and the resulting product
yields the desired result. It can be more clear if we express this relation,
f5
= Y5 ⋅ Z3 ⋅ Y1 ,
e1
as this form highlights how each represents an impedance or admittance relation from the system under study. The advan-
tage in this approach is that each of these terms can be readily found as follows:
Y5 = 1∕Z0 ,
Z3 = 1∕Y3 , where Y3 = Yb + Yc ,
Y1 = 1∕Z1 , where Z1 = Za + Z3 .
These are all simply algebraic relations in the operator s and the system parameters. The process is somewhat systematic
but can be tedious as systems get more complex. However, the availability of modern computer-aided symbolic processors
has made this a much more tangible and less error-prone task. This approach to deriving transfer function relations is based
on the classical use of transmission matrices, which has been applied in vibration analysis of complex structural systems
for many years. The use of the bond graph helps broaden this application to multi-energetic systems as well. More will be
said about this in a subsequent section. First, some examples are presented of a direct approach with impedance relations.

Za Zb Figure 6.14 Impedance form of bond graph for transfer function derivation.

e2 f2 e 4 f4

e1 e3 e5
1 0 Z0
f1 f3 f2
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
6.3 Impedance Methods Using Bond Graphs 313

Example 6.3 Torsional drive system

Consider the model of a torsional drive system shown in Figure 6.15, along with the bond graph in Figure 6.16. We would
like to derive a transfer function relating the output angular velocity 𝜔J to the input torque 𝜏(t).
First, form the impedance bond graph as in Figure 6.17. We can find the desired transfer function by noting from the
impedance bond graph that:
[ ] [ ] [ ]
𝜔J 𝜔5 𝜏 𝜔
= ⋅ 3 ⋅ 1 ,
𝜏 𝜏5 𝜔3 𝜏
where subscripts refer to the numbered bonds, and 𝜔5 = 𝜔7 = 𝜔J from the bond graph structure. The relation above gives,
𝜔J
= Y5 ⋅ Z3 ⋅ Y1
𝜏
and,
Y5 = 1∕(b2 + Js),
Cs(b2 + Js) + 1
Z3 = 1∕Y3 , where Y3 = Cs + Y5 = ,
b2 + Js
b (Cs(b2 + Js) + 1) + b2 + Js
Y1 = 1∕Z1 , where Z1 = b1 + Z3 = 1 .
Cs(b2 + Js) + 1
Multiplying the three transfer functions gives,
𝜔J 1 (b2 + Js) Cs(b2 + Js) + 1
= ⋅ ⋅
𝜏 (b2 + Js) Cs(b2 + Js) + 1 b1 (Cs(b2 + Js) + 1) + b2 + Js
which simplifies to,
𝜔J 1
= .
𝜏 JCb1 s2 + (b1 b2 C + J)s + b1 + b2
Note that the impedance and admittance relations in the earlier steps were consolidated into a single term (e.g., with com-
mon denominator). In this way, it is easy to see that some numerator and denominator terms will cancel, as seen in the
final step in this case.

Linear bearing Linear bearing

friction, b1 friction, b2

Input Shaft compliance, C = 1/k

torque, (t) Rotational
inertia, J

Figure 6.15 Torsional drive system schematic.

Figure 6.16 Torsional drive system bond graph. R : b1 C:C R : b2

τ (t)
Se 1 0 1 I:J
ωJ

Figure 6.17 Impedance form of torsional drive system bond graph. b1 1/sC b2

2 4 6
τ (s) τ3 τ5
1 ω3 0 ω5 1 sJ
1 ωJ (s)
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
314 6 Frequency Response and Impedance-Based Modeling

6.3.3 Impedance Across Coupling Elements

The impedance change across a transformer can be derived in a straightforward manner as

e1 n
·· e2
T Z
f1 f2

from which we see (adopting the convention in equations (3.20)) that,

e1 = e2 ∕n = (1∕n)Zf2 = (1∕n2 )Zf1 .
Thus, when looking into a transformer, there is an effective impedance,
e
Z1 = 1 = Z∕n2 , (6.14)
f1
which shows that a transformer does not change the nature of the impedance of a system having impedance Z1 but simply
transforms or scales it by a factor n2 .

Example 6.4 Impedance relations for a system with transformer

The impedance bond graph shown below can be used to derive transfer functions between variables of interest for the
system model.

Ra Rb

e2 f2 e6 f6

e1 e4 n
·· e5 e7
Se 1 T 1 Ib s
f1 f4 f5 f7
e3 f3

Ia s

Find f7 ∕e1 : The transfer function that relates the flow variable for Ib to the input e1 can be found by identifying the
following from the bond graph:
[ ][ ]
f7 f f4
= 5
e1 f4 e1
which makes use of the relations, f7 = f5 = f4 ∕n. Also note that f4 = f1 , thus we need to find the total impedance Z1 = e1 ∕f1
to determine the desired relation. This can be done as follows:
Step 1: Z5 = Z6 + Z7
Step 2: Z4 = Z5 ∕n2 = (Z6 + Z7 )∕n2
Step 3: Z1 = Z2 + Z3 + Z4 = Z2 + Z3 + (Z6 + Z7 )∕n2
Substituting the respective impedance relations:
Step 4: Z1 = Ra + I2 s + (Rb + Ib s)∕n2
Now, since f5 ∕f4 = 1∕n, and f4 ∕e1 = 1∕Z1 (since f4 = f1 ),
[ ]
f7 [ 1 ] 1 1∕n
= 2
= .
e1 n Ra + Ia s + (Rb + Ib s)∕n s(Ia + Ib ∕n ) + Ra + Rb ∕n2
2

This transfer function shows the system is first order, which is expected since the two I elements are coupled by the trans-
former. In the standard form,
f7 K
= ,
e1 𝜏s + 1
where 𝜏 = (Ia + n2 Ib )∕(Ra + n2 Rb ) and K = 1∕n(Ia + Ib ∕n2 )∕(Ra + Rb ∕n2 ).
This example demonstrates how to account for the scaling of impedance level in a system when there is a transformer.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
6.3 Impedance Methods Using Bond Graphs 315

rm
Vm ·· τm Vm 1 
G sJm Jm
im ωm im s 2
rm
Ideal EM transduction

(a) (b) (c)

Figure 6.18 (a) Ideal EM transduction with flywheel, (b) impedance bond graph, and (c) reduced impedance bond graph of equivalent
flywheel capacitor.

Impedance scaling by a transformer is useful in design of many types of power transmission systems across different
energy domains. The concept of impedance matching is used to maximize power transfer from one part of a system to
another or to match impedance and eliminate power reflection.6
Consider now the series connection of a gyrator with an impedance, Z, as shown below:

e1 r
·· e2
G Z
f1 f2

In this case, we have,

e1 = rf2 = re2 ∕Z = (r 2 ∕Z)f1
so that the effective impedance seen at the input to the gyrator is,
e1 r2
Z1 = = = r2 Y . (6.15)
f1 Z
In contrast to the transformer, the gyrator does change the nature of the impedance of a system. In particular, the
impedance is inverted and scaled by r 2 . The ability to convert the impedance characteristic using a gyrator has been used
in many areas. For example, a linear I element with rotational impedance sJm would appear as an impedance rm 2 ∕sJ
m
2
on the other side of a gyrator. The impedance rm ∕sJm has the same form as the impedance of a linear capacitance 1∕sC
with C = Jm ∕rm 2 . This concept can be used by combining electrical rotating machinery with large inertial flywheels to act

as equivalent electrical capacitors at the electrical port, as shown in Figure 6.18. Since the energy density of a flywheel
(energy/volume) can be appreciably higher than the energy density of an electrical capacitor, a flywheel can serve as an
efficient electrical energy storage device.
This was called “C-ing an I through a G” by Paynter. This concept can help explain why purely capacitive microelectronic
designs became popular in the early 1960s to realize “inductorless” filters. Since designers had conceived of ways to build
gyrators in silicon, a series connection of a gyrator–capacitor combination enabled realization of a functional inductor
without having to use bulky magnetic coil designs.

Example 6.5 Motor-driven torsional drive system

Let us now attach a permanent-magnet DC (PMDC) motor to the input of the torsional drive system from Example 6.3 as
shown in Figure 6.19(a). The associated impedance bond graph is shown in part (b) of that figure. We now seek a transfer
function relating the output angular velocity 𝜔 to the input voltage, V.
From the impedance bond graph, the desired transfer function is found from:
[ ] [ ] [ ] [ ] [ ]
𝜔J 𝜔J 𝜏k 𝜔m 𝜏 i
= ⋅ ⋅ ⋅ m ⋅ m ,
V 𝜏k 𝜔m 𝜏m im V
where a different approach is taken from the previous example to show how to quickly derive these forms without suc-
cessive numbering of all bonds. This requires being able to read the implicit relation of effort and flow variables at the

6 See Brown [7], for examples; impedance matching is used in circuit design, acoustics, and electromagnetic wave systems.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
316 6 Frequency Response and Impedance-Based Modeling

Linear bearing Linear bearing

friction, b1 friction, b2
pmdc
motor

Shaft compliance, C = 1/k

Rotational
inertia, J
(a)

sIm sJm k/s

VLm τJm τk ω k
rm
V Vm ·· τm
1 G ωm 1 0 1 ωJ sJ
im
τ1 τ2

Rm Bm + b1 b2
(b)

Figure 6.19 (a) Torsional system driven by PMDC motor and (b) impedance bond graph.

1- and 0- junctions and applying rules for how impedance and admittance relations add at those junctions, respectively.
The relations implied above are found as follows:
𝜔J [ ]−1
= b2 + sJ ,
𝜏k
[ ]
𝜏k 𝜔 −1
= s∕k + J ,
𝜔m 𝜏k
[ ]
𝜔m 𝜏 −1
= bt + sJm + k ,
𝜏m 𝜔m
𝜏m
= rm .
im
[ ]−1
im 2 𝜔m
= Rm + sLm + rm .
V 𝜏m
Note that each of these transfer function relations is either an impedance or and admittance, as needed to form 𝜔J ∕V as
defined above. Multiplying these transfer functions and simplifying (with the aid of a symbolic processor) gives,
𝜔J b0
= 4 ,
V s + a 3 s3 + a 2 s2 + a 1 s + a 0
where
krm
b0 = ,
JJm Lm
2 +R b +R b)
k(rm m 2 m t
a0 = ,
JJm Lm
2 + JR k + J R k + R b b + L b k + L b k
b 2 rm m m m m 2 t m 2 m t
a1 = ,
JJm Lm
2 + J R b + JR b + JL k + J L k + L b b
Jrm m m 2 m t m m m m 2 t
a2 = ,
JJm Lm
JJm Rm + Jm Lm b2 + JLm bt
a3 = ,
JJm Lm
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
6.4 Applications of Frequency Response 317

where bt = Bm + b1 . The fourth order form of the resulting transfer function shows that all of these energy storing elements
are independent in this system. If it was determined that the motor inductance (Lm ) was very small and could be neglected,
then a simplified model of third order form would be determined through this analysis. One way to determine which model
is suitable is to compare the system frequency response characteristics.
Lastly, make note of how the gyrator is taken into account in two different ways when deriving the transfer function of
interest. In forming the final transfer function, we require the gyrator relation, 𝜏m ∕im = rm . However, when expressing the
impedance looking into the gyrator to find the relation im ∕𝑣, we need to make sure to apply the proper scaling of rm as well
as the inversion of the impedance, as given in equation 6.15.

6.3.4 Summary
This section introduced the use of impedance relations for deriving transfer functions directly from bond graphs. Transfer
functions can be derived using basic impedance elements along with impedance and admittance relations implied through
the effort and flow junctions. Using impedance methods can provide insights into systems, in contrast to direct approaches
to building transfer functions (e.g., from a state-space model as shown in Chapter 5). Specifically, the impedance rela-
tions associated with coupling elements are fundamental to understanding analysis and design of a wide range of systems.
Another approach for finding transfer function relations with impedance forms uses methods borrowed from classical
transmission matrix methods. This approach along with two-port modeling is introduced in Section 6.5. Having multiple
ways to derive transfer functions for a system can be very helpful, since the method chosen can depend on the application
and information available. If the system model has already been derived and linear state-space equations are available,
then finding transfer functions between outputs and inputs of interest follows well established methods as discussed in
Chapters 4 and 5 and as shown earlier in this chapter. Using impedance methods might be more useful if several differ-
ent models are being investigated and for quick derivations, and also when it is desirable to build insight into how the
results relate to the system structure. Indeed, in some contexts, it is preferred to use impedance formulations for building
models rather than first deriving state-space equations.

6.4 Applications of Frequency Response

Frequency response functions (FRFs) of a system are used in many different ways across a broad range of applications.
Direct use of the FRF enables making estimates of the amplitude and phase response of a system, say, G(s), excited by
an input with a given amplitude and frequency. Recall, that the forced response (output) of G(s) to a sinusoidal input,
u(t) = uo sin(𝜔t) is,

y(t) = yo (𝜔) sin(𝜔t + 𝜙(𝜔)),

where the amplitude yo (𝜔) = |G( j𝜔)| ⋅ uo and 𝜙(𝜔) = ∠G( j𝜔). This practical relation has implications for providing ways to
estimate and specify system behavior. It is especially useful when characterizing sensor and actuator systems, for example.
Further, when combined with some information using a structured physical system model (like a bond graph), it is possible
to guide system identification. For many practical problems, such insight can be sufficient to guide parameter estimation.
The following are selected application areas to demonstrate the use of the methods described for deriving and using fre-
quency response functions.

6.4.1 Vibration Applications

Transfer functions are widely used to understand how vibratory forces propagate through mechanical systems. While
sources of vibration such as unbalanced rotating machines or engines can often be unavoidable, their effect on a sys-
tem can often be substantially reduced by properly designed spring and damper combinations, which are referred to as
isolators. Consider Figure 6.20(a), where F(t) = Fo sin 𝜔t represents a vibratory force applied onto a spring–mass–damper
system. The bond graph in Figure 6.20(b) indicates there are two states and an impedance bond graph is also given in
Figure 6.20(c). Two different approaches will be illustrated for obtaining a transfer function for this system.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
318 6 Frequency Response and Impedance-Based Modeling

Se Se

F (t) F
ṗm Fm
1 I 1 ms
vm vm

FT vm FT vm

FT Fb Fb
0 vm 1 vb R 1 vb b

FT vg = 0 Fk ẋk Fk vk

Sf C k/s
(a) (b) (c)

Figure 6.20 (a) Sinusoidal forcing of vibratory system, (b) causal bond graph, and (c) impedance bond graph (with 0-junction and
ground source removed).

The force transmissibility in a vibration isolator is the ratio of the force transmitted to the ground to the input force, or
FT ∕F. From the bond graph in Figure 6.20(b) note that, FT = Fk + Fb = kxk + b𝑣m . This relation can serve as an output
equation for the state equations:
[ ] [ ][ ] [ ]
𝑣̇ m −b∕m −k∕m 𝑣m 1∕m
= + F(t),
ẋ k 1 0 xk 0
[ ]
[ ] 𝑣m [ ]
y = FT = b k + 0 F(t).
xk
Given the ABCD terms from these equations, the transfer function is found as,
[ ]−1 [ ]
F [ ] s + b∕m +k∕m 1 (b∕m)s + k∕m
G(s) = T = b k +0= 2 .
F −1 s 0 s + (b∕m)s + k∕m
Letting 2𝜁𝜔n = b∕m and 𝜔2n = k∕m,
FT 2𝜁𝜔n s + 𝜔2n
= . (6.16)
F s2 + 2𝜁𝜔n s + 𝜔2n
Consider now the impedance bond graph in Figure 6.20(c). First, note that the impedance FT ∕𝑣m = b + k∕s = (bs + k)∕s,
and F∕𝑣m = ms + FT ∕𝑣m = ms + (bs + k)∕s = (ms2 + bs + k)∕s, so that,
[ ] [ ] [ ] [ ]
FT FT 𝑣 bs + k s
= ⋅ m = ⋅
F 𝑣m F s 2
ms + bs + k
or,
FT bs + k bs∕k + 1 2𝜁s∕𝜔n + 1
= 2
= = (6.17)
F ms + bs + k ms ∕k + bs∕k + 1 (s∕𝜔n )2 + 2𝜁s∕𝜔n + 1
2

which is equivalent to that found in equation 6.16.

By s = j𝜔 in this transfer function and rearranging, we get the transmissibility ratio as the magnitude function,
√
| FT | (2𝜁𝜔∕𝜔n )2 + 1
| |= √ . (6.18)
|F | [ ]2 [ ]2
| |
(1 − (𝜔∕𝜔n )2 + 2𝜁𝜔∕𝜔n
√
As shown in Figure 6.21, the amplitude
√ √ 𝜔∕𝜔n = 2, regardless of
ratios are all equal to |FT ∕F| = 1.0 at the frequency
damping. Note that for 𝜔∕𝜔n < 2, the value of |FT ∕F| is greater than 1.0, and √ for 𝜔∕𝜔n > 2, the value of |FT ∕F| is less
than 1.0. This shows that vibration isolation is achieved only when 𝜔∕𝜔n > 2. Note that better isolation is achieved in
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
6.4 Applications of Frequency Response 319

0
0 1 2 3 4 5

Figure 6.21 Linear plot of force transmissibility.

this region as 𝜁 is decreased. An undamped spring reduces transmissibility much better than a damped spring. Of course,
practically, there is a need to introduce damping when 𝜔 varies over the resonant region, as in a startup period.
These results provide insight when studying the related problem of a vibration absorber, as illustrated in the following
example. Two methods were demonstrated for deriving the required transfer function. The decision on which approach to
adopt in a given application can come down to many factors, only one of which is the availability of the state equations.
The following example illustrates use of the force transmissibility result.

Example 6.6 Vibration absorber

An electric motor of mass 50 kg is spring mounted with a natural frequency fn = 10 Hz and a damping of 𝜁 = 0.1. Assume
there is an imbalance in the motor that results in a harmonic force of F(t) of amplitude 100 N and frequency of 20 Hz.
Determine the force transmitted to the floor.
Solution
Using equation 6.18, we have,
√
|F | = √ (2(0.1)2)2 + 1
| T| ⋅ (100 N) = 35.5 N.
[ ]2
(1 − (2)2 + [2(0.1)2]2

Another method for achieving vibration isolation is to tune a spring–mass system k2 , m2 , shown in Figure 6.22(a), to the
frequency of the disturbing force 𝜔 such that 𝜔2n = k2 ∕m2 . In this case, the spring–mass system will act as a vibration absorber
to reduce the motion of the main mass m1 to zero. Insight into this requirement can be gained by using a transfer function
relating the position of the mass xm1 to F, where xm1 = 𝑣m1 ∕s. The transfer function required is 𝑣m1 ∕F, which can be found
from the sum of impedances,
F F F F k k2 m 2 s k (m s2 + k2 ) + m1 s2 (m2 s2 + k2 ) + k2 m2 s2
= 1 + m1 + a = 1 + m1 s + = 1 2 .
𝑣m1 𝑣1 𝑣m1 𝑣m1 s 2
m 2 s + k2 s(m2 s2 + k2 )
Thus,
𝑣m1 s(m2 s2 + k2 ) s(m2 s2 + k2 )
= = .
F k1 (m2 s2 2
+ k2 ) + m1 s(m2 s + k2 ) + k2 m2 s 2 (k1 + m1 s )(k2 + m2 s2 ) + k2 m2 s2
2

Now, the transfer function between the position of the mass 1 and the force F is found by xm1 ∕F = (1∕s)𝑣m1 ∕F, giving,
xm1 k 2 + m 2 s2
= . (6.19)
F (k1 + m1 s )(k2 + m2 s2 )
2 + k 2 m 2 s2
To get the frequency response, set s = j𝜔,
xm1 k2 − m2 𝜔2
= .
F (k1 − m1 𝜔 )(k2 − m2 𝜔2 ) − k2 m2 𝜔2
2
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
320 6 Frequency Response and Impedance-Based Modeling

C k1 /s

F1 ẋ1 F1 v1
F (t) F
Se 1 I 1 m1 s
vm1 vm1

Fa vm1 Fa vm1

F2 F2
0 C 0 k2 /s
ẋ2 v2

F2 vm2 F2 vm2

1 I 1 m2 s
vm2 vm2
(a) (b) (c)

Figure 6.22 (a) Vibration absorber, (b) bond graph, and (c) impedance bond graph.

If k2 and m2 are designed such that 𝜔2 = k2 ∕m2 then,

xm1
= 0 if 𝜔2 = k2 ∕m2 .
F
√
Note that the transfer function 6.19 has a zero at s = ±j k2 ∕m2 and we are simply tuning the system such that the input
has the frequency of this zero. This method of input cancellation is also used in control design under the name pole-zero
cancellation (see discussion in Chapter 7).
So far only the ratio of k2 ∕m2 has been determined. The transfer function between the absorber position xm2 and F is,
xm2 k2
=
F (k1 + m1 s2 )(k2 + m2 s2 ) + k2 m2 s2
and at 𝜔2 = k2 ∕m2 it can be shown that,
xm2 k2
= = −1∕k2
F (k1 − m1 𝜔 )(k2 − m2 𝜔2 ) − k2 m2 𝜔2
2

or
k2 xm2 = F.
From this result, we can see that the deflection of the spring exactly balances the input force. The allowable values for x2
will determine the range of acceptable spring values. The stiffer the spring k2 , the larger the value for m2 , and therefore the
softest acceptable spring value is chosen consistent with a maximum acceptable x2 .
This vibration absorber is only effective at one frequency, 𝜔2 = k2 ∕m2 , with resonant frequencies on each side of this
value. If the system parameters can appreciably vary, then an absorber may not be an effective solution for vibration
isolation. Many other techniques for vibration control and analysis are also available.7

6.4.2 Characterizing Sensors and Actuators

Insight into frequency response behavior is useful when characterizing systems and devices that act as sensors, being
those that monitor a system, and actuators, being those that impose a condition on a system. An ideal sensor draws zero
power when interacting with a system and conveys a signal that is directly proportional to a physical quantity of interest,
or y = K ⋅ u, where K is a constant value. Since most sensors are real physical devices the value of K is expected to vary
with frequency. The bandwidth is often defined by the frequency range over which the sensor would have G(𝜔) ≈ constant.

7 For a more detailed description of these methods, see the textbooks by Thomson [98] and Meirovitch [99].
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
6.4 Applications of Frequency Response 321

Many actuators can be described in a similar way, with the bandwidth now related to the frequency range over which the
device can deliver amplitudes that do not vary due to dynamic effects.
For either a sensor or actuator, of course, there may be variation in the amplitude ratio due to dynamics. Indeed, many
devices exhibit first- or second-order dynamic characteristics. It becomes evident, then, why so many devices have band-
width defined by a break or natural frequency. So for the case of a first-order system having time constant, 𝜏t , and break
frequency, 𝜔b = 1∕𝜏t , the bandwidth can be defined by this value, or a value less than 𝜔b if it is desired to define the band-
width as the range over which the frequency response is constant. This break or corner frequency for a first-order system
also corresponds to the point where the amplitude ratio has changed from the value a 𝜔 = 0 by 3-dB (so it is also called the
“3 dB point”).
A similar specification is given for second-order systems for which the natural frequency might form an approximate
upper limit for defining the bandwidth. More formally, the bandwidth is defined where the amplitude ratio has changed
by 3 dB from the value at 𝜔 = 0. It is clear, then, that a system model can be adopted to help guide design of a sensor or
actuator using the FRF to specify design parameters for key system components.

Example 6.7 Torque sensor response

A torsional rod is fitted with strain gauges to form a sensor that can measure an unknown applied torque, 𝜏1 , as shown
in Figure 6.23(a). Time varying torques are to be measured, 𝜏(t) = 𝜏1 sin(2𝜋ft), up to a certain frequency, fH . It is desired
to minimize the response to high frequency torque noise so the configuration in Figure 6.23(b) has been proposed,
where gears enable coupling a rotational damping element, B, and the original rod is now modeled by an ideal torsional
spring, K.
A static calibration test (with known constant torques applied) reveals the torsion rod can be modeled with an effective
spring constant K. Derive a transfer function that relates the torque measured at the strain gage location, 𝜏m , to the input
torque, 𝜏1 (t).
Solution
The bond graph in Figure 6.24(a) shows a single state, 𝜃5 , which represents the angular twist at the strain gage location.
The measured torque is then modeled by 𝜏m = K𝜃5 .

Gear ratio,

Unknown
torque

Gear ratio,
Strain gages

(a) (b)

Figure 6.23 (a) Torsion rod with strain gauges and (b) proposed sensing configuration with damper.

n1 τ3
·· n1 τ3
T R:B ··
τ2 ω3 T B
τ2 ω3
τ1 (t) ω2
Se 1 τ1 ω2
τ4 ω1 1
n2 τ4
ω4 ·· τ5 n2
C : 1/K ω4 ·· τ5
T K/s
θ̇5 T ω5
(a) (b)

Figure 6.24 (a) Bond graph with causality, first order. (b) Impedance bond graph form.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
322 6 Frequency Response and Impedance-Based Modeling

Thus, the desired transfer function can be found from a relation between 𝜃5 and 𝜏1 (t). First, the single state equation for
𝜃5 is,

𝜃̇ 5 = 𝜔4 ∕n2 = 𝜔2 ∕n2 = n1 𝜔3 ∕n2 = (n1 ∕n2 )𝜏3 ∕B = (n21 ∕n2 )𝜏2 ∕B = (n21 ∕n2 )(𝜏1 (t) − 𝜏4 )∕B
= −n21 ∕(n2 B)𝜏4 + (n21 ∕n2 B)𝜏1 (t)
= −n1 ∕(n2 B)𝜏5 + (n21 ∕n2 B)𝜏1 (t)
= −(n1 ∕n2 )(K∕B)𝜃5 + (n21 ∕n2 B)𝜏1 (t).

Thus,

𝜏t 𝜃̇ 5 + 𝜃5 = (n2 ∕K)𝜏1 (t),

where 𝜏t = (n1 ∕n2 )2 B∕K is the time constant. From this last form, transform into s-domain to get,
𝜃5 n ∕K 𝜏m n2
= 2 ⇒ = .
𝜏1 𝜏t s + 1 𝜏1 𝜏t s + 1
Contrast now with use of the impedance bond graph in Figure 6.24(b) to derive the transfer function, 𝜔5 ∕T1 . To find the
transfer function for the torque measured at the strain gauges, 𝜏m = K𝜃5 , we write,
[ ] [ ] [ ] [ ] [ ] [ ] [ ]
𝜏m K𝜃5 [ K ] 𝜔5 K 𝜔5 𝜔1 K 1 𝜔1 K
= = ⋅ = ⋅ ⋅ = ⋅ ⋅ = Y,
𝜏1 𝜏1 s 𝜏1 s 𝜔4 𝜏1 s n2 𝜏1 n2 s 1
where,
1 1 1 s
Y1 = = = = .
Z1 Z2 + Z3 B∕n21 + K∕(sn22 ) (B∕n21 )s + K∕n22
The result is a first-order form,
𝜏m K n2 ∕K n2
= ⋅ 2 =
𝜏1 n2 s 𝜏t s + 1 𝜏t s + 1
as found above.
The transfer function form shows the sensor bandwidth is directly set by the break frequency, 𝜔b = 1∕𝜏t = (n1 ∕n2 )2 (K∕B).
The case n2 = n1 = 1 shows that the amplitude ratio 𝜏m ∕𝜏1 is effectively a measure of the error in the measurement.
Extending the model to account for inertia in the gears will enable a design based on second-order system characteristics.
This will allow determination of how bandwidth depends on the damped natural frequency and thus the system parameters.

Example 6.8 Hydraulic pump with line filter

A hydraulic pump has an output flow of the form, Qp (t) = B + A sin(𝜔f t), in which sinusoidal “ripple” is superimposed on
a constant flowrate due to valves in the pump. Design a tank system to reduce this unwanted ripple by a factor of 10.
Solution
A hydraulic filter can be constructed as shown in Figure 6.25, where R2 is a fixed orifice, and R1 and C are to be chosen.
Assuming a flow source model for the pump, a bond graph model for this system is given in Figure 6.26(a) along with an
impedance form in Figure 6.26(b).

Figure 6.25 Noisy hydraulic pump with accumulator and load.

Accumulator

Pump
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
6.4 Applications of Frequency Response 323

C 1/sC

1 R 1 R1

Sf 0 R 0 R2
Qp Qo Qp Qo

Figure 6.26 (a) bond graph and (b) impedance bond graph.

If the modeling elements are taken as linear (this will only be approximately true for small deviations in flow), the transfer
function between the outlet flow Qo and the pump flow Qp can be derived by first noting, Yp = Yo + Ya , where Ya = 1∕Za ,
with Za = R1 + Cs being the accumulator impedance and Yo = 1∕R2 .
Qo R1 Cs + 1 𝜏 s+1
= = 1 ,
Qp (R1 + R2 )Cs + 1 𝜏2 s + 1
where 𝜏1 = R1 C and 𝜏2 = (R1 + R2 )C. At steady-state, the outlet flow is
[ ]
Qo = G(0)B + ℜ AG( j𝜔)e( j𝜔t+𝜋∕2) = G(0)B + A|G( j𝜔)| sin(𝜔t + 𝜋∕2).

In order to reduce the ripple by a factor of 10 we want |G( j𝜔)| = 0.1. An asymptotic plot of |G( j𝜔)| is approximated
as follows:

–20

High frequency
attenuation
–20

This plot implies that 𝜏2 needs to be at least 1 decade larger than 𝜏1 . This ensures that the attenuation is 20 log |G| < −20,
or |G| < 0.1. Setting 𝜏2 = 10𝜏1 results in R1 + R2 = 10R1 so R1 = R2 ∕9. The value of the capacitance is set to ensure that the
pump flow ripple frequency 𝜔f lies within the full attenuation region of the frequency response. The larger the value of C,
the farther 1∕𝜏1 and 1∕𝜏2 are shifted to the low-frequency end of the response. Since larger values of C imply larger tanks
and higher costs, the minimum value of C is chosen such that 1∕𝜏1 = 1∕R1 C = 𝜔f and C = 9∕(R2 𝜔f ).

6.4.3 Frequency Response Measurement and Transfer Function Approximation

Experimentally determined frequency response functions can be used to approximate a system transfer function.
This approach to model development might be motivated by a need to: (i) evaluate a theoretical model and/or use physical
testing results to estimate key physical parameters, or (ii) formulate a model of a system that can be time consuming or
impossible to develop from first principles. The latter is a type of system identification, which refers to an extensive field
comprised of a wide range of methods and approaches [100–102]. Along with the context and scope of this chapter, many
of the approaches assume the system behaves linearly for the range of applied inputs and frequencies.8 An experimentally
determined frequency response and associated transfer function model of a system can also represent a “snapshot” of the
system at a given time and under prescribed operating conditions. In this way, it can serve as a baseline for comparisons
made in the future as the system undergoes changes due to wear, degradation, or even retrofit. The primary goal and
intended scope of the following discussion is to relate practical considerations in measuring frequency response and

8 Linear transfer functions that have amplitude-dependence are referred to as describing functions (DF). DFs have been used extensively in the
past to study nonlinear response behavior and control design. The books by Graham and McRuer [103] and Gelb and Vander Velde [104] provide
a comprehensive overview.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
324 6 Frequency Response and Impedance-Based Modeling

methods used in system identification to the topics introduced in this chapter. The reader is referred to the cited works for
a thorough review of principles and methods in system identification.

6.4.3.1 Measuring Frequency Response

A requirement for measuring a system’s frequency response is that any inputs can be excited such that relevant and measur-
able output response(s) provide sufficiently rich information about the system dynamics over a defined region of operation.
Providing a gentle push to observe the behavior of a structure may be good enough in some cases while a precise impulse
force is required in others. The methods chosen for supplying inputs and measuring outputs must cover a system’s band-
width of interest. Similar to how assumptions are made when building theoretical models, there are implied assumptions
about how a system will behave subject to the types of inputs it is exposed to during its life of operation. Experimental
testing for these purposes requires many considerations. While there are some common aspects, specialized methods are
common in given applications (e.g., modal analysis of structures [90]) and, in some disciplines, frequency response testing
techniques are adopted as part of standards qualification of components (see, e.g., [105, 106]). Experimentally determining
a frequency response function generally requires merging not only domain knowledge on the system under study but also
measurement and instrumentation, signal processing, and system identification theory.
The simplest way to measure frequency response is by using a swept sinusoidal input, or swept-sine test. During such
a test, input and output signals are measured and used to determine the magnitude and phase response over a range of
frequencies of interest:
y |
̂ i )| = o |
|G(𝜔 , (6.20)
uo ||𝜔=𝜔i
̂ i) = [ 2𝜋
𝜙(𝜔 ] (units of rad), (6.21)
t̂m2,𝜔i − t̂m1,𝜔i

where the “ ̂ ” symbol indicates that these are estimates based on measurements and t̂m,𝜔 is an estimate of the time at
which a selected measure m of the signal is made at frequency 𝜔. Typical measures include peak values and level crossings.
When a linear system as in Figure 6.27(a) is excited by a sinusoidal input, u(t) = uo sin 𝜔t, an output will also be a sinusoid
with the same frequency but with a modified magnitude and phase shift that depend on the forcing frequency 𝜔, according
to equations 6.20 and 6.21. This now familiar concept of frequency response is illustrated in Figure 6.27(b).
An advantage of swept-sine testing is the inherent simplicity, and a direct frequency-by-frequency excitation of the system.
The input magnitude can be adjusted as needed as the frequency is changed to make sure there is sufficient output sig-
nal. It is sometimes also necessary to avoid driving the system into an inoperable region or to reduce the output during
a resonance. While it can be time consuming to generate data across the frequency range of interest, especially when the
system has a very low-frequency bandwidth, sometimes only a few points are needed to capture key information for making
approximations. For example, there may be only a few points needed to identify a low-frequency gain or a high-frequency
slope so that a break frequency can be estimated.

Example 6.9 Testing an unknown electrical component

A swept-sine test on the circuit shown in Figure 6.28(a) has produced the frequency response data shown below and plotted
in Figure 6.28(b). The input voltage amplitude was kept constant at Vin = 1 V as frequency was varied. The resistance was
measured and found to be R = 81.4 kΩ. Estimate the value of the capacitance in this circuit, C.

1
0.5

System 0
Time,
–0.5
–1
0 0.5 1 1.5

(a) (b)

Figure 6.27 (a) System input–output sinusoidal signals. (b) Relation of input and output magnitude and phase for typical linear
response, where phase is estimated based on zero crossings.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
6.4 Applications of Frequency Response 325

–10

–20
Vin Vout
–30

–40
101 102 103 104
f, Hz
(a) (b)

Figure 6.28 (a) Circuit under test. (b) Measured swept-sine frequency response data.

f (Hz) 10.0 20.0 50.0 100. 200. 400. 500. 1000. 2000. 5000. 10000
Vout (V) 0.12 0.36 0.626 1.180 2.50 4.15 4.62 5.75 6.27 6.58 6.64

Solution
For this relatively basic circuit, a model comparison makes it possible to determine the break frequency at the intersection
of the low-frequency asymptote with the high-frequency asymptotes, and Figure 6.29(a) shows a value of fb ≈ 450 Hz.
In this case, we are able to compare directly with a structured model as provided in the impedance bond graph of
Figure 6.29(b).
This impedance bond graph can be used to derive a TF in the form,
[ ][ ]
Vout Vout iR sRC
= = .
Vin iC Vin sRC + 1
One should recognize that there is a derivative factor in the numerator, a distinction of the high-pass filter. This effect also
introduces the upward slope in the magnitude function and cancels the high frequency downward slope, so the magnitude
response “flattens” and passes high-frequency inputs. Since 𝜏t = RC for this system (from the characteristic equation), we
can use the approximate value of 𝜏t to estimate the capacitance as, C = 1∕(2𝜋𝜏t R) ≈ 4.5 nF. Confirmation can be made by
plotting the magnitude plot for the model with the data.

Swept-sine tests are conceptually straightforward. However, implementing and conducting these tests can be time
consuming compared to other approaches. Depending on how many values of the amplitude are measured at different
frequencies, collecting the data may require long periods. Accurately generating harmonic signals can also introduce
unique challenges in a given system. The means for generating and measuring forces, torques, pressures, and other types

–10
Vout
Vin –20 C:C R:R

–30 VC iC VR iR

Vin Vout
–40 1 0
101 102 103 104 i=0
f, Hz
(a) (b)

Figure 6.29 (a) Using low and high frequency asymptotes to estimate time constant at break frequency and (b) impedance bond
graph.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
326 6 Frequency Response and Impedance-Based Modeling

Triangle pulse Chirp Random

1 1 50

0.8
0.5
0.6
0 0
0.4
–0.5
0.2

0 –1 –50
0 0.02 0.04 0.06 0.08 0.01 0 0.1 0.2 0.3 0.4 0.5 0 0.02 0.04 0.06 0.08 0.01
Time, sec Time, sec Time, sec

10
0 0 0
–10
–20 –20
–20
–40 –40 –30
–40
–60 –60
–50
–80 –80 –60
102 102 102
Frequency, Hz Frequency, Hz Frequency, Hz

Figure 6.30 Three typical test signals used to excite a system over a wide range of frequencies: impulse/triangular, chirp, random
(with bandwidth limit).

of physical quantities can be costly and time consuming. Unanticipated effects can be induced in the system response if
care is not taken in the design and realization of testing hardware. One consequence can be to introduce higher-order
harmonics which can corrupt a swept-sine experiment. In such cases, estimating an amplitude ratio cannot simply be
done by comparing peak values as implied in Figure 6.27(b). Higher-harmonics may arise not only because the input is
not purely harmonic but also because of nonlinear effects in a system. Indeed, nonlinear effects will induce a response at
higher-order harmonics even when the input is purely harmonic.9 Despite these difficulties, swept-sine testing continues
to be a viable option. The use of spectral analysis techniques (based on digital Fourier techniques) enable processing of the
data, regardless of anharmonic features.
Exciting a system’s frequency response across a wide range of frequencies can also be accomplished by using input sig-
nal types having a broad frequency content, in contrast to forcing at a single frequency during swept-sine testing. Among
many options used are chirps (or sweep signals), transient impulses, and random signals with specified broad-band or
shaped-spectral characteristics. Figure 6.30 shows typical time and frequency domain characterizations of these three sig-
nals as examples. The lower graphs are plots of the power spectral density of each test signal. These examples convey
the extent to which each of these signals delivers a relatively constant magnitude across a frequency range of interest.
While ideal impulses and “white” random noise both have constant power across all frequencies, both are mathematical
idealizations that cannot be practically realized.
A triangle impulse is shown in Figure 6.30 because it is a fairly good representation of a type of impulse signal that can be
practically realized. See, for example, the actual force signals measured for two types of impulse hammer strikes shown in
Figure 6.31, which are more reflective of a triangular pulse than a step pulse. Impulse hammer tip hardness is commonly
adjusted in order to change how the “energy” from the force is delivered across a frequency range of interest. Impulse
testing is commonly used when testing a broad range of physical systems [90, 107–109].
Analysis of the response data from transient and random forcing experiments requires Fourier analysis and more specif-
ically techniques that rely on digital Fourier transform analysis. A digital Fourier transform (via the Fast (digital) Fourier
Transform, FFT) provides a basis for estimating the spectral density functions for a system input, u(t), denoted Suu (𝜔),

9 Consider how the output from a basic nonlinear function, y = u2 , for u = sin 𝜔t, results in y(t) = [sin 𝜔t]2 = sin2 (𝜔t) = (1 − cos(2𝜔t))∕2,
showing a harmonic response with frequency 2𝜔t.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
6.4 Applications of Frequency Response 327

2.0 4.0
Force Soft hammer tip Force Hard hammer tip
signal signal
Volts Volts

–2.0 –4.0
0.0 0.02 0.04 0.06 0.08 0.0 0.002 0.004 0.006 0.008 0.010
Time, sec Time, sec

2e–02 2e–02

2e–05 2e–05
2.0 1024 32.0 16384
Frequency, Hz Frequency, Hz

Figure 6.31 Impulse induced by an impact hammer striking a cantilevered beam showing difference between a soft tip (left)
compared to a relatively hard tip (right) on the shape of the impulse (upper) and the corresponding frequency content (lower).

and output, y(t), denoted Syy (𝜔). The notation in the subscripts denotes these as auto-spectral density functions, which are
related in the fundamental linear relationship,

Syy (𝜔) = |G(𝜔)|2 Suu (𝜔), (6.22)

where,
Suy (𝜔)
G(𝜔) = (6.23)
Suu (𝜔)
and Suy (𝜔) is the cross-spectral density function between u and y. These relations provide the theoretical basis for making
̂
an estimate of the transfer function, G(𝜔), denoted G(𝜔) from estimates of indicated spectral density made using digitized
signals. Practically, these estimates are based on one-sided spectral density functions, the details of which can be found in
Bendat and Piersol [110, 111].10
To illustrate how these methods are used, consider the response of a standard linear second-order system,
y 𝜔2n
=
u s2 + 2𝜁𝜔n s + 𝜔2n
letting 𝜁 = 0.1 and 𝜔n = 2𝜋(75) rad/sec. For impulse testing, a triangular pulse is selected to deliver a constant power spec-
tral density magnitude over a relevant region of the system bandwidth of interest. Here, u(t) is chosen with T = 0.005
seconds, since this will give a power spectral density that only begins to dip (to a zero) at about 400 Hz (see, e.g., [107]).
The response y(t) is shown plotted along with u(t) in Figure 6.32(a), and the power-spectral density of u(t) is shown in
Figure 6.32(b). The result from using equation 6.23 to estimate the frequency response of the system transfer function is
shown in Figure 6.32(c), which compares very well for this simulated case.
It should be mentioned that physically generating test signals in a specific system requires specialized equipment and
knowledge of how to measure and analyze the data in order to construct accurate frequency response function measure-
ments that can then be used for system identification. There is an extensive literature base covering the background and
specific techniques used in measuring and estimating frequency response functions. For example, often the estimation of
power spectra requires averaging the results from many tests in order to achieve valid statistical estimates, especially when
random inputs are used. Reference textbooks such as those by Bendat and Piersol [110, 112] provide an extensive introduc-
tion to these methods and techniques. Experimental modal analysis of mechanical structures is an application area where
there has been considerable attention given to both measurement of frequency response functions and system identifica-
tion [90, 113]. Doebelin’s review of experimental and theoretical modeling, while dated, provides an excellent overview of
how these methods are applied to a variety of physical systems [114].

10 These functions can be practically computed using contemporary functions available in both open-source and commercial software packages.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
328 6 Frequency Response and Impedance-Based Modeling

1
Response, y

Input, u
–1
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1
Time, sec
(a)

0
–20 Spectrum for triangular input
–40
–60
–80
101 102 103
Frequency, Hz
(b)

20
0 Actual
–20
–40
Estimated
–60
1 102 103
10
Frequency, Hz
(c)

Figure 6.32 Example results illustrating estimation of experimental frequency response for a system transfer function from input
and output spectral density functions. (a) Plots of input and response signals, (b) spectrum for triangular input, and (c) comparison of
actual and estimated transfer function.

6.4.3.2 Approximating a Transfer Function

Assuming frequency response functions have been suitably estimated for a system, it is possible to construct approxima-
tions of system transfer functions. There are at least two different takes on this, depending on whether the emphasis is on
parameter estimation or system identification. If there is a reasonably good model of the system “structure,” then param-
eter estimation can follow traditional methods for fitting to measured data. If there is no model (“gray” or “black” box)
then system identification is needed and more insight is needed into how the model itself should be formulated. The latter
may be more data intensive, especially when it is desired to make comparisons with existing computational models. This
is the case, for example, with complex mechanical structures that require experimentally measured frequency response
functions for verifying and refining computational models and parameter data. The classic work of Ljung [100] forms a
popular reference for contemporary studies in this area of data-driven model development.
As shown earlier in this chapter, a desired transfer function can be derived and used to determine a theoretical frequency
response function. Comparison with frequency response test measurements can be used to estimate parameter values or to
determine whether the system has changed significantly over time. One way to make these comparisons is to use the basic
factors discussed in Section 6.2 and their associated parameters (time constants, damping ratios, etc.). Alternatively, general
functional approximations can be made to the overall frequency response function. To illustrate the former approach, con-
sider the systems shown in Figures 6.33(a) and (b). The measured frequency response data represent averaged amplitude
ratios between the angular velocity, 𝜔, measured using a tachometer and the controlled input current, is (t). The current was
supplied from a current amplifier using a random signal generator. The data points were selected from plots of estimated
spectral density functions averaged from 256 measured sample realizations.
The two systems differ by the addition of a 0-junction in case (b), which effectively decouples the motor rotational inertia
from that of the load. This is illustrated by the bond graphs in Figure 6.34. The tachometer is modeled as an ideal effort
source to indicate no torque is imposed in measuring 𝜔. Also, since current is a specified input, no electrical elements are
included in the motor model.11
11 If the model was going to be used to determine the back voltage on the current amplifier, then both motor inductance and resistance should
be included.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
6.4 Applications of Frequency Response 329

Rigid 20
shaft Tachometer 0
dc
motor –20
Inertia, J Measured –40
–60
100 101
Frequency, f, Hz
(a)

20
Spring-coupling
Tachometer 0
dc
motor –20

Inertia, J Measured –40

–60
100 101
Frequency, f, Hz
(b)

Figure 6.33 Schematic descriptions and measured frequency response data that relates angular velocity to drive current for a dc
motor that is (a) rigidly coupled versus (b) compliantly coupled with a spring to a load inertia.

I:J I:J C : 1/K I:J

ḣ ḣm ωm τK θ̇K ḣJ

τm τ ≈0 τm τ ≈0
Sf G 1 Se Sf G 1 0 1 Se
is ωm ω is ωm ω

τB τBm τB

R:B R:B R:B

(a) (b)

Figure 6.34 (a) Model proposed for rigidly coupled inertia indicates the first-order system. (b) Model proposed for compliantly
coupled inertia indicates the third-order system.

For relatively simple and low-order systems, it is often sufficient to use the basic factors introduced in Section 6.2 when
making these approximations, in a manner such as in Example 6.9. Depending on the requirements, the values for key
parameters in a given basic factor can be either graphically approximated or determined using a numerical curve fit-
ting process. There are many curve fitting tools available to aid the latter process. The results shown in Figure 6.35, for
example, were conducted by direct curve fitting for case (a). Then this model was used to initiate an iterative process to
fit the model in case (b). It is possible to extract factors, for example, so that each can be dealt with individually. This is a
simplified version of the approach used in interactive modal fitting software tools often applied to complex structural test
results [90].
In this approach to system identification, care should be taken to ensure that the system is minimum phase [115]. A system
that is not minimum phase is referred to as non-minimum phase and is characterized by having a transfer function with
zeros in the right-hand plane. The relevance to the discussion here is that a non-minimum phase system may have the
same magnitude function plot as a minimum phase system, but the numerators will be different. The difference would be
evident in the phase function, however. For example, a system with G1 (s) = (s + 2)∕(s + 1) will have the same magnitude
function as G2 (s) = (s − 2)∕(s + 1). The phase functions will be different, however, with G1 (s) exhibiting non-minimum
phase behavior. More will be said about non-minimum phase systems in Chapter 7. When approximations of transfer
functions are made from the magnitude function alone, one should first make sure the system is minimum phase before
interpreting trends based on basic factors. Estimating transfer functions using measured frequency response is a technique
commonly introduced in control system modeling and analysis [53, 61].
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
330 6 Frequency Response and Impedance-Based Modeling

Approximated
10 20

0
0
–10
Approximated –20 Approximated
–20
–40
–30

–40 –60
100 101 100 101
Frequency, f, Hz Frequency, f, Hz
(a) (b)

Figure 6.35 Results from approximating measured frequency response functions for the systems in Figure 6.33(a) and (b) with (a) a
single first order factor and (b) a first- and a second-order factor.

As systems become more complex, the methodology may require more sophisticated methods. In such cases, it is com-
mon to adopt polynomial fitting and other specialized techniques to construct transfer function models. Many of the
approaches are discussed by Ljung [100], Tangirala [102], and Pintelon and Schoukens [101]. Adoption of the methods
as described in these references, which emphasize the mathematical basis and signal processing methods, may be neces-
sary when the system complexity is high and the use and insights gained from the first-principles physical system modeling
perspective discussed here cannot yield adequate results.

6.5 Two-Port Models and Transmission Matrices

As discussed in previous sections, a one-port connection to a basic element or to a network of elements is defined
by the relationship between e and f variables at that port, as in Figure 6.36(a). This transfer function relationship
takes the form of an impedance, Z = e∕f , or driving-point impedance, or driving-point admittance, Y = f ∕e. A system
can have many ports at which power flow connections can be made with the environment and with other systems.
A two-port representation as shown in Figure 6.36(b) shows a basic extension of this concept that has practical sig-
nificance in many areas. Note that as done for any general system (recall in Chapter 1), power flow is defined as
positive into the system at both power bonds. Relationships between the different variables on both ports may take
the form of impedances (effort/flow), admittances (flow/effort), or to a suitable dimensionless ratio of interest, such
as e1 ∕e2 .
In the present context, we are concerned with two-ports composed of linear physical elements with time-invariant prop-
erties, in the same way impedance relations were defined earlier in this chapter. This section describes how two-port
models can be formulated from bond graph descriptions, alluding to methods commonly used in electrical circuit the-
ory [37, 116, 117]. These two-port systems will be composed of passive R, C, and I elements, as well as ideal transformers
and gyrators. Two-port models can also be used to represent active elements, which are commonly used in some circuit
network models [37]. For now, the intent is to show how these representations can be integrated into impedance bond
graph models, which can be a preferred way for modeling certain types of systems.

e e1 e2
one-port two-port
f f1 f2

(a) (b)

Figure 6.36 Basic (a) one- and (b) two-port systems with power bonds defined with positive power flow into the systems.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
6.5 Two-Port Models and Transmission Matrices 331

A two-port model is defined by four relations, or two-port parameters, which form the elements of a 2 × 2 matrix relating
the port effort and flow variables.12 The two-port parameters are ratios of polynomial functions having real coefficients
taking the form, N(s) = A(s)∕B(s), since the models are formed assuming lumped-parameter models are suitable for the
system study at hand. For example, the impedance form is given as,
[ ] [ ][ ] [ ]
e1 z z f1 f
= 11 12 =Z 1 , (6.24)
e2 z21 z22 f2 f2
where Z is the impedance matrix. Two-port models are a useful way to represent systems and devices and can be used to
systematically compose complex descriptions of interconnected physical subsystems. The two-port representations are com-
patible with frequency domain descriptions of gain, bandwidth, impedance, sensitivity, and stability and have been adopted
in a wide range of applications. These include modeling electrical and electronic systems, acoustical and transducer sys-
tems, and power transmission analysis in many forms (hydraulic, electrical, torsional, etc.). A two-port description may also
be required when integrating some existing models or experimentally determined approximations of impedance relations
for a subsystem component. Consequently, understanding how to integrate two-port representations within a bond graph
description of systems can be especially efficient and effective.

6.5.1 Definition of Linear Two-Port Elements

We categorize two-ports into causal and acausal forms, which makes it convenient in choosing which two-port to use in a
given application study.

6.5.1.1 Causal Two-Port Matrices

It can be convenient to adopt a two-port matrix form that relates the two causal output variables as a function of the
two causal input variables. This is the form used for the classical two-port (open-circuit) impedance matrix defined in
equation 6.24, where we define the driving-point impedances,
e |
z11 = 1 || = driving-point impedance,
f1 |f2 =0
e2 ||
z22 = = driving-point impedance
f2 ||f1 =0
and the transfer impedances,
e |
z12 = 1 || = transfer impedance,
f2 |f1 =0
e2 ||
z21 = = transfer impedance.
f1 ||f2 =0
These are commonly referred to as “open-circuit” since they are defined with the flow variable at one-port forced to zero (an
open circuit). Consequently, when deriving or measuring these relations at one-port, it is necessary to set proper conditions
on the other port. Recall that an “open-circuit” can be represented by an ideal flow source where the flow variable is set
to zero. Likewise, a “short-circuit” can be modeled by an ideal effort source with the effort set to zero. These dual model
concepts are illustrated for a two-port impedance in Figure 6.37.
Following on from the defined impedance form, a set of commonly used two-ports can be categorized according to the
four different causal combinations imposed at the two ports as shown in Table 6.1, again with power flow defined as positive
into the two-port. The impedance-based two-ports in this table (as in Figure 6.37) adopt causal strokes only to indicate the
causal definition for each case. To be clear, causality is not used in the same way for impedance-form models as it is used
to identify state variables for generally nonlinear, time-domain state equation formulations. Causal arguments are useful
in defining the impedance forms in Table 6.1 and in choosing which forms are convenient to use when formulating more
complex models.

Figure 6.37 Ideal flow and effort sources can be used to impose open-circuit (f = 0) and e e=0
short-circuit (e = 0) conditions, respectively, as needed in two-port definitions. Sf Z Se
f =0 f

12 The two-port parameters described here have also been referred to in the past as four-pole parameters or even four terminal parameters,
referring to the genesis of these methods in electrical circuit theory [114, 118].
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
332 6 Frequency Response and Impedance-Based Modeling

Table 6.1 Two-port elements with causal inputs and associated matrix designations. Note that causal strokes are not generally
needed for impedance bond graph forms and are used here only to emphasize the causal definition of each two-port. For example,
the impedance form has flow inputs at each port.

e1 e2 e1 e2
Z e1 z11 z12 f1 Y f1 y11 y12 e1
f1 f2 = f1 f2 =
Impedance
e2 z21 z22 f2 f2 y21 y22 e2
Admittance

e1 e2
     e1 e2     
ZY e1 h11 h12 f1 YZ f1 g11 g12 e1
f1 f2 = f1 f2 =
f2 h21 h22 e2 Adpedance
e2 g21 g22 f2
Immittance

The matrices for the ZY and YZ two-ports are also referred to as “hybrid” matrices in electrical network theory and may
also be designated by H (immittance) and G (adpedance), respectively.13 To avoid confusion with other multiport bond
graph elements (e.g., we use G for a gyrator in this text), the bond graph symbols shown in Table 6.1 may also be adopted
in this text. The two-port matrices, however, use nomenclature commonly adopted in the literature.

6.5.1.2 Transmission or Transfer Matrix

Another two-port form that is very convenient in model formulation is the classical acausal transmission matrix, M, which
relates the port variables on a two-port system by,
[ ] [ ][ ] [ ]
e1 m11 m12 e2 e
= =M 2 . (6.25)
f1 m21 m22 f2 f2
The M is the forward transmission matrix, but it is also sometimes referred to as the “ABCD” or transfer matrix (with
A = m11 , B = m12 , C = m21 , and D = m22 ). Reversing the direction of the bonds defines a backward transmission matrix, B.
The transmission two-ports have a (historical) sign definition where power flow is positive into one-port and positive out
the other, as shown in bond graph form in the following:

e1 e2
M
f1 f2

where M is adopted here as the bond graph symbol for a transmission two-port. Note how this is distinct from the impedance
forms in Table 6.1 where power is defined positive into both ports.
The transmission matrix relates the power variables at one-port to those at the other. Along with the sign convention
described, this form makes composing models of interconnected subsystems efficient. For example, for n subsystems con-
nected in series as shown in Figure 6.38, the composite transmission matrix is readily found by direct matrix multiplication,
[ ] [ ]
e1 e
= M1 M2 M3 · · · Mn−1 Mn n . (6.26)
f1 fn
So, for example, for two systems connected in series and shown in bond graph form,

e1 e2
Ma Mb
f1 f2

the resulting transmission matrix becomes,

[ ]
Aa Ab + Ba Cb Aa Bb + Ba Db
M = Ma Mb = .
Ab Ca + Cb Da Bb Ca + Da Db

e1 e2 e3 en−2 en−1 en
M1 M2 ··· Mn−1 Mn
f1 f2 f3 fn−2 fn−1 fn

Figure 6.38 A series connection of n subsystems represented by two-port transmission matrices.

13 See, for example, [1, 37, 119].

 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
6.5 Two-Port Models and Transmission Matrices 333

Table 6.2 Table for conversion of linear two-port parameters. Find the two-port matrix on left column from a given
two-port matrix in top row. Z = impedance, Y = admittance, H = immittance, G = adpedance, M = (forward) transmission
(or transfer), B = backward transmission.
– Z Y H G M B

y22 y12 |H| h12 1 g12 m11 |M| b22 1

+z11 + z12 + − + + + − + + + +
|Y| |Y| h22 h22 g11 g11 m21 m21 b21 b21
Z
y21 y11 h21 1 g21 |G| 1 m22 |B| b11
+z21 + z22 − + − + + + + + + +
|Y| |Y| h22 h22 g11 g11 m21 m21 b21 b21

z22 z12 1 h12 |G| g12 m22 |M| b11 1

+ − +y11 + y12 + − + + − + −
|Z| |Z| h11 h11 g22 g22 m12 m12 b12 b12
Y
z21 z11 h21 |H| g21 1 1 m11 |B| b22
− + +y21 + y22 + + −  + − + − +
|Z| |Z| h11 h11 g22 g22 m12 m12 b12 b12

|Z| z12 1 y12 g22 g12 m12 |M| b12 1

+ + + − +h11 + h12 + − + + + +
z22 z22 y11 y11 |G| |G| m22 m22 b11 b11
H
z21 1 y21 |Y| g21 g11 1 m21 |B| b21
− + + + +h21 + h22 − + − + − +
z22 z22 y11 y11 |G| |G| m22 m22 b11 b11

1 z12 |Y| y12 h22 h12 m21 |M| b21 1

+ − + + + − +g11 + g12 + − + −
z11 z11 y22 y22 |H| |H| m11 m11 b22 b22
G
z21 |Z| y21 1 h21 h11 1 m12 |B| b12
+ + − + − + +g21 + g22 + + + +
z11 z11 y22 y22 |H| |H| m11 m11 b22 b22

z11 |Z| y22 1 |H| h11 1 g22 b22 b12

+ + − − − − + + +m11 + m12 + +
z21 z21 y21 y21 h21 h21 g21 g21 |B| |B|
M
1 z22 |Y| y11 h22 1 g11 |G| b21 b11
+ + − − − − + + +m21 + m22 + +
z21 z21 y21 y21 h21 h21 g21 g21 |B| |B|

z22 |Z| y11 1 1 h11 |G| g22 m22 m12

+ + − − + + − − + + +b11 + b12
z12 z12 y12 y12 h12 h12 g12 g12 |M| |M|
B
1 z11 |Y| y22 h22 |H| g11 1 m21 m11
+ + − − + + − − + + +b21 + b22
z12 z12 y12 y12 h12 h12 g12 g12 |M| |M|

When relating the different forms presented, make sure to note how the transmission matrix has a sign change
since one bond has power “into” the two-port while the other has power “out of ” the two-port. Again, this is espe-
cially convenient in conventional bond graph modeling where a sign convention often flows across a system in one
direction. This sign distinction is implicit to the relationships between all the two-port parameters presented in
Table 6.2, which summarizes all the two-port element parameters and provides guidance in converting between the
different forms.
A transmission matrix form can be directly derived from the basic element relations and study of a bond graph (or
network) structure. For example, consider two basic model structures commonly used in two-port modeling, the series
impedance and shunt admittance shown in Figures 6.39(a) and (b).
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
334 6 Frequency Response and Impedance-Based Modeling

Z Y

e f e f

1 2 1 2
1 0
(a) (b)

Figure 6.39 Two basic model elements commonly used in two-port modeling. (a) Series impedance and (b) Shunt admittance.

For the series impedance, f = f1 = f2 and e1 = e + e2 = e2 + Zf2 , so,

[ ] [ ][ ]
e1 1 Z e2
= .
f1 0 1 f2
And for the shunt admittance, e = e1 = e2 and f1 = f + f2 = Y e2 + f2 , so,
[ ] [ ][ ]
e1 1 0 e2
= .
f1 Y 1 f2
Additional transmission matrices for other common model forms expressed in (impedance and admittance) bond graph
form are summarized in Table 6.3. These model forms can be very useful in deriving composite transmission matrix forms,
which can then be converted to any of the other forms given in Table 6.2. Note, however, that there are some cases where
one or more of the two-port forms may not exist. For example, the series impedance in Figure 6.39(a) has a transmission
matrix (as derived above), but a two-port impedance matrix does not exist. One way to see this is that in the conversion
from M to Z (see Table 6.2), there is a singularity in z21 = 1∕m21 , since m21 = 0. A similar situation arises for the shunt
admittance, as m12 = 0 in that case, so the two-port admittance does not exist. It may be possible to anticipate such cases
without relying on the table. For the series impedance, two independent flows could not be specified at the two ports,
indicating that an impedance form, e = Zf, is undefined. The same can be said for the shunt admittance which would
not permit a two-port Y since two efforts could not be independently specified. In these cases, however, it would be pos-
sible to define adpedance and immittance matrices. Once again, causality is not generally used in deriving impedance
forms but causal relations still provide guidance and can save time by helping to identify permissible relations in a given
application.

Example 6.10 Two-port relations for a spring–damper and R–C system

Consider the familiar spring–damper coupling and the analogous R–C circuit shown in Figures 6.40(a) and (b), respec-
tively. Confirm that these systems both have the bond graph structure shown in Figures 6.40(c) and then derive two-port
transmission matrices for each case which can be used in two-port analysis.
Solution
The spring and damper and RC elements are both connected with a common effort as reflected in the bond graph of
Figure 6.40(c). In each case, the flow relations will be f1 = fR + fC + f2 . Now, for the mechanical case, 𝑣1 = 𝑣b + 𝑣k + 𝑣2 ,

e1 e2
0
f1 f2

C
(a) (b) (c)

Figure 6.40 Examples for two-port matrix derivations in simple spring–damper and RC systems.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
6.5 Two-Port Models and Transmission Matrices 335

Table 6.3 Table of common model bond graph forms and associated transmission matrices
relating the effort-flow pairs a the 1 and 2 ports per equation 6.25.
Model bond graph form Transmission matrix, M
Z
 
1 Z
1 2
0 1
1
Series impedance
Y
 
1 0
1 2
Y 1
0
Shunt admittance
Z Y  
1+YZ Z
Y 1
1 2
1 0
Y Z  
1 Z
Y 1+YZ
1 2
0 1
Z1 Y Z2
 
1 + Y Z1 Z1 + Z2 + Y Z1 Z2
1 2
Y 1 + Y Z2
1 0 1
T-section
Y1 Z Y2
 
1 + ZY2 Z
1 2
Y1 + Y2 + ZY1 Y2 1 + ZY1
0 1 0
Π-section

where 𝑣k = sxk , where xk = Fk ∕k = F2 ∕k, and 𝑣b = Fb ∕b = F2 ∕b, so we can write,

[ ] [ ][ ] [ ]
F1 1 0 F2 F
= =M 2 , (6.27)
𝑣1 s∕k + 1∕b 1 𝑣2 𝑣2
where M is the transmission matrix. For this system two-port, the Z, H, and G matrices can be found, but the Y does not
exist. This can be seen in Table 6.2 where defining Y would require division by the element m12 = 0.
An alternative approach to deriving M is to combine the two elements into an equivalent shunt admittance, which add
at a zero junction so, Y = 1∕R + sC. The transmission matrix can then be found from Table 6.3. For the RC system, for
example,
[ ] [ ][ ]
V1 1 0 V2
= .
i1 sC + 1∕R 1 i2
As expected these transmission matrices are of the same form since these are analogous systems.

6.5.2 Characteristics and Properties of Two-Ports

When deriving and working with two-port models, it can be helpful to reference properties of two-port model relations.
First, according to the reciprocity theorem [37, 116, 120], two-ports composed of linear basic elements are reciprocal if:
f2 || f |
| = 1 ||
e1 |e2 =0 e2 |e1 =0
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
336 6 Frequency Response and Impedance-Based Modeling

Two-port Two-port

Two-port Two-port

(a)

Two-port Two-port

Two-port Two-port

(b)

Figure 6.41 Illustration of two-port reciprocity theorem using two-port electrical networks and equivalent bond graph
representation.

as illustrated in Figure 6.41(a), or,

e2 || e |
| = 1 ||
f1 |f2 =0 f2 |f1 =0
as illustrated in Figure 6.41(b). As such, two-port Z and Y matrices are symmetrical; i.e., z12 = z21 and y12 = y21 , respectively.
As a consequence, the determinant of the transmission matrix for a reciprocal two-port is,
|M| = m11 m22 − m12 m21 = 1.
which should be clear from studying the relationship between Z and M in Table 6.2. In contrast, hybrid two-ports, G and H,
are skew symmetric, so, h12 = −h21 and g12 = −g21 . Further, their determinants, |G| = g11 g22 − g12 g21 = z22 ∕z11 = y11 ∕y22
and |H| = h11 h22 − h12 h21 = |G|−1 , are impedance ratios and thus dimensionless. These properties are a consequence of
how the models are formulated, using Kirchhoff-law type relations and system variables typically of the effort-flow type.14
In cases where there is physical symmetry, z11 = z22 or y11 = y22 , which implies m11 = m22 . Note thus that while sym-
metric passive two-ports are generally specified by only three unique parameters, the physically symmetric cases will only
require two. In mechanical systems, this symmetry might be referred to as space (or spatial) symmetry, while in electri-
cal systems, it is electrical symmetry. An example where there is physical symmetry is the torsional drive system from
Example 6.5, shown in Figure 6.42(a) along with its electrical circuit analog in Figure 6.42(b). Both of these systems have
physical symmetry, reflected clearly in the bond graph structure of Figure 6.42(c).
If a system is reciprocal, then you can take advantage of certain properties in developing as well as evaluating the two-port
model relations. For example, derived impedance matrices can be quickly checked to confirm correctness. Further, these
relations provide an additional sanity check on experimentally measured transfer functions. The following subsection
reviews ways of deriving and evaluating two-port relations.

6.5.3 Deriving Two-Port Model Relations

Formulating a two-port model requires determining the four two-port parameters, which are linear transfer functions
relating the effort-flow variables at the two power ports. These relations can be determined in several ways, including
by: (i) direct formulation using the basic impedance relations introduced in Section 6.3, (ii) adopting and/or combining
known two-port relations, and (iii) using transfer functions approximated from measurements. The relations in Table 6.2
can be used to find the specific form needed in a given application.

14 See Guillemin [116] for a discussion of this concept as it relates to electric circuits (reference page 81).
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
6.5 Two-Port Models and Transmission Matrices 337

Linear bearing Linear bearing

friction, b1 friction, b2

Shaft compliance, C = 1/k

Rotational
inertia, J

(a) (b)

R C R

1 0 1
(c)

Figure 6.42 Physical symmetry in analog mechanical systems. There is analogous mechanical and electrical symmetry in the
systems contained within the indicated dashed box in parts (a) and (b), which both have the corresponding bond graph in (c).

6.5.3.1 Basic Two-Port Formulation

An example of deriving a two-port model from basic impedance relations was demonstrated in Example 6.10. In this case,
a system is specified in isolation so two-port connections were clearly defined. The bond graph of Figure 6.40(c) can then
be used to drive the two-port relations as repeated in Figure 6.43.
This example shows the results as a transmission matrix, which can be converted if necessary and if possible into one of
the other forms from Table 6.2. Transmission matrices can also be found using common forms summarized in Table 6.3.
For example, the systems in Figure 6.42 both feature a Π-section, where Z = R and Y1 = Y2 = sC, so the result from Table 6.3
can be used to give the transmission matrix as,
[ ] [ ]
1 + ZY2 Z 1 + sRC R
M= = .
Y1 + Y2 + ZY1 Y2 1 + ZY1 sC(2 + sRC) 1 + sRC
It is sometimes desirable to derive two-port relations by isolating part of an existing system. In doing so, the two-port
relations will dictate where the connections are made in partitioning an isolated two-port from the rest of a system.
In some cases, this may be relatively easy to do. Such partitioning is illustrated for analogous electrical and hydraulic
systems in Figure 6.44.
Most often, we need to rely on insight and knowledge about a system along with some idea about how the two-port
relations derived will be used in order to derive the two-port relations. Chua et al. [37] propose two ways to think about
how to identify points where a system can be partitioned for two-port analysis and model formation in electrical networks.
A “soldering iron entry” into a system refers to joining two terminals to two nodes, as in attaching a voltmeter to measure the
potential across those two points. A “pliers” approach refers to connections made by cutting branches to allow adding and
inserting a current (flow) source (or to make a current measurement). This preference ensures that the natural dynamics or

 
e1 e2 e1 e2 e1 1 0 e2
0 M sC + 1/R 1
f1 f2 f1 f2 f1 f2

(a) (b) (c)

Figure 6.43 (a) Basic RC bond graph and two-port with (b) transmission form symbol and (c) using explicit transmission matrix.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
338 6 Frequency Response and Impedance-Based Modeling

Pump
A′ B′ C′ D′

A′ B′ C′ D′

(a) (b)

Figure 6.44 Analogous (a) electrical and (b) hydraulic systems with dashed lines indicating points at which two-port subsystems may
be partitioned.

τ1 τ2 τL τw
Se Mc 0 ZL Sf
ω1 ω2 ωL ωw = 0
Rigid
shaft

(a) (b)

Figure 6.45 Rotational coupling and associated bond graph. (a) Mechanical schematic and (b) impedance bond graph.

initial condition response15 of a two-port system will not be influenced by how it is inserted into a system. These conceptual
constructs can also be interpreted as preferred ways for setting effort or flow “boundary conditions” at a given port so that
the order (as in number of independent states) of a two-port system will not change based on the imposed input. In other
words, the two-port dynamic states would maintain their causal form regardless of the causal nature of the connection. All
of this basically means that when a system is partitioned properly into two-port forms, the “two-port” elements will have
the same two-port dynamic structure regardless of the input specified. In addition, models formed this way can be more
easily reused.

Example 6.11 Dual-inertia rotational coupling with impedance load

The dual-inertia rotational coupling shown in Figure 6.45(a) (from Solved Problem A-3-6) is used to connect a torque source
𝜏1 (t) to a load characterized by a two-port impedance matrix ZL (s). The impedance formulation in Figure 6.45(b) shows a
bond graph representation. One side of the load is fixed to a wall, so 𝜔𝑤 = 0. Find relations that allow for quantifying the
torque imposed on the wall, 𝜏𝑤 , and the torque at the input to the load, 𝜏L , as functions of the input torque, 𝜏1 (t).
Solution
An impedance bond graph for the rotational coupling is shown in Figure 6.46. This bond graph is partitioned to indicate
three defined two-port subsystems, indicated by 
1 ,2 , and 3 , having the transmission matrices,
[ ] [ ] [ ]
1 Z1 1 Y 1 Z2
M1 = , M2 = , M3 = ,
0 1 0 1 0 1
respectively, where Z1 = B1 + sJ1 , Y = (B + K∕s)−1 , and Z2 = B2 + sJ2 . First, a composite transmission matrix is found by,
Mc = M1 M2 M3 .
Next, the load impedance two-port matrix, ZL , is converted to a transmission matrix, ML , using the relations from
Table 6.2, to allow building a total transmission matrix,
[ ] [ ] [ ]
𝜏1 𝜏𝑤 𝜏𝑤
= Mc ML = MT ,
𝜔1 𝜔𝑤 𝜔𝑤
where T in the subscript denotes the total transmission matrix. This relation can then be used in an expression with 𝜏1 and
𝜔𝑤 as inputs, which is an adpedance matrix, YZT , as in Table 6.1. Thus, we find,
[ ] [ ] [ ][ ]
𝜔1 𝜏1 gT11 (s) gT12 (s) 𝜏1
= YZT = .
𝜏𝑤 𝜔𝑤 gT21 (s) gT22 (s) 𝜔𝑤

15 Chua et al. [37] refer to this as “zero-input” response.

 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
6.5 Two-Port Models and Transmission Matrices 339

Figure 6.46 Impedance form of bond graph for dual-inertia rotational sJ1 sJ2
coupling, with vertical dashed lines marking partitions that define two-port
subsystems.
1 2 3

τ1 τ2
ω1 1 0 1 ω2

B1 1 B2

K/s B

From this relation, since 𝜔𝑤 = 0, we find the two transfer functions,

𝜔1 𝜏𝑤
= gT11 (s) and, = gT21 (s).
𝜏1 𝜏1
Now, in order to find the “interior” variable, 𝜏L , refer to Figure 6.45(b) to see we can use the load two-port impedance matrix
relation,
[ ] [ ] [ ][ ]
𝜏L 𝜔L z (s) zL12 (s) 𝜔L
= ZL = L11 .
𝜏𝑤 𝜔𝑤 zL21 (s) zL22 (s) 𝜔𝑤
However, we need a transmission matrix form so we can solve for 𝜏L , thus,
[ ] [ ] [ ][ ]
𝜏L 𝜏𝑤 mL11 (s) mL12 (s) 𝜏𝑤
= ML =
𝜔L 𝜔𝑤 mL21 (s) mL22 (s) 𝜔𝑤
and then,
𝜏L
𝜏L = mL11 𝜏𝑤 = mL11 (s) ⋅ gT21 (s)𝜏1 ⇒ = mL11 (s) ⋅ gT21 (s).
𝜏1
The transfer functions found provide the relations for the torque at the wall, 𝜏𝑤 , and the torque at the interconnection
between the coupling inertia J2 and the load, 𝜏L .

6.5.3.2 Coupling Element Transmission

In certain cases, it may not be possible to derive some of the two-port relations directly. For example, for a simple trans-
former, the relations are e1 = me2 and f2 = mf1 . Thus, a transmission matrix can be formulated directly,
[ ] [ ] [ ]
e1 m 0 e
= = 2 .
f1 0 1∕m f2
Neither an impedance or admittance matrix can be formed for an isolated transformer element since m12 = m21 = 0
(see Table 6.2). On the other hand, for a gyrator, it is possible to directly write an impedance matrix,
[ ] [ ][ ]
e1 0 r f1
e= = = Zf.
e2 r 0 f2
We usually consider both the transformer and the gyrator as ways to modify impedance, and so on their own are not
necessarily of interest as stand-alone two-ports. The following example shows application of a gyrator two-port.

Example 6.12 Field-excited dc machine model

A schematic for a separately- or field-excited dc machine is depicted in Figure 6.47(a), which has a value of rm that is
modulated by the induced field. Recall, for DC (PMDC) electrical machine (motor or generator), the electromechanical
transduction is modeled by, 𝜏g = rg ig and Vg = rg 𝜔g , where the ideal torque, speed, voltage, and current variables are related
by a constant parameter, rg . Use the bond graph shown in Figure 6.47(b) to develop a two-port transmission matrix model
of this system.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
340 6 Frequency Response and Impedance-Based Modeling

Bt + sJt Rg + sLg

rg
τt τg ·· Vg Vo
ωt 1 ωg G 1
ig io

(a) (b)

Figure 6.47 Field-excited dc electrical machine (a) electromechanical schematic and (b) impedance bond graph for generator.

Solution
Assume a constant voltage field source so that rg ≈ constant, as for a PMDC machine. This case can be modeled by the
bond graph shown in Figure 6.47(b), where power directions have been assigned assuming the machine is a generator. This
system is represented by a series connection of transmission elements, TM, as shown in bond graph form,

τt τm Vg Vo
Mt MG Ma
ωt ωm ig io

the resulting transmission matrix relating input and output becomes,

[ ][ ][ ]
1 Bt + sJt 0 rg 1 Rg + sLg
M = Mt MG Ma = .
0 1 1∕rg 0 0 1
Thus, the input–output relations are,
[ ] [ ] [ ][ ]
𝜏i Vo (Bt + sJt )∕rg rg + (Bt + sJt )(Rg + sLg )∕rg Vo
=M = .
𝜔i io 1∕rg (Rg + sLg )∕rg io
We can use relations from Table 6.2 to find other two-port parameters. for example, for impedance relations we find,
[ ] [ ]
m11 ∕m21 |M|∕m21 Bt + sJt rg
Z= = ,
1∕m21 m22 ∕m21 rg Rg + sLg
where we adopt the standard convention with power flow positive into both ports of the two-port impedance.

6.5.4 Interconnection of Two-Ports

One of the most common ways to interconnect two-port models is by cascading them as shown earlier in this section.
Table 6.4 summarizes the five possible ways that two-port elements might be interconnected along with their corresponding
composite relations. The cascade connections multiply as discussed earlier in Section 6.5.1 (e.g., see equation 6.26). Once a
given composite form is determined, the two-port form can be expressed in a related form using the relations in Table 6.2.
The composite relations in Table 6.4 are derived using the basic definitions and junction relations. For example, for the
Series–Series connection, we can write,
e1 = eA1 + eB1 = z11,A f1 + z12,A f2 + z11,B f1 + z12,B f2 .
Thus, the composite driving-point impedance is found by holding f2 = 0 to get,
e |
z11 = 1 || = z11,A + z11,B .
f1 |f2 =0
The remaining two-port relations (z12 , z21 , z22 ) can be derived in a similar manner to find the composite relation for the
Series–Series two-port impedance, Z. The other composite relations in Table 6.4 are indicated in the most convenient
two-port form (see defined two-port types in Table 6.1) for the corresponding interconnections. The composite relations
indicated can be proven using the same approach indicated here (left as an end of chapter problem on two-port intercon-
nections).
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
6.5 Two-Port Models and Transmission Matrices 341

Table 6.4 Summary of two-port interconnections and corresponding composite relations.

Connection Bondgraph Composite relation

A ··· B
Cascade 1 2 M = MA · · · MB

A
Series-Series 1 1 Z = ZA + ZB
1 2
B

A
Series-Shunt 1 0 H = HA + HB
1 2
B

A
Shunt-Series 0 1 G = GA + GB
1 2
B

A
Shunt-Shunt 0 0 Y = YA + Y B
1 2
B

6.5.5 Summary on Use of Two-Port Models

This section briefly reviews some areas where two-port models are convenient and thus commonly applied. Most applica-
tions of linear two-ports can be grouped into the following categories: (i) transfer problems – finding the effort or flow at a
downstream port in response to effort or flow at an upstream port, with known and ideal terminations on the two-port at
the downstream port; (ii) transmission – finding the power state at one-port in terms of the power state at the other port,
with unrestricted or specified port conditions; (iii) insertion – study the effect of inserting or retrofitting a two-port into
a system, such as in filtering and protection applications and where performance is measured in terms of the change in
power, effort, or flow after insertion relative to unaltered system.

6.5.5.1 Two-Port Analysis Problems

As mentioned earlier, certain type of problems lend themselves for analysis by two-port relationships. Consider a two-port
system excited by an effort or flow in Figure 6.48. A transfer analysis problem seeks to determine either the effort or the flow
at a location downstream of an upstream excitation. Typically, there would be specified conditions on both the upstream and
downstream connections, such as a source model and a load, respectively. A transmission analysis problem commonly refers
to those where the power state at one of the ports is of interest. Lastly, in an insertion problem, it is desired to understand
how the system response changes when a two-port is inserted or changed. Two-port insertion is usually done to improve
performance, to match conditions, or to filter and protect a load.
Many practical configurations for transfer analysis can take the forms shown in Figure 6.49(a) and (b). In case (a) of this
figure, the transfer function relation of interest is eL ∕ei while in (b), we would seek fL ∕fs .
Configuration (a) in Figure 6.49 shows a two-port impedance terminated by a one-port load impedance ZL . To begin
with, note the need to introduce a 0-junction to accommodate the fact that the two-port impedance is defined with power
flow positive into both ports. However, the attached load requires a sign change to indicate positive flow into the load
impedance. The 0-junction is used in this case so that eL = e2 but note that fL = −f2 . Now, we seek the transfer (function)
relation, e2 ∕es , where es (t) is an input effort. A Thevenin source model is formed by introducing the source impedance, Zs .
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
342 6 Frequency Response and Impedance-Based Modeling

Figure 6.48 Upstream and downstream definitions in two-port analysis.

Specified Two-port Output
input(s) conditions

‘upstream’ ‘downstream’

Two-port

Zs Ys

es fs es fs

ei (t) e1 e2 eL e1 e2 eL
Se 1 Z 0 ZL Sf 0 Y 1 ZL
f1 f2 fL fi (t) f1 f2 fL

(a) (b)

Figure 6.49 (a) Load-terminated Z with Thevenin source. (b) Load-terminated Y two-port with Norton source model.

From the two-port Z,

e1 = z11 f1 + z12 f 2 = z11 f1 − z12 e2 ∕ZL ,
where we used, f2 = −fL = −eL ∕ZL = −e2 ∕ZL . On the source side, e1 = es − Zs f1 . We can combine the two relations for e1
to find,
e + (z12 ∕ZL )e2
f1 = s .
z11 + Zs
Now, from the two-port relations,
es + (z12 ∕ZL )e2 −e
e2 = z21 f1 + z22 f2 = z21 + z22 2 .
z11 + Zs ZL
Now solve for e2 and simplify to find the desired relation,
e2 ZL z21
= (6.28)
es (Zs + z11 )(ZL + z22 ) − z12 z21
which is equivalent to eL ∕es . Since ZL is known, this relation makes it possible to solve the power transmission problem
as well.

Example 6.13 Turbine-driven, field-excited dc machine model

Consider the dc generator in Example 6.12 driven by a turbine source, with the generator also connected (terminated) by a
known impedance load, ZL (s). Determine the transfer function that relates the load voltage, VL , to the turbine input torque,
𝜏s (t). Assume a turbine source impedance is given as Zs (s).
Solution
Modeling the source using a Thevenin model allows the use of the configuration in Figure 6.49(a), as shown in Figure 6.50.
The load voltage to input torque transfer function is then given by equation 6.28. With the Z found in example 6.12,
[ ] [ ]
m11 ∕m21 |M|∕m21 Bt + sJt rg
Z= =
1∕m21 m22 ∕m21 rg Rg + sLg

Zs Figure 6.50 Field-excited dc electrical machine.

τs fs

τs e1 e2 eL
Se 1 Z 0 ZL
f1 f2 fL
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
6.5 Two-Port Models and Transmission Matrices 343

we can find,
VL ZL z21 ZL rg
= = .
𝜏t (Zs + z11 )(ZL + z22 ) − z12 z21 (Zs + Bt + sJt )(ZL + Rg + sLg ) − rg2
To complete a transfer analysis using this relation requires explicit relations for ZL (s) as well as the source impedance, Zs (s).

Any of the interconnections shown in Table 6.4 can be used in a similar way as in the previous example to formulate a
reduced two-port representation. This form can then be converted or used directly in a two-port analysis.

6.5.5.2 Experimental Testing and Characterization

Two-port models are often used to capture the port-based behavior of devices and components during testing. While a
two-port requires prescribing four different transfer functions (or parameters), a passive reciprocal two-port has only three
unique and independent parameters. This is relevant in testing, for example, where a given frequency response test only
requires three measurements at any given frequency. One of the ports is typically identified as an input and the other
an output, with prescribed conditions, and as such can be identified through experimental testing. Note that if you make
frequency domain measurements, you will have both magnitude and phase to deal with from the measurements.
For example, here are some typical steps in estimating impedance two-port elements:
1) Excite a system at port 1 with f1 while forcing f2 = 0. Measure response e1 and f1 .

e1 e2 = 0 e1
Sf Z Sf ⇒ ẑ11 = = driving-point impedance
f1 f2 f1 f2 =0

2) Excite the system at port 2 with f2 while forcing f1 = 0. Measure response e2 .

e1 e2 e2
Sf Z Sf ⇒ ẑ22 = = driving-point impedance
f1 = 0 f2 f2 f1 =0

3) Excite a system at port 1 again with f1 but now let e2 = 0, allowing “free response” at port 2. Measure response e1 , f1 ,
and f2 .

e1 e2 = 0 ∗ e1
Sf Z Sf ⇒ ẑ11 =
f1 f2 f1 e2 =0

Here, the three measured (estimated) impedances are indicated by the “hat” symbol; that is, ẑ 11 , ẑ 22 , and ẑ ∗11 . Given these
values (estimated, say on a frequency-by-frequency basis), the following relations can be used to determine the two-port
parameters. First, from step 3, we have the following relations:
e1 = z11 f1 + z12 f2 , and, e2 = 0 = z21 f1 + z22 f2 .
Thus, we can write from the second relation,
f2 || z
| = − 21 .
f1 |e2 =0 z22
Then, from the first relation, we can divide both sides by f1 ,
e1 || f | z
| = z11 + z12 ⋅ 2 || = z11 − z12 21 ≡ z11
∗
.
f1 |e2 =0 f1 |e2 =0 z22
If the system is reciprocal, z12 = z21 , and we can solve for this quantity in terms of the measured impedances, or,
[ ]1∕2
ẑ 21 = ẑ 22 (̂z11 − ẑ ∗11 ) . (6.29)
It is then possible to solve for other two-port parameters from these estimated elements of the impedance matrix. For
example, from Table 6.2, we can find: m ̂ 11 = ẑ 11 ∕̂z21 , m
̂ 21 = 1∕̂z21 , m
̂ 22 = ẑ 22 ∕̂z21 , and, finally, m ̂ z21 = ẑ ∗11 ẑ 22 ∕̂z21 .
̂ 12 = |Z|∕̂
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
344 6 Frequency Response and Impedance-Based Modeling

Note that assuming the system is reciprocal reduces the number of tests required to fully characterize system. Only three
tests were needed to fully characterize a two-port, rather than four. A system that is nonreciprocal (or anti-reciprocal) will
require four tests. Since we know that the presence of a gyrator causes a system to be nonreciprocal, it should be expected
that testing of systems that include electromechanical coupling, for example, will require more testing than, say, a purely
mechanical structural system.

Example 6.14 Measurement of resistive two-port

To demonstrate how a test and measurement procedure can be used to experimentally determine the two-port parameters,
a simple resistive two-port is considered here. For the purpose of simulating measurements and for assessment purposes
assume a known resistive two-port in the impedance form,
[ ]
1 2
Zm = .
2 3
The simulated tests are modeled using this impedance as follows:
[ ]
1) Inputs: f = f1 = 1, f2 = 0 . Measure response e1 .
[ ] [ ][ ]
e 1 2 1 e |
Simulate test: 1 = ⇒ Estimate: ẑ 11 = f 1 | =1
e2 2 3 0 1 |f2 =0

[ ]
2) Inputs: f = f1 = 0, f2 = 1 . Measure response e2 .
[ ] [ ][ ]
e 1 2 0 e |
Simulate test: 1 = ⇒ Estimate: ẑ 22 = f 2 | =1
e2 2 3 1 2 |f1 =0

[ ]
3) Inputs: u = f1 = 1, e2 = 0 (note mixed input). Measure response e1 . Given the mixed input form, we convert the model
impedance two-port, Zm , into its immittance from, Hm as per Table 6.2.
[ ] [ ][ ]
e1 −1∕3 2∕3 1 e |
Simulate test: = ⇒ Estimate: ẑ ∗11 = f 1 | = −1∕3
f2 −2∕3 1∕3 0 1 |e2 =0

Then estimate z21 from equation 6.29,

[ ]1∕2 √
ẑ 21 = ẑ 22 (̂z11 − ẑ ∗11 ) = 3(1 − (−1∕3)) = 2.

As expected for such a simple case, the model impedance is estimated exactly. This approach can be generalized to test
two-ports with dynamic characteristics. However, the estimates of the two-port parameters (as transfer functions) will
require testing over a frequency range of interest and using methods that can account for the complex nature of the response
results. Methods and techniques in system identification are discussed in [100, 110].

6.6 Additional Application of Impedance Formulations

Impedance-based models are extensively used in many areas, with methods and specific techniques developed to aid anal-
ysis and design in practical application. The impedance bond graph approach described in Section 6.3 and as demonstrated
throughout this chapter lends support to understanding these widely used impedance and frequency response methods.
This section briefly describes a few of these application areas.

6.6.1 Impedance and Multiports

Up to this point, most of the uses of impedance methods and two-port forms, such as Figure 6.48, have emphasized systems
of cascaded elements. These make up a large class of practical systems. Modeling of such systems can often be accomplished
by using cascaded transmission matrices. There are many instances, however, that require additional steps to simplify the
form, and the number of ports may be greater than two. It may be possible in some cases to use relations in Table 6.4, but this
is not always the case. References such as Chua et al. [37] review some specific ways to handle these problems for electrical
circuits. The discussion here emphasizes how to go about using the bond graph based impedance form to formulate a
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
6.6 Additional Application of Impedance Formulations 345

Input
torque

Bearing
friction B
(a) (b) (c)

Figure 6.51 Three different passive multiport networks: (a) generic passive filter, (b) bridged-T RC network, and (c) the rotational
coupling from Example 6.11.

Y2

Y2 (V2 − V4 )

V2
1 0
i2
Y1
Y6 1
Y6 (V2 − V3 )
Y1 (V1 − V4 )

V1 V3
1 0 1 0
i1 3 5 i3
Y3 (V4 − V3 ) Y4V3

V4
Y5 Y3 Y4
i4

Figure 6.52 Impedance bond graph for passive network in Figure 6.51(a). Note that the individual impedance one-port elements are
indicated by their admittance value, following the electrical network convention.

suitable model. In particular, consider the example systems shown in Figure 6.51. These are linear passive systems that can
be modeled using impedance methods, and each introduces some useful techniques.
The network in Figure 6.51(a) features several interconnected linear passive one-port elements, indicated by their admit-
tance, Yi . In this case, we seek an admittance matrix, Y, that relates the four currents, i, to voltages, V, at the four ports
of this network in the relation, i = YV. Why use an admittance form in this case? By assigning the implied causality, that
is, specified voltages (efforts) at the four inputs, it is found that four current relations can be specified without any arbi-
trary causal assignments. In contrast, if an impedance form was sought it would require three arbitrary assignments, which
typically requires solving algebraic equations to resolve the variables. Thus, using the impedance bond graph shown in
Figure 6.52 with admittance form, the current relations are found as follows:
i1 = + Y2 (V1 − V4 ),
i2 = + Y2 (V2 − V4 ) + Y6 (V2 − V3 ),
i3 = − (V2 − V3 ) − Y3 (V4 − V3 ) + Y4 V3 ,
i4 = − Y1 (V1 − V4 ) − Y2 (V2 − V4 ) + Y3 (V4 − V3 ) + Y5 V4 .
The resulting admittance relationship is then formulated as,
⎡ i1 ⎤ ⎡ Y1 0 0 −Y1 ⎤ ⎡ V1 ⎤
⎢ ⎥ ⎢ ⎥⎢ ⎥
⎢ i2 ⎥ = ⎢ 0 Y2 + Y6 −Y6 −Y2 ⎥ ⎢ V2 ⎥ . (6.30)
⎢ i3 ⎥ ⎢ 0 −Y6 Y3 + Y4 + Y6 −Y3 ⎥ ⎢ V3 ⎥
⎢ i ⎥ ⎢ −Y −Y2 −Y3 Y1 + Y2 + Y3 + Y5 ⎥⎦ ⎢⎣ V4 ⎥⎦
⎣ 4⎦ ⎣ 1
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
346 6 Frequency Response and Impedance-Based Modeling

This set of equations can be used to find transfer functions for specified port conditions formed by connecting sources or
loads at the terminals. It is left as an exercise to show that trying to find the impedance form (i.e., specify currents causally
at the ports) is less tractable.

Example 6.15 Transfer function from admittance matrix of multiport

For the network in Figure 6.51(a) with port conditions, V1 = known input, i2 = 0 (open terminal), V3 = 0 (grounded), i4 = 0
(open terminal), determine the transfer function V2 ∕V1 .
Solution
Imposing the given port conditions on the admittance relations given in equation 6.30 gives the following relations:
i1 = Y1 V1 − Y1 V4 ,
0 = (Y2 + Y6 )V2 − Y2 V4 ,
i3 = −Y6 V2 − Y3 V4 ,
0 = −Y1 V1 − Y2 V2 + (Y1 + Y2 + Y3 + Y5 )V4
which can be sorted into,
⎡1 0 0 Y1 ⎤ ⎡ i1 ⎤ ⎡ Y1 ⎤
⎢ ⎥⎢ ⎥ ⎢ ⎥
⎢ 0 Y2 + Y6 0 −Y2 ⎥ ⎢ V2 ⎥ = ⎢ 0 ⎥ V
⎢0 −Y6 1 Y3 ⎥ ⎢ i3 ⎥ ⎢ 0 ⎥ 1
⎢0 −Y2 0 Y1 + Y2 + Y3 + Y5 ⎥⎦ ⎢⎣ V4 ⎥⎦ ⎢⎣ Y1 ⎥⎦
⎣
and solved for the unknowns i1 , V2 , i3 , and V4 in terms of the known Yi values and the input V1 . The transfer function V2 ∕V1
thus follows.

A different approach is required for the bridged-T RC network of Figure 6.51(b). Testing for a preferred admittance or
impedance form shows that the former (voltage inputs at the two ports) leads to one arbitrary assignment in contrast to
three for impedance form. Again an admittance form as shown in Figure 6.53 is preferred, but there is no way to avoid the
algebraic problem.
In solving for the port current relations, we find:
i1 = i2 + i3 = Y12 (V1 + V10 ) + Y4 (V1 − V6 ),
i10 = i11 + i9 = Y12 (V1 + V10 ) + Y8 (V10 − V6 ).
The challenge here is in determining i3 and i9 , which both depend on the internal variable V6 . This voltage can be found
by noting, i6 = i5 + i7 = i3 + i9 , and i6 = Y6 V6 , thus,
V6 = i6 ∕Y6 = (i3 + i9 )∕Y6 .

Y12

12

1
2 11

V1 V10
0 1 0 1 0
i1 3 5 7 9 i10

4 6 8

Y4 Y6 Y8

Figure 6.53 Impedance bond graph for bridged-T RC network in Figure 6.51(b) with admittance indicated for the elements.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
6.6 Additional Application of Impedance Formulations 347

Z2 Z4 Z6

τ1 τ7
ω1 1 0 1 ω7

0 1 0

Z9 Z11 Z13

Figure 6.54 Impedance bond graph for an alternative model of the dual-inertia rotational coupling in Figure 6.51(c) with
impedances indicated for the elements; solving for a two-port impedance is the preferred approach for this system.

Thus, we can form the two equations,

[ ]
i3 = Y4 V1 − (i3 + i9 )∕Y6 ,
[ ]
i9 = Y8 V10 − (i3 + i9 )∕Y6 .
These equations can be solved for i3 and i9 in terms of Yi values and inputs V1 and V10 . Once found, the admittance matrix
is determined in the form:

i1 y11 y12 V1 V1 V10

= Y
i10 y21 y22 V10 i1 i10

This two-port admittance matrix can be converted to other two-port matrices as prescribed by the relations in Table 6.2.
Desired transfer functions can then be derived for insertion of this network into a given system.
Consider now the mechanical rotational coupling in Figure 6.51(c). For the model conveyed by the bond graph of
Figure 6.46, this system prefers formulation of an impedance two-port in contrast to the previous electrical systems
discussed. Consider instead the slightly more complicated model shown in Figure 6.54. In contrast to the model in
Figure 6.46, this model includes linear compliance in the main shaft, Z9 = 1∕sCs , as well as additional linear friction at the
end of this shaft in the bearing, Z11 = Bs (relative to ground). In this model, Z2 = sJ1 , Z4 = 1∕sCc , Z6 = sJ2 , and Z13 = B. As
for the bridged-T RC network, the loop structure in this model will require that an algebraic problem to be solved in order
to form the two-port impedance matrix relating ports 1 and 7.
While incorporating the impedance, Z9 = 1∕sCs , complicates the model, the transfer function relating, say, 𝜔7 and the
input torque 𝜏1 can be used to assess the influence of the shaft compliance, Cs , on the response behavior.

6.6.2 Impedance and Active Elements

Two-port models can also integrate active elements in systems [37, 117], such as dependent source models (see Chapter 4,
Section 4.7.6). An active element can be used to model physical effects that introduce energy into a system, as opposed
to passive systems formed by elements that can only represent the ideal transmission, dissipation, or storage of energy.
The examples in the preceding section were chosen to emphasize ways for modeling purely passive networks using
impedance methods. Effort and flow sources may be found within a network and would thus be modeled as active
elements. This makes it possible to introduce model elements that can account for the influence of amplifiers, as described
by Chua et al. [37] and Huelsman [117].
To illustrate one approach, consider a single amplifier filter, built by combining the passive network from Figure 6.51(a)
with an ideal amplifier, with gain K. This active filter is shown in Figure 6.55. A bond graph for this system is shown in
Figure 6.56, where a dependent effort source is used to represent a voltage-controlled voltage source (VCVS) model of the
ideal amplifier. In this case then, the amplifier output is V2 = K ⋅ V3 , and note that flow sources are used to enforce i3 = 0
(ideal amplifier assumption). Consider the specific case of a high-pass Sallen and Key filter [117], where port 4 is open and
Y1 = sC1 , Y3 = sC3 , Y2 = 1∕R2 , and Y4 = 1∕R4 . Also, this filter has Y5 = Y6 = 0.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
348 6 Frequency Response and Impedance-Based Modeling

Figure 6.55 A single amplifier filter.

V2 i2
V1 V3 V3 V2 V2
Se Y 0 Se 0 Sf
i1 i3 i2 i2 = 0
V4 i 4 = 0 i3 = 0

Sf Sf

Figure 6.56 Impedance bond graph for single amplifier filter with dependent source model.

To find the transfer function for this high-pass filter, V2 ∕V1 , the admittance matrix from equation 6.30 is used with the
applied constraints. Since V3 = V2 ∕K, the equation for current i3 is,
1
i3 = 0 = −Y6 V2 + (Y3 + Y4 + Y6 ) V2 − Y3 V4 .
K
Then, since i4 = 0,
1
i4 = 0 = −Y1 V1 − Y2 V2 − Y3 V + (Y1 + Y3 + Y3 + Y5 )V4 .
K 2
Solving for V4 from the first of these relations and substituting into the second allows us to find, after setting Y5 = Y6 = 0
and simplifying that,
V2 Ks2
= ,
V1 s + 2𝜁𝜔n s + 𝜔2n
2

where
1 1 1 K
2𝜁𝜔n = + + − ,
R2 C1 R4 C3 R4 C1 R2 C1
1
𝜔2n = .
R2 R 4 C 1 C 3
The methods illustrated for this active high-pass electrical filter can be extended to model other types of systems that are
best modeled using interacting passive and active elements. Frequency response analysis from such models can lend insight
into the effect of key parameters and are thus useful in design studies.

6.6.3 Using Two-Ports for Distributed-Parameter Models

Two-port modeling and impedance representations provide a very efficient way to explore the influence of distributed-
parameter model (DPM) elements in a system. A DPM introduces a spatial variable, say the distance along a fluid pipe or a
torsional shaft. The spatial variable introduces another independent variable in addition to time. Thus, physical variables
such as the angular twist in a shaft now depend on both x and t, that is, 𝜃 = 𝜃(x, t). These models make it possible to study
localized behavior in some physical components, which is not possible with lumped-parameter models. In addition, a DPM
can may also model variation in properties with the spatial variable. For example, a pipe might have varying parameters
so that fluid inertia and compliance vary by some known functions of x, or I(x) and C(x). Such cases can require relatively
complex analysis techniques.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
6.6 Additional Application of Impedance Formulations 349

I:m

‘heavy spring’ ṗ vm
F (t) Fk HEAVY Fw
Se 1 Sf
vm SPRING vw = 0

Fb vb

R:b
(a) (b)

Figure 6.57 (a) A mass–spring–damper system with a “heavy spring.” (b) Bond graph.

In lumped-parameter models, there is no way to solve for the dynamic response of physical variables at certain points
along a shaft, say to find the temperature at a particular location in a heat conduit. These capabilities are possible with DPMs,
which are based on partial differential equations. Solving for the transient response of systems that include DPM elements
will require specifying not only initial conditions but also boundary conditions. In the following, however, the model will
be formulated using impedance formulations, which takes advantage of Laplace transform solutions of the implied PDE.
Thus, impedance models and two-port formulations can be used. Boundary conditions are imposed by how the DPM ele-
ment fits within a given system model. Consider, for example, the mass–spring–damper system in Figure 6.57(a), depicted
schematically to emphasize a “heavy spring,” meaning that there is significant mass distributed along the spring’s length.
The bond graph in Figure 6.57(b) indicates a word bond graph element “HEAVY SPRING” to convey that the attention will
be given to the elastic as well as the mass effects distributed along the spring. In this form, the model is sufficiently specified
so that causality can be assigned, thus imposing velocity boundary conditions on each end. In the following, impedance bond
graphs will be formed but causality can still be used in this way to specify preferred boundary conditions on DPM elements.
In the following, we begin by showing how two-port models can be formulated.

6.6.3.1 Transmission Lines from Differential Elements

We first develop a representation that can serve as a basis for modeling the conveyance or transmission of energy in many
common physical and engineering systems. In particular, the concept of a transmission line is introduced for use in dif-
ferent energy domains. The motivation for such a modeling abstraction comes from the practical need to transmit power
using shafts, pipes, rods, ducts, wires, cables, etc., and also to couple models of these physical systems with other model
formulations (e.g., lumped-element devices). This section begins to form the basis for formulating and using these models.
We adopt impedance and admittance forms for these purposes.
Consider the differential element of a general transmission line as illustrated in Figure 6.58(a). This differential
element can be modeled as shown in circuit form in Figure 6.58(b), or in impedance bond graph form in (c). This
lumped-approximation is similar to that used in deriving the classical “telegrapher’s” equations (e.g., see [121] or [122]).

Z ∗ (x, s)dx

ex ex+dx
1 0
fx fx+dx

Y ∗ (x, s)dx
(a) (b) (c)

Figure 6.58 (a) A differential element of a transmission line, (b) model by lumped impedance circuit, and (c) equivalent impedance
bond graph form.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
350 6 Frequency Response and Impedance-Based Modeling

For physical components or devices characterized more accurately by DPMs, the dynamics between the end points are
governed by the impedance and shunt admittance per unit length, denoted by Z ∗ (x, s) and Y ∗ (x, s), respectively. In these
functions, x represents the distance along a DPM and s is the complex frequency or Laplace operator. These impedance and
admittance functions allow considering how these functions vary with x. For each differential element, we can write two
relations,
ex − ex+dx fx − fx+dx
= Z ∗ (x, s) ⋅ dx, and, = Y ∗ (x, s) ⋅ dx.
fx ex+dx
In the limit as dx → 0, two coupled differential equations take the form,
𝜕e 𝜕f
= −Z ∗ (x, s) ⋅ f , and, = −Y ∗ (x, s) ⋅ e. (6.31)
𝜕x 𝜕x
Note, that these are the classical Telegraphist’s or transmission line equations. The impedance and admittance can incorpo-
rate dissipative effects that also depend on x. Assuming linear behavior,
Z ∗ (x, s) = R∗ (x) + I ∗ (x) ⋅ s, and Y ∗ (x, s) = G∗ (x) + C∗ (x) ⋅ s,
where,
I ∗ (x) = inertance/length,
R∗ (x) = resistance/length,
C∗ (x) = capacitance/length, and
G∗ (x) = conductance/length.

It should be clear that the capacity to represent the generally dissipative and nonuniform transmission line is extremely
valuable in many practical engineering problems. However, it can be difficult (if not practically impossible) to derive solu-
tions for equations (6.31). For now, consider a general uniform transmission line where
Z ∗ (x, s) = Z ∗ (s), and, Y ∗ (x, s) = Y ∗ (s).
This simplification still permits us to handle “lossy” lines (with dissipation, not just energy storage). Indeed, even the case of
nonuniform distributed-parameter elements can be treated by cascading several uniform lines as described by Paynter and
Ezekiel [123]. It is also possible to handle the dissipation by introducing losses with “effective” resistive elements outside of
a lossless line element. This can be convenient, especially if it is desired to convert to time-difference equations that enable
time-domain simulation. Alternatively, transformation into time-domain forms is possible using approximations of the
resulting relations.
The uniform or U-line equations are then,
de df
= −Z ∗ (s) ⋅ f , and = −Y ∗ (s) ⋅ e, (6.32)
dx dx
which can be differentiated once with respect to x to find,
d2 e d2 f
= +Z ∗
(s)Y ∗
(s) ⋅ e, and, = +Z ∗ (s)Y ∗ (s) ⋅ f . (6.33)
dx2 d2 x
A propagation operator (a generally complex function) is now defined by,
√
Γ(s) = Z ∗ (s) ⋅ Y ∗ (s), (6.34)
and a characteristic impedance is,
√
Z0 (s) = Z ∗ (s)∕Y ∗ (s) = 1∕Y0 (s), (6.35)
where Y0 (s) is also called the surge admittance. These definitions arise when the two (de-coupled) U-line equations 6.33
are solved for e(x) and f (x) along the U-line. These equations suggest the solutions,
e(x) = c1 cosh(Γx) + c2 sinh(Γx),
and,
f (x) = d1 cosh(Γx) + d2 sinh(Γx).
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
6.6 Additional Application of Impedance Formulations 351

Note that these are s-domain relations as well. Assume that the boundary conditions at x = 0 are known as e(x = 0) = e0
and f (x = 0) = f0 , and then it is possible to solve for the unknown constants to form the two-port relations,
[ ] [ ][ ]
e(x) cosh(Γx) −Z0 sinh(Γx) e0
= . (6.36)
f (x) −Y0 sinh(Γx) cosh(Γx) f0
This transfer matrix relation expresses how the effort and flow variables at a distance x are related to those values at x = 0.
The relations are general for any uniform distributed-parameter element as the relations are in the (x, s) space. We will
refer to this element as a U-line (uniform and uni-power), and a simple bond graph representation is simply the two-port
U. The U-line forms the basis for two particular cases: (i) a line model for representing wave-like processes, and (ii) a line
model for representing diffusive processes, such as heat conduction. The former are processes modeled by hyperbolic PDEs
while the latter are parabolic [124]. These two PDE cases can also be shown to arise when the per unit impedance and
admittance take the form:

● Wave-like: Z ∗ = I ∗ s, Y ∗ = C∗ s (lossless wave transmitter)

● Diffusive: Z ∗ = R∗ , Y ∗ = C∗ s

These can be used in the differential bond graph model in Figure 6.58(c). The two model elements can be cast as basic
bond graph elements as follows:

e0 ex e0 ex
W H
f0 fx f0 fx
W-line H-line

EM, mechanical, Heat conduction

fluid, acoustical,etc. diffusion,etc.

These two line types can be used in a wide range of system modeling applications. As will be shown in the following,
frequency response models can be formulated very efficiently. In some cases, converting to time-domain representations
makes it possible to examine transient response as well.

6.6.3.2 Modeling with a W-Line

Consider first the W-line which can be used to model system components of the type shown in Table 6.5. The one-
dimensional (1D) power transmission in each case can be described by the classical wave equation, and a strong analogy
exists among the types shown.16
For purposes of introducing a way to use a W-line in a bond graph context, consider a long shaft under torsion.
In lumped-parameter modeling, a decision is either made to model a shaft as an ideal spring with negligible inertia, a
noncompliant shaft with significant inertia, or to construct a combination of ideal torsional springs and rotational inertias
to approximate the shaft’s dynamic behavior (e.g., as shown in previous sections with transmission matrices). Models of

Table 6.5 Table of W-line parameters (𝜌 = density, G = shear modulus, E = Young’s modulus, 𝛽 = bulk modulus, 𝜇 = permeability,
𝜖 = permittivity; A = characteristic area, R = characteristic radius, ro , ri = outer and inner radii).

√ √
Component e f I∗ C∗ c = 1∕ I ∗ C ∗ Z0 = I ∗ ∕C ∗
√ √
Longitudinal rod F 𝑣 𝜌A 1∕EA E∕𝜌 A 𝜌E
√ 𝜋R4 √
Rotational shaft 𝜏 𝜔 𝜌𝜋R4 ∕2 2∕G𝜋R4 G∕𝜌 𝜌G
2
√ √
Fluid line P Q 𝜌∕A A∕𝛽 𝛽∕𝜌 𝜌𝛽∕A
( ) √
𝜇 ro 2𝜋𝜖 √ ln(ro ∕ri ) 𝜇
Coaxial cable V i ln 1∕𝜖𝜇
2𝜋 ri ln(ro ∕ri ) 2𝜋 𝜖

16 A good discussion of the analogies for wave and diffusion dynamics in different energy domains is given by Moore [122].
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
352 6 Frequency Response and Impedance-Based Modeling

Load
impedance

τi τL
Applied Sf W ZL
ωi (t) ωL

(a) (b)

Figure 6.59 (a) Lossless shaft with specified end conditions. (b) W-line representation.

the purely torsional, distributed nature of such a shaft can show how the angular displacement (or “twist”) in the shaft
can be represented by a 1D wave equation, a classical partial differential equation of the form,
𝜕2 𝜃 𝜌 𝜕2 𝜃 𝜕2 𝜃 1 𝜕2 𝜃
= , or, = 2 2, (6.37)
𝜕x 2 G 𝜕t2 𝜕x 2 c 𝜕t
√
where x is the distance along the shaft’s length, 𝜌 is the material density, and G is the shear modulus, and c = G∕𝜌 is the
wave or phase velocity [125]. Solving equations of this form using Laplace transform methods [124] leads to results of the
form in equation 6.36. This rationalizes an impedance approach and use of a W-line to model the rotational shaft. Note
that in the following, 𝜔 is used to designate angular velocity and should be distinguished by context from the common
use of “𝜔 = 2𝜋f ” for circular frequency (rad/sec). To minimize confusion, angular velocity will often be designated with a
subscript.
Consider a shaft with circular cross section with an input angular velocity 𝜔i (t) on one end and a linear lumped-parameter
load impedance ZL at the other, as shown in Figure 6.59(a). This system is modeled using a W-line with an imposed flow
source as shown in Figure 6.59(b). Although this is an impedance bond graph, using causality helps identify proper bound-
ary conditions on the distributed element models. The causality on the impedance load is not specified.
First, to parameterize the W-line, consider the properties for the uniform shaft. The total shaft mass moment of inertia
is J = mR2 ∕2 = 𝜌(𝜋R2 )LR2 , so the inertia per unit length is I ∗ = 𝜌𝜋R4 ∕2. Thus, the impedance per unit length is,

Z ∗ = I ∗ ⋅ s = (𝜌𝜋R4 ∕2)s.

The total shaft compliance is Ct = L∕(GIz ), where G is the shear modulus of elasticity, and Iz = 𝜋R4 ∕2 is the polar area
moment of inertia. The admittance per unit length is then given by,

Y ∗ = C∗ ⋅ s = (1∕GIz )s = (2∕G𝜋R4 )s.

These values are shown in Table 6.5. The propagation operation, Γ, and characteristic impedance, Z0 , can then be deter-
mined using equations 6.34 and 6.35.
The bond graph can be used in combination with the transfer matrix relations 6.36 to find transfer functions of interest.
For the full shaft length, equations 6.36 give,

𝜏L = cosh(ΓL) ⋅ 𝜏i − Z0 sinh(ΓL) ⋅ 𝜔i , (6.38)

𝜔L = −Y0 sinh(ΓL) ⋅ 𝜏i + cosh(ΓL) ⋅ 𝜔i . (6.39)

Use the load impedance relation, 𝜏L = ZL 𝜔L , in equation 6.38, multiply equation 6.39 by ZL , then multiply this equation
from the first to get,

0 = cosh(ΓL)𝜏i − Z0 sinh(ΓL)𝜔i + ZL Y0 sinh(ΓL)𝜏i − ZL cosh(ΓL)𝜔i .

This equation can be used to find,

𝜏i Z sinh(ΓL) + ZL cosh(ΓL)
Zi = = 0 , (6.40)
𝜔i cosh(ΓL) + ZL Y0 sinh(ΓL)
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
6.6 Additional Application of Impedance Formulations 353

which is the input impedance for the shaft-load system. It is now possible to derive an expression for the impedance as a
function of x. From the two relations in equations 6.36,
𝜏x cosh(Γx) ⋅ 𝜏i − Z0 sinh(Γx) ⋅ 𝜔i cosh(Γx) ⋅ Zi − Z0 sinh(Γx)
Z(x) = = = ,
𝜔x −Yc sinh(Γx) ⋅ 𝜏i + cosh(Γx) ⋅ 𝜔i −Yc sinh(Γx) ⋅ Zi + cosh(Γx)
where 𝜏x = 𝜏(x) and 𝜔x = 𝜔(x). This expression can be rewritten as,
Z(x) Zi cosh(Γx) − Z0 sinh(Γx)
= . (6.41)
Z0 Z0 cosh(Γx) − Zi sinh(Γx)
Finally, to find a relation for how the angular velocity varies with x use the second relation in equation 6.36,
𝜔x = −Y0 sinh(Γx) ⋅ 𝜏i + cosh(Γx) ⋅ 𝜔i .
Dividing by 𝜔i ,
𝜔x
= −Y0 sinh(Γx) ⋅ Zi + cosh(Γx).
𝜔i
Substituting from equation 6.40 and simplifying we get an expression (transfer function) for the angular velocity at any
point x along the shaft,
𝜔x cosh(Γ(L − x)) + ZL Y0 sinh(Γ(L − x))
= . (6.42)
𝜔i cosh(ΓL) + ZL Y0 sinh(ΓL)
Again, note that all of these transfer functions can be used to find the frequency response for several cases of interest by
setting s = j𝜔 in the definition of the propagation operator, Γ. For problems of this type, certain cases of interest include:
● ZL = 0, the “free end” shaft case,
● ZL = Z0 , the matched load case, which results in no reflected power from the load,
● ZL ≠ Z0 , general mismatched load, and
● ZL = ∞, fixed end
To illustrate how the model can be used to analyze these cases, assume an aluminum shaft with the following parameters:
𝜌 = 2870 kg/m3 , G = 2.67 × 101 0 Pa, R = 0.01 m, and L = 5 m. The graphs in Figures 6.60(a) and (b) show three frequency
response plots of 𝜔L ∕𝜔i for the cases: (i) the free-end case, ZL = 0, (ii) the mismatched case with ZL = Z0 ∕5, and (iii) the
case where ZL = Z0 (matched impedance). First and foremost, some appreciation should be gained for the relative ease by
which frequency response results can be obtained through this approach. It is possible to gain practical insight into whether
the distributed nature of any components in a system will be relevant by simply examining such frequency response plots,
including whether “higher-order” modes are relevant, and what if any simplifications can be made.
Consider the free-end case when ZL = 0 in equation 6.42 and examine the velocity at the end, x = L, which gives the
transfer function,
𝜔L 1
= . (6.43)
𝜔i cosh(ΓL)
√ √ √
Since cosh(ΓL) = cosh( I ∗ C∗ j𝜔L) = cos( I ∗ C∗ 𝜔L), we see that there is a resonance when cos( I ∗ C∗ 𝜔L) = 0. This occurs
√
now for I ∗ C∗ 𝜔L = n𝜋∕2, (n = 1, 3, 5, 7, … ). These frequencies are indicated by the peaks in Figure 6.60(a), which for this
shaft occur at,
√ √
n𝜋 1 n𝜋 𝜌
𝜔= = = 958.2, 2874.7, 4791.1, 6707.5, …
2L I ∗ C∗ 2L G
Note that the phase for the free-end case shows a progressive decline in sharp steps of 180∘ as frequency increases. This
model assumes negligible damping in the shaft.17 In contrast, consider when the load impedance at the end of the shaft is
not zero. First, for ZL = Z0 ∕5, we see that the resonant peaks are more damped, as expected and the phase decays at the
resonant frequency for each “mode.” Each mode is effectively a “second-order” factor, which rationalizes approximating
distributed-parameter elements using products of first- and second-order factors.18

17 For this analysis, a very small non-zero linear damping, B∗ , was introduced, that is, Z ∗ = I ∗ s + B∗ . The computational routine that performs
the “unwrapping” of the phase results tends to fail when B∗ = 0.
18 See, for example, Oldenburger and Goodson [126], and the discussion and practical examples by Doebelin [114].
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
354 6 Frequency Response and Impedance-Based Modeling

6
1
0.8
4
0.6

2 0.4
0.2
0 0
102 103 104 0 0.2 0.4 0.6 0.8 1
Distance along shaft, x/L
(a) (c)
0 6
–200
4
–400
deg
–600
2
–800
–1000 0
102 103 104 0 0.2 0.4 0.6 0.8 1
Frequency, rad/sec Distance along shaft, x/L
(b) (d)

Figure 6.60 Plots of analysis of the shaft in Figure 6.59: (a) and (b) are amplitude and phase response for 𝜔L ∕𝜔i for three different
end load conditions, (c) response of 𝜔x for forcing frequency of 𝜔 = 3832.9 rad/sec, and (d) plot of the impedance along the shaft
length for free-end condition (ZL = 0) and a damped end (ZL = Z0 ∕5).

The results shown in Figure 6.60(c) show how the angular velocity varies along the shaft for three different end load
conditions. First, note that a “matched” case with ZL = Z0 is not shown on this graph. Equation 6.42 will reveal as expected
that 𝜔x ∕𝜔i = 1.0 for this case. This important result conveys how this condition results in the shaft not introducing any
dynamics: power is directly transferred from source to the matched load.
To answer the question: “when should I use a distributed-parameter model?”, it can help to establish criteria that will
quantify the conditions under which distributed effects are important when making predictions of the response. In this case,
wave propagation is characterized by the frequency of a disturbance, 𝜔, its wavelength 𝜆, and by the speed of propagation
in the material, c (also called the phase velocity). These parameters are related by,
2𝜋
𝜔 = 2𝜋f = k ⋅ c = ⋅ c, (6.44)
𝜆
where k = 2𝜋∕𝜆 is the wavenumber. A wavelike disturbance has a wavelength that is inversely proportional to its frequency.
To put this in a simplistic manner, small wavelengths will have high frequency, and large wavelengths of low frequency.
Now, consider the shaft under periodic forcing. The wave dynamics will not be important if the wavelength of the highest
√ 𝜆m = 2𝜋c∕𝜔m (i.e., the smallest wavelength disturbance), is large compared to the
frequency of significance in the forcing,
length of the shaft. For the shaft, c = G∕𝜌. So, a lumped-parameter model would be sufficient for practical purposes if,
shaft length L
= ≪ 1.
smallest wavelength 𝜆m
For a shaft within a system, it may not always be clear from this simple rule whether wave dynamics will be important.
However, this rule can be used in an initial model. If the system undergoes abrupt forces (step changes, etc.), then some
wave dynamics could be induced, and a DPM could provide useful insights.
Consider now Figure 6.60(c), which shows how the angular velocity will vary along the shaft for ZL ≠ Z0 . The particular
case shown is for a forcing frequency 𝜔 = 2𝜋f , where f was chosen so that the wavelength 𝜆 = L, or f = c∕L = 610 Hz,
so it is expected that “distributed effects” will be evident. For the free end case, ZL = 0, note how the velocity will have
variations along x. Results are also shown for a case where there is damping at the end with ZL = Z0 ∕5. Finally, consider
the case where ZL → ∞. The results shown for 𝜔x ∕𝜔o are scaled by a factor of 500 in this plot, reflecting the effect of the
standing wave set up within the shaft.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
6.6 Additional Application of Impedance Formulations 355

The specific case study on the rotational shaft can readily extended for other “wavelike” transmitters of power, such as
rods and bars with longitudinal waves, and pipes and tubes with fluids (see for example Solved Problem A-6.9. Table 6.5
provides guidance on key W-line parameters for some common physical system components that can exhibit wavelike
behavior.

6.6.3.3 Modeling with an H-Line

A similar approach as taken with the W-line for wavelike processes can be taken for processes such as electrical line leakage,
heat conduction, and diffusion. The latter can all be modeled using a parabolic PDE. In the following, a transmission line
model is rationalized for this type of process and will be referred to as H-line. This DPM is parameterized by a purely
capacitive shunt (or surge) admittance, Y ∗ = C∗ s, and a purely dissipative (lossy), series impedance, Z ∗ = R∗ s. So, while the
W-line is most often used to model lossless wavelike process, the H-line can be used when lossy wave processes need to be
modeled.
The W-line enables modeling of a wide range of lossless wave propagation. The use of effort and flow variables in each
energy domain makes the W-line compatible with a bond graph basis. When formulating the H-line, however, it may
be more convenient to adopt effort and flow variables that are not consistent with “true bond graphs.” For example, as
discussed in Chapters 2 and 3, it can be more convenient when modeling heat conduction to use temperature, T, as effort
̇ as the flow variable. While the product of these two variables is not power, there is a strong
and heat transfer flow rate, Q,
analogy with electrical circuits (as commonly used in introductory heat transfer courses). In this analogy, temperature
corresponds to voltage and heat transfer flow rate to current. A helpful advantage of this analogy is that linear heat transfer
relations take the form,
1
Q̇ = ⋅ ΔT,
Rh
where ΔT is the temperature difference and Rh is a linear resistance that can be derived from the relevant heat transfer law.
For example, for heat conduction, this relation can be derived from the (scalar) Fourier law,
𝜕T
Q̇ = −kΔA ,
𝜕x
where k is thermal conductivity and ΔA is the area over which the heat transfer occurs. In the following, it is more conve-
nient to use heat transfer flux (per unit area),

̇ 𝜕T
fQ = Q∕ΔA =k .
𝜕x
For the differential element in Figure 6.61(a), energy continuity (see Chapter 1) for this 1D case relates the rate of change
of (internal) energy in this differential element to the power flux (across uniform areas ΔA), as,
𝜕U 𝜕T ∑
𝜌(ΔAΔx)c = 𝜌(ΔAΔx)c = ΔA fQ ,
𝜕t 𝜕t

Z ∗ (x, s)dx

Tx Tx+dx
1 0
fx fx+dx

Y ∗ (x, s)dx
(a) (b)

Figure 6.61 (a) A differential element for modeling 1D heat conduction and (b) impedance bond graph form.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
356 6 Frequency Response and Impedance-Based Modeling

Insulated
Ts (t) TL
Se H Sf
fs fL = 0

(a) (b)

Figure 6.62 Heat conduction using H-line with pseudo-bond formulation.

∑
where 𝜌 is the material density, c is the specific heat, and fQ is the net heat transfer (flux) (along x only). Thus,
𝜕T
𝜌cΔx = fx − fx+Δx
𝜕t [ ]
𝜕f
= fx − fx +Δx
𝜕x
𝜕T 𝜕T 𝜕2 T
= −k +k + k 2 Δx
𝜕x 𝜕x 𝜕x
𝜕T 𝜕2 T
𝜌c =k 2
𝜕t 𝜕x
giving, as expected, the 1D heat conduction equation,
𝜕T k 𝜕2 T 𝜕2 T
= = 𝛼t 2 , (6.45)
𝜕t 𝜌c 𝜕x2 𝜕x
where 𝛼t = k∕𝜌c is the thermal diffusivity.19
It should now be clear that a representation for the heat conducted in a uniform 1D heat conductor can be formulated
using an H-line. An H-line is used in the example of Figure 6.62(a), with a temperature source Ts (t) imposed at one end
(x = 0) and an insulated wall at the other (x = L). These boundary conditions are reflected in the causality assigned on the
H-line bond in Figure 6.62(b). From the transfer matrix relations 6.36, the temperature at the insulated wall can be found
from,
√ √
TL (s) = cosh( 𝜏t s) ⋅ Ts (s) − Y0 sinh( 𝜏t s) ⋅ fs ,
which would lead to,
1
TL (s) = √ ⋅ Ts (s),
cosh( 𝜏t s ⋅ L)
where 𝜏t is an effective thermal line time constant. Definition of this parameter along with showing that equations 6.36
hold is left as an exercise for the reader (See Problem B-6.39).
Note that an equation of the form 6.45 can be derived for an electrical transmission line where there is negligible leakage
(i.e., G = 0) and inductance, and the equations take the form,
𝜕V 1 𝜕2 V 𝜕i 1 𝜕2 i
= and, = , (6.46)
𝜕t RC 𝜕x 2 𝜕t RC 𝜕x2
where R∗ and C∗ are the per unit length resistance and capacitance, respectively. For this H-line, Z ∗ = R∗ and Y ∗ = C∗ s,
and the line parameters are,
√ √
Γ = Z ∗ Y ∗ = 𝜏t s∕L,
√ √
Z0 = Z ∗ ∕Y ∗ = R∕ 𝜏t s.
The time constant 𝜏t = RC provides a measure of the characteristic time scale of the H-line in this case. Analogous types of
physical processes where an H-line can be used include Fickian diffusion and diffusion in semiconductor operation.

6.6.3.4 Lumped-Parameter Approximation of Distributed-Parameter Elements

Ad hoc approximation of distributed-parameter elements can make use of cascaded lumped-parameter models. A common
approach is to use the classical Π- and T-model forms, as discussed previously in this section. In addition, transmission

19 Heat transfer textbooks often define thermal diffusivity with 𝛼, here we use 𝛼t to distinguish from 𝛼 as the attenuation constant discussed
earlier.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
6.6 Additional Application of Impedance Formulations 357

I:J

ḣ3 ω3
τ1 τ2 τ4
Sf W 1 R:R
ω1 (t) ω2 ω4
(a) (b)

Figure 6.63 (a) Lossless shaft with rotational inertia and damping load. (b) Bond graph with W-line model of shaft.

matrices provide a way to systematically derive transfer functions from impedance based bond graphs. The form symmetric
of the Π- and T-models are sometimes chosen based on how they will interconnect with other elements or sources. Using
causality, it can be clear that one form is preferred over another based on how many “independent” states may be introduced.
In some cases, it may be desirable to accurately predict a certain number of mode frequencies. For example, Rocke [127]
showed how the “symmetric” forms can provide better approximations (lower error in frequency estimates) when build-
ing low-order lumped approximations in comparison with using asymmetric lumping as in Figure 6.58(b). On the other
hand, cascading n of the resulting matrices based on the Π- or T-model, and letting n → ∞, yields the same results as
the asymmetric form in Figure 6.58(b) [1]. So it is only when building lower-order models that care should be taken. The
following example illustrates how a W-line model compares with an approximated form. The example also shows how
distributed-parameter elements can be integrated with lumped-parameter models in an impedance bond graph form. A
comparison between a lumped approximation and DPM is also made.

Example 6.16 Shaft system with bearing/inertia load

We continue here with the uniform shaft problem with the added effect of a load. The load will have resistive characteristics
both with and without inertial effects, as illustrated in Figure 6.63. For the purely resistive load, we will simply make J = 0
in the final model. This example illustrates how easily transmission lines and lumped elements can be appended in a system
model, particularly for the purpose of predicting frequency response and especially for those cases where the elements are
linear.
There is only one state equations for this system,
B
ḣ 3 = 𝜏2 − 𝜏4 = 𝜏2 − h3 .
J
The s-domain equations for the W-line are,
T2 = cosh(ΓL) ⋅ T1 − Z0 sinh(ΓL) ⋅ 𝜔1 ,
𝜔2 = −Yc sinh(ΓL) ⋅ T1 + cosh(ΓL) ⋅ 𝜔1 .
Since we know 𝜔1 = 𝜔1 (t) (input), and 𝜔2 = h3 ∕J, we can determine an exact transfer function between 𝜔3 and 𝜔1 . Through
some algebraic manipulation and substitution,
𝜔3 (s) Z0
= .
𝜔1 (s) Z0 cosh(ΓL) + (Js + B) sinh(ΓL)
The amplitude and phase functions for this transfer function are given in Figure 6.64, where the results from using a W-line
and a Π-model are compared.

Lumped approximations can also be formed for diffusive processes using the two dual reversible microelement models
shown in Figure 6.65. The structure of 6.65(a) is a symmetrical model for heat conduction of the Fourier type and is also
applicable to charge diffusion, and other processes. The dual formulation shown in 6.65(b) would be more applicable to
magnetic hysteresis, eddy currents, and skin effects, as well as in secondary flows and boundary layers in fluid mechanics.

6.6.3.5 Using Infinite Product Expansion Approximations of DPMs

Another approach for building low-order lumped approximations is to use truncated series approximations. Infinite prod-
uct expansions are a proven way to build approximations of transfer functions for distributed-parameter systems [126, 128].
Doebelin [114] illustrates a practical application of this approach. Fortunately, using W- and H-line models with bond
graphs makes is relatively easy to formulate “exact” transfer function models for use in deriving frequency response func-
tions. It is then possible to focus on making the approximation in the cosh() and sinh() functions using well-established
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
358 6 Frequency Response and Impedance-Based Modeling

Pl
0 W

–50

–100

–150
100 101 102 103 104 105

–90
deg

–180

–270

–360

100 101 102 103 104 105

Frequency, rad/sec

Figure 6.64 Plots contrasting the amplitude and phase functions found by modeling the shaft in the system of Figure 6.63 using a
W-line compared to a Π-model.

R C R R I R

1 0 1 0 1 0

(a) (b)

Figure 6.65 Diffusive line microelement models. (a) RCR-line. (b) RIR-line.

approximations (e.g., see [126, 129, 130]). Modern computing software makes it possible to readily adopt these methods
and to experiment with various orders of approximation in a given application.
The two √ hyperbolic functions cosh(Γx) and sinh(Γx) commonly arise in the types of models under discussion, where
Γ = Γ(s) = Z ∗ Y ∗ is the propagation operator. The resulting compact functional form can be used to compute frequency
response functions directly, even for relatively complex Γ(s). Approximations as infinite product expansions, however,
make it possible to construct simplified transfer functions that are rational functions of s. These can be used in analysis
and design, an particularly for linear time-domain analysis as well. In general form, infinite product expansions of primary
interest are summarized as (see Goodson [130]),
∞ [ ]
∏ 4z2 ∕𝜋 2
cosh(z) = 1+ , (6.47)
k=0
(2k − 1)2
∞ [ ]
∏ z2
sinh(z) = z 1+ 2 2 . (6.48)
k=1
k 𝜋
√
For example, recall the heat conduction model in Figure 6.62 that resulted in the transfer function, TL ∕Ts = 1∕ cosh( 𝜏t s).
√
We can represent cosh( 𝜏t s) using an infinite-product expansion. The s-domain transfer function can then be
expressed as,
TL (s) 1
= ,
Ts (s) (1 + as)(1 + as∕9) … (1 + as∕(2k − 1)2 ) …
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
6.6 Additional Application of Impedance Formulations 359

where a = 4𝜏t ∕𝜋 2 . Truncating this series of “first-order factors” must be done to give a finite-order polynomial as desired.
Note that each product term contributes a unique time constant for this RC-type line. For W-line models, the products
appear as second-order factors, making it possible to identify how individual product terms contribute “modes” with dif-
ferent damping ratio and undamped natural frequency values.

6.6.3.6 Insights into Transient Response Analysis of DPMs

This chapter has focused on transfer function and frequency response analysis. For these purposes, models formulated with
distributed-parameter elements can be used directly to derive exact frequency response results. The models also provided
a basis for assessing the frequency domain characteristics of lumped-parameter approximations.
The lumped approximations derived using infinite product expansions provide one way to formulate rational polyno-
mial transfer functions that can then be used for analysis within contemporary linear system software tools. As shown in
Chapter 5, a linear transfer function form can be used directly to simulate system response using available functions (e.g.,
lsim(), step(), etc.).
In some cases, it is possible to use the DPMs to formulate time-domain simulations directly. √ For example, for the shaft
given a step input in Figure 6.59, we can use the relation cosh(Γ ⋅ L) ≡ 12 (eΓL + e−ΓL ). Since, Γ = G𝜌 ⋅ s,
[ √ √ ]
1 𝜌 𝜌
cosh(Γ ⋅ L) = exp( ⋅ Ls) + exp(− ⋅ Ls) ,
2 G G
√
and we can define the time for a wave to propagate across the shaft by T = L∕c, or T = 𝜌∕G ⋅ L (wave transit time). Now,
1 [ Ts ]
cosh(Γ ⋅ L) = e + e−Ts ,
2
and we can use this to get a simple time-domain solution for the response of 𝜔L to the input 𝜔i . From the transfer function
in the previous example,

𝜔L (s) ⋅ cosh(Γ ⋅ L) = 𝜔i (s).

This can be rewritten,

[ ]
𝜔L (s) ⋅ eTs + e−Ts = 2 ⋅ 𝜔i (s).

Recall the inverse Laplace transform, eTs ⋅ f (s) → f (t + T), where f (s) is the Laplace transform of f (t). Now, the above rela-
tionship between the angular velocities can be expressed using the time-difference equation,

𝜔L (t + T) + 𝜔L (t − T) = 2 ⋅ 𝜔i (t).

To use this equation recursively, we can shift the time delays to get,

𝜔L (t) = −𝜔L (t − 2T) + 2 ⋅ 𝜔i (t − T),

which for a step input will have the solution shown in Figure 6.66. A similar relationship can be found for 𝜏i (t) if it is
desired to simulate for the back torque. Different time-different relations can be found for the various combinations of
causality on a W-line, as shown in Table 6.6. In some cases, these types of simulations can be used for quick insight into a
system. Further, the delay operators are readily implemented in some graphical block diagram simulation programs (e.g.,
MATLAB/Simulink).

Figure 6.66 Time-domain response of shaft’s end to a step at the input, a.

Step input

“Exact” response

1 2 3 4 5 6 7 8 9 10
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
360 6 Frequency Response and Impedance-Based Modeling

Table 6.6 Summary of causal time-difference relations.

Causal form Time-difference relations
e1 e2
W e1 (t) − e1 (t − 2T ) = −2Z0 f2 (t − T ) + Z0 [f1 (t) + f1 (t − 2T )]
f1 f2 e2 (t) − e2 (t − 2T ) = +2Z0 f1 (t − T ) − Z0 [f2 (t) + f2 (t − 2T )]
e1 e2
W e1 (t) + e1 (t − 2T ) = +2e2 (t − T ) + Z0 [f1 (t) − f1 (t − 2T )]
f1 f2 f2 (t) + f2 (t − 2T ) = +2f1 (t − T ) − Yc [e2 (t) − e2 (t − 2T )]
e1 e2
W f1 (t) − f1 (t − 2T ) = −2Yc e2 (t − T ) + Yc [e1 (t) + e1 (t − 2T )]
f1 f2 f2 (t) − f2 (t − 2T ) = +2Yc e1 (t − T ) − Yc [e2 (t) + e2 (t − 2T )]
e1 e2
W f1 (t) + f1 (t − 2T ) = +2f2 (t − T ) + Yc [e1 (t) − e1 (t − 2T )]
f1 f2 e2 (t) + e2 (t 2T ) = +2e1 (t T ) Z0 [f2 (t) f2 (t 2T )]

6.6.3.7 Summary and Perspective

The U-line model was derived using classical differential elements, recast into an impedance bond graph form. The intent
of the brief introduction provided here is to show how impedance methods, bond graphs, and two-port formulations can
be used to introduce DPM elements. A much more extensive introduction is given by Brown [7], and more details can be
found in the various articles referenced throughout this discussion.

6.7 Summary
This chapter began by reviewing the basic concept of frequency response for linear systems and then described the classical
approach for representing magnitude and phase functions using basic factors. Understanding basic factors may have once
been most valuable because it allowed easily sketching composite forms of a system’s frequency response functions. With
contemporary computing tools, “Bode” plot functions are easily found. Nevertheless, it can still be very useful to be able
to see how basic factors comprise more complex systems. Such an insight finds use in modeling, analysis, and design.
Applications of frequency response were also reviewed to emphasize how these functions serve a key role in characterizing
all types of systems and in guiding experimental testing and model approximation.
This chapter also introduces the use of impedance methods directly within a bond graph context. Using impedance does
not require bond graph formulation; however, it is shown how this basis leads to efficient and effective ways for deriving
transfer functions. In addition, impedance forms provide a gateway for introducing two-port models. Two-port modeling
is used in many different areas. The introduction provided suggests how these representations can be integrated with bond
graphs allowing modeling and/or analysis when components of systems are dictated in two-port form.
Two-port models can also lend insight into how and when ideal transfer functions can be used within block diagram
descriptions without violating the inherent coupling intrinsic to some model descriptions. This will be discussed further
in Chapter 7. A generalized multiport concept as defined in bond graph modeling will be further described in Chapter 8.
Finally, the two-port formulations described in this chapter provide an effective way to introduce ways to introduce many
established model forms, such as active elements and distributed parameter elements.

6.8 Problems
Modeling, transfer functions, and FRF

Problem B-6.1 Second-order system – 1 A second-order system is subjected to sinusoidal forcing. Careful phase mea-
surements on the forced response provide the data in the table below.

Forcing frequency (Hz) 18 19 21 22

Phase angle (deg) 88.79 89.47 90.56 91.09

Use these results to determine 𝜁 and 𝜔n of the system.

 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
6.8 Problems 361

Problem B-6.2 Second-order system – 2 A second-order system is subjected to sinusoidal forcing. The phase lag is
observed to be 110∘ at 100 Hz, and 175∘ at 1000 Hz.
a) Find 𝜁 and 𝜔n of the system.
b) If at t = 0, this system is subjected to a step input of magnitude u0 , find the time of steepest slope of the response
and determine the value of this maximum slope.

Problem B-6.3 Water-buoy level meter The buoy shown in Figure B-6.3(a) is placed in a reservoir and moves only
along the z (vertical) axis, constrained by roller guides shown. The buoy has a mass m, having density 𝜌, area A, and
length 2L. When the buoy moves with velocity 𝑣m , there is linear damping with coefficient b.
The position of the buoy center of mass CG, zm , and the water level, z𝑤 , are measured relative to a datum. The schematic
shows the buoy at rest, where zm = z𝑤 and the buoyant force equal to the weight of the buoy due to gravity. As the water
level rises and falls over time, zm ≠ z𝑤 .
a) Show how the bond graph of Figure B-6.3(b) can be used to study the dynamic motion of the buoy due to dynamic
variations in the water level, z𝑤 (t). Fully justify each of the parameters of the elements as well as the forcing function
𝜌gAz𝑤 (t).
b) Derive a transfer function zm ∕z𝑤 .
c) Determine the damping ratio 𝜁 and 𝜔n from the transfer function in part (b). √
d) For the case where the water level varies sinusoidally, z𝑤 (t) = z𝑤o sin 𝜔t, it is expected that z𝑤o = L∕ 10. Determine
a value of b that will prevent the buoy from moving off the indicated scale (±L).

I : ρAL

ṗm vm
ρgAzw (t) Fz
Se 1 C : k = ρgA
ż

Fb vb
datum
buoy area, A
R:b

(a) (b)

Figure B-6.3 (a) Schematic of water-buoy level meter system. (b) Bond graph.

Problem B-6.4 Cam-driven spring–mass component The spring–mass mechanism shown in Figure B-6.4 is driven by
a cam that rotates at constant angular speed, 𝜔, about a fixed axle at point O. As it rotates, the cam forces the follower
to move horizontally, constrained by low-friction guides, with motion, xc = 𝜖 sin 𝜔t. So long as the contact force is
compressive, the follower will move with the prescribed motion. Otherwise, it will leave the surface and the mass, m,
will move in a way that disrupts the process. The loss of contact is prevented partially by applying constant force Fa at
the mass as shown in the figure.
a) Show that the motion can be approximated by xc (t) = 𝜖 sin 𝜔t, and thus 𝑣c (t) = dx1 ∕dt.
b) Develop a bond graph that assumes the prescribed motion input 𝑣c (t) as well as constant force Fa . For the purposes
of this problem, ignore damping in the system.
c) Use the bond graph to show that the states are spring deflection xk and mass velocity 𝑣m and derive the state
equations, including an output equation for the difference in applied forces, y = ΔF = Fa − Fk .
d) Derive a transfer function between ΔF and the input 𝑣c (t) and convert to ΔF∕xc .
e) For a given input amplitude, xco , find an expression for the amplitude of ΔF assuming xc = xco sin 𝜔t.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
362 6 Frequency Response and Impedance-Based Modeling

Cam
Follower

Figure B-6.4 Cam-driven spring–mass

mechanism.

f) Use the result from (e) to derive an expression for the lowest angular speed 𝜔 at which the follower will leave the
cam surface, which occurs when Fk = 0.
g) For the parameter values given below, solve for 𝜔: m = 0.454 kg, k = 43.78 kN/m, 𝜖o = 6.35 mm, Fo = 44.48 N.

Problem B-6.5 Velocity measurement device (revisited) Revisit the velocity measurement device from Problem A-6.2
and complete the following:
a) Express the derived transfer function 𝑣m ∕𝑣 as a standard second-order system, defining 𝜁 and 𝜔n .
b) For 𝜁 = 1.0, 𝜔n = 10 rad/sec, determine the amplitude 𝑣m if 𝑣o = 0.5 m/s and 𝜔 = 12 rad/sec.
c) What is the amplitude xm for same conditions as in (b)?

Problem B-6.6 Pneumatic joint actuator In Problem B-5-15, a configuration was shown for testing a pneumatically-
drive joint actuator.
a) Use the results from B-5-15 and develop a transfer function (TF) between the angular position, 𝜃, about the pivot
point and the input flowrate, Q(t). This TF should be put in the standard second-order form.
b) Derive the frequency response functions (amplitude and phase) for G(s) = 𝜔∕Q.

Problem B-6.7 Experimental thermal model A system has been experimentally found to have a thermal time constant
of 𝜏t sec. During a test, the system is forced by a known temperature, Ts (t) so a model can be formed: 𝜏t Ṫ + T = To sin 𝜔t.
During a test, the temperature from an ideal temperature sensor is observed to vary with a peak amplitude of Tm . Using
the following data, use the model to estimate the amplitude of the temperature source, To .
Available data: 𝜏t = 3 sec, f = 0.2 Hz, Tm = 2∘ C.

Problem B-6.8 Pipeline component characterization A pipeline component for transmitting viscous fluids has been
tested experimentally as in the configuration shown in Figure B-6.8(a). Measured values of the magnitude ratio and
phase for the transfer function G(s) = Q∕P over a frequency range of interest are shown in the plots of Figure B-6.8(b).
At low frequencies (𝜔 < 10−2 rad/sec), |G| = |Q∕P| tends toward a value of 0.14 × 10−6 and phase ∠G tends toward 0∘ .
Also note that at 𝜔 = 0.7 rad/sec the phase is about −45∘ . These characteristics suggest the component can be modeled
by a first-order system.
a) Construct a first-order transfer function model for G(s) by estimating the break frequency 𝜔b = 1∕𝜏t and a gain K
from the measured data provided.
b) Propose a bond graph for the pipeline component in the test configuration of Figure B-6.8(a). No need to include
the atmospheric pressure, Patm , since P(t) is gage.
c) Determine values of the parameters for each element in your bond graph model using the available data (e.g., if
there is an R element, what is the R value?).

Problem B-6.9 RC impedance bridge An RC impedance bridge is shown in Figure B-6.9(a).

a) Develop a bond graph of the system in Figure B-6.9(a) and show its relation to the impedance form in
Figure B-6.9(b).
b) Find a transfer function G = Vo ∕Vs , where Vo = V1 − V2 . Show that Vo ∕Vs = Z2 ∕(Z3 + Z4 ), where Z1 = R, Z2 =
1∕(sC), Z3 = 1∕(sC), and Z4 = R.
c) Let R = 1 and C = 1 and show that Vo ∕Vs = (1 − s)∕(1 + s).
d) Plot the magnitude and phase plots for the transfer function in part (c).
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
6.8 Problems 363

10–6
0.15
|G| = | Q–P | 0.10
P(t) = Po sin ωt
0.05
Fluid flowrate, Q
Patm 0.00
10–3 10–2 10–1 100 101 102
pump 10
0
G, deg
–45
Pipeline component under test
–90
–100
10–3 10–2 10–1 100 101 102
Frequency, rad/sec
(a) (b)

Figure B-6.8 (a) Pipeline component test configuration. (b) Measured frequency response functions (magnitude
and phase) for transfer function, G(s) = Q∕P.

(a) (b)

Figure B-6.9 (a) RC impedance bridge. (b) Equivalent

impedance form.

Sketching of approximate Bode plots

Problem B-6.10 .Sketch the Bode plots for the system,

(3∕50)(s∕3 + 1)
G(s) =
(s∕2 + 1)((s∕5)2 + 2s∕25 + 1)
and compare with the results from using a computer-aided software tool (e.g., bode() in MATLAB).

Problem B-6.11 .Sketch the Bode plots for the system,

20
G(s) =
s(s2 + 10s + 7)
and compare with the results from using a computer-aided software tool (e.g., bode() in MATLAB).

Problem B-6.12 .Sketch the Bode plots for the system,

4(1 + 0.5s)
G(s) =
s2 (1 + 0.125s)(1 + 0.1s)
and compare with the results from using a computer-aided software tool (e.g., bode() in MATLAB).

Problem B-6.13 .Sketch the Bode plots for the system,

100(1 + s∕2.5)(1 + s∕0.25)
G(s) =
(1 + s∕25)(1 + s∕0.025)
and compare with the results from using a computer-aided software tool (e.g., bode() in MATLAB).
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
364 6 Frequency Response and Impedance-Based Modeling

Problem B-6.14 .Sketch the Bode plots for the system,

2000
G(s) =
s(s + 10)(s2 + 8s + 100)
and compare with the results from using a computer-aided software tool (e.g., bode() in MATLAB).

Problem B-6.15 Motion sensor modeling The system shown in Figure B-6.15(a) is a motion sensor in which a seismic
mass, m, which includes a coil moves relative to a permanent magnet that is attached to the fixed case (see Problem
A-3.7). The relative motion of the magnet and coil induces a voltage V = rs 𝑣r , where 𝑣r is the relative velocity between
the mass and the case. The value of the transduction parameter, rs , has units of voltage/velocity. Assume that the flexible
supports have effective stiffness, k, and damping, b. These components are mounted between the mass and the rigid
case. Neglect the effect of gravity.
a) Develop an impedance bond graph model. Note that it is not important here to consider the mass the case, which is
rigidly mounted to the moving ground. Assume that a measurement system is attached to measure V and that this
device can be represented by impedance Zm (s).
b) Find a transfer function that relates the output voltage, V, to the ground motion velocity, 𝑣g .
c) If Zm is very large, show how this affects the transfer function.
d) Assume that this device is designed to have critical damping. Qualitatively sketch the amplitude and phase plots
for the case described in (c).

Electrical
connections
Inertial
mass, m
Leaf
springs Coil (attached to
inertial mass)

Ground PM (attached
motion to case)

(a) (b)

Figure B-6.15 (a) Schematic or geophone (or seismometer) and (b) detail on modeling of
transduction.

Experimentally determined TFs

Problem B-6.16 Transfer function approximation A swept-sine test frequency response test has resulted in the data
provided below. The phase values are in degrees and 𝜔 is in rad/sec. Estimate the transfer function.

𝝎 |G(𝝎)| 𝝓(𝝎)

0.1 5.0 −12

0.2 4.8 −25
0.4 3.8 −45
0.8 2.6 −73
1.0 2.3 −83
2.0 1.25 −122
4.0 0.63 −173
6.0 0.40 −218
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
6.8 Problems 365

Problem B-6.17 Estimating experimental transfer functions Three different sets of experimentally measured magni-
tude and phase data are provided below. All frequency values are in rad/sec and phase in degrees, but note that some
magnitude values are in dB. Determine a transfer function for each case and plot the magnitude and phase functions
together with the data provided.
Case (a)

𝝎 |G(𝝎)| dB 𝝓(𝝎) 𝝎 |G(𝝎)| dB 𝝓(𝝎)

0.10 46.02 −179.1 14.0 −33.56 −192.8

0.50 18.14 −175.8 20.0 −40.34 −207.9
1.0 6.34 −171.8 40.0 −55.74 −234.2
2.0 −4.90 −166.1 80.0 −72.92 −251.2
4.0 −15.0 −162.9 140.0 −87.29 −259.1
8.0 −24.60 −173.1 200.0 −96.53 −262.4
10.0 −27.96 −180.0 240.0 −101.3 −236.6

Case (b)

𝝎 |G(𝝎)| 𝝓(𝝎) 𝝎 |G(𝝎)| 𝝓(𝝎)

0.1 9.95 −96.9 3.0 0.092 −192.5

0.3 3.19 −110.1 5.0 0.028 −213.7
0.6 1.42 −127.8 8.0 0.0082 −230.9
0.8 0.963 −137.8 12.0 0.0027 −242.6
1.0 0.693 −146.3 16.0 0.0012 −249.0
2.0 0.21 −175.2 20.0 0.0006 −253.1

Case (c)

𝝎 |G(𝝎)| dB 𝝓(𝝎) 𝝎 |G(𝝎)| dB 𝝓(𝝎)

1.0 −0.08 −92.9 15.0 −26.80 −239.0

2.0 −5.71 −95.9 20.0 −36.00 −251.6
4.0 −10.77 −103.4 30.0 −47.75 −259.4
6.0 −12.55 −115.1 40.0 −55.65 −262.4
8.0 −12.68 −138.0 50.0 −61.63 −264.1
10.0 −13.98 −180.0 55.0 −64.17 −264.6

Problem B-6.18 Estimate transfer function from magnitude plot Given the straight-line asymptotic log plot shown in
Figure B-6.18, determine a transfer function assuming that the angle at 𝜔 = 0.8 rad/sec is −129.7∘ and that this system
is minimum-phase (i.e., there are no zeros in the right-hand plane).
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
366 6 Frequency Response and Impedance-Based Modeling

30
–20 dB/decade
20

10

0
10–1 100 101 102
–10

–20
–26 –40 dB/decade
–30

–40

Figure B-6.18 Asymptotic straight-line approximation of magnitude plot.

Modeling for Frequency Response Functions

Problem B-6.19 Cascaded LRC circuit For the circuit shown in Figure B-6.19, do the following:
a) Develop a bond graph model.
b) Use impedance methods to derive a transfer function between the output and input voltages, VC2 ∕Vs .
c) Plot the magnitude and phase plots. Use the following values for inductance, capacitance and resistance for the
cascaded systems: (a) first stage: L1 = 0.1 H, C2 = 0.001 F, R2 = 1 Ω, L2 = 0.2 H, C2 = 0.002 F, R = 92Ω. Use a
computer-aided software tool (e.g., bode() function in Matlab or Python).

Figure B-6.19 Cascaded LRC circuit system.

Problem B-6.20 Loud-speaker frequency response A voice-coil type loudspeaker is shown in Figure B-6.20. An induced
current ic interacts with the radial magnetic field from the permanent magnet source to induce a force, Fc = rc ic . Any
axial velocity of the moving coil, 𝑣c , induces a back-emf, Vc , in the coil circuit, Vc = rc 𝑣c . Assume the mass of the moving
elements of the speaker, including the effect of the air moved by the speaker cone, is mc .
a) Develop a bond graph model of this system.
b) Derive a transfer function relating the relating the speaker cone velocity to the input voltage, 𝑣c ∕Vs , using either
state equations or an impedance approach from the bond graph found in part (a).
c) For the parameter data given below, plot the magnitude and phase plots (Bode plots) for the 𝑣c ∕Vs relation over a
frequency range of 10 Hz to 30 000 Hz.
Parameter data: rc = 0.05 oz/mA, mc = 0.2 oz, R = 100 ohms, L = 0.1 H, kd = 16 oz/in.
d) Look up the audible range for humans and comment on the quality of this loudspeaker. Is it a “good” loudspeaker?

Problem B-6.21 Experimental determination of analog meter parameters from frequency response measurements
Solved Problem A-2.8 describes a model for an analog electromechanical voltmeter.
a) Refer to Solved Problem A-2.8 and develop a bond graph model, however neglect the coil inductance. Apply causal-
ity and derive linear state equations, identifying the ABCD matrices for the case where y = 𝜃 is the output variable
of interest.
b) Derive a transfer function, 𝜃∕Vin .
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
6.8 Problems 367

c) Plot the magnitude and phase functions (Bode plots) for the case where: Rm = 15 085 ohm, rm = 0.03 Nm/A, Jm =
2 × 10−7 kg m2 , Km = 200 × 10−6 Nm/rad, and Bm = 12 × 10−7 Nms/rad.

Permanent-magnet

Figure B-6.20 Loudspeaker system and circuit.

d) Measurements have been made of the magnitude function, and the following data are available:
Frequency measurements 𝜔 (rad/sec): 0.1, 0.5, 1, 2, 3, 4, 5, 6, 7, 9, 12, 15, 18, 21, 24
Amplitude ratio magnitude, |G(𝜔)| (rad/V): 0.105, 0.105, 0.105, 0.105, 0.105, 0.11, 0.123, 0.133, 0.14, 0.15, 0.12, 0.073,
0.0467, 0.033, 0.025
Explain using this data why it is justified to ignore the inductance in the meter coil, especially if the meter will only
be used up to forcing frequencies of about 3–5 Hz.
e) Plot the theoretical second-order model from (a), (b), and (c) to the measured magnitude data. Iterate on adjust-
ments to parameters such as rm , Bm , and Km in order to improve the fit of the theoretical model magnitude plot to
the measured data. Summarize the final parameter values.
f) For the final parameters, determine the system 𝜁 and 𝜔n .
g) Use the MATLAB step() function with the transfer function model to find the response to a step voltage of 5 V.

Problem B-6.22 Response of electromechanical shaker and table The electromechanical shaker shown in Figure B-6.22
is similar to the one shown in Figure 6.4 (and as studied in Problem B-5-31), except now the shaker table mass, mt and
the drive coil of mass mc are coupled by the spring-damper element, ktc -btc . The table is also supported by suspension
stiffness and damping, ks and bs . The permanent magnet is rigidly attached to a fixed foundation.
a) Develop a bond graph model of this system, having input voltage Vs (t).
b) Derive a transfer function between the velocity of the table, 𝑣t , and the input voltage, Vin (t).
c) Compute the frequency response of table acceleration for voltage input using the system parameters given below
(from Chapman [131]). Plot both the magnitude and phase functions (e.g., using a bode() function in MATLAB or
Python).
Lc = 0.0012 H, Rc = 3.0 Ω, ktc = 8.16 × 108 N/m, rs = 190 N/amp (shaker force constant), btc = 3850 N/(m/s), mc =
1.815 kg, mt = 6.12 kg, ks = 6.3×105 N/m, bs = 1120 N/(m/s).
d) Use a time-domain simulation of the shaker to compute the response of the table to a triangular voltage pulse, Vin (t),
with a peak voltage of 1 V and a base time width of t𝑤 = 0.01 seconds. Compute as outputs the table displacement,
velocity, and acceleration.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
368 6 Frequency Response and Impedance-Based Modeling

Ground

Permanent-magnet

Ground

Moving coil

Figure B-6.22 Electromechanical shaker and structurally coupled test specimen

table.

Problem B-6.23 Pressure-based depth-rate detection In the system shown in Figure B-6.23, an L-frame rotates by angle
𝜃 about the pivot as shown. The L-frame motion responds to forces transmitted through a spring and damper connected
to a C-frame. The C-frame has negligible mass and moves in response to pressure, Ps , from the exterior of an underwater
vehicle hull. The L-frame is sealed from water by the bellows elements. The system is designed to detect the rate of
change of pressure with time, dPs ∕dt, which is related to the angular position of the L-frame [81].
a) Develop a bond graph model of this system, making reasonable assumptions about what components should be
included. On the bond graph, number the bonds, and use the bond numbers for subscripts on effort/flow variables
and parameters as needed. Assume all elements can be modeled using linear elements and that the L-frame angular
rotational angle, 𝜃, remains small (to ensure linear behavior).
b) Develop a transfer function using the bond graph with impedance methods between the angle 𝜃 and the pressure
Ps (t). Comment on the form of the transfer function and whether the system response is as expected; that is, that 𝜃
is a measure of dPs ∕dt.

Open to fluid
(sea water)

area over which pressure

is applied

C-frame

L-frame
Pivot

Figure B-6.23 Mechanical detection of pressure derivative.

 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
6.8 Problems 369

Two-Port Modeling and systems

Problem B-6.24 Two-port interconnections Two-port composite relations are summarized in Table 6.4. Derive each case
using basic impedance relationships to prove the table contents.

Problem B-6.25 Two-port model of mechanical subcomponents Part of a system under development will incorporate
subsystems A and B as shown in Figure B-6.25.
a) Sketch an impedance bond graph for each subsystem A and B.
b) Write the transmission matrix for each subsystem (refer to Table 6.3).
c) Determine the transmission matrix for the interconnected system AB.
d) For the interconnected system AB, and with F2B = 0 (no applied force) and for an applied velocity 𝑣1A (t), determine
the transfer functions F1A ∕𝑣1A and 𝑣2B ∕𝑣1A .
e) Reconsider (d), but now apply F1A (t) and find the transfer functions, 𝑣1A ∕F1A and 𝑣2B ∕F1A .
f) For the interconnected system, what is F2B ∕𝑣1A if 𝑣2B = 0?

Subsystem A Subsystem B

Figure B-6.25 Composing a system from two mechanical

sub-components using two-port models.

Problem B-6.26 Two-port modeling of cascaded RC filters Part of a system under development will incorporate subsys-
tems A and B as shown in Figure B-6.26.
a) Sketch the impedance bond graph for each isolated filter and determine the transmission matrix.
b) Determine the transfer function relating V2A ∕Vs when filter B is not connected.
c) Determine the transfer function relating Vo ∕Vs , where Vo = V2B and with A and B connected as cascaded filters,
and assume i2B = 0 (i.e., terminals at the output of B are open).
d) For the cascaded filters of part (c), determine the transfer function V2A ∕Vs and compare to the result from part (b).
e) Under what conditions would it be reasonable to assume that,
[ ] [ ]
Vo V2A V
= ⋅ 2B ?
Vs Vs V1B

Filter A Filter B

Figure B-6.26 Modeling cascaded RC filters using two-port

models.

Problem B-6.27 Two-port modeling of coupled RC filters Consider a coupling between the cascaded RC filters in
problem B-6-26, inserted as in Figure B-6.27.
a) Find the total transmission matrix relating the input variables at the source to the output when the coupling element
is an ideal transformer with modulus n.
b) For the case in part (a), determine a transfer function relating Vo ∕Vs when i2B = 0.
c) Repeat part (a) for the case where the coupling is a gyrator having modulus r.
d) Repeat part (b) for the case where the coupling is a gyrator having modulus r.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
370 6 Frequency Response and Impedance-Based Modeling

Filter A Coupling Filter B

Figure B-6.27 Modeling cascaded RC filters using two-port models.

Problem B-6.28 RC impedance bridge Refer to the RC impedance bridge in Figure B-6.9(a).
a) Determine the two-port transmission matrix relating the outputs (Vo , io ) to the inputs (Vs , is ).
b) Find the transfer function Vo ∕Vs for the case, where n identical bridges of the form given are connected in tandem.
c) Derive the magnitude and phase functions for the transfer function in part (b).

Problem B-6.29 Bridged-T RC network A bridged-T RC network is shown in Figure B-6.29.

a) Develop an impedance bond graph.
b) Determine the admittance matrix.
c) For the case when a resistor, R, is placed across the output terminals, find the driving point impedance, Vi ∕ii = z11 .

Figure B-6.29 Bridged-T

RC network.

Problem B-6.30 Dual-inertia rotational coupling An impedance bond graph for the dual-inertia rotational coupling in
Figure 6.51(c) is given in Figure 6.54. Derive the two-port impedance matrix for this coupling in the form,
[ ] [ ][ ]
𝜏1 z z 𝜔1
= 11 12 .
𝜏7 z21 z22 𝜔7

Problem B-6.31 Impedance bond graph to two-port impedance transfer function A system is modeled by the
impedance bond graph shown in Figure B-6.31. Derive an impedance two-port matrix with e1 and e16 as inputs and f1
and f16 as outputs. Assume Z2 = 1∕sC2 , Z6 = R6 , Z7 = sI7 , Z9 = 1∕sC9 , Z11 = sI11 , and Z15 = 1∕sC15 .

Z2 1 Z15

2 4 13 15
e1 e16
1 0 1 0 1 0 1
f1 3 5 8 10 12 14 f16

7 9 11

Z7 Z9 Z11

Figure B-6.31 Impedance bond graph for two-port derivation.

 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
6.8 Problems 371

Problem B-6.32 Twin-T RC network A circuit schematic for a twin-T RC network is shown in Figure B-6.32.
a) Develop an impedance bond graph.
b) Determine the admittance matrix.
c) Find the driving point impedance, Vi ∕ii = z11 , when a resistor, R, is placed across the output terminals

Figure B-6.32 A twin-T RC

network.

Problem B-6.33 Two-port active high pass amplifier An active high-pass filter is shown in schematic form in
Figure B-6.33, with the ideal amplifier modeled using a current-controlled voltage source element with gain 𝜌m .
a) Develop a bond graph model using a dependent source model to represent the current-controlled voltage source
element.
b) Show how impedance methods can be used with the bond graph of part (a) to derive the transfer function Vo ∕Vs .
c) Consider the case where R1 = 100 kΩ, C1 = 100 μF, C2 = 10 μF, and 𝜌m = 1000, and show that,
Vo 10s2
= G(s) =
Vs (s + 1)(s + 100)
d) Plot the magnitude (in dB) and phase (in degrees) plots for this filter.
e) What is the gain (in dB) in the pass band range, and what is the low frequency −3 dB point (in rad/sec)?

Figure B-6.33 Active high-pass filter.

Problem B-6.34 Cascaded 𝚷 network model A distributed model is to be formed of a system using n cascaded Π net-
works of the form shown in Figure B-6.34.
a) Derive the overall transmission matrix for a model formed from n of these networks.
b) Derive the frequency response, eo ∕ei (output over input), in terms of the parameters L, C, and n, assuming an
open-circuit at the output (fo = 0).
c) Assume that L = 1, C = 1, ei = Es sin 𝜔t, and that the output is short circuited. Determine an expression for the
current at the short-circuited output.

Figure B-6.34
Classic Π network.

Problem B-6.35 Cascaded dual-𝚷 network model A distributed model is to be formed of a system using n cascaded
dual-Π networks of the form shown in Figure B-6.35 (dual of network in Figure B-6.34).
a) Derive the overall transmission matrix for a model formed from n of these networks.
b) Derive the frequency response, eo ∕ei (output over input), in terms of the parameters L, C, and n, assuming an
open-circuit at the output (fo = 0).
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
372 6 Frequency Response and Impedance-Based Modeling

c) Assume that L = 1, C = 1, ei = Es sin 𝜔t, and that the output is short-circuited. Determine an expression for the
current at the short-circuited output.

Figure B-6.35
Dual of the classic
Π network.

Problem B-6.36 Two transmission lines in series Two uniform transmission lines with series impedances Z1 and Z2 per
unit length and shunt admittances Y1 and Y2 have been connected in series. The lengths are L1 and L2 , respectively.
a) Determine the overall transmission matrix for the combined lines.
b) An ideal effort source, es (t) is applied at the input of line 1, and the termination end of line 2 is “grounded” at eo .
Determine an expression for the transfer function fo ∕es .
c) For the case when the end of line 2 is “open,” find the transfer function eo ∕es .

Problem B-6.37 Approximating an RC distributed-parameter element with lumped-parameter RC T-models A uniform

fluid line,

P1 P2 = 0
UNIFORM LINE Se
Q1 Q2

has a total (linear) fluid resistance of Rt and a total capacitance of Ct . It is desired to approximate the input (driving
point) impedance by the form,
P1 K(𝜏1 s + 1)
= .
Q1 𝜏2 s + 1
Find the values of K, 𝜏1 , and 𝜏2 for the following cases:
a) Assuming a lumped-parameter model of the form shown in Figure B-6.37(a).
b) Assuming a lumped-parameter model of the form shown in Figure B-6.37(b).
c) Assuming the uniform line is modeled by a uniform distributed-parameter element.

R : Rt /2 C : Ct R : Rt /2 R : Rt /4 C : Ct /2 R : Rt /4 R : Rt /4 C : Ct /2 R : Rt /4

P1 P2 = 0 P1 P2 = 0
1 0 1 1 0 1 1 0 1
Q1 Q2 Q1 Q2

(a) (b)

Figure B-6.37 T-model approximations for an RC distributed-parameter element (a) one-lump RC and (b) two-lump RC.

Problem B-6.38 Avoiding vibration in a pump-driven long line and valve Consider the system studied in Problem A-6.9.
If the line is “mismatched,” there will be a standing wave set up in the fluid. This can induce undesirable vibrations in
the pipe. This can be avoided using a matched condition. Excessive vibrations when in the unmatched condition can
be avoided by careful selection of the line length. This will avoid establishing a standing wave in the fluid.
Assume a hydraulic fluid with 𝜌 = 858 kg/m3 and for an effective bulk modulus of 𝛽 = 1.6 × 109 Pa. The fluid pipe has
radius R = 20 mm and length L = 5.04 m. Also consider a pump with Np = 9 pistons and driven with a constant speed
of np = 1800 rpm.
a) Plot the magnitude and phase plots for the pressure at x = L for the blocked end and matched load end conditions.
b) Determine how the impedance varies along x for the blocked and matched load end conditions and plot Z(x)∕Zc
versus x∕L.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
6.8 Problems 373

c) Plot the pressure as a function of x for the blocked and matched end conditions and plot versus x∕L.
d) Plot Q(x) for the blocked end condition.
e) Comment on how these results could provide insight into mitigating structure vibration of the hydraulic circuit
support system.

Problem B-6.39 Transfer matrix for heat conduction equation The 1D heat conduction equation is given in equation
6.45 as,
𝜕T k 𝜕2 T 𝜕2 T
= = 𝛼 , (6.49)
𝜕t 𝜌c 𝜕x2 𝜕x2
For the uniform 1D H-line, show the transfer equations are given by
⎡ T(x) ⎤ ⎡ cosh(Γx) −Zc sinh(Γx) ⎤ ⎡ Ts ⎤
⎢ ⎥=⎢ ⎥⎢ ⎥,
⎢ ⎥ ⎢ ⎥⎢ ⎥
⎣ f (x) ⎦ ⎣ −Yc sinh(Γx) cosh(Γx) ⎦ ⎣ fs ⎦
where Ts and fs are the variables at x = 0, with f () being the heat transfer flux. The parameters Γ and Zc depend on the
per unit length resistance and capacitance relations, R∗ = 1∕k and C∗ = 𝜌c, respectively. Define the thermal line time
constant as 𝜏t = R∗ C∗ .

Problem B-6.40 Heat conduction using H-line Consider 1D conduction of heat in the solid depicted in Figure B-6.40,
which has thermal conductivity k and thermal diffusivity 𝛼t = k∕(𝜌c), where 𝜌 is the density of the solid and c is the
specific heat. The right side (x = L) is insulated and an area on the left, As , is exposed to the ambient temperature, Ts .
The left side has a convection coefficient hs , and the Biot number is Bi = hl∕k, where l is the thickness of the solid (into
page).
a) Use an H-line model for the solid and sketch a bond graph, applying causality to indicate the indicated boundary
conditions.
b) Determine the transfer function, T𝑤 ∕Ts , in terms of the given constant parameters.
c) Develop a second-order approximation for this transfer function, T𝑤 ∕Ts .
d) Consider a case for an aluminum alloy with L = 0.25 m, l = 0.01m, exposed area length is ds = 0.1 m (so As = ds ⋅ l).
Use representative material and thermal properties and plot the frequency response of T𝑤 ∕Ts .

Uniform solid
Ts
Insulated

T1 over
area As

Figure B-6.40 Heat conduction in a solid,

with convected heat on left side.

Problem B-6.41 Pressure-driven hydraulic piston via a long line with accumulator The hydromechanical system in
Figure B-6.41 has a pressure-source, Ps (t), driving two long fluid lines (A and B), with a gas-bag accumulator in between

Gas-bag
Hydraulic cylinder
accumulator
piston area, Ap
Ca
Ps

‘long fluid line A’ ‘long fluid line B’

Seal friction b

Figure B-6.41 A pressure-driven hydraulic piston via a long fluid line with
accumulator.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
374 6 Frequency Response and Impedance-Based Modeling

the lines that has an effective hydraulic compliance of Ca . The hydraulic cylinder has piston with area Ap and mass
m, and is subject to spring and damping forces from a spring of stiffness k and seal friction b, respectively. Line A has
length LA , line B has length LB , and both have area A. Assume the fluid has bulk modulus 𝛽 and fluid density 𝜌.
a) Write expressions for the propagation operator Γ and characteristic impedance Z0 for each line A and B. Assume
the lines have negligible friction and no leakage.
b) Develop a bond graph model of the system using a W-line for each line A and B.
c) Develop a transfer function that relates 𝑣p to the input Ps .
d) Use product expansions for cosh() and sinh() functions in the result from (a) to compose a model using rational
polynomial transfer functions
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
375

Modeling Feedback Control Systems

Previous chapters have focused on modeling physical systems, many of which may form part of controlled systems.
Figure 7.1(a) shows a highly generalized diagram where physical outputs from a system of interest are measured,
processed, and used to inform controlled inputs to the system. This diagram is meant to describe how systems are inter-
connected, using both power bonds and signals along with conceptual functional blocks of key subsystems comprising
controlled systems. Feedback control systems more often make use of block diagrams of the form in Figure 7.1(b).
This diagram represents the form of classical single-input–single-output (SISO) systems used in feedback control system
modeling and analysis. These block diagram descriptions use functional blocks as discussed in Chapter 4. Diagrams of
the type shown in Figure 7.1 are useful in the steps taken toward analyzing and designing feedback control systems.
Often, the model of the plant is simplified to enable application of control design analysis. However, a more complex
plant model may be used in simulation studies when evaluating a controller’s effectiveness. These are some examples
of common ways that modeling plays a role in feedback control studies. This chapter reviews concepts useful for these
purposes.
Modeling insights play a key role in control system design, development, and evaluation. For example, a first step in
formulating a control system is to identify and categorize controllable inputs, u, and disturbance inputs, 𝑤, as illustrated
in Figure 7.2. However, control analysis and design tasks aim to meet specified response characteristics in output(s) of
interest, y, by manipulating controllable inputs in the presence of disturbances and any significant variations in system
parameters. The system should be stable at all times, and the output(s) of interest should either remain within a certain
range (regulated) or must track a command input signal.
Thus, this chapter will also review techniques commonly used in feedback control analysis. The emphasis will be on
linear, scalar SISO systems, but a brief discussion is also provided on state-space methods. There are many excellent text-
books such as by Ogata [53], Dorf and Bishop [61], and Franklin et al. [93] that provide a thorough introduction to feedback
control and analysis methods, and these will be referenced throughout this chapter.

7.1 Feedback Control Representations

Block diagram descriptions provide a way to represent integration of system components both conceptually and in physical
system model form. As previously shown in Chapter 4, block diagrams can be used with bond graphs to form a hybrid
representation of signal and power flow. Note that block elements can represent causal relations between input and output
variables. Block diagrams may include nonlinear functional forms, in addition to all forms of linear function elements.
These representations were adopted early on in describing analog and digital computing [40, 88, 132] and formed the
basis for the block diagram descriptive approach commonly used in contemporary computer-aided analysis and design
environments (e.g., MATLAB/Simulink, Scilab, and AMESim).
The block diagram in Figure 7.1(b) shows interconnected blocks representing typical and essential elements in feedback
control. Based on this diagram, some common definitions can be defined as1 :

1) Plant – any physical object/system to be controlled

2) Process – any operation to be controlled

1 See also the classical standard from AIEE in 1951 [133].

Modeling of Physical Systems: Simulation and Control, First Edition. Raul G. Longoria and Joseph J. Beaman.
© 2025 John Wiley & Sons Ltd. Published 2025 by John Wiley & Sons Ltd.
Companion Website: www.wiley.com/go/Modeling_of_Physical_Systems/1e
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
376 7 Modeling Feedback Control Systems

Input(s) Output(s)
Physical
system

Reference Distrubance Output

Actuation and Sensing and +
manipulation + c +
measurement Controller Plant
–

Power conversion Signal processing

or amplification controls, and Feedback
Measurement
decision

Power or energy
source
(a) (b)

Figure 7.1 (a) Diagram of general approach for a controlled system. (b) Canonical single-input–single-output feedback control
diagram.

Figure 7.2 A system to be controlled has inputs that can be manipulated, u, as well as disturbances,
System to be 𝑤, that cannot be controlled. Outputs of interest are designated here as y.
controlled

3) System – of physical, biological, economic, etc. form with discernible boundary

4) Inputs – variables prescribed by environment or other system
5) Outputs – variables of interest and produced by the system
6) Disturbance – a signal or variable that adversely affects the output of a system
In this context, feedback control is an operation that, in the presence of disturbances, ud , tends to reduce the difference,
or error e, between the output of a system, y, and a reference input, r. This is achieved by a controller responding to the error
signal. The controller output, um , and any disturbance inputs combine to form a total input to the plant, u, here a single
input.2 Figure 7.1(b) only becomes useful for analysis and design when specific interconnections of signals and quantifiable
and functional blocks replace the conceptual description. In other words, each block should contain a valid function that
relates the input and output variables. Linear feedback control system analysis and design, for example, can be conducted
using open and closed-loop transfer functions (CLTFs) derived from these diagrams (as in [53, 61, 93]).
Transformation into useful formulations for most contemporary and practical controlled systems requires making
assumptions that help define the signal and power flow, as illustrated at a high-level in Figure 7.1(a). An example is shown
by the diagram in Figure 7.3 which emphasizes bilateral power flow between key elements of a system. Distinguishing
when an interconnection between two system components involves conveyance of signal (i.e., relatively low or insignif-
icant power flow, only information) or power is important. Standard block diagrams assume signal input/output forms.
Once formed, block diagram algebra can be used to manipulate and simplify the system for analysis and design. Mixed or

Power
Reference Output
Comparator Controller
Plant
multiport /actuation

Feedback
Measurement
Signal level

Figure 7.3 Diagram illustrating typical bilateral signal flow in a feedback control system, with dashes used to indicate that relatively
low back effects. For example, a “good” measurement system will not typically load the plant and the back effect (dashed line) can be
neglected.

2 Plants (systems) with multiple inputs are discussed in Section 7.5.

 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
7.1 Feedback Control Representations 377

Supply of
fluid Time
scale

Float Controlled
valve flow,
+
Regulating –
vessel Valve Regulating
vessel

(a) (b)

Figure 7.4 (a) Simplified schematic of a Ktesibios water clock based on the reconstruction by Diels shown in [134]. (b) Block diagram
of the Ktesibios water clock per. Source: [134]/Massachusetts Institute of Technology.

hybrid model formulations that combine block diagrams with bond graphs provide a way to systematically formulate such
models, allowing system component boundaries to be carefully assessed. This approach to a controlled system model
analysis can reveal any possible shortcomings in assumptions and provides a way to revisit a system if and/or when
changes need to be made. This approach was illustrated for a motor-drive system in Chapter 4 with Example 4.12.
Proper modeling not only identifies how components in a controlled system interact but can also help determine whether
a system achieves a feedback control function. This approach is demonstrated well by Mayr [134] in his comprehensive
study on the historical origins of feedback control. Mayr defined and applied the following criteria in his study of systems
that had been built throughout history and considered to embody feedback control:
1) The purpose of a feedback control system is to carry out commands; the system maintains the controlled variable equal
to the command signal in spite of external disturbances.
2) System operates as a closed loop with negative feedback.
3) The system includes a sensing element and a comparator, at least one of which can be distinguished as a physically
separate element.
Functional block diagram descriptions were developed for a wide range of physical control systems and used to determine
to what extent a system satisfied the proposed criteria for feedback control. This study was challenging, since many of
these systems are known only through limited and often incomplete historical records. Nevertheless, nonlinear functional
forms were assumed in order to represent the relations between key variables and thus enabled block diagram formulation.
This is common practice in block diagram modeling. In addition to a general nonlinear block function, say, y = f (u) (as
shown in Chapter 4), specific nonlinear functions are often useful; for example, absolute value, saturation, sign, as well as
basic mathematical functions.3 The role of any nonlinear behavior needs to be recognized. Sometimes a system relies on
nonlinearity to achieve an essential function, while in many cases, it is undesirable and/or parasitic.
Constructing a block diagram helps reveal the system structure, as in Figure 7.1(b), including the inherent causal relations
[135]. Block diagrams also make it possible to identify whether the key criteria proposed by Mayr are satisfied and if the
system provides feedback control. The historical systems analyzed by Mayr revealed insight into the physical mechanisms
used in regulating processes with feedback. For example, Mayr determined that the water clock of Ktesibios c. 250 BCE
was the earliest documented feedback control system [134]. This system is shown in Figure 7.4(a) and illustrates how a
float valve regulates flowrate, Qo . The float regulator, composed of a valve and vessel, sets Qo by virtue of a commanded
height hr .
The block diagram constructed by Mayr is shown in Figure 7.4(b). This diagram satisfies all the criteria of a feedback
control system: (i) a command in terms of hr can be specified, and Qo ∝ h; (ii) negative feedback is used to control the
regulated flow, Q𝑣 ; (iii) the float regulator acts as a sensing element and a comparator. The block diagram features nonlinear
functions to convey a qualitative understanding of the elements.

3 These functions are standard elements in software programs that enable user-specified block diagram models.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
378 7 Modeling Feedback Control Systems

Figure 7.5 A mixed block diagram/bond graph description of the Ktesibios water clock.

Figure 7.5 shows part of the block diagram integrated with bond graph models. The models show how the functional
block diagrams for the valve and regulating vessel could have been derived from the physical models. This system provides a
good example for how signal flow interconnections are used for modulation and to construct elements that may not interact
directly with the significant power flow in a system. For example, the float valve is effectively a modulated hydraulic resistive
element where the pressure-flow constitutive relation depends on an implied error adjustment between a reference height,
hr , and the actual height, h, in the regulating vessel. The effect of this error, e, an essential part of the feedback control
function, is modeled by the modulated R𝑣 .
Because the valve flow, Q𝑣 , can only flow into the regulating vessel, a signal flow conveys Q𝑣 to the model of the regulating
vessel. There is no “back flow” effect. The regulating vessel has a hydraulic capacitive element Cr𝑣 that models the accumu-
lation of water and thus dictates the pressure Pr𝑣 . The bond graph more clearly shows how the regulating vessel height, h,
depends on the pressure, Pr𝑣 , which is (causally) determined by the volume state of the hydraulic capacitive element, Cr𝑣 .
An additional signal bond conveys the controlled flowrate, Qo , which could be used for coupling to the actual clock water
vessel.4
Identifying true feedback control is more challenging when employing physical feedback, as in classical/historical feed-
back control systems studied by Mayr [134]. This is not expected to be the case for those systems intentionally designed
following the standard form in Figure 7.1(b). The main issue is ensuring that proper system boundaries are formed to
identify and model the control system components. This will enable formulation of a proper block diagram and application
of standard analysis methods, as will be briefly described in later sections and reviewed in detail by Ogata [53], Dorf and
Bishop [61], and Franklin et al. [93].
Many early feedback control systems such as those studied by Mayr [134] used mechanical and fluid mechanisms to
realize key control functions. The somewhat ingenious use and integration of these devices with the “plant” is what can
make it difficult to determine whether there is true feedback control at work. A critical model assessment, for example,
may reveal that there is actually only a resistive-type element that provides “self-regulation” rather than feedback control.
Modeling such systems and applying the criteria from Mayr [134] helps identify such effects. While construction of control
systems using “classical” means may not be as prevalent nowadays, they can still be found in some use cases. Understanding
how they work can provide insight into the evolution of control system technology. Contemporary control systems are more
likely to use system components that readily integrate into a standard feedback control architecture of Figure 7.1(b).
Eventually, the goal may be to build a complete block diagram description, which can incorporate a wide range of model
element types. For systems where a single controlled variable is of interest and where linearization is valid, transfer function
representations relating key control system variables can be obtained and linear control system analysis and design methods
are applied. In some cases, a state-space approach to control design may be preferred. The rest of this chapter describes how
modeling plays a role in helping get to the preferred approach. As shown above, an effective approach is to develop models of
controlled systems by merging bond graph models with block diagram descriptions of functional components. Taking this
step can help clarify assumptions and identify where difficulties may arise.
One more example that illustrates how some thought may need to go into the model is a basic lead-screw table drive con-
figured as a feedback control system as shown in Figure 7.6. The shaft from an electric motor (servodrive) is directly attached
to the lead-screw of the table, and a feedback control system is formed by closing the loop with measurement of the table

4 The Ktesibios water clock is modeled further in solved Problem A-7-3.

 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
7.2 Modeling Control System Elements 379

Forces
Workpiece
+ Bearing Table Bearing
Controller
– Lead screw drive

Servo drive

Measurement

Figure 7.6 Servo-controlled lead-screw-driven table.

position, xt , the controlled variable. The controller output is a signal, u, sent to a servo/motor drive unit. The controller and
error-forming function in this diagram are digital. The interaction between the motor and the lead-screw/table needs to be
finalized, so the coupling is indicated by a single power bond in this diagram. The lead-screw drive may be non-backdrivable.
This means the motor easily drives the table but the table dynamics and load cannot necessarily drive the motor. This nonlin-
ear behavior may be a desirable effect from a design standpoint, but it can present challenges in formulating an ideal linear
model. In such cases, controller design/tuning may be better accomplished by turning directly to nonlinear simulation
methods rather than linear control design methods. A quick model assessment can often aid in making such a decision.
The sections of this chapter further describe how physical system modeling can provide insight into understanding
feedback effects. In addition, modeling of the system plant as well as other parts of a feedback control system can make sure
that a proper model is formulated. Emphasis is placed on reviewing how linear methods of analysis reviewed in Chapters 5
and 6 are used, along with related techniques often used in linear control system analysis and design.

7.2 Modeling Control System Elements

Models of control systems components should be sufficiently detailed so that key effects and functionality are properly
represented and parameterized. Block diagrams enable this to be accomplished and provide a basis for analysis and design.
When the elements are all linear, it is possible to derive linear transfer function relations. Block diagrams are also used
to represent a wide range of system types and often used for direct numerical simulation methods. This section reviews
this approach for modeling the types of system elements typically used in feedback control systems, such as shown in
Figure 7.1(b). We first describe the role of linear transfer functions and general block diagram usage, incorporating bond
graphs as needed. Examples of how representative control system components are modeled are also provided.

7.2.1 Transfer Function Models in Control Systems

Linear control system elements can be represented either in factored pole-zero form (as in Chapter 5),
N(s) K(s + z1 )(s + z2 ) … (s + zk )
G(s) = = , (7.1)
D(s) (s + p1 )(s + p2 ) … (s + pn )
or using combinations of basic factors (as introduced in Chapter 6):
● Gain factor, K
● First-order factors, (𝜏t s + 1)±1
[ ]±1
● Second-order factors, (s∕𝜔n )2 + 2𝜁s∕𝜔n + 1
● Integral and derivative factors, s±1
These transfer function models can be formulated from “first principles,” as discussed in Chapters 4–6, including an
impedance approach with bond graphs (see Section 6.3). The standard forms above are amenable to linear analysis and are
commonly the basis for building models through experimental testing. Both time-domain and frequency domain methods
are used for this purpose (see Chapter 6, but also extensive literature base [100, 114], etc.). Even an approximation of a model
based on the standard forms makes it possible to conduct initial control analysis. Results can help identify characteristics
of system elements needed to meet desired performance.
It should be remembered that these models are intentionally simplified to facilitate analysis and design of systems and
controllers. Neglecting any poles, for example, that are far away from “dominant” poles, is a common form of model reduc-
tion. Indeed, their presence signifies that assumptions made in the modeling process may have given importance to effects
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
380 7 Modeling Feedback Control Systems

that should actually be neglected. Care should be taken, of course, that such poles (which have high-frequency dynamics)
don’t lead to instability.
Models that take the form of the transfer function of equation 7.1 are used in stability analysis and help identify whether
the system is physically realizable. A physically realizable system has a proper transfer function if there are at least as many
poles n as there are zeros m; that is, n ≥ m. If n < m, the transfer function is referred to as improper, in this sense. A transfer
function is said to be strictly proper if n > m.
One way to understand the meaning of these terms with respect to equation 7.1 is to think about how the function
behaves as s gets very large. Recall that we can relate s to j𝜔 in the frequency domain. For very large s, G(s) → Ksm−n .
Thus, a physical system would tend to have a frequency response at large 𝜔 that has a magnitude tending to |G( j𝜔)| =
|K( j𝜔)m−n | = K𝜔m−n . If the system has m > n, then a system with an input u(t) = uo sin 𝜔t will produce an output at high
frequency, y(t) = K𝜔m−n uo sin(𝜔t + 𝜙). For m > n, the amplitude will have “infinite amplification” at “infinite” frequency,
which is not possible for physical systems. Thus, for most physical systems, n < m and the response tends to zero at high
frequencies, as shown in Chapter 6. Note, however, that controllers and compensators often have m = n (e.g., see Table 7.3).
Consider the relatively simple case of a mass with linear friction and an applied force as shown in Figure 7.7(a). It is
desired to find a transfer function relating the mass position, xm , to applied force, Fa . The bond graph in Figure 7.7(b)
illustrates assumptions made about the ideal applied force and an ideal sensing of xm , modeled by integrating the mass
velocity. The dynamic equations in the bond graph are converted into block diagram form as shown in Figure 7.7(c).
Block diagram algebra gives,
1 1 1∕ms x 1
xm = 𝑣m = ⋅ F ⇒ m = GxF (s) = .
s s 1 + b∕ms a Fa s(ms + b)
Consider a force actuator system will be used to provide Fa . The actuation system may have a transfer function, Ga (s) =
Fa ∕ua , where ua is a control signal. The actuator and mass–damper system can be interconnected to form a composite
model Gxu = Ga GxF as below:

A key assumption in this formulation with block diagrams is that the actuator system can provide Fa regardless of the
mass velocity, 𝑣m , inherent to the coupling. This assumption is conveyed in the bond graph model of Figure 7.7(b) with a
signal setting the effort at a 0-junction. This assumes that the force actuator system can generate any force over the range
of expected 𝑣m .
Interconnecting components in block diagrams in this way is common, especially because it can be convenient to reuse
modular models. The interactions should always be verified. To illustrate further, consider a common example of modeling
two low-pass RC-circuit modules, as shown in Figure 7.8(a). Simply cascading the RC filters as shown in Figure 7.8(b) by
just using the block diagrams calls for multiplying the transfer functions to give,
1 1 1
G(s) = G1 (s) ⋅ G2 (s) = ⋅ = ,
𝜏1 s + 1 𝜏2 s + 1 R1 R2 C1 C2 s2 + (R1 C1 + R2 C2 )s + 1
where the subscripts denote the first and second module component parameter values.

+
–

(a) (b) (c)

Figure 7.7 (a) Mass–damper with applied force, Fa . (b) Hybrid diagram with input–output signals. (c) Block diagram.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
7.2 Modeling Control System Elements 381

1 2

1 1 2
1 2

1 1 1 1
+1 +1 +1

(a) (b)

Figure 7.8 (a) Basic RC low-pass filter and linear block diagram element, where 𝜏t = RC; (b) two cascaded RC low-pass filters
incorrectly modeled with cascaded block elements.

A proper model formulation of the cascaded RC filters, however, will show that,
Vo 1
=
Vi R1 R2 C1 C2 s2 + (R1 C1 + R1 C2 + R2 C2 )s + 1
indicating there is an extra term in the denominator not found in the block diagram form. While this seems like an obvious
error to avoid, it can arise in many computer-aided block diagram environments when two blocks are simply cascaded. It is
more valid to construct the mathematical model from first principles, such as using a bond graph. Adopting an impedance
approach as shown in Chapter 6 (e.g., by using a two-port description or a bond graph form) also inherently captures the
coupled interaction. In the end, drawing a block diagram connection with a signal (bond) should always elicit the need to
“check” the coupling between the two elements: the absence of loading or back effect should be rationalized.

Example 7.1 Integrating a permanent-magnet DC (PMDC) motor with rotational load

Torque–speed curves and other steady-state operating characteristics provided by manufacturers can provide informa-
tion needed for developing a control system model. A schematic for a steady-state PMDC model is shown in Figure 7.9.
Torque–speed curves for these motors take the form,
𝜏o = kV Vs − k𝜔 𝜔m , (7.2)
where kV = rm ∕Rm and k𝜔 = 2 ∕R
rm m+ Bm . With these parameters, the stall torque, 𝜏s , and no-load speed, 𝜔nl , can be deter-
mined, as shown for an input voltage, Vs = V3 .
Develop a transfer function that integrates a torque–speed curve model with a rotational inertia that relates output rota-
tional speed to the input voltage; that is, G(s) = 𝜔∕Vs .

Figure 7.9 A PMDC schematic with associated representative trends in

torque–speed curves.
3

= 3

Solution
Since the torque–speed relation requires two inputs, it is necessary to develop an integrated model with the inertia. This can
be done in two ways. From a first principles model and given the motor parameters, a bond graph can be constructed as
shown below:

A single state is indicated for the rotational inertia momentum, ḣ = J 𝜔̇ = 𝜏o , where 𝜏o = 𝜏m − 𝜏Bm , which is effectively
the same as equation 7.2. By causality, take 𝜔m = 𝜔. We then write,
J 𝜔̇ = 𝜏o = 𝜏m − Bm 𝜔
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
382 7 Modeling Feedback Control Systems

= rm im − Bm 𝜔
= rm (Vs − rm 𝜔)∕Rm − Bm 𝜔
[ 2 ]
rm rm
= V − + Bm 𝜔
Rm s Rm
= kV Vs − k𝜔 𝜔.

Thus, we can now write the desired transfer function from,

(Js + k𝜔 )𝜔 = kV Vs ,

or,
𝜔 kV
= . (7.3)
Vs Js + k𝜔
An alternative approach is to use equation 7.2 to construct a block diagram. For the rotational inertia, we know J 𝜔̇ = 𝜏o
and with no other loads, 𝜔∕𝜏o = 1∕Js. We can then integrate the torque–speed equation and inertia as shown below:

This block diagram construction using the basic torque–speed relation can arise from a purely empirical model of the
motor. With no other torques in this model, the block diagram algebra gives,
1 1 [ ] k k
𝜔= 𝜏o = kV Vs − k𝜔 𝜔 = V Vs − 𝜔 𝜔 ⇒ (Js + k𝜔 )𝜔 = kV Vs ,
Js Js Js Js
which provides the same transfer function as before. In either approach, if the motor will be driving an additional load it
may need to be accounted for in the model development. A similar approach can be used in Problem B-7.6 which asks for
integrating the steady-state characteristics for a two-phase ac motor with a load.

The rest of this chapter will continue to demonstrate ways to examine the models typically adopted in control systems.
Section 7.2.2 discusses how block diagrams help form model relations.

7.2.2 Using Block Diagram Descriptions

Block diagram algebra provides a way not only to construct a diagram but also to analyze and reduce into forms use-
ful for analysis, such as linear transfer functions. The blocks represent causal, functional relationships between an out-
put and an input signal, and can generally be nonlinear. It is common to use available modular descriptions, especially
as block diagrams become complex. A system may involve multiple loops, and even different signal types (continuous,
discrete). While the emphasis here is on continuous-time systems, the extension and interconnection with discrete-time
(DT) elements is well established [136]. Some common equivalent block diagram forms useful in reduction techniques
are summarized in Table 7.1, and more extensive discussions of reduction and manipulation methods can be found in
Ogata [53]. Perhaps the most important practice to follow when using block diagrams is to always make sure the physical
component models are constructed carefully, avoiding the issues described earlier. As a control system block diagram is
constructed, it is important to clearly partition the system and to only form block diagram elements once the input–output
forms have been confirmed. Once this is done, many useful techniques can be readily applied.
Consider, for example, the relations in row 5. The output is determined by y = Ge = G(u − Hy) = Gu − GHy,
G
y= ⋅ u = Gcl ⋅ u.
1 + GH
Thus, Gcl is identified as a closed-loop transfer function relation. This is a very useful relation for converting loops in a block
diagram into a single input–output block.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
7.2 Modeling Control System Elements 383

Table 7.1 Linear block diagram equivalents useful in manipulation and

reduction, where u and 𝑣 are input signals and y is an output.

Original block diagram Equivalent block diagram

1 + – + – –
– –
1

4 + +
1
– –

5 +
1+
–

Example 7.2 Permanent-magnet DC (PMDC) motor driving rack and pinion load
Instead of connecting to a single rotor, the PMDC motor in Example 7.1 will be used to move a table through a pinion
connection, as shown in Figure 7.10. Assume all shafts and gears are rigid. Propose a feedback control diagram to model
the system by identifying the plant. Assume a transfer function, G1 (s) for the controller and that the table position, xt , is
measured through H(s).

Pinion with rotational

inertia,

Table/rack of mass

Pinion radius,

Figure 7.10 Integrating the PMDC motor with a pinion/rack-table load.

Solution
Begin by finalizing a plant model that integrates the PMDC motor and the pinion/rack-table load. Assume the table has
mass mt and that the rotational parts of the plant include inertia on the motor side, the pinion (gear), and any shaft elements.
Referring these elements to the rotational side will allow us to use the transfer function from Example 7.1. A total (effective)
rotational inertial for the transfer function is found from (using equivalent energy),
[ ]
1 1 1 1 1
Jeff 𝜔2m = Jp 𝜔2m + mt 𝑣2t = Jp + mt rp2 𝜔2m .
2 2 2 2 2
Thus, the transfer function for the plant after integrating the PMDC torque–speed curve model and the pinion/rack-table
will take the form,
𝑣t 𝜔 rp k V
= rp ⋅ m = .
Vs Vs Jeff s + k𝜔
Now, a basic closed-loop control can be configured as shown in Figure 7.11.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
384 7 Modeling Feedback Control Systems

Figure 7.11 Block diagram.

It is now possible to write a closed-loop transfer function for this system relating the table position, xt , to the reference
command xr . The relation from Table 7.1 on row 5 can be applied, with G = rp kV ∕s(Jeff s + k𝜔 ) and H = 1, to give,
r k
xt G1 s(J ps+k
V
G1 rp kV G1 r p k V
𝜔)
= Gcl = eff
= = .
xr r kV
1 + G1 s(J ps+k ⋅ 1 s(Jeff s + k𝜔 ) + G1 rp kV Jeff s2 + k𝜔 s + G1 rp kV
eff )
𝜔

This system is a simplified form of the system in Figure 7.6. Working the simpler case provides some insight into
what to expect when a different type of motor drive is used. Also, a lead screw introduces nonlinear behavior because of
non-backdrivable characteristics.

When the elements of a feedback system are linear and can be numerically defined, available software programs can con-
struct and derive transfer function relations. These can then be analyzed using available linear system analysis programs.
Software tools of this type can be very useful in analysis and design iteration. Some of the key categories of functions that
are of particular use include means for:
● constructing linear system models (transfer functions, state space, etc.)
● interconnecting model elements (appending elements, forming feedback constructions, etc.)
● conducting time-domain simulation (forced response with steps, impulse, general inputs, etc.)
● plotting frequency domain representations (Bode plots, Nyquist, etc.)
● conducting specific control system analyses (root locus, pole-zero maps, stability margins)
Both commercial and open-source software tools of this type are continuously extended and improved. Consequently,
the extensive documentation and example databases available through online resources provide the most effective tutorial
guides to their use.5 Some of the most useful functions available in these software tools enable creating system mod-
els as transfer functions using either polynomial coefficients or zero-pole pairs. Once these system elements are created,
additional functions are used to form connections, such as series or feedback form, to reflect a block diagram description.
For example, the closed-loop system in Example 7.2 was formulated analytically. However, if the need is to conduct
numerical studies on specific models, a series of commands as below can produce a closed-loop system:

1 rp = 0.1; % define pinion radius

2 Jeff = 1; % define effective rotational inertial
3 % define torque--speed curve constants
4 kw = 0.25; kV = 10;
5 G1 = 1; % specify a controller TF
6 % create the plant TF relating x to voltage
7 % note: the polynomial coefficients are specified
8 GxV = tf([rp*kV],[Jeff,kw,0]);
9 % connect the forward path TFs
10 G = series(G1,GxV);
11 % define the feedback TF
12 H = 1;
13 % close the loop with feedback in H
14 Gcl = feedback(G,H);

5 In MathWorks MATLAB software, the Control System Toolbox provides a comprehensives set of such tools. An open source library is also
available in Python.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
7.2 Modeling Control System Elements 385

In this example, the system object created, Gcl, can now be analyzed using linear analysis routines. For example, the
response of the table position to a step change in the reference command can be readily found using a step() function.

7.2.2.1 Nonlinear Elements

Block diagrams may also include nonlinear elements that arise in the physical modeling process or during revisions of the
model. There are a diverse set of nonlinear elements that can arise, and users of computer-aided block diagram modeling
environments will be familiar with ways that the relationship between a block’s input and output can be related. Standard
mathematical functions that are commonly used take the sign, absolute value, square roots, etc. of inputs. Alternatively,
some modeling functions are specific such as Coulomb friction, hysteresis, saturation, and more. A transport delay is often
used in some system models to represent pure delays, such as y = f (t − T), meaning that the output y is simply delayed by
a time T. In the s-domain, this function is simply a multiplication by e−Ts . Nonlinear elements cannot be used in linear
transfer function forms and require linearization of components or of the entire system. Another approach is to directly
simulate the system, which is the approach taken by computer-aided software programs. Dealing with nonlinear effects in
control systems is a specialized field and will not be extensively dealt with in this book, aside from examples showing how
to apply linearization.
For some classes of nonlinear elements, however, a linearized model can provide suitable representation. Linearization,
which is introduced in Chapter 4 of this book, seeks to find a system representation using linear function relations. A critical
first step is to identify an equilibrium condition of interest about which the system is assumed to operate in a way that can
be predicted by linear or linearized models. In the present discussion, this would allow use of linear block diagram models
as well as reduction and analysis procedures.
There may be different approaches that provide a suitable linearized model. For example, consider a block diagram where
there is a one nonlinear function block. This block could be linearized and simply replaced by a linear element. If the
operating point is zero, then for such a simple case, the block diagram can be considered a complete response. Issues arise,
however, if the linearization is about a nonzero equilibrium value of the block input. In this case, when the block diagram
is interpreted in linear form then it should be as deviations about the equilibrium. This subtle and practical point can be
illustrated with the tank shown in Figure 7.12(a). The bond graph in (b) of this figure uses signal bonds to define input and
output connections for the system, and the block diagram in (c) simply√ conveys the physical model equations that follow
from the bond graph. The flow exiting the tank is given by, Qo = K |V c |, where K is assumed to be a constant.6
Consider the case where this system will be integrated into a linear control system, so it is desired to find a transfer
function relating the output y = Qo to input flow u = Qs . One approach is to linearize the system directly from the state
equations derived from the bond graph. Here there is only one state equation,
√
V̇ c = Qs (t) − K |V c |,

Constant tank
area,

Nonlinear resistance
(a) (b)

(c) (d)

Figure 7.12 (a) Tank with outlet flow through nonlinear resistance; (b) bond graph; (c) nonlinear block diagram; (d) linearized block
diagram.

6 Note that K is either determined empirically or can be shown to be a composite parameter related to fluid properties and geometry of the flow
passage.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
386 7 Modeling Feedback Control Systems

and the output equation is,

√
y = Qo = K |V c |.
Methods for finding G(s) = y∕u as discussed in Chapter 4 are followed. Alternatively, we can focus on using the block dia-
gram and linearizing the Qo function directly. The two approaches are substantially the same. In both cases, an equilibrium
for the system is found by,
√
V̇ c = 0 = Qse − Qoe ⇒ Qoe = K |V ce | = Qse ,
thus, V ce = (Qse ∕K)2 . Now, approximate the function for Qo by a tangent linearization about the equilibrium state and input,
𝜕Qo ||
Qo ≈ Qoe + ̃c ,
⋅ (V c − V ce ) = Qse + 𝛼 V (7.4)
𝜕V c ||V ce ,Qse
√ √
where the equilibrium flow is Qoe = Qo (V ce ) = K (Qse ∕K)2 = Qse . It can then be shown that, 𝛼 = K∕2 V ce = K 2 ∕2Qse .
Use equation 7.4 to define the flow variation or linearized flow relation, Q ̃ o = 𝛼V
̃ c , where the value of 𝛼 depends on Qse .
These results enable construction of a linearized block diagram in Figure 7.12(d), where variables are indicated with the
“tilde” to emphasize these are for the linear case. It should be emphasized also that the actual flow from the system is then
given by Qo = Qoe + Q ̃ o . That is, the linearized system provides response variations about the equilibrium state. This explicit
description of this process should provide guidance for more complex cases.

7.2.3 Measurement and Actuation Systems

Most control system models require representation of how controlled variable(s) are measured as well as how physical
control inputs are realized and connected to the plant. In some cases, highly idealized models may be used, as in some of
the previous examples in this section. When it is assumed that a measurement can be made ideally, a simple signal bond
on a bond graph has been used. In some cases, however, the measurement and actuation systems can introduce static and
dynamic characteristics that need to be accounted for in an overall control system model. The permanent-magnet dc motor
is a good example of this case.
It can be helpful to identify distinct functional elements, as suggested in Figure 7.1(a). The sensing and measurement
function might be broken up into typical elements, as suggested by the word bond graph in Figure 7.13, while the ampli-
fication and actuation system can be modeled as in Figure 7.14. These conceptual models help identify possible areas that
may be relevant. Generally there is power flow between the components, but in some cases, decisions can be made to use
signal flow.
Key functions in both measurement and actuation systems are achieved by transducers, which are devices that transform
energy from one form to another. Processes do not typically fall under this definition. Sensors are transducers that extract
information from a system to measure a physical quantity, or measurand. If not directly sensed, a measurand may be inferred
using a sensor’s functional structure. For example, strain induced in a mechanical element using strain gages may be used to

Physical Sensor Signal Recorder/indicator

system system conditioning processing
controller

Power source Power source

Figure 7.13 Word bond graph of typical sensor and measurement systems.

From Amplifieror Actuator Physical

controller modulator system system

Power source Power source

Figure 7.14 Word bond graph of typical amplifier and actuator system.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
7.2 Modeling Control System Elements 387

measure force. Sensors are usually designed to convey relatively low levels of power flow. In contrast, actuators are typically
designed to handle relatively high levels of power. Actuators inject energy into a system for the purpose of configuring a
desired state. These dual concepts are consistent with Figures 7.13 and 7.14.
If deemed essential, models of either sensors or actuators can begin with word bond graphs and block diagram descrip-
tions. The aim of such a process is to properly convert to a SISO block diagram form that fits into a control system block
diagram. The need for additional power flow in some of these devices should not be overlooked and is sometimes embedded
in block diagram element models.
It is not always possible nor necessary to reticulate these models. The model should match the available information.
It may also be necessary to integrate with the plant, as was shown with the permanent-magnet dc motor in Example 7.2 of
Section 7.2.2. It is most common to adopt highly simplified representations in early-stage control system analysis and design.
In such cases, it is best to reduce a component model to its most basic form so that the number of parameters introduced
into the final model form is minimized. It is also possible to take advantage of the fact that many sensors and actuators are
designed and/or operated so that they exhibit desirable static and dynamic characteristics. As such, in linearized form, they
can often be described as in Section 7.2.2.
Many of the transducers used in controlled systems are electromechanical. Sensors are typically resistive, capacitive, or
inductive in nature. Many are realized using micro-electromechanical technology and thus are dealt with as ideal sensors.
This means their representation in block diagram form can be readily converted into a transfer function form, Gs (s) = ys ∕y,
where y is the measurand and ys the output signal. The signal ys is often an analog electrical voltage or current. Generally,
however, signals might be encoded in different forms in measurement systems. The signals may be discretized and encoded
digitally or put in time-based forms such as pulse width, frequency, or phase. For the most part, the discussion here will
be restricted to continuous time signals in controllers. Chapter 8 discusses electromechanical device modeling that can be
used to model a wide range of sensor and actuator types. Any model should be suited for the control system task. If the
dynamics of a transducer is of interest, the model should allow including sufficient information about key physical elements
and parameters.

7.2.4 Error Detection and Controller Functionality

Another category in control system models provides the key functional role of comparing the variable to be controlled
to a variable representing a reference (or command). Typically, this takes the form of an error function represented by
the summing block, which directly conveys a purely algebraic form, say, e = r − y. Indeed, the “comparator” in item 3 of
Mayr’s list, which is typically an error variable, is now often realized as a line of computer code, even though it might still
be conceptualized by the a summation block element. As such, the block diagram representation for the error function in
many cases is somewhat trivial.
In contrast, the systems critically evaluated by Mayr used purely physical constructions to form error functions in
feedback control. Many of these constructions are still commonly found in introductory feedback control books even
though they may not be as commonly used to actually build modern control systems. Some of these constructions are
summarized in Figure 7.15.7
Physical mechanisms that can serve a comparator or error-detecting function are likely the most innovative part of some
feedback control systems. These were essential before analog and digital electronics and can still be found in some industrial
systems. Models of these mechanisms can be used to prove their function. Some of the simpler examples are those shown

Block Code Mechanical Electrical Fluid (pressure)

r = read(sensor1)
+ y = read(sensor2) +
=
– // compute error Bellows
–
e = r–y – ≈ –
+ –
= –

Figure 7.15 Detection of error as represented by the block summation element is realized in many ways, with examples shown here
ranging from computational code to physical forms found in many classical feedback systems.

7 See Appendix B in Thaler and Brown [137] for a description of error detectors from different energy domains.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
388 7 Modeling Feedback Control Systems

in Figure 7.15. Some classical feedback control texts such as Ogata [53] and Van de Vegte [138] provide good descriptions of
these system elements. One of the key features that is common among these devices as part of an overall feedback control
system is that, like sensors, they accomplish their function with relatively low power flow. As such, it is almost always
possible to represent them in a block diagram form.
The controller or control functionality includes those parts of the feedback control system that act on reference and error
signals in order to generate a control signal, say u, which is used to drive an actuator or a plant directly (when an actuator
model is integrated into the controller). This topic will be discussed in more detail in Section 7.3.

7.3 Closed-Loop Feedback Control

Closed-loop control commonly makes use of error-based and specifically negative feedback. Despite having been formulated
in 1951, the original definition by the AIEE [133] continues to hold: “A feedback control system is a control system which
tends to maintain a prescribed relationship of one system variable to another by comparing functions of these variables and
using the difference as a means of control.” It is intended that:
● the system will remain stable at all times
● the system output will track the command input signal
● the system output should not respond (too much) to disturbance inputs
These goals should be achieved even if there is not a complete understanding of the system to be controlled. If the elements
of the control system can be specified by proper transfer functions, then it is possible to analyze and design the controller,
particularly to achieve desired steady-state and transient performance measures of the controlled (output) variable. In this
section, it is shown how useful feedback control relations are defined and derived. The standard form of the SISO closed-loop
block diagram is taken as shown in Figure 7.16. In this diagram, G2 (s) is a plant or physical system transfer function, G1 (s)
is a controller transfer function, H(s) is a sensor or measurement transfer function, and r is a reference or command signal.
This system includes both disturbance 𝑤 and noise 𝑣 (at the output). The control objective is to have ys , the output as scaled
or processed by the measurement, be equal to r.

7.3.1 Feedback System Relations

The block diagram in Figure 7.16 can be used to derive several key relationships that are used when specifying or evaluating
controller performance. Since all the elements are assumed to be linear in this form, basic block diagram algebra is used to
derive the relations. Identify the following from the block diagram:
e = r − Hys = r − H(c + 𝑣) (Closed-loop error).

u = G1 ⋅ e = G1 (r − Hc − H𝑣) (Controller output).

[ ]
y = G2 [u + 𝑤] = G2 G1 (r − Hc − H𝑣) + 𝑤 (Output).

7.3.1.1 Output, y
Using the relations above, the controlled output, y, can be found as,
G1 G2 G2 −G1 G2
y= ⋅r+ ⋅𝑤+ ⋅ 𝑣. (7.5)
1 + G1 G2 H 1 + G1 G 2 H 1 + G1 G2 H
Note that y depends on the three specified inputs: r, 𝑤, and 𝑣, and that the characteristic equation (CE) for the closed-loop
system is 1 + G1 G2 H = 0. If we neglect noise, 𝑣 = 0, and,
G1 G2 G2
y= r+ 𝑤, (7.6)
1 + G1 G2 H 1 + G1 G 2 H
and if H = 1 (unity feedback), we assume a perfect measurement of y. This latter relationship is commonly used in many
basic control system analyses.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
7.3 Closed-Loop Feedback Control 389

Reference Disturbance Output

+ Controller Plant

Measurement Noise
+

Figure 7.16 Single-input–single-output (SISO) form of a linear closed-loop control diagram with disturbance, 𝑤, and noise inputs, 𝑣.

7.3.1.2 Error, e
A measure of how well a system is tracking a command or regulating the output is the error signal, e. For this reason, it is
useful to derive transfer functions relating e to any of the inputs to the control system. From the closed-loop error relation,
[ ]
G1 G 2 G2 −G1 G2
e=r−H ⋅r+ ⋅𝑤+ ⋅ 𝑣 − H𝑣
1 + G1 G2 H 1 + G1 G 2 H 1 + G1 G2 H
1 + G1 G2 H G1 G2 H G2 H G G HH (1 + G1 G2 H)H
= ⋅r− ⋅r− ⋅𝑤+ 1 2 ⋅𝑣− ⋅ 𝑣.
1 + G1 G 2 H 1 + G1 G2 H 1 + G1 G2 H 1 + G1 G2 H 1 + G1 G2 H
Thus, the error as a function of all the inputs is,
1 G2 H H
e= ⋅r− ⋅𝑤− ⋅ 𝑣. (7.7)
1 + G1 G 2 H 1 + G1 G2 H 1 + G1 G2 H
This relation can be used to examine how e depends on, say, disturbances, 𝑤, while r = 𝑣 = 0, which gives the transfer
function,
e G2 H
=− . (7.8)
𝑤 1 + G1 G2 H
This transfer function provides a measure of controller regulation. A measure of controller tracking is found by,
e 1
= , (7.9)
r 1 + G1 G2 H
where 𝑤 = 𝑣 = 0. It is common to use these transfer functions to solve for transient and steady-state response of the error
subject to specific types of reference and disturbance inputs. Similar relations can be found to study the effect due to noise.
The e∕r relation is also referred to as the control system sensitivity, S, in the context of frequency domain design in particular.

7.3.1.3 Control Effort, u

Given the error, it is possible to find the corresponding control effort, u, that a closed-loop controller will generate under
specified inputs. Since, u = G1 (s) ⋅ e, the control effort is,
G1 G1 G2 H G1 H
u= ⋅r− ⋅𝑤− ⋅ 𝑣. (7.10)
1 + G1 G2 H 1 + G1 G2 H 1 + G1 G2 H
This relation can be used to quantify the actual control effort required during operation of the closed-loop controller.
While one should bear in mind that the model is linear (or linearized), this relation provides insight into the required
range of actuation. From a model basis, the results can also be used to inform selection of candidate actuator systems. It is
important to bear in mind that these results have been derived with assumed linear transfer function descriptions.
As an example, consider a simplified model of a ship undergoing forced roll motion due to waves, as shown in
Figure 7.17(a). A proposed feedback stabilization control might incorporate a sensor to monitor roll angle 𝜃 and fin
actuators. Ideal gain factors are indicated to model the sensor and actuator models as K1 and Ka , respectively. A suitable
ship transfer function would be needed, G(s). The block diagram description allows formulation of key relations that can
be used to, say, estimate the level of torque, 𝜏f , needed from the fin actuator to regulate roll given wave disturbance inputs,
𝜏d . To regulate with r = 𝜃d = 0, from equation 7.10, with 𝑣 = 0 (no noise), the desired transfer function for control effort is,
𝜏f Ka K1 G
=− .
𝜏d 1 + Ka K1 G
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
390 7 Modeling Feedback Control Systems

+ Fin control and + + Ship

actuator,
–

Fin Roll sensor

1
Fin

(a) (b)

Figure 7.17 (a) Ship with roll angle 𝜃. (b) Block diagram for fin-actuated roll stabilization.

7.3.2 Beneficial Effects of Feedback

“I suddenly realized that if I fed the amplifier output back to the input, in reverse phase, and kept the device from
oscillating (singing, as we called it then), I would have exactly what I wanted: a means of canceling out the distortion
in the output.” – H.S. Black [139]

The relations derived in Section 7.3.1 can be used to examine the influence and benefits of “closing the loop” with negative
feedback. In this section, closed-loop relations are compared to corresponding open-loop relationships. These relations
show that closed-loop systems with negative feedback,

1) provide for disturbance rejection,

2) reduce sensitivity to parameter variations,
3) enable using error signal for dynamic tracking, and
4) enhance accuracy and extend bandwidth.

Introducing feedback can also cause a system to oscillate or become unstable, conditions can be avoided through analysis
and design with derived transfer functions.

7.3.2.1 Force-Control Case Study

The system in Figure 7.18(a) is proposed to create a desired force state in a compliant material under test, represented
here by the spring, k. A force sensor is attached to the material base in order to directly measure the force (against a rigid
support). A force actuator will drive the mass so as to achieve the desired force in the spring. The force sensor is constructed
by a bar with cross-sectional area, A, Young’s modulus, E, and length L, which has a strain gage mounted to detect axial

: : 1/

Force Force
Strain
actuator sensor
gage
1 SG
0 1 0
EA Amp

Actuator Strain gage

command amp

(a) (b)

Figure 7.18 (a) Mass–spring–damper forced by actuator with wall-mounted force sensor at spring base. (b) Block diagram of sensor
and actuator elements integrated with bond graph of plant.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
7.3 Closed-Loop Feedback Control 391

+ + +
+ +
–

Gamp /EA

Figure 7.19 Feedback for controlling spring force.

strain.8 Changes in the strain, 𝜀, are transformed into measurable voltage changes, Vo , using signal conditioning and an
amplifier. The bond graph in Figure 7.18(b) models the system as a mass–spring–damper. The block diagram models of the
sensor and actuator convey key assumptions. For the force sensor, a signal bond from the 0-junction conveying Fk implies
the sensor stiffness, ks , is very high compared to that of the material under test, k, or k∕ks ≪ 1. This assumption is essential
for “decoupling” the sensor from the plant.
A similar assumption has been made for how the actuator interacts with the mass. The output Fa from the actuator
transfer function, Ga (s) = Fa ∕ua , sets the effort of a 0-junction.9 In this way, effort is imposed on the power bond attached
to the mass; however, the velocity of the mass, 𝑣m , has no influence on the force actuator. This type of assumption should
be justified since it implies an ideal force (effort) source with unlimited power. If this assumption is not valid, the actuator
model would need to be integrated with the plant model.
If the sensor and actuator models proposed are justified, a closed-loop control diagram can be formed as shown in
Figure 7.19. The plant model is taken as Gp (s) = xm ∕F = 1∕(ms2 + bs + k). Since the sensor is assumed not to deflect,
xk = xm . Thus, this forms an output, Fk = kxk = kGp (s) ⋅ F. Since the feedback provides a voltage, we define a reference
voltage Vr = Ks Fkr , where Ks is the sensor gain. In this way, the error is e = Vr − Vo , where Vo = H(s) ⋅ Fk and the measure-
ment transfer function is H(s) = Gamp (s)∕EA. In this relation, Gamp (s) is a transfer function that accounts for how the strain
gage circuit is formed as well as any dynamic characteristics in the amplifier.

7.3.2.2 Effect of Feedback on Gain

In many applications, it is desired to manage the overall system gain. This was one of the goals in Black’s work referred to
above [139]. To simplify the concept, assume that G and H are constant “gains.” The open-loop or forward gain is simply
y = G1 G2 r = Gr, indicating a gain of G. By closing the loop,
G
y= r, (7.11)
1 + GH
the “overall system gain” changes by a factor of 1 + GH. This shows how the general effect of feedback is to increase or
decrease the gain, since the quantity GH may be made positive or negative.
The gains G and H, of course, are not just constants, as they represent system transfer functions with dependence on
frequency (as discussed in Chapter 6). This means that the value of 1 + GH can take on different values as the frequency
of signals conveyed through the system vary. Indeed, for some frequency ranges, this quantity can be greater than unity
(amplify) and for others less than unity (attenuate).

7.3.2.3 Sensitivity of Output Relations

Table 7.2 provides relations between variations in the controller output y to variations in the system disturbance or system
component parameters. Note that the open-loop case takes H = 0.
Studying these variations in the output from a closed-loop system directly shows the benefit of feedback compared to the
open-loop case.
It is also common in control system analysis to define sensitivity functions for a control system. Sensitivity is effectively
the ratio of the percentage change in a quantity to a percentage change in a parameter or transfer function. A specific system
sensitivity function can be determined by [61, 142],
𝜕M p
SpM = ,
𝜕p M
for a given system transfer function, say M, where p is the system parameter or transfer function of interest.

8 Various strain gage configurations are typically used (e.g., in Wheatstone bridge) in these applications, but a single gage is shown here for
simplicity [140, 141].
9 An alternative to the 0-junction is to use an effort source, Se .
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
392 7 Modeling Feedback Control Systems

Table 7.2 Variations in the output y to variations in system components and inputs.

Variations in… Open-loop 𝚫y Closed-loop 𝚫y

G2
Disturbance, Δ𝑤 Δy = G2 Δ𝑤 Δy = Δ𝑤
1 + G1 G 2 H
Δr = 0 (can’t change G2 ) If we make G1 G2 ≫ 1, Δy can be made small

G2 r
Plant, ΔG2 Δy = G2 ΔG2 ⋅ r Δy = ΔG2
1 + G1 G 2 H
Δy G ΔG ⋅ r ΔG2 Δy 1 ΔG2
𝑤 = 0, Δ𝑤 = 0 = 1 2 = =
y G1 G2 ⋅ r G2 y 1 + G1 G2 H G2
Δr ≠ 0 Adjust G1 H to
minimize effect of ΔG2 ∕G2

Δy ΔG1 Δy 1 ΔG1
Controller, ΔG1 = =
y G1 y 1 + G1 G2 H G1
𝑤 = 0, Δ𝑤 = 0 Adjust G1 H to
Δr = 0 minimize effect of ΔG1 ∕G1

Δy −G1 G2
Measurement, H n/a = ΔH
y 1 + G1 G2 H
𝑤 = 0, Δ𝑤 = 0 If G1 G2 ≫ 1,
Δy∕y ≈ −ΔH∕H
Good H → ΔH ≈ 0

For an open-loop system, where, say, y = G ⋅ e, sensitivity to changes in G is,

G dG
SGG = = 1,
G dG
which shows what is expected: any variation in the parameters of an open-loop system results in changes in the output. On
the other hand, for a closed-loop system, the system transfer function is y∕r = Gcl = G∕(1 + GH) (G = G1 G2 , 𝑤 = 0), and
the sensitivity is,
𝜕Gcl G (1 + GH) − GH G(1 + GH) 1
SGG = = ⋅ = .
cl 𝜕G Gcl (1 + GH)2 G 1 + GH
We thus see the role of sensitivity functions in the variation relations shown in Table 7.2. These results show that the
sensitivity (to variations in G) for a closed-loop system can be made very small by increasing GH (as long as the system
remains stable). You can similarly derive the sensitivity due to H in this way to assess sensitivity to variations in the
feedback path.
Tracking sensitivity functions is practical for many reasons. Components of systems can have properties that change
over time according to environmental conditions, aging, use, etc. Those effects can make themselves known through cor-
responding changes in physical parameters used to model the system. For example, friction in bearings and coil resistances
in electrical motors. A good control system should be insensitive to any such parameter changes at the same time that it is
able to follow commands reliably.

Example 7.3 Sensitivity to system parameters in a closed-loop system

Consider the unity feedback control system shown in Figure 7.20.

a) Compute the sensitivity of the closed-loop transfer function, Gcl , to changes in the parameter A

+ A Figure 7.20 Closed-loop feedback sensitivity example.

+
–
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
7.3 Closed-Loop Feedback Control 393

Solution (a)
First find the closed-loop transfer function as,
A
Gcl (s) = 2 .
s + as + A
G 𝜕Gcl A 1(s2 + as + A) − A(1) A s2 + as
SAcl = = = .
𝜕A Gcl (s2 + as + A)2 Gcl s2 + as + A
How do you interpret this function? For low frequencies, the sensitivity is zero but as frequency gets large sensitivity to
A actually goes to 1.
b) Compute the sensitivity of the closed-loop transfer function to changes in the parameter a
Solution (b)
A
Gcl (s) = .
s2 + as + A
G 𝜕Gcl a 0 − As a −as
Sa cl = = 2 = 2 .
𝜕a Gcl (s + as + A)2 Gcl s + as + A
This sensitivity function indicates that the rate of change of a will be important.
c) If the unity gain in the feedback changes to a value of 𝛽 ≠ 1, compute the sensitivity of the closed-loop transfer function
with respect to 𝛽
Solution (c)
In this case, the CLTF will change to,
A
Gcl (s) = 2 .
s + as + A𝛽
G 𝜕Gcl 𝛽 −A𝛽
S𝛽 cl = = 2 .
𝜕𝛽 Gcl s + as + A𝛽
At low frequency, there is a lot of dependence on variability in 𝛽, but this goes down with frequency. Using methods from
Chapter 6, these sensitivity functions could be sketched or plotted as functions of frequency.

In the following, we briefly summarize the effect of feedback on stability and the tracking function, and how feedback
can extend the linear range of operation of a closed-loop system.

7.3.2.4 Effect of Feedback on Stability

A system is stable when it has somewhat predictable and “bounded” behavior. A system that begins oscillating with ever
increasing amplitude is considered unstable. Feedback can affect stability. For the purposes of a simplified discussion, let
G = G1 G2 and consider 𝑤 = 𝑣 = 0. Now, since the closed-loop gain is G∕(1 + GH), we see that if GH = −1, then the gain
goes to infinity so the system will produce an infinite output. So even though a system to be controlled, G, is stable, as soon
as the loop is closed it is possible that the system can become unstable. A main goal of control system analysis and design
makes use of stability characteristics of the open and closed-loop system. We bear in mind that this discussion is using
simplified relations, and the system may be unstable for other cases beyond just when GH = −1. Further, these insights are
based on model representations which assume G and H as models that are valid for specified regions of operation.
Feedback can also stabilize a system that is unstable. Say the previous system in Figure 7.20 was unstable because
GH = −1. By closing the loop with an outer loop system that has a feedback gain of, say, F, it can be shown F can be
chosen to make the system stable.

7.3.2.5 Effect of Feedback in Tracking a Reference Signal

Closing the loop makes it possible to track a command or reference input r. To show this consider again the output relation,
y G
= ,
r 1 + GH
where H = 1 (that is, unity feedback). Assume G is just a constant so by making G ≫ 1, it is clear that one can drive y → r,
and thus set the error to a desired value since e = r − y. If positive feedback is used, then y will be 180∘ out-of-phase and
this can cause instability.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
394 7 Modeling Feedback Control Systems

7.3.2.6 On the Range of Operation Over Which a System Behaves Linearly

Consider a system with a static plant that has the characteristic shown. The function is linear for |x| < 1 with y = x, and
then for |x| ≥ 1, y = sgn(x). This is a saturation function, which can be expressed, y = sat(x).
The system shown in Figure 7.21 has this static (nonlinear) plant, which is linear for |x| < 1 with y = x, and then for
|x| ≥ 1, y = sgn(x). Call this a saturation function, y = sat(x). Saturation is the source of distortion that Black sought to
address [139].
For the open-loop case, we have only the controller and the plant, or process. Assume the disturbance is 𝑤 = 0. For a
controller Gc = K, u = K ⋅ r, then y = sat(K ⋅ r). Note that here the system output will be linear for r ∈ [−1, 1]. Now close
the loop so that e = r − y (unity feedback). In this case, the output is y = sat(K(r − y)), so we need to solve for y from an
algebraic loop. For |K(r − y)| < 1, the output is found to be y = Kr∕(K + 1), and when |K(r − y)| ≥ 1, then y = sgn(K(r − y)).
First note that y can be made to track r closely for large K and the region over which this is true increases compared to the
open-loop case. In other words, the closed-loop system is linear as long as |r| < (K + 1)∕K, which is a range greater than
the open-loop case which required r to be within ±1 for linear behavior.
Consider the following alternative derivation to how feedback extends linear behavior (based on discussion from Kuo
[143]). Assume a linear system that has a forward gain, G. The output is y = G ⋅ r as long as |r| < 1, but if |r| ≥ 1 then
y = sat(G ⋅ r).
Now, again, for error feedback, e = r − y. If |e| < 1, then y = G ⋅ e = G(r − y). Solve for y (algebraic loop), so y = G ⋅ r −
G ⋅ y ⇒ y(1 + G) = G ⋅ r, and y = G ⋅ r∕(1 + G). Otherwise, if e ≥ 1 then y = ±1 (again based on sign). This shows again that
the linear range is extended. The system output is linear if e = r − y < 1 therefore r < 1 + y = 1 + Gr∕(G + 1) so we can
solve for the linear range of the reference input, r < (G + 1)∕G. Thus, by “cranking up the gain” a larger linear range of
performance is achieved with closed-loop control.

7.3.3 Open-Loop and Feedforward Control

Table 7.2 includes results for how the output y is influenced by variations in the open-loop case, H = 0. In this case, it was
assumed,
y = G1 G2 r + G2 𝑤, (7.12)
and the variation due to disturbance, Δ𝑤, results in output variations in the output, Δy = G2 Δ𝑤, Disturbance-related vari-
ation is directly proportional to plant gain |G2 |. For a variation in plant ΔG2 , we can also find the change in output Δy as,
Δy = ΔG2 𝑤 + G1 ΔG2 r, (7.13)
when r ≠ 0. Open-loop control plays a central role in many applications where it may not be possible or practical to close
the loop. Closing the loop can add cost and complexity to a system, not to mention the propensity for instability. There are
many applications where the model required for effective open-loop control can be readily determined, and where any
indicated variations due to disturbances or parameter uncertainty are tolerable or correctable within reason. Many of the
classical systems studied by Mayr [134], for example, were shown to be “open loop.”
An open-loop controller is most effective when a model relationship between the output and the input is well understood
and can be used to inform the controller. Any changes in the system behavior that cannot be characterized or modeled as
well as any external disturbances can lead to large deviations in the output. To some extent, open-loop control is effectively
passive design of a system’s response. Many timed processes and machines effectively use open-loop control to achieve
required operation.
A form of “open-loop” control has been traditionally applied to classical analog voltage meters. A schematic of a typical
analog voltage meter movement is shown in Figure 7.22(a). The meters are meant to provide a measurement of the applied

Figure 7.21 Basic feedback system – static plant.

1
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
7.3 Closed-Loop Feedback Control 395

Coil and needle

inertia

Coil spring Analog meter

stiffness
Analog
+ Open-loop
Effective meter
–
damping gain
Galvanometer movement

(a) (b)

Figure 7.22 (a) Schematic of an analog voltage meter where 𝜃m is meant to reflect Vdc . (b) Open-loop block diagram.

voltage Vs (t) through the angular deflection, 𝜃m . For measuring dc voltages, a direct connection is made, while for ac voltage,
a rectifying circuit is needed (to give root-mean-square values). For the dc voltage case, it is simply desired that 𝜃m reach
a steady-state value that correlates with Vs (dc). The design of an open-loop gain can be done empirically: simply measure
the steady-state angular displacement over a range of dc voltages and a correlation. Often, these systems are fairly linear so
a gain is determined as, KV𝜃 , relating to the slope of voltage to angle measurements.
An open-loop controller is then constructed as shown in Figure 7.22(b) which allows a desired angle, 𝜃md to be specified.
The control voltage is then Vs = KV𝜃 𝜃md . For dc meters, there is an expected error, e𝜃 = 𝜃m − 𝜃md , due to various sources. In
order to have minimal error, the device is characterized and there is often a need for a series resistor, Rs , to adjust the meter
current, im .
The meter is a good example of how basic open-loop control can be accomplished and how it can be insufficient in cases
where there can be more significant error due to disturbances, variation in parameters, etc. In addition, most control prob-
lems of interest are also concerned with dynamic response either in time or frequency domain. It is evident that this meter is
constructed using physical elements that will introduce dynamic effects: coil in inductance, Lm , rotational inertia, Jm , and a
coil spring, Km , which serves to return the needle to a zero position when there is no induced torque. All of these parameters
will influence the static and dynamic characteristics of the analog meter response to voltage inputs. If a reasonable model of
the analog meter is available, it is possible to modify the open-loop gain so that it accounts for dynamic effects. For example,
assume the analog meter was known to have a second-order transfer function, say, G𝜃V = K𝜃V 𝜔2n ∕(s2 + 2𝜁𝜔n s + 𝜔2n ), where
K𝜃V , 𝜁, and 𝜔n have been measured or are known. It would be possible, although not necessarily practical for the meter case,
to implement an open-loop controller as an inverse of G𝜃V . This would have the effect of canceling the meter dynamics, to
the extent that the model is accurate. Here is the issue, of course, with open-loop control: it is only as good as the model.
In some cases, however, open-loop control may be preferred, especially if a closed-loop control introduces dynamics
as a result of the controller and the measurement systems that introduce design complexity. It is also possible that an
open-loop control can be used effectively in combination with closed-loop control. In cases, say, where disturbances or
system variations are well characterized, they can be compensated for by feedforward control. This can be considered a
form of open-loop control integrated with a closed-loop control system, as shown in Figure 7.23. This combined control
architecture is able to use the feedforward effect, uFF , to achieve a reasonable control of the plant. Whatever error continues
to exist or arise can be compensated for by the error-based feedback control effort, ue .
Feedforward is effective in compensating for forces induced by gravity, for example, on a mechanical mechanism (linkage,
etc.). This is a nonlinear effect. However, if the geometry and inertia properties are well known, and the positions are

Feedforward
control
+ + + + +

Figure 7.23 Feedback combined with one form of feedforward control.

 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
396 7 Modeling Feedback Control Systems

being measured, then a feedforward control can provide a compensating force or torque that reduces the error. This leaves
primarily dynamic effects to be dealt with by error-based feedback control. The same idea applies to other nonlinear effects,
as long as there is a reasonable and predictable model. Note that the feedforward control, uFF , might then be a nonlinear
function of the reference and other measured variables.

7.4 Linear Feedback Controllers and Compensators

Up to this point, we have discussed general feedback control loop configurations. Examples with proportional control,
G1 = K, have been described, but this simple form may not be able to meet desired performance and stability requirements
in many cases. More flexibility and capability are available by using dynamic controllers or compensators. A PID controller
is one of the most common type of controllers, defined by a combination of linear proportional, derivative, and integral
control efforts based on error feedback. The PID controller can take a standard form such as,
[ ]
1
G1 (s) = K 1 + + 𝜏d s , (7.14)
𝜏i s
where 𝜏d and 𝜏i are derivative and integral time constants, or a parallel form,
Ki
G1 (s) = Kp + + Kd s. (7.15)
s
Equation 7.14 is used in the block diagram form shown in Figure 7.24 using distinct P, I, and D blocks. This diagram is
meant to convey how combinations of P, I, and D actions are fundamental to most controllers and compensators. Some typ-
ical combinations are summarized in Table 7.3, along with how each can influence steady-state error and relative stability.
Relative stability refers to how close the poles of a system are located to the j𝜔 axis, as systems become more oscillatory as
poles approach this region and possibly cross into the right-hand real-imaginary plane. Additional discussion about selec-
tion and design of these controllers is given in Section 7.4.5, but the reader is referred to introductory controls texts for a
thorough description of such methods [53, 61, 93].

1
+
+ 1 + +

+ +
–

Figure 7.24 PID closed-loop control block diagram including disturbance, 𝑤, and noise inputs, 𝑣.

Table 7.3 Common types of compensation and effects on error and stability, I = integral, P = proportional.

Compensator type
and connection Transfer function Effect on steady-state error Effect on relative stability

K
I, series Gc (s) = Greatly improved Greatly reduced
s
K(s + a)
PI, series Gc (s) = Improved Reduced
s
PD, series Gc (s) = K(1 + 𝜏d s) Minimal to none Improved
K(s + a)
Phase lag, series Gc (s) = ,b<a Improved Reduced; can increase response time
s+b
K(s + a)
Phase lead, series Gc (s) = ,a<b Some improvement or worse Increased; can increase effects due to high
s+b frequency noise
Rate, feedback H(s) = 1 + 𝜏t s Improved Increased
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
7.4 Linear Feedback Controllers and Compensators 397

+ +
1 2
1 2 +
– – –

(a) (b)

Figure 7.25 (a) Series compensation with G1 = Gc . (b) Parallel compensation Gc across plant G2 .

Block diagrams such as in Figure 7.24 are often used to conceptualize control systems and highlight signal flow paths.
Different blocks can be added or combined before a final configuration is made and then reduced for analysis. Of special
note is the derivative term in the forward path, 𝜏d s, which indicates differentiation of the error. A pure derivative function
can be practically problematic, especially in the presence of disturbances and noise that change rapidly. Impulses can arise,
as when there is step change in the reference input. For these reasons, the derivative function is often replaced by a transfer
function of the form,
𝜏d s
G(s) = , (7.16)
𝛾𝜏d s + 1
rather than using a pure derivative term such as 𝜏d s. The value of 𝛾 is usually chosen so that signals with relatively high
frequency are filtered. Taking a derivative action in the feedback path is another way to avoid this issue, as suggested by the
rate feedback form in Table 7.3. The relation between D-control and rate feedback is discussed further in Section 7.4.2.
Compensation refers to the modification of the system dynamics to satisfy performance specifications. The transfer func-
tions for controllers and compensators are used to represent combinations of passive and active components. Compensators
may also be formed using purely passive hardware (e.g., electrical filters and mechanical spring–dampers). Since compen-
sation in many contemporary systems may also be achieved purely using digital techniques [136], it makes sense that the
terms controller and compensator are often used interchangeably. One key dividing line, however, can be where plant
inputs and outputs are defined, so as to clearly identify any disturbance and/or noise inputs.
Figure 7.25 shows how compensators, Gc (s), are commonly inserted in series or parallel with a plant. The lag and lead
compensators shown in Table 7.3 are commonly used in a series configuration, and their design (parameterization) is
extensively discussed in linear control references such as Ogata [53] and Franklin et al. [93].
It is often the case that a proportional control is first attempted in a control design. If the performance specifications
cannot be achieved and/or relative stability is not acceptable, then more complex forms of compensation are adopted.
Generally, the effect of each is as follows:

● P – provides “faster response”

● I – “better steady-state response”
● D – “faster convergence”

Table 7.3 conveys similar and related features. In the following, P, D, and I control actions are applied to various example
systems.

7.4.1 Application of PID Control

One of the simplest forms of control is error proportional control, where the controller transfer function is G1 = Kp (using
the direct form of equation 7.15). The control effort is then proportional to the difference in the error signal. Consider
the control system shown in Figure 7.26(a) configured to regulate the position of a mass, xm , that is subjected to dis-
turbance force, Fd . A block diagram is given in Figure 7.26(b), where G1 = Kp is the controller proportional gain and
H = 1 (unity feedback), so xs = xm (note, noise 𝑣 = 0). Assuming negligible mechanical damping in the plant the transfer
function is,
xm 1∕m
G2 = = 2 .
F s + k∕m
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
398 7 Modeling Feedback Control Systems

+ +
1 2
+
–
Force
actuator

Controller =1
Position
sensor

(a) (b)

Figure 7.26 (a) Mass-spring system with attached force actuator and position sensing to control motion due to imposed disturbances,
Fd . (b) Feedback control diagram.

The response to both the reference and disturbance inputs is then (from equation 7.6),
(Kp ∕m)∕(s2 + k∕m) (1∕m)∕(s2 + k∕m)
xm = xr + F
1 + (Kp ∕m)∕(s2 + k∕m) 1 + (Kp ∕m)∕(s2 + k∕m) d
Kp ∕m 1∕m
= xr + F . (7.17)
s2 + (k + Kp )∕m s2 + (k + Kp )∕m d
The change in the CE shows that P-control changes the effective stiffness of the system by an amount Kp , and thus the
√
undamped natural frequency is 𝜔n = (k + Kp )∕m. Unfortunately, P-control alone will not have an effect on the system
damping, which would be desirable in offsetting disturbances. Setting xr = 0 defines the control system as a regulator, which
has a transfer function relating xm to Fd ,
xm 1∕m
= 2 .
Fd s + (k + Kp )∕m
This transfer function can be used to study the response of xm to different disturbance inputs.
If the control force was determined only by derivative control, say, G1 = Kd s, then Fc = Kd ⋅ e. ̇ Note that if again we set
xr = 0, so e = 0 − x, then Fc = G1 ⋅ e = 0 − Kd ẋ m . This force essentially introduces an “active” linear damper on the mass in
parallel with the spring. In this case, the disturbance response transfer function (xr = 0) is,
xm G2 1∕m
= = 2 .
Fd 1 + G1 G2 H s + (Kd ∕m)s + k∕m
If there had been any mechanical damping, the total effective damping (coefficient of the “s” term) would appear to be
increased by an amount Kd . We can now expect combing P+D control would result in,
xm 1∕m
= 2 . (7.18)
Fd s + (Kd ∕m)s + (k + Kp )∕m
Since the CE is the second order in form, and Kp and Kd directly influence the undamped natural frequency and damping
ratio, respectively, the response can be made to meet desired specifications. Recall (from Chapter 5) how time-domain
specifications are related to both of these system parameters.
In the case where xm is to be regulated in the presence of disturbances, it can also be useful to look at the error response
relations. Equation 7.8 for the case of P or PD control in this example leads to:
e 1∕m
=− 2
𝑤 s + (k + Kp )∕m
e 1∕m
=− 2 .
𝑤 s + (Kd ∕m)s + (k + Kp )∕m
If we apply the Final Value Theorem [53], we can see that for a step disturbance in Fd , the steady-state error will be
ess = 1∕(k + Kp ). Thus, we can solve for a value of Kp that will give a desired value of ess , or even make ess → 0 if Kp → ∞.
For either case, it may be a good idea to also solve for the control effort required by using equation 7.10 to make sure that
the force, Fc , required can be achieved by a given force actuator. Further, it may be found that the necessary gain Kp itself
is not achievable given resource constraints.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
7.4 Linear Feedback Controllers and Compensators 399

If integral control is introduced, the CE is again changed and in this case the order will be increased (typical when adding
I control). Continuing the example with the mass–spring system, the controlled response to disturbances now with PID
becomes,
xm G2
=
Fd 1 + G1 G2 H
(1∕m)∕(s2 + k∕m)
=
1 + (Kp + Ki ∕s + Kd s)(1∕m)∕(s2 + k∕m) ⋅ 1
1∕m
=
s2 + k∕m + Kp ∕m + Ki ∕(ms) + Kd s∕m
xm s∕m
= 3 2
.
Fd s + (Kd ∕m)s + (k + Kp )s∕m + (Ki ∕m)
The system is now the third order and a zero has been introduced (the s term in the numerator). In this case, the response
to a step input goes to zero at steady state, showing how adding integral control helps with convergence (and error control).
On the other hand, it is now necessary to more carefully select three gains to make sure the system remains stable. For this
reason, either Routh–Hurwitz (see Chapter 4) or a root locus analysis [53, 61] may need to be used to guide gain selection
so that performance specifications are met while ensuring the system remains stable.

Example 7.4 Speed control of a dc motor

It is proposed to apply proportional control to regulate the speed of a dc motor as described by the diagram in Figure 7.27.
Assume a tachometer will be used to directly measure the motor/inertia shaft speed, 𝜔m , and that the tachometer output
voltage is given by, VT = KT 𝜔m , where KT is a constant. It is assumed that the dc motor inductance can be neglected, and
parameters are known as: rm = gyrator modulus, Rm = electric resistance, Jm = total inertia, and Bm = linear rotational
friction coefficient. The amplifier can be modeled as an ideal gain Ka .
a) Develop a block diagram integrated with a bond graph of the dc motor, clearly identifying signal and power flow inter-
actions. Apply causality to the motor model to show that the motor speed, 𝜔m , is the only system state.
b) Derive a transfer function relating 𝜔m and the motor input voltage Vs .
c) Derive a closed-loop transfer function relating the controlled motor speed 𝜔 and the input reference voltage Vr .
d) Determine for the closed-loop steady-state speed, 𝜔m,ss .
e) How is the system CE changed by closing the loop?

Controller
Summing, proportional Plant
power amplifier
Measurement
+
Tachometer
Summer Amp

Total effective damping

Figure 7.27 Schematic of dc motor with speed feedback and proportional control.

Solution (a)
Assume that the amplifier can deliver the motor voltage Vs regardless of the current level, causally determined by the
motor, as indicated by the bond graph in Figure 7.28. Thus, a signal bond is used to set the voltage at a zero junction. The
tachometer is also assumed to not load the motor shaft, so an ideal speed measurement is indicated by a signal flow from
the 𝜔m 1-junction. The motor bond graph is thus wrapped by the block diagram with the amplifier and tachometer blocks
indicated. The summing junction is used to model the error forming function in the controller, e = Vr − VT .
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
400 7 Modeling Feedback Control Systems

Plant
:
Controller Measurement

+
0 1 G 1
–

: :

Figure 7.28 A physical-functional model description using a bond graph for the dc motor plant in combination with the control
system in block diagram form.

Solution (b)
The bond graph indicates hm is a state, thus we can take 𝜔m = hm ∕Jm as the system state. We can thus write,

Jm 𝜔̇ m = 𝜏m − 𝜏B = rm im − Bm 𝜔m ,

where im = (Vs − Vm )∕Rm = Vs ∕Rm − rm 𝜔m ∕Rm . Thus,

Jm 𝜔̇ m = −(rm
2
∕Rm + Bm )𝜔m + rm Vs ∕Rm .

Transform to s-domain and rearrange to show that the first-order plant transfer function between output speed 𝜔m and
the input motor voltage Vs is,
𝜔m Km
G2 (s) = = (7.19)
Vs 𝜏t s + 1
2 ∕R + B ) and time constant, 𝜏 = J ∕(r 2 ∕R + B ).
with gain Km = (rm ∕Rm )∕(rm m m t m m m m

Solution (c)
The summing amplifier combines both the error forming function and gain Ka , shown distinctly in Figure 7.28 to form a
P-controller for the motor drive voltage,

Vs = Ka e = Ka (Vr − VT ) P-Controller.

Equation 7.6 with H = KT (𝑤 = 0, 𝑣 = 0) can be used to determine the closed-loop transfer function between output speed
𝜔m and set point voltage Vr ,
Ka Km
𝜔 𝜏t s + 1 Ka Km
GcL = m = = .
Vr Ka Km 𝜏 t s + 1 + Ka K m K T
1+ ⋅ KT
𝜏t s + 1
Solution (d)
For a step change in set point voltage Vr , the steady-state output speed can be found by letting s = 0 (Final Value Theorem)
in the closed-loop transfer function as,
Vro Ka Km
𝜔m,ss = lim 𝜔m (s) = lim s ⋅ GcL (s) ⋅ = V ,
s→0 s→0 s 1 + Ka Km KT ro
where Vro is the magnitude of the step. As the control gain K is made large, we have,
𝜔m,ss ≈ KT ∕Vro , Ka → ∞,

VT,ss ≈ Vro

or the tachometer voltage will be identical to the commanded voltage. When this happens, we say that the system has zero
steady-state error. For this system, error is defined to be the difference between the reference and tachometer voltages.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
7.4 Linear Feedback Controllers and Compensators 401

Solution (e)
The closed-loop transfer function is first order, so closing the loop with P-control does not change the order of the system,
as expected. Closing the loop, however, does change the speed of response of the system, which can be judged from the
closed-loop time constant of GcL which is,
𝜏t
𝜏cL = .
1 + Ka Kp KT
Thus, as control gain Ka increases, we have 𝜏cL → 0, K → ∞. So, not only does the system get more accurate as control
gain increases, but also the system responds faster. Assuming the model is correct, an optimal control solution would be
to make the proportional gain as large as possible. This design would have accuracy and fast response and be insensitive to
plant variations and disturbances. Step response for increasing gains is shown in Figure 7.29.

0.5
,

0.4
Tachometer voltage,

0.3

0.2

0.1
Increasing
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Time, sec

Figure 7.29 Trends in step response of controlled PMDC motor with increasing control gain, Ka .

From the analysis in this example, it might be concluded that control design is a simple matter of increasing a proportional
gain. Besides the cost of building high gain power amplifiers, there is another snag to this utopian situation: there is a strong
possibility for instability in very high gain feedback systems.
As will be shown later, it can be prohibitive to simply make the proportional gain extremely large in order to achieve zero
steady-state error. It was shown how integral action can achieve zero steady-state error with lower proportional gain, but
there is an impact on relative stability. Consider an integral controller for the dc motor, but now with the transfer function
(from equation 7.14),
( )
1
Gc (s) = K 1 + .
𝜏i s
Taking the first-order dc motor plant, the closed-loop transfer function becomes,
𝜔m Ka Km (𝜏i s + 1)
GCL = = .
Vr 𝜏i 𝜏t s2 + (Ka Km 𝜏i + T𝜏i )s + KKm KT
At steady state, we have,
1
Gcl (0) = ,
KT
or, ess = 0. But, note that the closed-loop system order has increased to the second order and has the potential for oscillation.
Integral action always raises the order of the closed-loop system by one which can increase the chance of stability problems.
Section 7.4.2 reviews additional factors that can affect a system’s response and stability characteristics.

7.4.2 Comparing Derivative Action and Rate Feedback

In order to improve stability in a control system, derivative feedback action can be added to the proportional control.
As previously discussed, derivative action can be implemented in one of two ways. A direct D-control approach extends
the proportional controller transfer function to the form (see equation 7.14),
G1 = K(1 + 𝜏d s).
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
402 7 Modeling Feedback Control Systems

Recall how this adds a damping effect in the characteristic equation for the mass–spring system discussed in Section 7.4.1.
In this way, D-control is sometimes called error rate damping. Also recall that practical implementation of this form may
require adding a filtering element as in equation 7.16.
Instead of using the error to improve damping (and thus relative stability) in a system, it is also possible to use the output
signal for this effect. For the motor speed control, for example, one approach involves modifying the sensor transfer function
such that,
H(s) = KT (1 + 𝜏d s).
This is a rate or derivative feedback approach (as in Table 7.3). One benefit of this approach is that the output won’t
generally vary as widely as the error signal. Further, any sharp changes in the reference won’t be seen in the output whereas
they will define the error.
Introductions to feedback control that also include classical servomechanisms often provide a comparison between these
two approaches for introducing a damping effect (see [53]). Here, we can see that for the closed-loop transfer function for
derivative feedback in Example 7.4 gives,
𝜔m Ka Kp
Gcl = = ( )2 ( ) , (7.20)
Vr s∕𝜔n + 2𝜁∕𝜔n + KKp KT 𝜏d s + 1 + KKp KT
and for error rate damping, the closed-loop transfer function is,
( )
𝜔 KKp 1 + 𝜏d s
= GCL = ( )2 ( ) .
Vr s∕𝜔 + 2𝜁∕𝜔 + KK K 𝜏 s + 1 + KK K
n n p T d p T

As can be seen, these closed-loop transfer functions have identical characteristic equations and therefore identical poles.
However, note that error rate damping introduces an additional zero at z = −1∕𝜏d . This additional zero adds an additional
impulsive effect to the closed-loop response but the stability considerations are the same for the two types of derivative
controllers.
The closed-loop natural frequency for both controllers is,
√
𝜔nCL = 𝜔n 1 + KKp KT ,

and the closed-loop damping ratio is,

( )√
𝜁CL = 𝜁 + KKp KT 𝜏d ∕2 1 + KKp KT .

From these two expressions, we see that the natural frequency and damping can now be changed independently of each
other, and therefore high gain can be used when the plant is the second order. For third- and higher-order plants, infinitely
high gains will cause instabilities. However, proportional plus derivative control allows higher gains to be used than just
proportional control.

7.4.3 Error Constants and Linear Stability Analysis

The closed-loop feedback relations in Section 7.3.1 can be used to define transfer functions relating the output (controlled)
variable, y, error, e, and control effort, u, to the control system reference, r, disturbance, 𝑤, and noise, 𝑣, inputs.
These transfer functions enable model-based comparisons to be made to specifications or requirements of a closed-loop
system controller in the time or frequency domain. The error response in particular is used to define error constants, which
can indicate steady-state response to certain types of inputs. These relations also reveal the closed-loop CE, 1 + GH, from
which stability properties can be determined directly. In many cases, however, the open-loop transfer function, GH, is
sufficient for assessing stability.

7.4.3.1 Error Constants

Error constants are one way to assess how well a control system can track an input. For a SISO, equation 7.7 can be used to
assess improvements gained by introducing feedback. While transient error dynamics can be important, often steady-state
error alone is a key measure of performance that can be used as a specification. For example, it is often desired to have a
system converge to a steady-state value within a specified error. How steady-state error depends on system parameters can
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
7.4 Linear Feedback Controllers and Compensators 403

be determined from the linear error relations. In such cases, the steady-state error can be determined from the Final Value
Theorem (making 𝑣 = 0 in equation 7.7),
1 G2 H
ess = lim e(t) = lim s ⋅ e(s) = lim s ⋅ ⋅r− ⋅ 𝑤. (7.21)
t→∞ s→0 s→0 1 + G1 G2 H 1 + G1 G 2 H
So the reference and/or the disturbance must be specified. Typically error may be studied subject to only one of these
inputs but, of course, the effects are (linearly) additive.
It is also possible to classify control systems according to their ability to follow standard input signals, and this is done
with relation to error constants. The magnitudes of the resulting errors are then taken as measures of the “goodness of the
control.” This relationship between error and response to input type helps form a system type characterization. The different
types of inputs, r, of particular interest are: step, ramp, parabolic, and polynomial. The unit input types have related time
and s-domain representations as shown below, which can be scaled by a given magnitude as necessary:

Input type Step Ramp Parabolic Polynomial

Time domain 1(t) t t2 tN

s-domain 1∕s 1∕s2 2∕s3 N!∕sN+1

Consider the error relation for the case where 𝑤 = 𝑣 = 0 and with G(s) = G1 (s)G2 (s). This gives the error response to r,
1
e= ⋅ r,
1 + GH
where GH = G(s)H(s). When necessary, we may write this open-loop transfer function as GH(s), to emphasize “GH” is a
function in s-domain.10
Now, a general formula for assessing the steady-state error makes use of the Final Value Theorem,
1 1 1 1
ess = lim e(t) = lim e(s) = lim s ⋅ ⋅ = lim ⋅ , (7.22)
t→∞ s→0 s→0 1 + GH sN+1 s→0 1 + GH(0) sN

where GH(0) denotes setting s → 0. A system is classified as Type N if it converges to a non-zero constant given an input of
tN ∕N!. Taking GH expressed in pole-zero form,
K(𝜏a s + 1)(𝜏b s + 1) … (𝜏m s + 1)
GH(s) = ,
sN (𝜏1 s + 1)(𝜏2 s + 1) … (𝜏p s + 1)
a series of error constants can be defined as in Table 7.4. Note that the error constants in this table are each defined from
equation 7.22,
KN = lim sN GH(s) (7.23)
s→0

Thus, a Type 1 system will have a nonzero steady-state error for a ramp and a zero steady-state error for a step input.
However, Type 1 will give infinite error for parabolic and higher form inputs. In the factored form, we can often identify
the type from free integrators (that is, presence of s).

Table 7.4 System error constants for given types.

Step Ramp Parabolic t N ∕N!

Type 0 e0 = 1 +1 K ∞ ∞ ∞
0
Type 1 0 e1 = K1 ∞ ∞
1
Type 2 0 0 e2 = K1 ∞
2
Type N 0 0 0 eN = K1
N

10 Some controls books may refer to the loop function, L(s) = G(s)H(s).
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
404 7 Modeling Feedback Control Systems

Throttle controller Engine and vehicle Figure 7.30 Block diagram for closed-loop control of a vehicle’s
longitudinal speed.
+
1
+1
–

P-control PI-control

1 1
Step response

0.8 0.8 =0
0.6 1 0.6
=
0.4 1+ 0 0.4
0.2 0.2
0 0
0 0.5 1 1.5 2 0 0.5 1 1.5 2
2 2
Ramp response

1.5 1.5

1 1
= 1
0.5 0.5 1+ 1

0 0
0 0.5 1 1.5 2 0 0.5 1 1.5 2
Time, sec Time, sec

Figure 7.31 Comparison of P- versus PI-control for the vehicle speed control in Figure 7.30 to contrast Type 0 versus Type 1 response
to step and ramp inputs (K𝑣 = 1, 𝜏t = 1, Kp = 5, and Ki = 10).

So, consider the example of selecting a controller using a linear model of and engine/vehicle system, shown in Figure 7.30.
For this system,
K𝑣
GH = G1 ⋅ ,
𝜏t s + 1
So if it was required to have the speed, 𝑣x , achieve a finite error with a commanded ramp input, 𝑣xr , then G1 should be
selected so that GH that has one free integrator and is thus Type 1. This can be done by selecting a PI controller, G1 = Kp +
Ki ∕s. Note that if Ki = 0 (P-control), then the error constant for a unit step input would be K0 = lims→0 GH(s) = Kp K𝑣 , so
ess = 1∕(1 + Kp K𝑣 ), a finite value but not zero. Changing to PI control would make ess = 0 for a step input and finite for a
ramp input. Figure 7.31 shows graphs of results comparing P- and PI-control to illustrate the difference between Type 0
and Type 1 response to step and ramp inputs.
Application of error constants assumes that GH is stable, otherwise the Final Value Theorem would not apply (see
Chapter 2 [53]). A controller that introduces additional integrators (that is, more free “s” in the denominator) can improve
the ability to track certain inputs, one should be mindful that stability issues can arise in some cases. Being able to predict
these trends for first- and second-order plants is relatively tractable, but for more complex systems, it is necessary to adopt
additional analysis tools.

7.4.3.2 Stability Analysis

As discussed previously, tools for assessing stability of open- and closed- loop systems are essential. For linear cases, it is
possible to apply Routh–Hurwitz criterion or root locus methods to determine stability from the location of closed-loop
poles, in particular, which must have negative real parts to ensure stability. The Routh or Routh–Hurwitz stability criterion
was discussed in Chapter 4. Routh stability allows determination of absolute stability; that is, given the CE of a system’s
transfer function, the Routh criterion can allow determination of whether a system is stable or not. This is done by con-
structing a Routh table (see Chapter 4) and examining the sign changes in elements of this table. The method can be used
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
7.4 Linear Feedback Controllers and Compensators 405

for characteristic equations of the form,

an sn + an−1 sn−1 + · · · + a1 s + a0 = 0, (7.24)
and the coefficients can be either numerical or purely in parameters. Thus, stability can be determined when the CE polyno-
mial coefficients are in parameter and/or numerical form. Absolute stability, of course, says nothing about relative stability,
which refers to “how stable is a system,” or how close are any roots to crossing over into the right-hand plane. For these
purposes, control analysts turn to root locus plots as well as to frequency domain measures.
Root loci reveal how poles of a system trace trajectories when plotted in the real-imaginary plane as a parameter is varied.
Recall that only the poles derived from the CE inform system stability, since they determine unforced response behavior,
while both poles and zeros are used in solving for forced response (see Chapter 5). Being able to determine how the poles vary
with parameters of interest provides insights into stability and response behavior. Some examples were given in Chapter 5
where the poles of a system were computationally determined and plotted as a system parameter was varied.11 W.R. Evans
developed a classical approach for “sketching” root locus plots in the 1940s. The method relies only on knowledge of a
control system’s open-loop transfer function, GH. In many such cases, a parameter K of interest can be isolated from GH
in expressing the CE in the form,
N(s)
1 + K ⋅ G(s)H(s) = 1 + K ⋅ . (7.25)
D(s)
The parameter to vary, K, may be a controller gain or any parameter of interest.
There are some basic concepts and definitions that are worth remembering about root locus plots:
● n open-loop poles, pi , are the roots of D(s).
● m open-loop zeros, zi , are the roots of N(s).
● The number of loci is equal to the maximum of n and m, and each locus describes how the poles vary as K is varied from
K = 0 to K → ∞.
● The loci always start from the open-loop poles and end at the open-loop zeros, or infinity when m < n.
● Loci only lie on the real axis in regions to the left of an odd-number of open-loop poles and zeros.
● As K increases, the poles trace trajectories in the real-imaginary (𝜎 − j𝜔) plane, and the loci are always symmetrical about
the real axis.
● There are n − m asymptotes, which are lines toward which loci converge as K → ∞ infinity (when they do not tend to an
open-loop zero). All asymptotes intersect the real axis.
Any and all such rules are based on the need for the loci to satisfy GH = −1, and as such can provide guidance on
quickly checking a root locus plot. These rules are used in sketching root loci, a skill than can take practice to learn and
apply. Fortunately, guided introductions are available in textbooks such as Ogata [53], Franklin et al. [93], Dorf and Bishop
[61], and Nise [144]. Most root locus plots, however, can be quickly produced using functions found in control design and
analysis software packages (typically called “rlocus()”). It is necessary to formulate the GH function and to identify the gain
parameter of interest. Nevertheless, to this day, practicing methods introduced by Evans [145] provides insights into how
the poles of a system change when a parameter of interest is varied. For this reason, sketching root loci is a useful skill for
the controls engineers. For the purposes of this book and chapter, it will be assumed that the reader has access to means
for generating root locus plots.
It should be made clear that a root locus can typically be plotted for any CE that can be put in proper form. Say it was of
interest to understand how varying parameter a1 parameter in the CE of equation 7.24 affected the root locus. The classical
root locus sketching/plotting methods (or programs) can be applied by first arranging in the form,
s
1 + a1 ⋅ = 0.
an s2 + an−1 sn−1 + · · · + a2 s2 + a0
Algebraic manipulations of a CE in this way allows isolating a model or controller parameter of interest. For example,
say it was of interest to study the variation of 𝛼 in the equation,
s3 + (3 + 𝛼)s2 + 3s + 6 = 0.
The “Evans root locus form” can be obtained by simply extracting the 𝛼 term,
s2
s3 + 3s2 + 𝛼s2 + 3s + 6 = 0 ⇒ 1 + 𝛼 ⋅ = 0,
s3 + 3s2+ 3s + 6
11 See Solved Problem A-5-11 on dynamics of an electromechanical shaker.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
406 7 Modeling Feedback Control Systems

3 10

2
5
1

0 0

–1
–5
–2

–3 –10
–4 –3 –2 –1 0 1 2 –2 –1.5 –1 –0.5 0 0.5 1
(a) (b)

Figure 7.32 Root locus for characteristic equations in equation 7.26, varying first 𝛼 (a) and then 𝛽 (b).

so K = 𝛼, N(s) = s2 , and D(s) = s3 + 3s2 + 3s + 6. It is also possible to study how variations in more than one parameter
influence the root locus.
For example, consider a CE,
s3 + s2 + 𝛽s + 𝛼 = 0, (7.26)
both 𝛼 and 𝛽 can vary or be varied. First arrange as if to study the effect of varying 𝛽, that is,
s
1+𝛽⋅ 3 = 0, (7.27)
s + s2 + 𝛼
but note that s3 + s2 + 𝛼 = 0 is the CE for 𝛽 = 0. For this equation, the root locus is studied from the form,
1
1+𝛼⋅ = 0. (7.28)
s2 (s + 1)
As shown in Figure 7.32(a), we first select 𝛼. Note, the indicated poles here are for a value of 𝛼 = 3.2 and indicate an
unstable system since they are on the right-hand plane. But these are poles for 𝛽 = 0 only.
Use the selected value of 𝛼 in equation 7.27 to then study varying 𝛽. This root locus is based on equation 7.28 and is
plotted in Figure 7.32(b). Now 𝛽 is varied to give a stable system when poles are in the left-hand plane. The value of 𝛽 in
this case is 11.5. We could successively iterate on 𝛼 and 𝛽 to achieve desired closed-loop poles for this system.12
Studying root locus can allow us to determine if adjusting gain with or without a given controller can achieve required
performance or stability. If gain changes alone do not provide the desired changes, it may be necessary to add other forms
of compensation or control action. This may particularly be the case when the system is not stable despite gain changes.
A root locus can help show how the addition of poles and/or zeros influences the loci, and thus how changes can be made
to influence both performance and stability.

7.4.4 Influence of Model Dynamics on Response and Stability

As a controlled system reacts to controller commands, the system may respond in ways that a model may fail to predict.
For example, the assumptions made in developing the model of the dc motor plant in Example 7.4 may not be valid when
the system responds at high rates, if the model was overly simplified. Previously ignored effects due to motor inductance,
shaft compliance, and motor inertia could all become important contributing effects. Models that include these effects are
shown in Figure 7.33 using bond graphs.
Adding motor inductance, Lm to the model gives a plant transfer function of,
𝜔m Km 𝜔2n
= , (7.29)
Vs s2 + 2𝜁𝜔n s + 𝜔2n

12 Remember, while we use the open-loop poles and zeros to sketch the root locus, the root locus is for a CE of 1 + GH, which is for the
closed-loop system.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
7.4 Linear Feedback Controllers and Compensators 407

: : : : : 1/ :

S 1 G 1 S 1 G 1 0 1

: : :

(a) (b)

Figure 7.33 (a) Plant model with inductance and (b) plant model with shaft compliance 1∕Ks and distinct motor and load inertia.

where
2
Rm B m + r m
𝜔n = .
Lm Jm
√ √
Bm Lm ∕Jm + Rm Jm ∕Lm
𝜁= √ .
2 Rm Bm + rm
In this case, the closed-loop transfer function becomes,
𝜔m Ka Km 𝜔2n
GcL = = .
Vr s2 + 2𝜁𝜔n s + (1 + Ka Km KT )𝜔2n
As shown previously the steady-state error goes to zero as Ka → ∞. For the transient response, we have a closed-loop
natural frequency 𝜔ncL and damping ratio 𝜁cL of,
√
𝜔ncL = 𝜔n 1 + Ka Kp KT .

𝜁
𝜁cL = √ .
1 + Ka K p K T
As Ka → ∞, the closed-loop natural frequency 𝜔ncL → ∞ and damping ratio 𝜁cL → 0. Since the closed-loop damping
is approaching zero, the control system will become increasingly oscillatory with more overshoot as the control gain is
increased as shown in Figure 7.34.
If shaft compliance, Cs = 1∕Ks , and motor inertia, Jm , are included in the model (but not inductance), the plant transfer
function becomes
𝜔m Km
= , (7.30)
Vs a 3 s3 + a 2 s2 + a 1 s + 1

1.8
1.6 Increasing
,

1.4
Tachometer voltage,

1.2
1
0.8
0.6
0.4
0.2
0
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08
Time, sec

Figure 7.34 The second-order system step response of dc motor tachometer with increasing control gain.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
408 7 Modeling Feedback Control Systems

where now
rm J L
Km = ,
Rm B L J m
( 2 )
rm 1 J
a1 = + Bm + L ,
Rm Ks J m B L J m
( 2 )
1 rm JL
a2 = + + Bm ,
Ks Rm Ks B L J m
JL
a3 = ,
Ks B L
and Ks = shaft stiffness, Bm = motor friction coefficient, Jm = motor inertia, BL = load friction coefficient, and JL = load
inertia.
The closed-loop transfer function then becomes
𝜔m Km
GcL (s) = = 3 2
.
Vs a 3 s + a 2 s + a 1 s + 1 + Ka K m K T
The steady-state error is zero as Ka → ∞. As in the previous cases, the system can develop oscillatory response and pos-
sibly go completely unstable. The closed-loop system is stable as long as the eigenvalues have negative real parts. The CE
for the closed-loop system,
a3 s3 + a2 s2 + a1 s + 1 + Ka Kp KT = 0,
can be used to construct a Routh table (see Chapter 4):

s3 a3 a1 0
2
s a2 1 + Ka Kp KT 0
a3
s1 a1 − (1 + Ka Kp KT ) 0
a2
s0 1 + Ka Kp KT 0 0

According to the Routh–Hurwitz criterion, the system is stable if and only if a3 > 0, a2 > 0, 1 + Ka Kp KT > 0, and,
a2 a1 > a3 (1 + Ka Kp KT ).
As Ka → ∞, this last inequality will be violated. This implies that the system will become unstable as Ka gets large.
Alternatively, this result can be shown by a root locus plot, where we are interested in showing the effect of increasing Ka
in a P-control. The form required for root locus analysis now becomes, using the plant transfer function from equation 7.30,
and assuming H = 1,
Km KT
1 + Ka ⋅ .
a 3 s3 + a 2 s2 + a 1 s + 1
Note that there are n = 3 open-loop poles and m = 0 open-loop zeros. Root locus plots, of course, typically require numerical
parameters (if using rlocus()), while a Routh analysis can provide inequalities solely from parameter variables. If we assume
here a case where there are two complex-conjugate poles and one real pole, Figure 7.35 shows how a root locus can be
“qualitatively sketched.”
Figure 7.35 Root locus plot of controlled third-order system.

=0

=
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
7.4 Linear Feedback Controllers and Compensators 409

The three open-loop poles mark the start of the loci, where Ka = 0. The arrows on the curves represent the trajectory of
the poles as Ka increases. Since m − n = 3, there are three asymptotes as Ka → ∞. As seen in this plot, the pair of complex
conjugate poles cross into the right-hand plane, indicating the system becomes unstable as Ka increases. Theoretically,
the response would grow exponentially for gains greater than the stability gain at this neutral point, Ku (gain for unstable
response). In actuality, unmodeled nonlinear effects, such as amplifier saturation, would prevent large Ka and thus limit the
response. This example shows how increased complexity in a system can often make it more difficult to stabilize a system.

7.4.4.1 Relative Stability and Stability Margins

Root locus plots are one way to provide insight into relative stability. The plots of root loci provide a fairly clear description
of how a system’s stability will depend on a parameter. In many cases, however, a frequency domain description of GH is
available and it may not be desirable or possible to determine the poles explicitly. The latter might be the case if the GH were
measured experimentally. Frequency domain plots of GH such as Nyquist plots, log-magnitude plots, and Bode diagrams
all provide ways to assess relative stability [53]. Bode plots are composed of the magnitude |GH| plotted in decibels (dB) and
phase 𝜙 = ∠GH in degrees. Relative stability in the frequency domain is measured by gain and phase margins defined using
both of these plots. These definitions can be effectively introduced in relation to a root locus plot by example. Consider the
system shown in Figure 7.36, which has an open-loop transfer function GH = K ⋅ 1∕s(s + 1)2 . The open-loop poles are −1,
−1, and 0, as shown in the root locus plot in Figure 7.37(a). As the gain K is increased from a value of 0, one of the poles
that originates at 𝜎 = −1 moves toward higher negative values. The other pole from -1 moves toward the right and meets
the pole that originates at 0. As K increases, these two poles meet then leave the real axis. These two loci will cross the j𝜔
axis when K = 2, defining a point of neutral (or marginal) stability. For K > 2 the two poles will move into the right-hand
plane, indicating that the closed-loop system will be unstable.

Figure 7.36 Block diagram of a plant with unity feedback. +

+1 2
–

3 100

50
Magnitude, dB

2
0

–50
1
Neutral –100
stability
–150 –2
0 10 10–1 100 101 102
Zero phase
–90 margin
(neutral stability)
–1
Positive phase
Phase, º

margin (stable)
–180
Negative phase
–2 margin (unstable)

–270
–3
–5 –4 –3 –2 –1 0 1 2 10–2 10–1 100 101 102
Frequency, rad/sec
(a) (b)

Figure 7.37 (a) Root locus plot for GH = K∕s(s + 1)2 , indicating point of “neutral stability” where K = 2. (b) Bode plots for different
values of K showing the progression of the magnitude plot as K increases. Note that phase is not influenced when only gain is
changed, so the same phase plot holds for all three cases.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
410 7 Modeling Feedback Control Systems

Gain crossover
frequency

0 0
Negative
Positive gain
gain margin
margin
GH, ° GH, °
–90º –90º
–180º –180º
Positive
Phase
–270º phase –270º
crossover Negative phase
margin
frequency margin

Stable system Unstable system

(a) (b)

Figure 7.38 Phase and gain margins for stable (a) and unstable (b) systems.

The corresponding Bode plots are shown in Figure 7.37(b). Note that at the neutral stability point, the gain |GH| crosses
the 0 dB line right when the phase crosses −180∘ . These Bode plots define the system in neutral stability. If K < 2, the phase
at the point where |GH| crosses 0 dB, called the gain crossover frequency, 𝜔c , indicates that the phase 𝜙 is less than −180∘ .
We define a phase margin by this difference, or, 𝛾 = 180∘ + 𝜙. Systems with 𝛾 > 0 are stable. On the other hand, if K > 2,
we see that the phase margin is negative for the unstable case.
The amount of phase margin is a measure of relative stability. If 𝛾 is positive and close to zero, the system is less stable
than when 𝛾 is higher. It can also be said that the phase margin is that amount of additional phase lag at the gain crossover
frequency, 𝜔c , required to bring a system to the verge of instability. A gain margin is defined by the reciprocal of the mag-
nitude |GH| at the phase crossover frequency, 𝜔1 , which is the frequency at which the phase is equal to −180∘ . It can also
be denoted,
1
Kg = .
|G( j𝜔1 )|
If gain margin is negative, a system is unstable. These definitions and terms are illustrated in Figure 7.38. The gain margin
for stable minimum-phase systems13 provides a measure of how much the gain can be increased before a system becomes
unstable. Thus, for an unstable system, the (negative) gain margin tells us how much the gain should be decreased to make
the system stable.
Relative stability should be assessed with both margins. Stable systems must have a positive gain margin and a positive-
phase margin. For the latter, it is generally considered desirable to have 30∘ < 𝛾 < 60∘ . A gain margin of at least 6 dB is
desirable. In such cases, the stability of a minimum phase system will typically be assured even if there are variations in
the estimated system parameters used in the Bode plots. Both commercial and open-source programs have functions that
extract these stability margins from frequency domain models.14

7.4.4.2 Relating Time and Frequency Domain Response and Stability Metrics
The transient response characteristics of the first- and second- order systems (see Chapter 5) are useful in defining metrics
for performance in feedback control. Since design can be done both in time and in frequency domain, it can also be helpful
to relate metrics from these contexts. For example, a system with dominant first-order behavior can be described by its time
constant, 𝜏t , which is sufficient to specify the corner frequency, 𝜔b = 1∕𝜏t in magnitude and phase plots. So the relationship
between time- and frequency-domain metrics for a first-order system is relatively easy to understand. A closed-loop system
that would result in a standard first order form is shown in block diagram form in Figure 7.39(a); as clearly, Gcl = y∕r =
(1∕𝜏t s)∕(1 + 1∕(𝜏t s)) = 1∕(𝜏t s + 1).
The relationships for the second- order systems are a bit more involved but help when it comes to dealing with
higher-order systems as well. Consider the feedback system in Figure 7.39(b), which leads to a closed-loop transfer

13 Stable non-minimum phase systems, which have zeros in the right-half plane, have negative phase and gain margins.
14 Other means for assessing stability such as Nyquist are beyond the scope of this brief overview, but the reader can refer to Ogata [53] or
Franklin et al. [93] for discussions on this topic.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
7.4 Linear Feedback Controllers and Compensators 411

+ +
– –

(a) (b)

Figure 7.39 Closed-loop feedback systems resulting in standard: (a) the first- and (b) the second-order form.

function that is a standard second-order form; that is, Gcl = y∕r = 𝜔2n ∕(s2 + 2𝜁𝜔n s + 𝜔2n ). Feedback systems of practical
value can take this form, as shown in previous examples (e.g., see Example 7.4).
Section 5.3.2 reviews key transient response characteristics for the second-order systems. Measures such as over-
shoot percentage and settling times are of particular relevance, and for second-order systems, these relate to 𝜁 and
𝜔n . In particular, the maximum overshoot ratio, Mp , which occurs at time t = tp = 𝜋∕𝜔d , is found to be only depen-
√
dent on 𝜁, Mp = e−(𝜁 ∕ 1−𝜁 )𝜋 . This value is commonly used as a time-domain specification. Another key metric for a
2

standard√second-order system is the peak of the amplitude function |G( j𝜔)|, which occurs√at the resonant frequency,
𝜔r = 𝜔n 1 − 2𝜁 2 . The resonant peak amplitude is found to be (see Chapter 6), |G|r = 1∕2𝜁 1 − 𝜁 2 . Plotting Mp versus
𝜁 and |G|r versus 𝜁, as in Figure 7.40(a), provides one way to relate time- and frequency- domain specifications for the
second order systems.
Figure 7.40(a) shows that as 𝜁 decreases both Mp and |G|r increase. There is a good correlation between these metrics
for 0.4 < 𝜁 < 0.71, which is considered a range for “good design.” Note that the value of |G|r for small and large 𝜁 gives
diverging trends in Figure 7.40(a).
Consider now the open-loop GH for the standard second-order system (H = 1) of Figure 7.39(b),
𝜔2n
GH = . (7.31)
s(s + 2𝜁𝜔n )
The magnitude will cross a value of 1 at the gain crossover frequency, 𝜔c , thus we can find this value of frequency as
follows:
𝜔2n [√ ]1∕2
|GH(𝜔c )| = √ = 1 ⇒ 𝜔c = 𝜔n 1 + 4𝜁 4 − 2𝜁 2 .
𝜔4c + 4𝜁 2 𝜔2n 𝜔2c
Recall that the phase margin is defined at this frequency,

𝛾 = 180∘ + 𝜙(𝜔c ) = 180∘ + ∠GH(𝜔c ),

3 90

2.5
Phase margin, γ, °

2 60

1.5

1 30

0.5

0 0
0 0.2 0.4 0.6 0.8 1 0 0.4 0.8 1.2 1.6 2

(a) (b)

Figure 7.40 Time- and frequency-domain specifications for a standard second-order system can be related through their dependence
on damping ratio, 𝜁, as shown in plots of: (a) peak overshoot ratio, Mp , and peak resonant magnitude, |G|r , versus 𝜁; (b) phase margin, 𝛾,
versus 𝜁.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
412 7 Modeling Feedback Control Systems

where
[ ]1∕2
⎛ √ 4 − 2𝜁 2 ⎞
1 + 4𝜁
∘ −1 ⎜ ⎟
∠GH(𝜔c ) = −∠j𝜔c − ∠( j𝜔c + 2𝜁𝜔n ) = −90 − tan ⎜
2𝜁 ⎟,
⎜ ⎟
⎝ ⎠
So the phase margin for a standard second-order system is solely dependent on 𝜁,
[ ]1∕2
⎛ √ 4 − 2𝜁 2 ⎞ ⎛ ⎞
1 + 4𝜁
∘ −1 ⎜ ⎟ −1 ⎜ 2𝜁 ⎟
𝛾 = 90 − tan ⎜ ⎟ = tan ⎜ [√ ]1∕2 ⎟ . (7.32)
⎜ 2𝜁 ⎟ ⎜ ⎟
⎝ ⎠ ⎝ 1 + 4𝜁 4 − 2𝜁 2 ⎠
This relationship, plotted in Figure 7.40(b), provides a way to directly relate phase margin and damping ratio for a
second-order system. The relationship is reasonably approximated for 𝜁 ≲ 0.7, by a linear relation, 𝛾 = 100𝜁 (degrees).
The ability to analytically relate time-domain characteristics of the second-order systems to frequency-domain charac-
teristics provides useful insights when confronted with “mixed” specifications; that is, some that are more easily thought
about in time domain than in frequency domain.

Bandwidth Related to 𝜻 The system bandwidth, the range of frequency over which system response is deemed satisfactory,
defined as the frequency 𝜔b at which the value of the gain magnitude, |G|, drops by 3 dB from the value at 𝜔 = 0, or,
can be √
|G| ≈ ( 2∕2)|G(0)|.
For a second-order system, we can determine a relationship between a bandwidth frequency, 𝜔b and the undamped
natural frequency, 𝜔n , by setting the magnitude equal to 0.707 (the value 3 dB down),
√
2 1
|G( j𝜔b )| = = √ ,
2 [ ]2 [ ]2
2
1 − ((𝜔b ∕𝜔n )) + 2𝜁𝜔b ∕𝜔n
Solving for the ratio, 𝜔b ∕𝜔n gives,
𝜔b [ √ ]1∕2
= 1 − 2𝜁 2 + 2 − 4𝜁 2 (1 − 𝜁 2 ) . (7.33)
𝜔n
For a second-order system, when 𝜁 = 0.707, 𝜔b = 𝜔n .
Now, how do you select a desired crossover frequency given a desired bandwidth? Again, following guidance from a
standard second-order system, we can show that 𝜔c ≈ 0.635𝜔b . First, remember that the 𝜔c is defined for an open-loop
system, which for the standard second-order system takes the form, GH = 𝜔2n ∕s(s + 2𝜁𝜔n ). Thus, we can use this to estimate
a relation for how the crossover frequency for this case relates to 𝜔n . That is, we can solve for when |GH| = 1, which occurs
when
𝜔c [√ 4 ]1∕2
= 4𝜁 + 1 − 2𝜁 2 .
𝜔n
Using this relation, we can now estimate how 𝜔c relates to 𝜔b as follows:
[√ ]1∕2
4𝜁 4 + 1 − 2𝜁 2
𝜔c 𝜔 𝜔
= c ⋅ n = [ √ ]1∕2 .
𝜔b 𝜔n 𝜔b
1 − 2𝜁 + 2 − 4𝜁 (1 − 𝜁 )
2 2 2

When this relation is plotted over values of 𝜁 from 0 to 1, it ranges from 0.64 to 0.7 and is often just estimated as 0.635;
that is, 𝜔n ≈ 0.635𝜔b for a second-order system.
Remember, while these are all approximations based on standard second-order systems, they are useful if we can justify a
particular mode of a system as being dominant. These initial approximations then give us a close estimate on selection of
a controller or compensator design, which can then be refined through computational analysis. A tabulation of the relations
reviewed for a second-order system such as in Table 7.5 is commonly used to summarize and relate specifications in time
and frequency domain.
For systems of higher order (n > 2), it can become more difficult to relate transient time domain and frequency domain
characteristics or specifications. The relations found above for second-order systems can be used to approximate those
cases that have dominant closed-loop poles. Dominant poles are those close to the j𝜔 axis that have real parts about five
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
7.4 Linear Feedback Controllers and Compensators 413

Table 7.5 Correlations between bandwidth and crossover frequency with damping
and overshoot for the standard second-order system.

𝜻 𝜸 (rad) 𝜸 (∘ ) 𝝎c Mp 𝝎c ∕𝝎b 𝝎c ∕𝝎n 𝝎b ∕𝝎n

0.0 0.00 0.00 1.00 100.00 0.64 1.00 1.55

0.1 11.42 11.42 0.99 72.92 0.64 0.99 1.54
0.2 22.60 22.60 0.96 52.66 0.64 0.96 1.51
0.3 33.27 33.27 0.91 37.23 0.63 0.91 1.45
0.4 43.12 43.12 0.85 25.38 0.62 0.85 1.37
0.5 51.83 51.83 0.79 16.30 0.62 0.79 1.27
0.6 59.19 59.19 0.72 9.48 0.62 0.72 1.15
0.7 65.16 65.16 0.65 4.60 0.64 0.65 1.01
0.8 69.86 69.86 0.59 1.52 0.67 0.59 0.87
0.9 73.51 73.51 0.53 0.15 0.71 0.53 0.75
1.0 76.35 76.35 0.49 0.00 0.75 0.49 0.64

times that of other poles in the system, and with no zeros close by. The effect of dominant poles will tend to dominate the
transient response of the system, as the other poles (with large time constants) are assumed to decay at a much faster rate.
If the dominant poles are complex-conjugate, then the transient response can be approximated using the aforementioned
second-order system metrics now defined by the dominant pole damping ratio and undamped natural frequency. For this
case, it can be assumed that:

a) if 1 < |G|r < 1.4, this implies damping between 0.4 and 0.7, and thus good relative stability;
b) the system will effectively have a bandwidth defined by the 𝜔r and the transient rise time (for step inputs) will be approx-
imately 1∕𝜔r , with the transient damped natural frequency lying in the range of 𝜔r ; and
c) if |G|r > 1.5, the system will have high transient oscillations, and any inputs with frequency components close to 𝜔r will
be highly amplified, and transients will have a damped natural frequency close to 𝜔r .

In some cases, a higher-order control system gain may be increased in order to cause some poles to become dominant
and complex-conjugate. As long as the system is stable, this is one way to minimize the effect of hard nonlinearities such
as dead zone, backlash, and stiction [53].
Intuitively, this would seem to have the effect of using the controlled oscillations as a form of “natural dithering.” In
contrast, an overly damped (lower gain) might be more sensitive to these nonlinear effects.

7.4.5 Controller and Compensator Selection and Design

Proportional control directly adjusts loop gain15 and thus influences performance and relative stability but often only in a
limited manner. Dynamic controllers and compensation, such as those in Table 7.3, provide additional means for improving
performance, accuracy, and stability. Their transfer functions are not overly complex, but their design (tuning) requires
balancing gain and a suitable combination of poles and/or zeros to achieve desired closed-loop system characteristics.
For example, achieving satisfactory relative stability may translate into transient response where overshoot is not severe
and excessive oscillations do not endure. Both root locus and frequency domain analysis and design methods may be used
for these purposes, and topics from Chapters 5 and 6 are helpful in understanding these approaches. This section uses
examples to illustrate these concepts and methods, emphasizing how modeling plays a role in control design.
Open-loop analysis using a model can reveal ways that “passive” adjustments (damping, stiffness, etc.) can improve a
system and possibly make application of “active” control methods easier and more effective. It can also be helpful to use a
model to make early decisions about actuator and sensor requirements. The influence can often be understood by examining
pole-zero plots and root loci as well as frequency domain representations of the open-loop transfer function.

15 Loop gain, or the product of forward gain, say, G(s), and feedback gain, H(s), is also designated by L = GH in many references.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
414 7 Modeling Feedback Control Systems

7.4.5.1 Effect of Added Poles and Zeros

If gain alone cannot achieve the needs, it can be necessary to add poles and/or zeros using passive adjustments or dynamic
compensation/control. There are many ways to make such changes. A couple of basic rules can be very useful in many
practical applications. Consider the root locus in Figure 7.41(a), which represents a system with two real poles under pro-
portional control. As gain is increased, the two poles move along the real axis, becoming equal before splitting into opposite
positive and negative directions as complex conjugate pairs. Further increasing gain does not change the real values, but the
imaginary parts increase. That is, changing gain may influence effective damping and/or stability; however, gain change
alone will not change the shape of the loci, which can sometimes be necessary to influence system relative stability. For the
case in Figure 7.41(b), a pole is added and effectively this has the effect of pushing the loci away from that pole. In this case,
the poles move toward the j𝜔 axis, decreasing relative stability of the system. Alternatively, adding a zero as in Figure 7.41(c)
has the effect that the root loci will be pulled toward that zero. In this way, a zero can help improve relative stability by pulling
root loci away from the imaginary axis and into the left-hand plane. Remembering these trends can provide guidance when
trying to decide on dynamic compensation/control.
Reshaping the loci with dynamic compensation in the form of PID, lead/lag compensators, or feedback compensation,
such as those shown in Table 7.3, can be achieved by introducing poles and/or zeros into the loop gain, so it is necessary to
understand how the placement will influence response characteristics. The effects of poles and zeros on transient response
are also discussed in Section 5.5. An example of how controller zeros can shape the transient dynamics is discussed with
respect to the error rate damping in Section 7.4.2. Another useful example is basic PI control of a first-order system, so
consider the case shown in Figure 7.42(a).16 The closed-loop transfer function relating y to the reference input, r, takes the
form,
y N(s) K(s + z)
Gcl (s) = = = 2 .
r D(s) 𝜏t s + (1 + K)s + Kz
Now, study the effect of varying the zero, z, keeping poles fixed at −2 ± j. Thus, make the CE, D(s) = s2 + 4s + 5, so,
𝜏t = 1 and K = 5∕z. Choosing to examine z = 4, 2, 1.5, 1.3 requires K = 5∕z. The poles and the varying zeros are shown in
Figure 7.42(b), and the step response results are in 7.42(c).
Varying z as shown in Figure 7.42(b) from left to right increases the time to reach a peak value and increases the overshoot.
It is important to note that the typical correlations between key response specifications of the first- and second-order systems
that otherwise depend solely on 𝜏t or 𝜁 are not valid when there are zeros present. Choosing the zero location for the PI
controller in this case is critical in shaping the transient response.

7.4.5.2 Pole-Zero Cancellation

Pole-zero cancellation refers to adding poles and/or zeros that will effectively “cancel” corresponding zeros and/or poles
in GH that are in undesirable locations. The process can be used in either open- or closed-loop control configurations. For
example, placing a zero in a compensator in the vicinity of a pole of the system GH will result in the closed-loop system
having a shorter “small transient” due to that pole. A compensator may also be introduced solely to cancel undesirable
poles, especially when feedback cannot change the poles or minimize their effect.

Added zero
Added pole

(a) (b) (c)

Figure 7.41 Effect on root locus from (a) gain change brings two poles to meet and become complex conjugate pair, (b) adding a pole
can “push” other poles toward imaginary axis, and (c) adding a zero can “pull” poles toward it, and away from imaginary axis to
improve relative stability.

16 Using form from Table 7.3; compare to equations 7.14 and 7.15.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
7.4 Linear Feedback Controllers and Compensators 415

+
–

(a)

1.2

0.8
–1.5 –1.3

y(t)
0.6
–4 –3 –2 –1 0.4

0.2

0
0 0.5 1 1.5 2
Time, sec
(b) (c)

Figure 7.42 (a) PI control of a first-order system, (b) closed-loop poles and the zero, shown at different locations, with varying K, and
(c) step response for different zero values, with gain varied to keep poles at −2 ± j.

For example, consider design of a system G(s) = 1∕(s + 1)(s + 4), which is to be controlled using an open-loop compen-
sator, Gc = s + z, in block diagram form:

Compensator Plant

The open-loop transfer function is,

y (s + z)
Gol = = .
r (s + 1)(s + 4)
Pole-zero cancellation can be used to make the system appear like either 1∕(s + 1) or 1∕(s + 4) simply by choosing z to be
1 or 4, respectively. However, we can also show the influence of z by examining the transient response, say, to a step input,
u = 1∕s. The response is, y = Gol ⋅ (1∕s), expanded by partial fractions as,
s+z a b c
y= = + + .
s(s + 1)(s + 4) s s+1 s+4
The residues a, b, and c can be found from
s + z = a(s + 1)(s + 4) + bs(s + 4) + cs(s + 1)
giving the three relations,
0 = a + b + c.
1 = 5a + 4b + c.
z = 4a.
The results are: a = z∕4, b = (1 − z)∕3, and c = (z − 4)∕12, and we find,
z (1 − z)∕3 (z − 4)∕12 1 (1 − z) 1 z−4 1
y= + + = − + .
4s s+1 s+4 s 3 s+1 12 s + 4
The step response is thus,
z (1 − z) −t z − 4 −4t
y(t) = − e + e . (7.34)
4 3 12
In this form, the effect of pole-zero cancellation on the response is evident. If z = 1 in equation 7.34, the effect due to the
pole at s = −1 is canceled. A similar effect is seen when z = 4 with the pole at s = −4. Small errors in the selection of z may
not sufficiently cancel a pole’s contribution to the response.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
416 7 Modeling Feedback Control Systems

Purposely introducing a pole-zero cancellation through a compensator or controller is a common way to minimize the
effect of undesirable poles. As shown in the previous example using residues in a partial fraction expansion, the zero influ-
ences the magnitude of both of the response components due to each pole. In this way, if the zero is close to a pole, then the
“transient” effect from that pole will be minimized or canceled. This makes pole-zero cancellation a very useful method.
For example, if there are complex conjugate poles close to the imaginary axis, then these may dominate system response.
Introducing a canceling set of zeros can reduce the effect of the transients due to these poles.
Note, however, that pole-zero cancellation is not recommended for unstable poles, for reasons evident from the examples
2
discussed. If there is error in the model then the system will remain unstable. For example, the unstable system, Gp = (s−5) ,
has an unstable pole, p = +5, that will show up in GH, unless it is canceled by Gc . Consider, Gc = (s − 5)∕(s + 3),
s−5 2 2
GH = Gc Gp = ⋅ = ,
s+3 s−5 s+3
and this is stable, as long as the plant model is accurate. Otherwise, the unstable pole will not be canceled. With an implied
error, Gc = (s − 5)∕(s + 3), but the open-loop transfer function is actually,
2(s − 5)
G = G c Gp = ,
(s − 5 + 𝜖)(s + 3)
which remains unstable. The pole/zero cancellation does not work. This is a strong argument against using pole-zero can-
cellation to stabilize an unstable plant with open-loop control. Other control or compensation approaches should be used
for stabilization.

Example 7.5 Pole-zero cancellation for non-minimum phase

Discuss the viability of using pole-zero cancellation for a system with plant transfer function, G2 (s) = (s − 1)∕(s + 1).
Solution
The plant transfer function is stable, with a pole at s = −1, but the system is non-minimum phase. We can try to cancel
the zero by using a compensator of the form, G1 = K∕(s − 1), based on the model given. However, if the model has some
error in the zero, say, G2 = (s − 1 − 𝛿)∕(s + 1). The unstable pole in the controller cannot cancel the zero so the closed-loop
transfer function becomes,
K s−1−𝛿
⋅
s−1 s+1 K(s − 1 − 𝛿) K(s − (1 + 𝛿))
Gcl = = = .
K s−1−𝛿 (s − 1)(s + 1) + K(s − 1 − 𝛿) s2 + Ks − [1 + K(1 + 𝛿)]
1+ ⋅
s−1 s+1
For this system to be stable requires K > 0 and K < −1∕(1 + 𝛿), which cannot be satisfied by any K for 0 < |𝛿| ≪ 1
(small 𝛿).

7.4.5.3 Studying Model-Structure for Controller Selection

Studying the dual systems shown in Figures 7.43(a) and (b) gives rise to common issues that can arise when trying to select
a controller or compensator. These systems are generalized classic T- and Π-section models that can be used to represent
a wide range of physical systems or parts of more complex systems. The models are presented using generalized physical
variables (e, f , q, p), and all elements are assumed to be linear. The system in (a) has effort inputs for control, ec , and
disturbance, ed , while the dual in (b) has flow inputs, fc and fd , for control and disturbance, respectively. Two different
measurements are considered in each case, with ideal sensing from 1 and 0 junctions, and the H1 and H2 blocks represent
sensor transfer function models.
There are various cases that can be examined depending on which sensor is used, so it is efficient to derive the possible
transfer functions using a state-space form of the equations. Alternatively, the impedance methods described in Chapter 6
can be used, especially for a specific transfer function.
Causality assigned on the bond graphs indicates the states for the ICI system in Figure 7.43(a) can be taken as f1 = p1 ∕I1 ,
f2 = p2 ∕I2 , and q, resulting in state equations:
⎡ḟ 1 ⎤ ⎡−R∕I1 −1∕(I1 C) +R∕I1 ⎤ ⎡f1 ⎤ ⎡1∕I1 0 ⎤ [ ]
⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥ ec
ẋ = ⎢ q̇ ⎥ = ⎢ +1 0 −1⎥ ⎢ q ⎥ + ⎢ 0 0 ⎥ . (7.35)
⎢̇ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥ ed
⎣f 2 ⎦ ⎣ 0 +1∕(I2 C) 0⎦ ⎣f2 ⎦ ⎣ 0 −1∕I2 ⎦
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
7.4 Linear Feedback Controllers and Compensators 417

(a) (b)

Figure 7.43 Bond graphs showing control and disturbance inputs on generalized (a) ICI, or T-section model, and (b) CIC, or classic
Π-section model.

Consider two output equations as,

[ ] [ ⎡0 ] 0⎤ [ ]
f1 ⎢
1 0 0 ⎥ ec
y= = x + ⎢0 0⎥ ,
f2 0 0 1 ⎢ ⎥ ed
⎣0 0⎦
so the transfer functions relating both outputs (f1 and f2 ) to inputs ec and ed are given by the transfer function matrix,
⎡ f1 f1 ⎤ ⎡ I2 Cs2 + RC + 1 −(RCs + 1) ⎤
⎢e ed ⎥ ⎢ D(s) D(s) ⎥
Gfe (s) = ⎢ ⎥=⎢ ⎥,
c
(7.36)
⎢ f2 f2 ⎥ ⎢ RCs + 1 −(I1 Cs2 + RCs + 1) ⎥
⎢ ⎥ ⎢ ⎥
⎣ ec ed ⎦ ⎣ D(s) D(s) ⎦
where D(s) = s(I1 I2 Cs2 + RC(I1 + I2 )s + I1 + I2 ).
Transfer functions relating displacement output variables to the effort inputs can be found by letting H1 = H2 = 1∕s,
resulting in,

⎤ ⎡ I2 Cs + RC + 1 ⎤
q q1 2
⎡ 1 −(RCs + 1)
⎢ ec ed ⎥ ⎢ sD(s) sD(s) ⎥
Gqe (s) = ⎢ q ⎥ = ⎢ ⎥. (7.37)
⎥ ⎢⎢ −(I1 Cs2 + RCs + 1) ⎥
q2
⎢ 2 RCs + 1
⎥
⎣ ec ed ⎦ ⎣ sD(s) sD(s) ⎦
The dual form in Figure 7.43(b) can be used to find a similar set of transfer functions, in this case relating effort or
displacement states of the C elements to the input flows. Note also that other transfer functions can be derived for either
system.

Example 7.6 Two-car train

The two-car train shown in Figure 7.44 can be modeled by the bond graph in Figure 7.43(a), with I1 = m1 , I2 = m2 , C = 1∕k,
and R = b = 0 (negligible damping). Assume the velocity or position of mass m2 can be measured and find a relation of each
to the input force Fc and disturbance force Fd . Formulate open-loop control diagrams for each case.

Figure 7.44 A two-car train with controlled drive force, Fc , on car 1 and disturbance or draw-bar
disturbance, Fd , on car 2.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
418 7 Modeling Feedback Control Systems

Solution
From equation 7.36, we can write, with R = 0,
1 (m1 ∕k)s2 + 1
𝑣2 = ⋅ Fc − ⋅F
s(m1 m2 s2 ∕k+ m1 + m2 ) s(m1 m2 s2 ∕k + m1 + m2 ) d
[ ]
(k∕m1 m2 ) (k∕m1 m2 ) (m1 ∕k)s2 + 1
= ⋅ Fc − ⋅ Fd
s(s2 + 𝜔2n ) s(s2 + 𝜔2n )
(k∕m1 m2 ) [ ( ) ]
𝑣2 = Fc − (m1 ∕k)s2 + 1 Fd .
s(s + 𝜔n )
2 2

From this last expression, open-loop diagrams can be drawn as in Figure 7.45. The diagram for x2 would simply require
appending 1∕s at the 𝑣2 output.

Figure 7.45 Open-loop diagram for two-car train system.

–
+

The diagram of Figure 7.45 can now be integrated in a closed-loop control design.

The two-car train system of Example 7.6 can illustrate the difference between having a sensor non-collocated versus
collocated with a control input. Having a collocated sensor in this case would measure the velocity or displacement of mass
m1 . For the purpose of this discussion, remove the disturbance, Fd = 0, and set m1 = m2 = 1, C = 1∕k = 0.2, and b = 0.
The two transfer functions for displacement outputs with sensor and actuator collocated and non-collocated then become,
respectively,
x1 s2 + 5 x2 5
= 2 2 (collocated) and, = 2 2 (non-collocated). (7.38)
Fc s (s + 10) Fc s (s + 10)
√
The√ CE indicates two open-loop poles at the origin, and two at ±j 10. The collocated case also has two imaginary zeros
at ± 5. Applying proportional control alone to either of these systems alone results in poorly controlled systems. The
collocated system with P-control alone will result in purely oscillatory behavior to a step response for any proportional
gain, while the non-collocated system will be unstable for any value of gain. Thus, an alternative controller is required.

7.4.5.4 Using Root Locus for Controller Assessment

With PD control, K(1 + 𝜏d s), applied to these systems, the root locus for each will be as shown in Figure 7.46. Now the collo-
cated case PD controller can be tuned to give much better results, while PD control is still ineffective for the non-collocated
system. In the latter, any gain still leads to an unstable system. The results from the collocated case (a) suggests that the
non-collocated case may benefit
√ from a form of compensation that similarly introduces complex zeros that can effectively
keep the unstable poles at ± 10 from moving into the right-hand plane.
The basic structure of the two systems in Figure 7.43 arises in many types of practical systems. The systems have iden-
tical transfer function forms between outputs and inputs. Even if transformers are inserted to represent power-conserving
transformations within or across energy domains, the same model forms will arise.17 As such, modeling insights gained
from this case study are transferable to a wide range of applications with physical plants having a model structure as in
Figure 7.43, and specifically when complex-conjugate poles lead to stability issues.

7.4.5.5 Application of Frequency Domain Methods

We can also use the systems in Figure 7.43 to describe how key performance requirements specified in the frequency
domain can be achieved using compensators and controllers. The frequency-domain concepts and methods discussed in
Chapter 6 are relevant for this discussion, as are the relationships between time- and frequency-domain metrics discussed
in Section 7.4.4.

17 This is not necessarily the case if gyrators are included.

 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
7.4 Linear Feedback Controllers and Compensators 419

5 5
4 4
3 3
2 2
1 1
0 0
–1 –1
–2 –2
–3 –3
–4 –4
–5 –5
–5 0 5 –5 0 5
(a) (b)

Figure 7.46 Root locus with PD control, K(1 + 𝜏d s) applied to (a) the collocated system and (b) non-collocated systems in
equation 7.38 with 𝜏d = 1.

Frequency-domain representations provide a useful perspective on how a system behaves across a range of frequen-
cies, and this can be especially useful for certain feedback control relationships, such as sensitivity, bandwidth, and key
system variables. In some cases, a system may be best understood using experimental methods and represented using
frequency-domain models. Indeed, there are some performance metrics that arise in a frequency-domain context. For
example, filtering of system signals in low, high, or band-pass regions is often an important measure of performance, not
as effectively conveyed using time-domain descriptions.
Desired performance measures in the frequency domain design are related to the open-loop gain, GH, and help guide
selection of controller or compensator forms such as those in Table 7.3. The trends inherent to the basic factors discussed in
Chapter 6 are particularly useful in understanding how common measures and guidelines are defined. Frequency-domain
control design approaches, such as loop shaping, seek to manipulate the open-loop frequency response so as to achieve:
(i) stability; (ii) desired performance, meeting any time and frequency domain specification; and (iii) sufficient robustness
in terms of phase and gain margins. Some common guidelines are typically shown as in the magnitude and phase plots in
Figure 7.47, as described in the following.

Performance,
disturbance attenuation

Robustness
Allowed and performance
error
Noise
attenuation

Robustness and
noise rejection

–90

Robustness
–180

–270
0.1 10

Figure 7.47 Performance and sensitivity goals through shaping of the open-loop magnitude and phase.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
420 7 Modeling Feedback Control Systems

Estimating Relative Stability Measures from GH The integration of compensator and plant transfer functions into GH will
result in magnitude and phase that can be readily computed using software tools. However, it is helpful to understand how
the transfer function forms lead to certain trends. Achieving a desired level of relative stability, however, is associated with
the phase margin, as previously discussed (see Figure 7.37). Consider a case where |GH| features a −20 dB/decade slope over
a broad range at the crossover frequency, 𝜔c , as shown in Figure 7.48(a). If the system is minimum phase, it can be assumed
that |GH| is dominated by an integral factor 1∕s. An estimate for the phase margin, 𝛾, is then 𝛾 ≈ 180∘ − 90∘ ≈ +90∘ , which
would indicate high relative stability. In contrast, if the slope is closer to −40 dB/decade as in Figure 7.48(b), the system
has a dominant 1∕s2 factor at 𝜔c which would imply the system is nearly unstable. If the region of the crossover frequency
features apparent combinations of basic factors, such as in Figure 7.48(c), known trends can reveal by inspection the level
of stability. For example, in this case, the shape indicates an integral factor and a first-order factor leading to a phase margin
of 45∘ , and thus a stable system. It should be identified how the 1∕s factor in this latter case reduces the phase margin.
It is important to also remember that these correlations between the magnitude and the phase for basic factors assume
the system is minimum phase. If the case in Figure 7.48(c) was altered by a zero in the right-hand plane, that is, GH =
(s − 1)∕(s∕𝜔c )(s∕𝜔c + 1), forming a non-minimum phase system, then the expected correlations between magnitude and
phase that help guide design do not hold. If the numerator was s + 1 the phase would increase toward +90∘ while for the
s − 1 (non-minimum phase) case, the phase would actually decrease further. While the magnitude plots looks the same, the
phase plots do not.

Steady-state Accuracy Error constants convey a way to assess transient errors, but we can also aim to meet accuracy require-
ments by making sure that the low-frequency asymptote of |GH| ≈ K∕( j𝜔)N (where N is the type number) does not fall
below a certain level, and thus it is desired to raise it as indicated in Figure 7.47. Error constants also ensure that zero
steady-state error can be achieved for certain types of inputs.

Accuracy Across Frequency Range of Interest It should be ensured that the sensitivity, or error, function, S = e∕r = 1∕(1 + GH),
meets desired specifications over the expected range of operating frequencies.

Noise Rejection At the high-frequency end of the operating range, a desired minimum attenuation of noise can be specified
at a frequency such as 𝜔g in Figure 7.47. A specification can be related to the ratio y∕𝑣 (equation 7.5), which is sometimes
referred to as the complementary sensitivity, T(s). Note that to make T(s) small (good noise attenuation) requires |GH| to
be small, which is at odds with the desire to make S (e∕r) small which requires |GH| to be large. Indeed, the constraint
is inherent to the fact that T(s) + S(s) = 1, so both cannot be “small” at the same time. This is why the shape indicated in
Figure 7.47 for |GH| ends up being a desirable goal in loop shaping. It is important to identify that where it is desired to have
|GH| large at low-frequencies (below 𝜔a ), the error (or sensitivity) to variations in model parameters and disturbances will
be small. At higher frequencies above 𝜔d and beyond the bandwidth 𝜔b the open-loop gain |GH| is preferred to be small.
Any high-frequency plant dynamics that have low gain can thus be ignored, provided there are no significant resonances
that can reach levels of 0 dB. Any such effects could introduce additional crossover frequencies into |GH| that could have

60 60 60
40 40 –40 dB/dec 40
–20 dB/dec –20 dB/dec
20 20 20
0 0 0
–20 –20 –20 –40 dB/dec
–40 –40 –40
–60 –60 –60
–80 –80 –80
0 0 0

–90 –90 –90

º

º
º

–180 –180 –180

–270 –270 –270

0.1 10 0.1 10 0.1 10

Figure 7.48 Approximate trends in the gain and their relation to phase margin for minimum phase systems.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
7.4 Linear Feedback Controllers and Compensators 421

dB dB

–20 dB/decade
º º –20 dB/decade

(a) (b)
dB dB
+20 dB/decade
+20 dB/decade

º º

(c) (d)

Figure 7.49 Frequency domain form of dynamic compensators: (a) PI control, (b) lag compensator, (c) PD control, and (d) lead
compensator.

negative phase margins. These should not be ignored and would require a more complex plant model. In addition, this
would impact the actuator and sensor selection and controller design.
Setting aside cases where the system is open-loop unstable or of non-minimum phase type, the process of design using
frequency domain methods makes use of techniques meant to shape the open-loop gain as in Figure 7.47. This can be
accomplished using dynamic controllers or compensators, such as those in Table 7.3. Figure 7.49 compares Bode plots
of PI and PD controllers to associated lag and lead compensators, respectively. Key parameters are adjusted to meet the
control specifications. For example, for lag or lead compensators, 𝜏1 and 𝜏2 are used to set the amount of gain attenuation
or gain difference (a) and also to set the maximum phase 𝜙m center frequency for phase lag or lead. These adjustments will
effectively shape the open-loop transfer function GH as needed.
Using a phase lead compensator, which approximates PD control, tends to be stabilizing, as it adds positive phase margin,
and acts to speed up response by lowering rise time thus decreasing transient overshoot. Phase lag compensators, which
approximate PI control, are usually used to improve steady-state accuracy, essentially adding a 1∕s factor at low frequencies
relative to the crossover.
Consider again the two-train example system from Example 7.6 for the collocated and non-collocated cases. As shown
before, the collocated case can be stabilized, as shown in Figure 7.50(a). The phase plot shows how the phase margin is
increased effectively using a PD controller. This is not the case for the same controller applied to the non-collocated system,
as shown in Figure 7.50(b). In particular, the crossover frequency at which a resonance occurs shows the system is on the
edge of instability. However, a more complex compensator could be used to stabilize this system.
More complex compensators may be needed when there are more complex dynamics, and possibly additional modes to
be controlled. In addition, some systems may be better controlled using a PID controller or a related lag-lead compensator,
both compared in Figure 7.51.
Systematic design processes for all of these types of controllers and compensators are well described in texts by Ogata
[53], Franklin et al. [93], and others.

7.4.5.6 Dealing with Open-Loop Unstable and Non-Minimum Phase Plants

For plants that are open-loop unstable or non-minimum phase, it is generally recommended that root locus or Nyquist
criterion be used to assess stability (e.g., recall Figure 7.46). This was illustrated earlier in this section using the two-train
system from Example 7.6. Root locus design methods can help determine gain and/or compensation methods that will mod-
ify the pole-zero configuration. The following example further describes the preliminary steps in a model-based approach.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
422 7 Modeling Feedback Control Systems
º

(a) (b)

Figure 7.50 Bode plots of (a) the collocated system and (b) non-collocated systems with PD control, Gc = K(1 + 𝜏d s), with 𝜏d = 1.

dB dB

º º

(a) (b)

Figure 7.51 Frequency domain form of dynamic compensators: (a) PID control and (b) lag-lead compensator.

A common form of an unstable system plant, G2 = 1∕(s2 − a2 ), which arises in problems like inverted pendula, unstable
attitude systems, and other specific physical problems, is used for this purpose.

Example 7.7 Magnetic levitation

It is proposed to stabilize a magnetic levitation system as described by the block diagram shown in Figure 7.52. Assume
Ko and a are constants derived during linearization of the system model.18 This system seeks to regulate the gap xg about
the equilibrium by setting the reference value, xgr = 0. Variations arise given small motion of the mass, m, induced by
disturbance forces Fd . A compensator of the form G1 = K(s + z)∕(s + p) is proposed to model how the electromechanical
control force induced by drive current, i, that is, Fem = Ko i is related to error, e.
a) Derive the closed-loop response relating error e to the disturbance, 𝑤 = Fd ∕m, for input reference, xgr = 0.
b) What are the conditions for stability of the closed-loop system in terms of the system parameters, a, Ko , K, z, and p.

+ – + +

Figure 7.52 Block diagram for magnetic levitation.

18 Problem B-4-31 describes development of a linearized plant model for this system and its stability characteristics.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
7.4 Linear Feedback Controllers and Compensators 423

Solution (a)
The error transfer function, e∕𝑤 is given by,
1
− 2
e −G 2 H (s − a2 ) −(s + p)
= = =
𝑤 1 + G1 G2 H KK o (s + z) 1 (s + p)(s − a2 ) + KKo (s + z)
2
1+ 2 2
⋅1
s + p (s − a )
e −(s + p)
= .
𝑤 s3 + ps2 + (KKo − a2 )s + (KKo z − a2 p)
Solution (b)
To determine conditions to guarantee stability, build Routh table from CE, D(s),
D(s) = s3 + ps2 + (KKo − a2 )s + (KKo z − a2 p).
The Routh table is:

s3 1 KKo − a2 0
2 2
s p KKo z − a p 0
1
p(KKo − a2 ) − (KKo z − a2 p)
s 0
p
s0 KKo z − a2 p 0

For stability, K > a2 p∕(Ko z). In addition, p(KKo − a2 ) − (KKo z − a2 p) > 0 provides the relation, z < p. This guidance from
Routh stability specifies how to place the compensator pole, p, and zero, z, which in this case suggests that a lead compen-
sator is required for stabilization. Using rlocus() can help estimate the required gain value, which has a lower bound value,
K > a2 p∕(Ko z).

As system complexity increases, there are various methods that employ root locus techniques that can be effective in
controller selection and design. These methods are extensively described in controls textbooks such as [53, 93].
For non-minimum phase systems, there is no longer a unique relationship between magnitude and phase plots which
helps guide design. In particular, the relation between an “adequate length” of −20 dB/dec slope in |GH| at the crossover
frequency and phase margin is no longer a valid guideline. The presence of a non-minimum phase factor will influence the
open-loop phase, notably decreasing phase margin. As such, it is likely a control gain that would meet desirable performance
in a minimum phase system would lead to instability in a non-minimum phase system. Thus, it is usually expected that the
presence of a non-minimum phase zero in the vicinity of a crossover frequency will require lower gain and thus reduced
speed of response, for example, in order to maintain stable behavior.

7.4.6 Ziegler–Nichols PID Controller Tuning

A PID controller can still be tuned when a model is not available. It may be necessary or desirable to tune the PID control
gains either after the controller has been installed or even when a PID controller is applied to a plant that is not necessarily
linear. In some cases, it is possible to tune the PID gains empirically. One of the earliest set of methods that remain popular
are those set forth by Ziegler and Nichols [146] in 1942. There are two basic methods chosen based on observed system
behavior. Both aim to achieve a transient step response in the controlled variable y with maximum overshoot of about 25%.
Assume the PID controller takes the form of equation 7.14 (similar relations can be found using equation 7.15) for control
of a plant as in Figure 7.53.
One method is based on a plant’s open-loop response to a unit step input (the controller is off in this test). It is assumed
the system has an S-curve (reaction) response as in Figure 7.54(a), typical of many process control applications where a

Figure 7.53 PID control of a plant, unity feedback.

+
Plant
–
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
424 7 Modeling Feedback Control Systems

= max slope
(at inflection point)

“Lag”
(a) (b)

Figure 7.54 (a) S-curve step response, showing variable definitions based on line intersections with line of maximum slope, R (at
inflection point). (b) Response with sustained oscillations in y when adjusted to a critical gain.

Table 7.6 Recommended gains based on Ziegler–Nichols tuning rules for controller of the
form in equation 7.14.

Controller Based on step response Based on critical gain and period

P K = T∕L, 𝜏i = ∞, 𝜏d = 0 K = 0.5Kcr , 𝜏i = ∞, 𝜏d = 0
PI K = 0.9T∕L, 𝜏i = L∕0.3, 𝜏d = 0 K = 0.45Kcr , 𝜏i = Pcr ∕1.2, 𝜏d = 0
PID K = 1.2T∕L, 𝜏i = 2L, 𝜏d = 0.5L K = 0.6Kcr , 𝜏i = Pcr ∕2, 𝜏d = Pcr ∕8

Source: Adapted from Ziegler and Nichols [146].

plant might be modeled by the form, G2 (s) ≈ Ae−Ls ∕(Ts + 1). The exponential term is used to capture transport lag effects.
The response curve can be obtained either from an experiment or from simulation study of plant. The S-curve is used to
estimate the delay (or lag), L, and the effective time constant, 𝜏t , as shown in Figure 7.54(a). For the form of the controller
in equation 7.14, Ziegler and Nichols determined these estimated parameters can be used to estimate controller gains that
give the desired “optimum” response. Table 7.6 summarizes the formulas for this case in the first column for each type of
controller.
A second method considers the closed-loop controller with only proportional control, K. The value of K is incrementally
increased (in a test or in simulation) up until the point that the system begins to oscillate. The value of K at this point is
set as a critical gain, Kcr . Presumably, this would be a safe state of operation if the test was being conducted experimentally.
The period of the sustained oscillations at the critical gain value is measured and set equal to Pcr (critical period), as shown
in Figure 7.54(b). Given these two values, Ziegler and Nichols recommended the controller gains summarized in the second
column of Table 7.6.
The Ziegler–Nichols tuning rules provide a way for controllers to be applied with limited knowledge of control theory
and/or the system plant. The recommended gains provide a good starting point and are usually refined with additional
tests as needed. If a PID controller is to be applied to a complex system in simulation, “virtual” experiments can provide the
results needed for applying the formulas in Table 7.6. It should emphasized that the rules apply for systems that are stable
and of Type 0 when modeled as a linear transfer function. The methods may work for but are not guaranteed for Type 1
and higher.

7.5 State Variable Control Methods

Classical control system design evolved with the consideration of controlling a single system variable and the use of
input–output relations. The single-input-single-output (SISO) methods make use of transfer functions as has been
discussed throughout this chapter. The emphasis in this book, however, has been on deriving physical system models
primarily in state space form, not only because of the systematic way in which the models can be derived but as has been
shown the ability to solve for response using digital simulation methods. State-space models are also used in many control
analysis and design methods that have become prevalent since the 1950s. These methods permit flexibility by feeding back
some or all of the system state variables, in contrast to methods that rely on a model and feedback in terms of a single
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
7.5 State Variable Control Methods 425

output variable. Different control strategies can then be adopted that would be difficult or not possible using classical
(SISO) methods. For example, root locus methods allow you to understand system behavior using the poles and zeros and
to study the effect of changing a single gain to effect change in system response and relative stability. State variable control
methods, also referred to as modern control theory [94, 147], include design techniques such as pole placement, that can
arbitrary set the poles of a system. Pole placement requires a system to be state controllable, and all states must be available
as feedback or by estimation. The latter would require the system to be observable. State variable methods are also essential
when dealing with control of multiple-input-multiple-output (MIMO) systems. This section provides a brief introduction
to these concepts, restricting the discussion to linear or linearized systems.

7.5.1 Controlling a System with Full-State Variable Feedback

Before reviewing concepts from the area of state space control, consider a system modeled in transfer function form as
y 20
G(s) = = 2 . (7.39)
e s (s + 1)
While this system is open-loop stable, a root locus shows that closing the loop with a simple proportional controller will
not result in a stable system for any gain value K. Alternative controllers can be used, but in the following a state variable
feedback system is developed.
Form a third-order system by first decomposing the transfer function as,
[ ][ ] [ ]
y y z 1
G(s) = = = [1] ⋅ 2 (7.40)
e z e s (s + 1)
[ ]T
these two transfer functions can be converted into a state-space form with state vector x = x1 x2 x3 . The open-loop
system is then,
⎡ẋ 1 ⎤ ⎡0 1 0 ⎤ ⎡x1 ⎤ ⎡0⎤
⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥
ẋ = ⎢ẋ 2 ⎥ = ⎢0 0 1 ⎥ ⎢x2 ⎥ + ⎢0⎥ e. (7.41)
⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥
⎣ẋ 3 ⎦ ⎣0 0 −1⎦ ⎣x3 ⎦ ⎣1⎦
[ ]
y = x1 = 20 0 0 x.
A full-state feedback control can be constructed using an error defined by,
[ ]
e = r − Kx = r − k1 k2 k3 x.
Substitute now into equation 7.41 to form a closed-loop control system with state variable feedback:
⎡ẋ 1 ⎤ ⎡ 0 1 0 ⎤ ⎡x1 ⎤ ⎡0⎤
⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥
ẋ = ⎢ẋ 2 ⎥ = ⎢ 0 0 1 ⎥ ⎢x2 ⎥ + ⎢0⎥ r. (7.42)
⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥
⎣ẋ 3 ⎦ ⎣−k1 −k2 −(1 + k3 )⎦ ⎣x3 ⎦ ⎣1⎦
[ ]
y = x1 = 20 0 0 x.
A closed-loop transfer function can now be derived using the relation,
y
= G(s) = C[sI − A]−1 B + D.
r
−1
⎡ 0 −1 0 ⎤ ⎡0⎤
y [ ] ⎢ ⎥ ⎢ ⎥
r
= G(s) = 20 0 0 ⎢ 0 s −1 ⎥ ⎢0⎥ + [0] .
⎢ ⎥ ⎢ ⎥
⎣k1 k2 s + (1 + k3 )⎦ ⎣1⎦
⎡ 0 −1 0 ⎤ ⎡0⎤
[ ]⎢ ⎥⎢ ⎥
20 0 0 ⎢ 0 s −1 ⎥ ⎢0⎥
⎢ ⎥⎢ ⎥
⎣k1 k2 s + 1 + k3 ⎦ ⎣1⎦
= + [0] .
s3 + (1 + k3 )s2 + k2 s + k1
y 20
= 3 .
r s + (1 + k3 )s2 + k2 s + k1
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
426 7 Modeling Feedback Control Systems

+
–
–
–

Figure 7.55 Block diagram of state-variable feedback system.

1
y(t)

0.5

0
0 1 2 3 4 5 6 7
Time, sec

Figure 7.56 Closed-loop response of system in Figure 7.55 to step change in reference, r.

Now it is necessary to select the three gains (k1 , k2 , k3 ) to give desired performance. Consider the case where it is desired
to have zero steady-state error in y for a step input in r. Also, two of the poles are to be placed at −1 ± j to give an effective
damping of 𝜁 = 0.707.
We assume the system can be stabilized and apply the Final Value Theorem for y,
y 1 20
y(∞) = lim y(t) = lim s ⋅ = = 1 (desired),
t→∞ s→0 rs k1
which implies k1 = 20.
To find the other two gain values, equate the derived CE for the closed-loop system to the desired set of poles:
s3 + (1 + k3 )s2 + k2 s + k1 = (s + p1 )(s + p2 )(s + a),
where p1,2 are the two desired poles and p3 is to be found. This equation provides three equations in three unknowns:
1 + k3 = a + 2
k2 = 2 + 2a

20 = 2a,
leading to: a = 10, so p3 = −10, and k2 = 22, k3 = 11. To emphasize the feedback of each state, we can draw a block diagram
for the closed-loop control system as in Figure 7.55. With the gains determined, the step response is as shown in Figure 7.56.
This example shows a somewhat ad hoc approach to applying state variable feedback. Sections 7.5.2–7.5.11 review some
of the basic principles and methods used in state-space control are reviewed.

7.5.2 Controllability and Observability

As systems become more complex (higher order), it can become difficult to use “ad hoc” approaches for design. It is also
helpful to know whether the system under study has certain desirable properties. For applying methods commonly used
in state-space design, it is important to test for controllability and observability. Several definitions are introduced in the
following, along with several example applications.

7.5.2.1 Definition of Controllability

A system is said to be controllable at time t0 if and only if it is possible to transfer the system by an unconstrained control
vector u from any initial state x(t0 ) to another state x(t) in a finite time period T = t − t0 . A system for which this is true is
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
7.5 State Variable Control Methods 427

said to be completely state controllable. Controllability is a property of a system that depends on the coupling between the
input u and the state x implied by the A and B matrices.

7.5.2.2 Definition of Observability

An unforced system is said to be observable at time t0 if and only if it is possible to determine any (arbitrary initial) state
x(t0 ), by using only a finite record of the output over (𝜏) for t ≤ 𝜏 ≤ T. Observability is a property of a system that depends
on the coupling between the state x and the output y implied by the A and C matrices.
Observability is a key requirement for estimating the states of a system, notably in feedback control applications where
the number of measured states is less than the order of the system, m < n. When a system is observable, y (the measure-
ments) contains sufficient information about the system states so that state feedback can be implemented, for example, by
construction of an observer.
The conditions for controllability and observability are functions of state and cannot be determined from a transfer
function model of a system.

7.5.2.3 Condition for Complete State Controllability

R.E. Kalman showed in the 1960s that complete state controllability requires a controllability gramian (an integral equation)
be nonsingular for some time t and interval T > t (see, e.g., [147]).
Complete state controllability requires that the n × n controllability matrix
[ ]
Pc = B AB A2 B … An−1 B (7.43)

has rank n.
Output controllability is defined for transferring the output with the input, so the controllability matrix is
[ ]
Pc = CB CAB CA2 B … CAn−1 B ,

and must have rank m (since C is order m × n).

7.5.2.4 Condition for Complete State Observability

A system that is completely observable has an observability matrix
⎡ C ⎤
⎢ ⎥
⎢ −− ⎥
⎢ ⎥
⎢ CA ⎥
⎢ −− ⎥
Po = ⎢ ⎥ (7.44)
⎢ CA2 ⎥
⎢ ⎥
⎢ ⋮ ⎥
⎢ ⎥
⎢ −= ⎥
⎢ n−1 ⎥
⎣CA ⎦
with rank n.
The foregoing discussion provides guidance and relations for checking controllability and observability conditions for
state-space systems. If a system is SISO, controllability and observability conditions are satisfied when the determinant of
the matrix is not zero. Further, if a system is not completely state controllable, it may consist of a part that is controllable
and uncontrollable. If the uncontrollable part is stable the entire system is said to be stabilizable. Every controllable system
is stabilizable but not every stabilizable system is controllable. Similarly, a system that is not completely observable can
be divided into a subsystem that is observable and another that is not. A system is said to be detectable if the part that is
unobservable is stable. All observable systems are detectable. More thorough discussions on these topics can be found in
Ogata [53], Franklin, et al. [93], or Dorf and Bishop [61].

Example 7.8 Controllability and observability of a mass–damper system

The system in Figure 7.57 has an applied control force Fc . (a) Determine if the system is controllable and (b) determine
which state(s) should be measured to form an observable system.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
428 7 Modeling Feedback Control Systems

Solution
a state equations for this system are,
The
m𝑣̇ = +Fc − b𝑣,
ẋ = 𝑣,

so,
[ ] [ ]
−b∕m 0 1∕m Figure 7.57
A= and, B = . Control force on a
1 0 0
mass sliding with
a) To check controllability, let m = b = 1 so the controllability matrix is, damping.
[ ]
[ ] 1 −1
Pc = B AB = ,
0 1
which has rank 2, so the system is controllable.
b) To determine which state(s) should be measured to ensure observability, let,
[ ]
y = c1 c2 x.

Now form the observability matrix,

[ ] [ ]
C c1 c2
Po = = .
CA c2 − c1 0
If we measure 𝑣 and not x the system is not observable, but if we measure x and not 𝑣 the system is observable. So
measuring x alone will give an observable system.

Example 7.9 Controllability and observability of a second-order system

Consider the system with the matrices given below:
[ ] [ ]
1 1 0 [ ]
A= B= C= 1 0 .
−2 −1 1
Is this system controllable and observable?
Solution
Create the controllability matrix,
[ ]
Pc = B AB
[ [ ] [ ]]
0 1 1 0
Pc =
1 −2 −1 1
[ ]
0 1
Pc = .
1 −1
This matrix has rank 2, so the system is completely state controllable. You can show that it is also completely output
controllable.
The observability matrix is
⎡C⎤
⎢ ⎥
Po = ⎢ −− ⎥
⎢ ⎥
⎣CA⎦
[ ]
⎡ 1 0 ⎤
⎢ [ ]⎥
Po = ⎢[ ] 1 1 ⎥
⎢1 0 ⎥
⎣ −2 −1 ⎦
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
7.5 State Variable Control Methods 429

[ ]
1 0
Po = .
1 1
This matrix has a rank of 2 so the system is completely state observable.

Example 7.10 Finding conditions on parameters to satisfy controllability

Consider the system with the matrices given below.
[ ] [ ]
−2 0 1 [ ]
A= B= C= 1 0 .
d −3 0
Determine the condition(s) for controllability.
Solution
Create the controllability matrix,
[ ]
Pc = B AB
[ [ ] [ ]]
1 −2 0 1
Pc =
0 d −3 0
[ ]
1 −2
Pc = .
0 d
For this SISO system, |Pc | = d, so as long as d ≠ 0 the system is controllable.

Example 7.11 Controllability and observability example

Consider the system with the matrices given below.
[ ] [ ]
2 0 1 [ ]
A= B= C= 1 1 .
−1 1 −1
Determine whether this system is controllable and observable.
Solution
Create the controllability matrix,
[ ]
Pc = B AB
[ [ ] [ ]]
1 2 0 1
Pc =
−1 −1 1 −1
[ ]
1 2
Pc = .
−1 −2
This matrix has rank 1 < n so the system is not controllable. It can also be shown that this system is not observable.

Example 7.12 Controllability of a third-order system

Consider the SISO system with,
⎡ 0 1 0 ⎤ ⎡0⎤
⎢ ⎥ ⎢ ⎥ [ ]
A=⎢ 0 0 1 ⎥ B = ⎢0⎥ C= 1 0 0 .
⎢ ⎥ ⎢ ⎥
⎣−a0 −a1 −a2 ⎦ ⎣1⎦
Thus,
[ ]
Pc = B AB A2 B
2
⎡0 ⎡ 0 1 0 ⎤ ⎡0⎤ ⎡ 0 1 0 ⎤ ⎡0⎤⎤
⎢ ⎢ ⎥⎢ ⎥ ⎢ ⎥ ⎢ ⎥⎥
Pc = ⎢0 ⎢ 0 0 1 ⎥ ⎢0⎥ ⎢ 0 0 1 ⎥ ⎢0⎥⎥
⎢ ⎢ ⎥⎢ ⎥ ⎢ ⎥ ⎢ ⎥⎥
⎣1 ⎣−a0 −a1 −a2 ⎦ ⎣1⎦ ⎣−a0 −a1 −a2 ⎦ ⎣1⎦⎦
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
430 7 Modeling Feedback Control Systems

⎡0 0 1 ⎤
⎢ ⎥
Pc = ⎢0 1 −a2 ⎥ .
⎢ 2 ⎥
⎣1 −a2 a2 − a1 ⎦
For this (SISO) system, |Pc | ≠ 0 so the system is controllable.

Example 7.13 Controllability of a sewer system with controlled pump flow

Consider a two tank sewer system with an interconnecting pipe as shown in Figure 7.58. A controlled pump is proposed to
draw fluid from tank 2. Assess controllability of this system.

Figure 7.58 Storm sewer system with controlled pumping.

Pump
Long pipe

Solution
A bond graph for this system is shown in Figure 7.59. Consider the simplified case where Ip = C1 = C2 = 1, the state
equations become
⎡V̇ 1 ⎤ ⎡ 0 −1 0⎤ ⎡V 1 ⎤ ⎡ 0⎤ ⎡1⎤
⎢̇ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥ ⎢ ⎥
ẋ = ⎢Qp ⎥ = ⎢+1 0 −1⎥ ⎢Qp ⎥ + ⎢ 0⎥ Qc + ⎢0⎥ Qd . (7.45)
⎢̇ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎣V 2 ⎦ ⎣ 0 +1 0⎦ ⎣V 2 ⎦ ⎣−1⎦ ⎣0⎦
[ ]
y = V 1 = 1 0 0 x.
We identify the matrices as
⎡0 −1 0⎤ ⎡ 0⎤
⎢ ⎥ ⎢ ⎥ [ ]
A = ⎢1 0 −1⎥ B = ⎢ 0⎥ C= 1 0 0 .
⎢ ⎥ ⎢ ⎥
⎣0 1 0⎦ ⎣−1⎦
Thus,
[ ]
Pc = B AB A2 B
2
⎡ 0 ⎡0 −1 0⎤ ⎡ 0⎤ ⎡0 −1 0⎤ ⎡ 0⎤⎤
⎢ ⎢ ⎥⎢ ⎥ ⎢ ⎥ ⎢ ⎥⎥
Pc = ⎢ 0 ⎢1 0 −1⎥ ⎢ 0⎥ ⎢1 0 −1⎥ ⎢ 0⎥⎥
⎢ ⎢ ⎥⎢ ⎥ ⎢ ⎥ ⎢ ⎥⎥
⎣−1 ⎣0 1 0 ⎦ ⎣−1⎦ ⎣0 1 0⎦ ⎣−1⎦⎦
⎡ 0 0 −1⎤
⎢ ⎥
Pc = ⎢ 0 1 0⎥ ,
⎢ ⎥
⎣−1 0 1⎦
which has rank n = 3, thus the system is controllable.

Figure 7.59 Sewer system bond graph.

 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
7.5 State Variable Control Methods 431

(a) (b)

Figure 7.60 (a) Mass–spring-mass with internal control force, Fi ; (b) bond graph.

7.5.3 How Do Uncontrollable or Unobservable Systems Arise?

There are a few ways to relate the assessment of controllability and observability to physical system modeling.

7.5.3.1 Redundant State Variables

If the A matrix is over-specified, it should be clear that some of the state equations will not be linearly independent and
thus controllability and observability criteria will not be satisfied. This is an advantage of the bond graph method which
includes a systematic approach for finding the minimum states to model a physical system using causality. To illustrate
how redundant states create an issue with a state-space system,
ẋ = Ax + Bu,
add states z = Fx. Then,
ż = Fẋ = F [Ax + Bu] = FAx + FBu.
Form a new “metastate,” x̂ , and after transforming it is found that ż = 0, which represents a set of uncontrollable states
[147]. Thus, writing down differential equations without a methodology for ensuring a minimal set can lead to a model
with uncontrollable states.

7.5.3.2 Physically Uncontrollable States

A system can be uncontrollable when only “internal efforts” are available. For example, in the system of Figure 7.60(a), an
actuator applies a force Fi between the two masses, m1 and m2 . The force can control the distance between the two masses,
which affects the spring, but it cannot control the actual velocity of position of the masses themselves.
From the bond graph in Figure 7.60(b), the state equations are,
⎡𝑣̇ 1 ⎤ ⎡0 −k∕m 0 ⎤ ⎡𝑣1 ⎤ ⎡−1∕m⎤
⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥
⎢ẋ k ⎥ = ⎢1 0 −1⎥ ⎢xk ⎥ + ⎢ 0 ⎥ Fi .
⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥
⎣𝑣̇ 2 ⎦ ⎣0 +k∕m 0 ⎦ ⎣𝑣2 ⎦ ⎣+1∕m⎦
Setting k = m = 1 gives,
⎡0 −1 0⎤ ⎡−1⎤
⎢ ⎥ ⎢ ⎥
A = ⎢1 0 −1⎥ and, B = ⎢ 0⎥ ,
⎢ ⎥ ⎢ ⎥
⎣0 +1 0⎦ ⎣+1⎦
from which it can be shown that the controllability matrix is,
⎡−1 0 2⎤
⎢ ⎥
Pc = ⎢ 0 −2 0⎥ ,
⎢ ⎥
⎣ 1 0 −2⎦
which has rank 2, showing the system is not controllable.

7.5.3.3 Too Much Symmetry

For the RC-bridge shown in Figure 7.61(a), if R1 C1 = R2 C2 (balanced bridge) then |Pc | = 0 and the system is not control-
lable. This means Vs cannot be used to achieve a desired state x. It can also be shown that this system is not observable.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
432 7 Modeling Feedback Control Systems

(a) (b)

Figure 7.61 (a) Resistive–capacitive impedance bridge; (b) bond graph.

Alternatively, an unbalanced bridge is both controllable and observable. To illustrate the symmetry in this system, the bond
graph shown in Figure 7.61(b) is used to derive the model with the following results:
State equations:
q̇ 1 = i1 − iL .
q̇ 2 = iL + i2 .
[ ]T
x = q 1 q2 .
( )
[ ] ⎡− 1 + 1 1 1 ⎤[ ] ⎡ 1 ⎤
q̇ 1 ⎢ R R C1
+
R C ⎥ q ⎢R ⎥
=⎢ ) ⎥
1
1 L
( L 2
+ ⎢ 1 ⎥ Vs (t).
q̇ 2 ⎢ 1 1 1 1 ⎥ q2
1
⎢ ⎥
⎢ + − + ⎥ ⎣
⎣ RL C 1 R2 R L C 2⎦ R2⎦

Say the measurement is y = VL = V1 − V2 ,

[ ]
y = 1∕C1 1∕C2 x.
With the balanced bridge case, the system will be found to be uncontrollable and unobservable.

7.5.4 Full-State Feedback

If a system is found to be state controllable as discussed in Section 7.5.2, then full-state feedback can be applied. This section
reviews some of the methods that can be applied given a linear state-space system.

7.5.4.1 Basic Control Formulation

Full-state feedback of a system,
ẋ = Ax + Bu,
adopts the control law,
u = −Kx, (7.46)
which in this case is a scalar and where it is assumed that all the states x are available and measurable. This control archi-
tecture is illustrated by the block diagram in Figure 7.62. Note that there is no reference input, r, as this case represents a
regulator system, r = 0. As such, the control maintains the system at an equilibrium subject to nonzero initial conditions
and/or disturbances, ud . The latter would be introduced by an additional input vector, Bd , not shown in this figure.
For this single-input case, state-space control design seeks to determine a 1 × n gain matrix, K. For cases where some
of the states are not measured, an observer is required, as discussed in Section 7.5.7.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
7.5 State Variable Control Methods 433

Figure 7.62 Full-state feedback block diagram, r = 0 (no reference input). System model

Control law

7.5.4.2 Finding Control Gain Matrix: n ≤ 3 Order Systems

For systems of order n ≤ 3, it is relatively easy to find a gain matrix analytically. First, the controlled system will have a CE
given by,
|sI − (A − BK)| = 0.
One way to set up a solution is to compare a specific CE in terms of k values to a standard form. For example, for n = 3,
D(s) = (s2 + 2𝜁𝜔n s + 𝜔2n )(s + 𝜁𝜔n ) = 0. (7.47)
As an example, taking the A and B matrices from the sewer system in Example 7.13, the characteristic equation becomes,
s3 − k3 s2 + (2 + k2 )s − (k1 + k3 ) = 0.
Comparing to the standard form and matching terms gives three equations in the three unknown gains:
−k1 − k3 = 𝜁𝜔3n
2 + k2 = 2𝜁 2 𝜔2n + 𝜔2n
−k3 = 3𝜁𝜔n ,
which can be solved given specified 𝜁 and 𝜔n that correspond to desired time-domain specifications.
For example, for 𝜁 = 0.8 and 𝜔n = 1, the desired roots are,
[ ]
p = −0.8 + j0.6 − 0.8 − j0.6 − 0.8000 .
The relations above give K = [1.60 0.28 − 2.40]. It can be shown that the same control gain matrix is found using built-in
functions called acker() or place(), which are available in control system design software packages.

7.5.4.3 Ackermann’s Formula for Pole-Placement

For a system with state feedback u = −Kx having the form,
ẋ = (A − BK)x = Ac x,
the resulting CE is given by,
|sI − Ac | = sn + a1 sn−1 + · · · + an−1 s + an = 0,
The Cayley–Hamilton theorem19 states that a matrix A satisfies its own CE. So, given the scalar result above, we can now
write [53],
𝜙(A) = An + a1 An−1 + · · · + an−1 A + an I = 𝟎. (7.48)
This result is used in the derivation of the state feedback gain matrix known as Ackermann’s formula,

c 𝜙(A),
K = [0 0 · · · 0 1] P−1 (7.49)
where Pc is the controllability matrix.

Example 7.14 Ackerman’s formula example

The system with matrices
⎡ 0 1 0⎤ ⎡0⎤
⎢ ⎥ ⎢ ⎥
A = ⎢ 0 0 1⎥ and, B = ⎢0⎥
⎢ ⎥ ⎢ ⎥
⎣−1 −5 −6⎦ ⎣1⎦
is found to be controllable. Find the control gain matrix to place poles at s = −2 ± j4 and s = −10.

19 See Ogata [53], p. 729.

 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
434 7 Modeling Feedback Control Systems

Solution
Ackermann’s formula is available as acker() and can be implemented as illustrated by the following MATLAB code:

1 % original system
2 A = [0,1,0;0,0,1;-1,-5,-6];
3 B=[0;0;1];
4 % desired poles
5 sd = [-2+1i*4,-2-1i*4,-10];
6 % create the desired denominator
7 sys=zpk([],sd,1);
8 [numd,dend] = tfdata(sys,’v’);
9 % apply Ackermann’s formula
10 Pc = ctrb(A,B);
11 phiA = A^3 + dend(2)*A^2 + dend(3)*A + dend(4)*eye(3);
12 K = [0,0,1]*inv(Pc)*phiA

For this case, the result is K = [199 55 8], as also found using acker().
An initial condition response for the designed regulator system is shown below. Arbitrary initial conditions are set at
x0 = [1 0 0]T , and the control returns all the states to zeros.

1 sys_c = ss((A-B*K),[0;0;0],[1,0,0],0);
2 [y,t,x] = initial(sys_c,[1;0;0]);
3 figure(1)
4 plot(t,x),legend(’x_1’,’x_2’,’x_3’)

5 x1
x2
x3
0

–5

–10
0 0.5 1 1.5 2 2.5 3
Time, sec

7.5.5 State Feedback with a Reference Input

The full-state feedback control presented up to this point assumes the reference input is zero, r = 0, as the control law
is u = −Kx. The examples have shown how this control can take a system from nonzero initial conditions to zero states,
x = 𝟎. In many practical cases, of course, it may be desired to have states or the output y converge to certain values. This can
be achieved by introducing a reference input that is either constant or time varying. In the end, a formulation is required
that will specify a control law that drives the states and output to the desired values. The following is one approach for
accomplishing this task.
Consider again the state-space system to be controlled,

ẋ = Ax + Bu
y = Cx + Du
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
7.5 State Variable Control Methods 435

(note: D is shown as scalar here since y and u are scalar) We want the value of yss to be equal to r, thus,
yss = r = Cxss + Duss .
Divide by r to give one equation in the form,
C(xss ∕r) + D(uss ∕r) = 1,
and for equilibrium,
ẋ = 𝟎 = Axss + Buss ,
so we can form the set of equations,
[ ][ ] [ ]
A B xss ∕r 𝟎
= .
C D uss ∕r 1
These equations can be solved to provide scaling constants for the reference input. First,
[ ] [ ] [ ]−1 [ ]
Nx xss ∕r A B 𝟎
= = .
Nr uss ∕r C D 1
Now, consider that the control law is extended as follows:
u = uss − K(x − xss ),
with the steady-state values now dictated using the relations derived and r as follows:
xss = Nx ⋅ r.
uss = Nr ⋅ r.
The updated control law with reference input is now,
u = Nu r − K(x − Nx r).
The state-space feedback with reference input equations can now be shown to take the form (with D = 0),
[ ]
ẋ = (A − BK)x + B Nu + KNx r. (7.50)
y = Cx.
This system can be illustrated in block diagram form as shown in Figure 7.63.
To illustrate application of this methods, consider applying this full-state feedback control to the three-state sewer system
from Example 7.13. Now a reference input is applied and set to a desired level, r = ro . Here, this is the desired reference
value for V 1 (tank 1). Assume the system has reached a steady state xo , which can be solved from,
[ ]
ẋ = 𝟎 = (A − BK)xo + B Nu + KNx ro ,
that is,
[ ]
xo = −(A − BK)−1 B Nu + KNx ro ,
where Nx and Nr can be solved using the relations given above and for this case using the ABCD matrices from the
two-tank/pipe sewer system.

+ System model
+

–
+

Figure 7.63 Full-state feedback with reference input.

 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
436 7 Modeling Feedback Control Systems

0.08
r
V1
0.06
V2
0.04

0.02
0 2 4 6 8 10
0.01
Qp

–0.01
0 2 4 6 8 10

4.2
Control inputs

uc
–ur
4

3.8
0 2 4 6 8 10
Time, sec

Figure 7.64 Results from simulated three-state sewer system with periodic changes in the reference level for V 1 .

The sewer system was defined with “normalized” parameter values, so numerical values may not correspond to practical
values. In this example, let ro = 0.05 which corresponds to the desired volume in tank 1. To test the control feedback, let a
simulation run with the initial ro , then test the controlled response with a step change either above or below ro .
The MATLAB code that follows uses pole-placement then sets up the reference input scaling constants (Nu and Ns ).
The results in Figure 7.64 show the response of step changes about ro . Some iteration was done with the settings of 𝜁 and
𝜔n . It can be challenging to get shorter rise times because the amplitude of changes in r causes large fluctuations in the
volume of tank 2 to “help” change the volume in tank 2. A step change in the reference input is an extreme test for most
systems, and likely ramp or sinusoidal changes would be controlled more effectively.

1 % FSFB with reference input for three-state (hydraulic) system

2 % no disturbance input
3 clear all
4 A = [0,-1,0;1,0,-1;0,1,0];
5 B = [0;0;-1];
6 C = [1,0,0];
7 D = 0;
8 % for adding in disturbance into tank 1
9 Bd = [1;0;0];
10 % uncontrolled system
11 sysu = ss(A,B,C,D);
12 % set poles
13 syms s
14 % specify desired zeta and wn (iterate as needed)
15 zeta = 0.8; wn = 0.25;
16 Ds = expand((s*s + 2*zeta*wn*s + wn*wn)*(s+zeta*wn));
17 ps = coeffs(Ds,s);
18 pn = double(ps);
19 p = roots(pn);
20 K = place(A,B,p);
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
7.5 State Variable Control Methods 437

21 sysc = ss((A-B*K),B,C,D);
22 % set initial conditions based on reference input
23 % First, solve for scaling constants
24 N = inv([A,B;C,0])*[0;0;0;1];
25 Nx = N(1:3); Nu = N(4);
26 % set initial reference value
27 ro = 0.05;
28 % solve for initial conditions for given reference value
29 % (from equilibrium of the closed-loop system equations)
30 xo = -inv(A-B*K)*B*(Nu+K*Nx)*ro;
31 % note, 1 gallon = 0.003785 m^3
32 % time array
33 t = linspace(0,10,1000);
34 % compute the desired reference values over time
35 % note: these need to be realistic and could cause
36 % system states or control input to
37 % take values that are not realistic
38 T = 4;
39 tstart = 1;
40 for i = 1:length(t)
41 if t(i)>tstart
42 r(i) = xo(1)*(1+0.05*square(2*pi*(t(i)-tstart)/T));
43 else
44 r(i) = xo(1);
45 end
46 end
47 % compute the reference for lsim() simulation
48 % NOTE: This ur will multiplied by B defined above
49 ur = (Nu + K*Nx)*r;
50 uro = (Nu + K*Nx)*ro;
51 uco = -K*xo;
52 % simulate the controlled system with reference input
53 % and specified initial conditions xo
54 [yc,tc,xc] = lsim(sysc,ur,t,xo);
55 % compute the control inputs
56 uc = -K*xc’;
57 ucm = max(uc);
58 sprintf(’ucm = %4.2f’,ucm)
59 u = uc + ur;
60 %
61 figure(1)
62 subplot(3,1,1), plot(tc,r,tc,xc(:,1),tc,xc(:,3))
63 legend(’r’,’V_1’,’V_2’)
64 %ylim([0.04,0.06])
65 subplot(3,1,2), plot(tc,xc(:,2))
66 legend(’Q_p’)
67 subplot(3,1,3), plot(t,uc,t,-ur), legend(’u_c’,’-u_r’)
68 %ylim([5,6.5])
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
438 7 Modeling Feedback Control Systems

7.5.6 Internal Model Approach for Tracking Reference Input

An alternative way to control with reference tracking is referred to as an “internal model,” or integral, approach [61, 148].
This method can provide better reference tracking in some cases than the method described in Section 7.5.5. This approach
seeks to track and control error between the output, y, and a reference input, r. In a way, the basic idea as presented by Dorf
and Bishop [61] and also by Astrom and Murray [148] is similar to what is classically referred to as a servo design approach
by Ogata [53].
Consider again control of the state-space system,
ẋ = Ax + Bu,
y = Cx,
for which we want to have a reference input, r. Define the tracking error, e = y − r (note the difference from classical
feedback control diagram where e = r − y). The derivative is, ė = ẏ − r,
̇ but if r is constant so,
ė = ẏ = Cx.
̇
We now define two new variables:
z = ẋ ̇
and ua = u.
This is done so we can create the augmented system,
[ ] [ ][ ] [ ]
ė 0 C e 0
= + ua , (7.51)
ż 0 A z B
where we note the order of z as n × 1. This system should be checked for controllability in the usual manner. If controllable
then a state feedback control input ua can be found,
[ ]
e
ua = −K = −K1 e − K2 z, (7.52)
z
such that the system is stable.
Since the system is stable, the tracking error will be stable and can be made to go to zero in the steady state. Note also
that,

u= ua dt = − K1 edt − K2 zdt,
∫ ∫ ∫
so the input to the controlled system becomes,
t
u = −K1 e(𝜏)d𝜏 − K2 x. (7.53)
∫0
It is now clear that the integral action [148] in the first term is as found in the internal model approach described by Dorf
an Bishop [61]. The overall control system configuration is illustrated in Figure 7.65.
Now substitute ua from equation 7.52 into the augmented state equations of equation 7.51,
[ ] [ ][ ]
ė 0 C e
= . (7.54)
ż −BK1 A − BK2 z
Note that the influence of the reference input, r, is embedded in the e term, e = y − r. So if you want to simulate, say, a
reference with r(t), then this needs to be reformulated in these equations, otherwise it is assumed that r is a constant.

+ + System model
– –

Figure 7.65 State feedback using internal model (or integral action) with reference input.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
7.5 State Variable Control Methods 439

1
e = y–r

–1
0 5 10 15 20 25 30 35 40 45 50
Control input

2 uc

0
–2
0 5 10 15 20 25 30 35 40 45 50
1.5
r
1 V1
V2
0.5
0
0 5 10 15 20 25 30 35 40 45 50
0.5
Qp

–0.5
0 5 10 15 20 25 30 35 40 45 50
Time, sec

Figure 7.66 Results from simulated three-state sewer system with periodic changes in the reference level for V 1 using internal
model reference tracking control.

7.5.6.1 Simulating the Internal Model Approach

One way to implement a simulation of the internal model approach for state feedback with reference input is to set up the
entire model as:

ė = z,
ż = −BK1 ⋅ e + A − BK2 z,
ẋ = Ax + Bu + Bd ud ,
u̇ = −K1 e − K2 z,
y = Cx,
e = y − r,

where the full (augmented) (order 2(n + 1)) state vector is now X = [e, z, x, u].
When put in this form, this set of equations is readily integrated numerically with a Runge–Kutta type solver. Note also
that a disturbance term has been added to allow studying the effect of an uncontrolled input. This integral action, or internal
model form, can be applied to the three-state hydraulic (“sewer”) system presented in earlier examples. The results are
shown in Figure 7.66, with the details of implementation illustrated by the following MATLAB code. For simulation, the
system and controller equations are first written into an m-file function and simulation file as illustrated by the examples
below:

1 function [xdot,yo] = three_state_IM(t,x)

2 % Internal model formulation for three-state system
3 global K A B C ro
4 e = x(1);
5 z = [x(2);x(3);x(4)];
6 xp = [x(5);x(6);x(7)];
7 u = x(8);
8 y = C*xp;
9 if (t>1 && t<10)
10 r = ro + 0.25;
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
440 7 Modeling Feedback Control Systems

11 elseif t≥=10 && t<30

12 r = ro - 0.5;
13 elseif t≥=30 && t<40
14 r = ro + 0.25;
15 else
16 r = ro;
17 end
18 e = y - r;
19 edot = C*z;
20 zdot = -B(:,1)*K(1)*e + (A-B(:,1)*K(2:4))*z;
21 xpdot = A*xp + B(:,1)*u;
22 udot = -K(1)*e - K(2:4)*z;
23 xdot = [edot;zdot;xpdot;udot];
24 % Outputs
25 yo(1) = e; % error
26 yo(2) = r; % reference input

Simulation file:

1 % Simulation of the generic 3-state system with internal model

2 % form for reference input tracking
3
4 clear all
5 % the following parameters are used in the model function file
6 % for simulation
7 global K A B C D ro
8
9 % Construct the open-loop system
10 A = [0,-1,0;1,0,-1;0,1,0];
11 B = [0,1;0,0;-1,0]; % inputs are uc and ud (disturbance)
12 C = [1,0,0]; % single-output state x1
13 D = [0];
14
15 % This design uses an internal model (IM)
16 % for IM form (see eq. 11.81 in Dorf & Bishop)
17 Aim = [0,C;zeros(3,1),A]; Bim = [0;B(:,1)];
18 % NOTE that the IM only uses the 1st column of B for the control input
19 % The second input is a disturbance
20 % form internal model controllability matrix
21 Pimc = ctrb(Aim,Bim);
22 % confirm system is completely controllable; det(Pc) not zero
23 det(Pimc)
24
25 % Select poles for IM
26 p = [-1+j*0.8,-1-j*0.8,-8,-6];
27 % select gains
28 K = place(Aim,Bim,p)
29
30 % Set up for simulation using RK4
31 % define the initial states of the plant
32 % make sure they are compatible
33 xo = [1;0;1]; % plant initial states
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
7.5 State Variable Control Methods 441

34 yo = C*xo; % initial output

35 ro = yo; % let the reference initialize equal to yo
36 eo = yo - ro; % initial error state
37 uo = 0; % initial control input
38
39 % initialize z; z = xdot, so zo comes from xdoto
40 zo = A*xo+B(:,1)*uo;
41 % construct initial condition vector for IM system; include plant
42 xco = [eo;zo;xo;uo];
43 % 2(n+1) = 8 states
44 % Simulation parameters
45 dt=0.001; % time interval for fixed-step simulation
46 t0 = 0.0; % start time
47 tf = 50; % final time
48 N = floor((tf-t0)/dt); % number of steps
49
50 % call RK4
51 [tc,xc] = rk4fixed(@three_state_IM,[t0 tf],xco,N);
52 % Compute defined outputs
53 for i = 1:N
54 [xd,y] = three_state_IM(tc(i),xc(i,:));
55 e(i) = y(1);
56 r(i) = y(2);
57 end
58
59 % Plotting
60 figure(2)
61 subplot(4,1,1), plot(tc,e), legend(’e = y-r’)
62 subplot(4,1,2), plot(tc,xc(:,8)), legend(’uc’)
63 subplot(4,1,3), plot(tc,r,tc,xc(:,5),tc,xc(:,7))
64 legend(’r’,’x_1’,’x_3’)
65 ylim([0,1.5])
66 subplot(4,1,4), plot(tc,xc(:,2)), legend(’x_2’)
67 ylim([-0.5,0.5])

7.5.7 Linear Observer Design for State Estimation

Full-state feedback as introduced and applied in Sections 7.5.4–7.5.6 requires information on all the system states. For
various reasons, not all states may be measured directly. For example, some states in a system may not be accessible or
cannot be measured at all. Methods such as pole-placement require information from all the system states for feedback.
When we write y = [1, 0, 0] x, this means only a measurement of x1 is available. If you could invert C, then x = C−1 y, but
C is often singular so we cannot find x just from y. It turns out you’d need to integrate over time to get such an estimate
[147], but it would be error-prone and not practical. D. Luenberger introduced an alternative way [149, 150]. If a system is
completely state observable, then it is possible to estimate states that are not directly measured. Luenberger [149] developed
a full-state observer for the linear system,
ẋ = Ax + Bu, (7.55)
y = Cx, (7.56)
in the form,
x̃̇ = Ax̃ + Bu + L(y − Cx̃ ), (7.57)
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
442 7 Modeling Feedback Control Systems

+
Observer
–

Figure 7.67 Observer system.

where x̃ is the estimate of x and L is the observer gain. As shown in Figure 7.67, an observer has two inputs, in this case, u
and y, with output x̃ .
An observer should be designed to provide state estimates that converge about five times faster than the fastest response
time of the modeled system. In this way, the estimates can be effectively used for feedback control purposes.
To design an observer, a defined tracking error,
e = x − x̃ (7.58)
can be made asymptotically stable if a system is completely state observable, enabling us to determine the observer gain
matrix L.
The error dynamics are defined by taking the derivative of the error above,
ė = ẋ − x̃̇ = Ax + Bu − Ax̃ − Bu − L(y − Cx),
̃
or,
ė = (A − LC)e, (7.59)
Thus, we can use the observer CE,
|sI − (A − LC)| = 0, (7.60)
to determine an observer gain matrix L that gives a desired response in the observer dynamics. Pole-placement can be used
if the system is completely state observable.
We can express the observer system equations as in equation 7.57 or in the form,
x̃̇ = (A − LC)x̃ + Bu + Ly, (7.61)
In equation 7.57, the term ỹ = y − Cx̃ is commonly referred to as the residual [147]. The value of ỹ will go to zero when the
observer error, e, goes to zero.

7.5.7.1 Observer Design: Low-Order Systems

For low-order systems (n ≤ 3), observer design can be accomplished analytically, as in the following example.

Example 7.15 Example of low-order observer design

Design and test an observer for the second-order system,
[ ] [ ]
2 3 0
ẋ = + u,
−1 4 1
[ ]
y = 1 0 x,
where only x1 is observed.
Solution
The observability matrix is,
[ ]
⎡C⎤ ⎡ 1 0 ⎤ [ ]
⎢ ⎥ ⎢ [ ]⎥ 1 0
Po = ⎢ −− ⎥ = ⎢[ ] 1 1 ⎥= .
⎢ ⎥ ⎢ 1 0 ⎥ 1 1
⎣ ⎦ ⎣
CA −2 −1 ⎦
This matrix has a rank of 2 so the system is completely state observable.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
7.5 State Variable Control Methods 443

The desired observer will have a CE given by,

|sI − (A − LC)| = s2 + (L1 − 6)s − 4(L1 − 2) + 3(L2 + 1),

The gains L1 and L2 can be found by comparing to a standard second-order form, s2 + 2𝜁𝜔n + 𝜔2n = 0. The damping and
undamped natural frequency can be specified to give desirable time-domain specifications. In this case, use 𝜁 = 0.8 and
𝜔n = 10 rad/sec. Solving for L1 and L2 ,

L1 = 2𝜁𝜔n + 6.
L2 = (𝜔2n + 4L1 − 11)∕3.

For the desired parameter values, L = [22 59]T .

To test the observe, the error response dynamics can be simulated using,

ė = (A − LC)e,

where e = x − x̃ . The code and results below show the transient response to initial conditions in errors e1 and e2 .

1 A = [2,3;-1,4];
2 B = [0;1];
3 C = [1,0];
4 Po = obsv(A,C)
5 rank(Po)
6 syms s L1 L2
7 CE=det(s*eye(2)-(A-[L1;L2]*C));
8 collect(CE,s)
9 zeta = 0.8; wn = 10;
10 L1 = 6 + 2*zeta*wn;
11 L2 = (wn*wn + 4*L1 - 11)/3;
12 L = [L1;L2];
13 Ao = (A-L*C);
14 sys = ss(Ao,B,[1,0;0,1],[0;0]);
15 [ye,te,xe]=initial(sys,[1;-2],0.8);
16 figure(1)
17 plot(te,xe), legend(’e_1’,’e_2’)

–1

–2

–3

–4

–5
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
Time, sec

7.5.7.2 Ackermann’s Formula for Observer Design

The observer gain matrix L can be found from Ackermann’s formula in a manner similar to state feedback control problems
where now,

L = 𝜙(A)P−1
o [0 0 · · · 0 1]
T
(7.62)
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
444 7 Modeling Feedback Control Systems

System model

+
Observer

Figure 7.68 System with observer, no feedback control.

with Po the observability matrix and,

𝜙(A) = An + a1 An−1 + · · · + an−1 A + an I = 𝟎. (7.63)
When using the MATLAB acker() function make sure to transpose the A and C matrices. The MATLAB acker() function
can be applied to the system in Example 7.15 as follows:

1 A = [2,3;-1,4];
2 B = [0;1];
3 C = [1,0];
4 p = roots([1,2*zeta*wn,wn*wn]);
5 acker(A’,C’,p)

This MATLAB script returns the same L as found before.

7.5.8 Using Observers Without Feedback Control

Prior to showing how an observer can be integrated with a system to provide state estimates for feedback control, consider
use of an observer solely for estimating unmeasured states in an open-loop system. For example, this could be a system for
which only certain measurements are made and it is desired to estimate the others. The system looks as in Figure 7.68.
Integrate the observer,
x̃̇ = (A − LC)x̃ + Bu + Ly
with the system to be observed, and construct an augmented state-space system,
[ ] [ ][ ] [ ]
ẋ A 𝟎 x B
= + u. (7.64)
̇x̃ LC A − LC x̃ B
[ ]
[ ] x
e = x − x̃ = In −In . (7.65)
x̃
The errors in the estimates can be calculated here since all the states are presumed known. If the actual measurements, y,
were made from a system, this, of course, would not be possible. Nevertheless, an estimate of the error can be made using
equation 7.59.

Example 7.16 Estimating spring deflection in a rotational drive

Consider the section of a rotational drive shown in Figure 7.69(a). A torque 𝜏(t) drives a rotor with damping through a spring
into a bearing and load torque, 𝜏L . The speed of the rotor, 𝜔1 is measured and both torques are also known or measured.
Design an observer that can estimate the spring rotational deflection, 𝜃.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
7.5 State Variable Control Methods 445

(a) (b)

Figure 7.69 (a) Rotational drive component; (b) bond graph.

Solution
From the bond graph in Figure 7.69(b), the state equations are derived:
[ ] [ ][ ] [ ][ ]
𝜔̇ 1 −B1 ∕J1 −K∕J1 𝜔1 1∕J1 0 𝜏(t)
= + . (7.66)
̇𝜃 1 −K∕B 𝜃 0 −1∕B 𝜏L
[ ]
[ ] 𝜔1
y= 1 0 . (7.67)
𝜃
To demonstrate the design of an observer for this system, set all parameters to 1.0, giving:
[ ] [ ]
−1 −1 1 0 [ ]
A= B= C= 1 0 .
1 −1 0 −1
This system is completely state observable. Select the poles to be −1 ± j, and the following code finds the observer gain
matrix L = [0.5, −0.25].

1 Simulated states
1

0.5

0
0 1 2 3 4 5 6 7 8 9 10
1
Measured
0.5

0
0 1 2 3 4 5 6 7 8 9 10

1 Estimated
states 1obs
obs
0.5

0
0 1 2 3 4 5 6 7 8 9 10
Transient period
0.6 e1
0.4 Errors e2
0.2
0
–0.2
0 1 2 3 4 5 6 7 8 9 10
Time, sec

Figure 7.70 Estimation results.

 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
446 7 Modeling Feedback Control Systems

This observer is tested by setting the system in a steady-state equilibrium with known inputs: 𝜏(t) = 𝜏o = 1 and 𝜏Lo = 0.5.
For equilibrium, set ẋ = 𝟎 and then solve for the initial states by,
[ ] [ ]−1 [ ][ ]
𝜔1o −1 −1 1 0 𝜏o
= .
𝜃o 1 −1 0 −1 𝜏Lo
A simulation of this system is initialized with these initial conditions and the initial inputs. The observer will respond
in a transient period, given arbitrary initial conditions, and then reach a steady state. For the given design, the errors will
stabilize to zero. This is shown in Figure 7.70.
At t = 5 seconds, the input torque is stepped up to further evaluate the observer response. As shown, both the estimated
states compare well with the actual system states (errors remain close to zero).
The MATLAB code is provided on the following page.
This example illustrates well how observers can be useful in sensing applications alone. In the following, the use of
observers for full-state feedback is discussed.

1 clear
2 J1 = 1; Ks = 1; B1 = 1; B2 = 2;
3 A = [-B1/J1,-Ks/J1;1,-Ks/B2];
4 B = [1/J1,0;0,1/B2];
5 C = [1,0]; % measuring Q1
6 % Note: observability checked
7 n=2;
8 % desired poles
9 po = [-1+1i,-1-1i];
10 L = place(A’,C’,po)’;
11 % Create augmented system
12 Aa = [A,zeros(n);L*C,A-L*C];Ba = [B;B];
13 % output is the error x-xobsv
14 Ca=[eye(n),-eye(n)];
15 Da=[0,0;0,0];
16 sysa = ss(Aa,Ba,Ca,Da);
17
18 % Simulate system first for initial conditions
19 % For observer, use zeros or use random values for ICs
20 % (initial error in observer not necessarily known)
21 %
22 tau_ino = 1; tau_Lo = 0.5;
23 xo = -inv(A)*B*[tau_ino;tau_Lo];
24
25 xao = [xo(1);xo(2);0;0];
26 % create time array
27 ta = linspace(0,20,2000);
28 % after ts seconds, introduce input
29 ts = 5;
30 for i=1:length(ta)
31 if ta(i) ≥ ts
32 tau_in(i) = 1.5*tau_ino; %*(1+1.5*sin(2*pi*(ta(i)-ts)/3));
33 tau_L(i) = tau_Lo;
34 else
35 tau_in(i) = tau_ino;
36 tau_L(i) = tau_Lo;
37 end
38 end
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
7.5 State Variable Control Methods 447

39 u = [tau_in;tau_L];
40 [e,t,xa]=lsim(sysa,u,ta,xao);
41 % compute the measurements
42 y = C*xa(:,1:n)’; % transpose needed
43
44 figure(1)
45 subplot(4,1,1), plot(ta,xa(:,1:2))
46 legend(’\omega_1’,’\theta’)
47 subplot(4,1,2), plot(ta,y)
48 subplot(4,1,3), plot(ta,xa(:,3:4))
49 legend(’\omega_{1obs}’,’\theta_{obs}’)
50 subplot(4,1,4), plot(ta,e)
51 legend(’e_1’,’e_2’)

7.5.9 Full-State Feedback Control with an Observer

We now integrate full-state feedback with an observer, enabling specification of the control, u, using the observed (or esti-
̃ It is necessary to evaluate (typically by simulation) whether using estimated states x̃ rather than x
mated) state vector, x.
achieves a desired closed-loop design.
The integrated system now takes the form shown in Figure 7.71. Now apply u = −K̃x (note that the feedback uses
estimated states),20
x̃̇ = Ax̃ + Bu + L(y − Cx̃ ) = (A − BK − LC)x̃ + Ly. (7.68)
The observer error dynamics are the same as before: the estimated error does not depend on the input (input terms
cancel). So as before,
e = x − x̃
and,
ė = (A − LC)e.
Integrate with the system model,
ẋ = Ax + Bu,
y = Cx,

System model

+
Observer

Figure 7.71 Integrated full-state feedback with full-state observer.

20 Sometimes it can be desirable to use a reduced-order observer, being one where only non-measured states are estimated. Refer to Friedland
[147] or Brogan [94] for more on this topic.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
448 7 Modeling Feedback Control Systems

̃ so,
in which we now insert, u = −Kx,
ẋ = Ax − BKx̃
= Ax − BK(x − e)
ẋ = (A − BK)x + BKe,
and the closed-loop integrated (estimated) state feedback and observer system is expressed,
[ ] [ ][ ]
ẋ A − BK BK x
= . (7.69)
ė 𝟎 A − LC e
In this form, the CE is found to be,
D(s) = |sI − (A − BK)| ⋅ |sI − (A − LC)| = 0, (7.70)
so the roots of the full-state feedback and the observer can both be specified in the left-hand plane and both systems are
asymptotically stable.
Note that the two systems can be designed independently, taking advantage of the separation principle,21 which allows
us to first design the feedback control and then the observer system in a sequential manner.

Example 7.17 Example of state feedback with observer

For the state-space system,
⎡−2 −3 1⎤ ⎡5⎤
⎢ ⎥ ⎢ ⎥
ẋ = ⎢ 5 0 0⎥ x + ⎢0⎥ u. (7.71)
⎢ ⎥ ⎢ ⎥
⎣ 0 1 0⎦ ⎣0⎦
[ ]
y = 1 0 0 x. (7.72)
a) determine controllability and observability, then
b) if controllable, design a full-state feedback controller, determining K, and
c) if observable, design a state observer
d) if both (b) and (c) are true, sketch/draw the integrated full-state feedback and observer system and simulate the inte-
grated system
Solution (a)
For the given state-space system, the controllability matrix is,
[ ]
Pc = B AB A2 B
2
⎡5 ⎡−2 −3 1⎤ ⎡5⎤ ⎡−2 −3 1⎤ ⎡5⎤⎤ ⎡5 −10 −55⎤
⎢ ⎢ ⎥⎢ ⎥ ⎢ ⎥ ⎢ ⎥⎥ ⎢ ⎥
Pc = ⎢0 ⎢ 5 0 0⎥ ⎢0⎥ ⎢ 5 0 0⎥ ⎢0⎥⎥ = ⎢0 25 −50⎥ .
⎢ ⎢ ⎥⎢ ⎥ ⎢ ⎥ ⎢ ⎥⎥ ⎢ ⎥
⎣0 ⎣ 0 1 0⎦ ⎣0⎦ ⎣ 0 1 0⎦ ⎣0⎦⎦ ⎣0 0 25⎦
We see that rank(Pc ) = 3 so this system is controllable. The observability matrix is,
[ ]
⎡ 1 0 0 ⎤
⎢ ⎥
⎢ ⎡ ⎤
−2 −3 1 ⎥
⎢ [ ] ⎢ ⎥⎥
⎡ C ⎤ ⎢ 1 0 0 ⎢ 5 0 0⎥ ⎥ ⎡ 1 0 0⎤
⎢ ⎥ ⎢ ⎢ ⎥⎥ ⎢ ⎥
Po = ⎢ CA ⎥ = ⎢ ⎣ 0 1 0⎦ ⎥ = ⎢ −2 −3 1⎥ .
⎢ 2⎥ ⎢ 2⎥ ⎢ ⎥
⎣CA ⎦ ⎢ ⎡−2 −3 1⎤ ⎥ ⎣−11 7 −2⎦
⎢[ ]⎢ ⎥⎥
⎢ 1 0 0 ⎢ 5 0 0⎥ ⎥
⎢ ⎢ ⎥⎥
⎣ ⎣ 0 1 0⎦ ⎦
This matrix has a rank of 3 so the system is completely state observable.

21 First demonstrated for discrete control theory by Joseph and Tou [151] and then later generalized for continuous-time systems [93].
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
7.5 State Variable Control Methods 449

Solution (b)
The system is controllable, but note that it is also open-loop unstable as the eigenvalues of A are −1.16 ± j3.8 and 0.318.
There is one unstable pole in the right-hand plane. Propose a state feedback controller with poles to be placed at −1 ± j0.5
and −2. By using the place() function, a control gain is found to be K = [0.4, −0.39, 0.3].
Solution (c)
Since the system is found to be observable, an observer can be designed. Poles are chose to be five times those specified in
(b), giving L = [30, 1285, 4176].
Solution (d)
The integrated full-state feedback and observer system is as shown in Figure 7.72. This system can be simulated using
equations 7.69. The results in Figure 7.73 show that based on only a measurement of y = x1 , estimates for all three states
can provide a reasonable basis for full-state feedback that stabilizes the system and provides nearly zero error in state
estimates after about 1.5 to 2 seconds.

System model

+
Observer
–

Figure 7.72 Integrated full-state feedback with full-state observer.

Full-state feedback Estimated state feedback

4 4
x1 x1
x2 x2
Estimated states

2 x3 2 x3
States,

0 0

–2 –2
0 2 4 6 8 0 2 4 6 8
1 1
uc = –Kx

0 0
Input,

Input,

–1 –1

–2 –2
0 2 4 6 8 0 2 4 6 8
Time (sec) Time (sec)
(a) (b)

Figure 7.73 Comparison of initial condition response for the system in equation 7.71[ with] (a) (direct) full-state feedback control and
(b) with estimated full-state feedback. Result shown have initial conditions of xo = 1, 1, 1 with assumed 10% error in the initial
estimated states, x̃ o .

7.5.10 Reference Inputs with Full-State Feedback and Observer

An approach for adding a reference input for full-state feedback control was described in Section 7.5.5. When a system
includes an observer, an approach described in [61, 93] is commonly adopted. Now, the reference is introduced by adding
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
450 7 Modeling Feedback Control Systems

+ System model

+
Observer

Figure 7.74 State feedback and observer compensation with reference input.

terms to the observer and control law equations as follows:

x̃̇ = Ax̃ + Bũ + L̃y + Mr, (7.73)

u = ũ + Nr = −Kx̃ + Nr, (7.74)

where M is an n × 1 matrix in the observer equation and N is a scalar for r. The selection of M and N will influence the
transient response of the system but not the system characteristic equations. We choose the controller and observer gains,
K and L, to influence the poles of the system and then M and N (feedforward gains) will influence zeros and thus transient
behavior. The general case for the integrated system with reference input is shown in Figure 7.74. If M = 𝟎 and N = 0, then
this becomes a system with no r as presented in Sections 7.5.7 and 7.5.9.
If there is a reference input, then the design task falls to selecting the appropriate M and N, assuming the control and
observer designs follow the approaches previously reviewed. We can consider two different cases based on Figure 7.74 [61].

7.5.10.1 Case 1: Observer Error Dynamics Do Not Depend on r

To get good estimator performance, we want to make it independent of the reference input. This can be done by selecting
M = BN. To see why this is the case write the error dynamics using equations 7.73 and 7.74 from above,

ė = ẋ − x̃̇ = Ax − Bu − Ax̃ − Bũ − L̃y − Mr

= (A − LC)e + (BN − M)r,

which shows that making M = BN will give error dynamics free of r. This is similar to the previous case examined with no
reference input, and the compensator takes the same form but the control input now has the effect of r; that is,

x̃̇ = Ax̃ + Bu + L̃
y.
u = −Kx̃ + Nr.

The system then becomes as shown in Figure 7.75, where the observer is driven directly by y and u. The observer and control
law in this case are in the feedback path.

+ System model

Observer

Figure 7.75 State feedback with compensation in feedback path with reference input.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
7.5 State Variable Control Methods 451

7.5.10.2 Case 2: Estimator Driven by Tracking Error, y − r

An alternative case allows the observer to be driven by the tracking error. In this case, selecting N = 0 and M = −L gives
the compensator equations (observer and control law),
x̃̇ = Ax̃ + Bu + L̃y − Lr = (A − BK − LC)x̃ + L(y − r),
̃
u = −Kx,
and the observer is driven by the tracking error, e = y − r. This design has the compensation in the forward path. A block
diagram for this form of state feedback with compensation is shown in Figure 7.76.

7.5.11 Linear Quadratic Optimum Control

This brief introduction is meant to guide application of linear quadratic optimal control methods for solving basic prob-
lems. In particular, emphasis is placed on understanding how to use tools such as the lqr() function found in control
design toolboxes. These functions determine optimal control gains for linear, stable, and controllable systems. Normally,
pole-placement for state-variable feedback is achieved by specifying pole locations, and as long as the system is controllable,
various methods can be used to come up with the control gain matrix, K, for state feedback, u = −Kx. Pole locations are
typically chosen based on trying to achieve certain time-domain specifications (settling time, overshoot etc.), and insights
gained through study of classical control methods.
Another way to go about determining the gain matrix for state feedback control is by associating performance measures
with a scalar performance index,
∞
J= L(x, u, t)dt, (7.75)
∫0
where the function L() depends on system states and inputs. A control, u (or for single-input, u), is sought that can give
the best trade-off between performance (states) and “cost of control” (inputs). An optimal solution arises by minimizing J.
This approach can be applied to both closed-loop and open-loop control problems.
For linear state-space systems and for many cases of practical interest, we adopt quadratic forms,
∞
J= (xT Qx + uT Ru)dt, (7.76)
∫0
where Q and R are weighting matrices on states and input(s), respectively.

7.5.11.1 Quadratic Sums

Typically Q and R are diagonal and, which lead to the scalar quadratic sums:
∑
n
∑
r
xT Qx = qi xi2 and uT Ru = rj u2j .
i i

Thus, when J is minimized, we are thinking about the contribution over time of the weighted sums of the states and
input(s). The “best” response then is one that has the minimum weighted area under the xi2 curves, as illustrated in
Figure 7.77. We choose Q and R to emphasize importance of individual states and inputs.
You can also weight the output and then use,
y = Cx → Q = CT QO C → xT Qx = yT QO y,
It is very often the case that Q and R are chosen by trial and error. Herein is the real problem for the engineer, since the
mathematical solution has effectively been done as outlined in detail in many books [53, 61, 147].

+ Observer System model

Figure 7.76 State feedback with compensation in forward path with reference input.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
452 7 Modeling Feedback Control Systems

Figure 7.77 Optimal control will minimize the areas under state trajectories.

7.5.11.2 Optimal Control Problem

The problem is to design state feedback that minimizes the performance measure J for the state-space system,
ẋ = Ax + Bu (7.77)
y = Cx (7.78)
with control law, u = −Kx. As usual, we substitute u into equation 7.77, and here assume the resulting closed-loop system
is stable so that at steady state, x(∞) = 𝟎.
The key results stem from a formulation taking advantage of several assumptions, a first that there exists an exact differ-
ential such that,
d T
xT Qx = − (x Px), (7.79)
dt
where P is a symmetric matrix that simplifies the algebra; that is, elements pij = pji . When equation 7.79 is expanded and
equation 7.77 is applied, then it is found that (for details, see Ogata [53], p. 917),
AT P + PA = −Q. (7.80)
For low-order systems, the elements of P can be determined from this matrix algebraic equation [53, 61].

7.5.11.3 Relevance of P
A key result is that J can be expressed as,
∞
d T
J= − (x Px)dt = −xT Px|∞ T
0 = x (0)Px(0), (7.81)
∫0 dt
where one of the key assumptions is that x(∞) → 𝟎. Now, as seen by equation 7.80, P will be composed of relevant system
parameters. Thus, the minimization of J can be accomplished relative to those parameters. Notably, when the A matrix
in equation 7.80 is actually the closed-loop form A − BK, this provides a basis for finding a controller gain matrix, K, that
minimizes J in a desired way.

7.5.11.4 Optimal Control of State-Space System

For the closed-loop control problem and specifically a linear state-space system with u = −Kx (regulator), for which appli-
cation to equation 7.77 results in a stable A − BK, the performance index becomes,
∞ ∞
J= (xT Qx + xT KT RKx)dt = xT (Q + KT RK)xdt. (7.82)
∫0 ∫0
This form is treated in a way similar to the result that gave algebraic equation 7.80 for P. It can then be shown that J is
minimized when [53],
K = R−1 BT P, (7.83)
so for the scalar control law, we write,
u = R−1 BT Px, (7.84)
although this extends to the multi-input case, u, as well [53]. Now, the P matrix must satisfy,
AT P + PA − PBR−1 B−1 + Q = 𝟎, (7.85)
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
7.6 Simulation of Controlled Systems 453

0.08
r
0.06 V1
V2
0.04

0.02
0 2 4 6 8 10 12 14 16 18 20

5
Qp

–5
0 2 4 6 8 10 12 14 16 18 20
Control inputs

54 uc
–ur
52

50
2 4 6 8 10 12 14 16 18 20
Time, sec

Figure 7.78 Results from simulated three-state sewer system with periodic changes in the reference level for V 1 with optimal control
gains.

which is a reduced-matrix Ricatti equation. The more general matrix Ricatti differential equation is considered one of the
most famous in the literature of modern control theory, as extended by Kalman [152].
For many practical problems, the controller gain matrix K can be obtained using readily available algorithms found in
contemporary control design software. Given a completely state controllable system (A, B) and properly selected Q and R,
the gain for a linear quadratic regular (lqr) is returned by the algorithm, lqr(A,B,Q,R) (found in MATLAB as well as
Python control toolbox).
Revisiting the full-state feedback control of the three state sewer system from Example 7.13 and Section 7.5.5, optimal
gains can be found by:

1 Q = [1,0,0;0,1e-4,0;0,0,.1]; R = 1e-6;
2 K = lqr(A,B,Q,R)

This gives: K = [−730.4, 682.06, −318.4], compared to K = [−68.6, 46.0, −11.4] from before. Note the much higher gains
and also control input arise because the R value was made nearly zero so effectively the control was unconstrained.
Simulated results are shown in Figure 7.78. This managed V 2 a little better (less dip) at the cost of more control (almost
10 times). Note that there was also less weighting put on V 2 itself.
The Q and R matrices could be tweaked a bit more to achieve different trends. Note that using the internal model approach
from Section 7.5.6 with optimal control can give even better results than these.

7.6 Simulation of Controlled Systems

Early stages in control system analysis and design may employ linear analysis tools given an approximated (linear) plant
model to facilitate design. However, prior to actually building a real system, it can be prudent to evaluate a controlled system
using nonlinear digital simulation with more complete models of the plant and any critical control system components.
Such a task may also be accomplished for testing control system design with a more complete model of the plant. These
simulations should be able to incorporate nonlinear models and thus employ standard ordinary differential equation (ODE)
solvers. Modeling and analysis of discrete-time effects may also be important, and, in some cases, essential given that most
controlled systems are implemented in this form [136].
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
454 7 Modeling Feedback Control Systems

This section describes two simulation approaches. Modern simulation environments exist that allow block diagram
descriptions to be used for computer-aided analysis of control system. These approaches use the same numerical solutions
of continuous-time state equations to be discussed in the following, which emphasizes a text-based approach to simula-
tion as first discussed in Chapter 2. These methods can be implemented in any commercial or open-source programming
environment.

7.6.1 Continuous-Time State Equations and Feedback Control

7.6.1.1 Direct Simulation of P-Control
Consider proportional control of a SISO system illustrated as a block diagram in Figure 7.79. In this diagram, the physical
system plant is represented in nonlinear state-space form,
ẋ = f(x, u, 𝑤),
subjected to a single control input, u, as well as disturbances, 𝑤. The controlled variable is the output,
y = g(x, u, 𝑤),
and a measurement system provides feedback signal ys = fs (y), where fs () is a sensor function. A direct proportional control
then directs u = K ⋅ e = K(r − ys ), where r is the reference input or command.
This system can be simulated using standard nonlinear ODE solvers, and the only new considerations are: (i) dictating the
reference command, (ii) specifying the form of disturbance(s), and (iii) computing the dynamic control input, u = K(r − ys ).
This last step requires a suitable model for ys , which is assumed to be part of the overall control system modeling effort,
but also specification of the control gain, K. Thus, setting up a direct simulation of P-control for any such system follows
conventional methods. Indeed, such a simulation can be configured fairly quickly using state of the art software programs
that use a block diagram modeling basis (e.g., MATLAB/Simulink and Scilab).
In the following, the simulation approach will be described in a way that enables implementation using any text-based
computer code environment, assuming availability of ODE solvers (e.g., fixed and adaptive time-step ODE solvers). The key
steps that convey integration of the P-control can be seen in a basic pseudocode such as the following:

Steps in Direct P-control

1 Update values of the n system states, xi , i … n

2 Update any disturbance input(s), 𝑤
3 Update reference (command) input, r
4 Determine plant output, y (function of states)
5 Determine the feedback signal from measurement function, ys
6 Compute error, e = r − ys
7 Set the controller gain, K (constant or scheduled)
8 Compute the controller output, u = K ⋅ e
9 Compute state rates: ẋ = f(x, u, 𝑤)

To illustrate these steps, this algorithm is applied to control the position of a mass in Figure 7.26(a) subjected to distur-
bance forces, Fd (t). It is desired to keep the mass at a desired position, xmr . The mass, m, is attached to the spring, assumed

Figure 7.79 Block diagram of feedback-controlled

single-input–single-output system.
+
–
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
7.6 Simulation of Controlled Systems 455

here to have a linear spring constant, Fk = kxk , xk being the spring deflection. A position sensor can directly measure xm .
The plant equations are summarized below along with the indicated initial conditions.

State equations Initial conditions

ẋ k = 𝑣m xk (0) = xko

𝒗̇ m = −Fk (xk ) − Fb (𝑣m ) − Ff (𝑣m ) + Fd (t) 𝑣m (0) = 0

ys = lo + xk ys (0) = lo + xk (0)

The regulated position will be xmr = lo , and the sensor output is ys , modeled as an ideal measurement of the distance
to the edge of the mass, which is equal to the sum of the initial length of the spring and spring deflection. The following
pseudocode shows how a simulation function file can integrate the plant state equations with the control system algorithm:
def mass_spring_function(x, t, m, k, K, lo, Fdo, fd):
xk, v = x[0], x[1]
r = lo # reference command
ys = lo + xk # feedback signal model
e = r - ys # error calculation
Fc = Kp*e # PID control force
Fk = k*xk # spring force
Fb = b*v # linear damping force
Fd = Fdo*sin(2*pi*fd*t) # disturbance force
# state equations
xdot = v
vdot = (Fc - Fk - Fb + Fd)/m
return [xdot, vdot])
A simulation code with an ODE solver (fixed or adaptive time step) can be used to directly simulate these equations.
Plots of simulation results shown in Figure 7.80 (top) show the mass position as the mass is subjected to a sinusoidal dis-
turbance force. The P-controller is turned on at t = 10 seconds and the motion subsides. The simulation solves numerically
what is expected from the analytical result in equation 7.17: the P-control makes it appear that the spring was stiffened by
an amount equal to the proportional gain, K. Since P-control only adds stiffness, there are sustained oscillations evident
in the velocity (middle plot, Figure 7.80), and the control force, Fc , has to introduce high-frequency oscillations to try and

0.20
Controller off P-control on r
0.15 ys
0.10 xk
0.05
Spring deflection,
0.00
–0.05
0.0 2.5 5.0 7.5 10.0 12.5 15.0 17.5 20.0

0.2 v
Mass velocity
0.0

–0.2

0.0 2.5 5.0 7.5 10.0 12.5 15.0 17.5 20.0

Fc
5 Fd
Fk
0
–5

0.0 2.5 5.0 7.5 10.0 12.5 15.0 17.5 20.0

Time, sec

Figure 7.80 Results of simulation for P-control of mass–spring system in Figure 7.26.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
456 7 Modeling Feedback Control Systems

regulate the position. This ideal simulation shows the force levels and bandwidth that the force actuator would need to
be capable of achieving. Both integral and derivative control could be added to improve the performance, but also some
(passive) linear damping in the system would also manage the oscillations.
The basic simulation algorithm described is similar to the one used to study any nonlinear dynamic system. In this way,
it is possible to analyze, tune, and/or refine a controlled system design. Models for any feedback control elements, sensors,
actuators, etc. are assumed to progress in time synchronized with the physical plant. It has become common to realize this
approach using block diagrams, and software programs for constructing models for simulation are prevalent. All of these
solutions are, of course, discrete numerical approximations. The concept most certainly finds it roots in analog computing
[40, 88], and a good reference that bridges analog to digital simulation is the work by Kochenburger [132].
Depending on the feedback control function or algorithm, there may be some steps that need to be taken to fully imple-
ment a controller function properly. Many of these methods have been implemented and demonstrated in contemporary
software programs (e.g., MATLAB and MATLAB/Simulink). It is always possible to construct the simulation platform using
text-based programming, however. Extension of the PID algorithm is described in the following. It is assumed here that the
plant is SISO so there is one control input and one output.

7.6.1.2 Implementing a PID Algorithm

Before showing how a PID compensator can be simulated in the time domain with a plant, each of the terms should be
examined. As shown previously, the proportional (P) term can be directly computed from the error, e. The integral (I) and
derivative (D) terms, however, require special consideration since they are dynamic. With the integral control effort defined,

ui = Ki edt = Ki Ie .
∫
Ie is the error integrated state. This state evolves according to the state equation,
dIe
= e = r − ys , (7.86)
dt
r being the reference input and ys is the feedback which depends on plant output, y.
Implementing the derivative of error is more involved. In general, taking a direct derivative of a signal, de∕dt, can be
problematic because any abrupt changes or noise can generate very large values. This can cause problems in controllers,
leading to large errors and possibly causing a simulation (or controller) to fail.22 For this reason, the derivative control effort,
ud = Kd De
is best implemented in a digital simulation using a form suggested by equation 7.16. The derivative of the error, De = de∕dt,
is then approximated by a transfer function,
De s
= , (7.87)
e 𝜏f s + 1
where 𝜏f is a low-pass filter time constant. This time constant should be chosen to match the expected bandwidth of working
signals, 𝜔f ≈ 1∕𝜏f (1st order cutoff).
Now, to use this form in a digital (time) simulation, we could inverse transform from the s-domain,
De s
= ⇒ De (𝜏f s + 1) = se ⇒ 𝜏f Ḋ e + De = de∕dt.
e 𝜏f s + 1
However, this is not a desirable ODE for De since it requires taking the derivative of e. For this reason, the transfer function
is redefined using two subsystems,
[ ] [ ]
De [ ] De [ zd ]
s s 1
= = = ,
e 𝜏f s + 1 1 𝜏f s + 1 zd e

where zd is an intermediate variable. Now, the two transfer functions are defined,
De
= s ⇒ De = dzd ∕dt, (7.88)
zd

22 These large variations in e can also cause the value of Ie to quickly “wind-up” and thus cause the integral term to inflate.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
7.6 Simulation of Controlled Systems 457

Table 7.7 Steps in PID algorithm for simulation with system ODEs.

Steps in PID simulation function

1 Update values of the n system states, xi , i … n

2 Update values of the controller states, Ie , zd
3 Update reference (command) input, r
4 Determine plant output, y (function of state)
5 Determine the feedback signal, ys
6 Compute error, e = r − ys
7 Compute the controller output, uc = Kp e + Ki ⋅ Ie + Kd De , with De from equation 7.90
8 Compute total input to the plant, u = uc + ud , where ud is the net disturbance input
9 Compute time rates of change using the state equations with controller input, u,
ẋ = f(x, u), and the two controller state equations,
dIe ∕dt = e, and dzd ∕dt = (−zd + e)∕𝜏d

and,
zd 1
= ⇒ zd (𝜏f s + 1) = e ⇒ 𝜏d ż d + zd = e.
e 𝜏f s + 1
The second equation allows for propagating zd without having to take de∕dt directly. A new state equation can be defined
for the zd state,
dzd
= (−zd + e)∕𝜏f , (7.89)
dt
and then De is determined from equation 7.88, that is,

De = (−zd + e)∕𝜏f . (7.90)

An algorithm that integrates these PID controller relations with a generally nonlinear system model for digital simulation
is summarized in Table 7.7. The algorithm can be implemented directly in a text-based code (e.g., MATLAB, Python, and
C) using standard ODE solver algorithms to propagate the dynamic states.

7.6.1.3 Application of the PID Algorithm

Setting up simulation of a PID controller in the way described provides a useful testing and evaluation framework. It should
be emphasized that while the PID controller was described using ideal linear definitions, it is possible to extend the con-
troller as needed to include other effects. For example, many PID controllers will include gain-scheduling, anti-windup,
and other features. See Astrom and Hagglund [153] for more on PID controllers and their application.
To illustrate application of the method described and outlined in Table 7.7, consider PID control of drive torque on a
wheel that rolls without slip, shown in Figure 7.81(a). The intent is to control the longitudinal speed, 𝑣x . From the bond
graph in Figure 7.81(b), causality shows there is only one state, here chosen to be the forward velocity, 𝑣x . The forward
velocity can be estimated from a rotational speed sensor on the wheel, since 𝑣x ≈ r𝑤 𝜔𝑤 . The state equation is,

ṗ m = m𝑣̇ x = F − Fa − Fr + Fd ,

where Fa and Fr are the aerodynamic and rolling resistance forces, respectively, which are generally nonlinear. A distur-
bance force, Fd , is also indicated. Since J 𝜔̇ 𝑤 = 𝜏c − r𝑤 F, we can express the (no slip) traction force,

F = 𝜏c ∕r𝑤 − J 𝑣̇ x ∕r𝑤
2
.

Thus, the state equation can be expressed,

𝑣̇ x = (𝜏c ∕r𝑤 − Fa − Fr + Fd )∕meff ,

 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
458 7 Modeling Feedback Control Systems

(no slip)

(a) (b)

Figure 7.81 (a) Torque-driven wheel with no slip. (b) Bond graph showing loads on wheel, disturbance force, and input/output signals
for controlled drive torque, 𝜏c , and sensed velocity, 𝑣x .

+ PID + +
algorithm +
–

=1

Figure 7.82 Block diagram of PID controlled, torque-driven wheel.

where meff = m + J∕r𝑤 2 , F = c 𝑣 |𝑣 |, F = f m g ⋅ sgn(𝑣 ), with c a constant that incorporate drag force properties and f
a d x x r r 𝑤 x d r
is the effective rolling resistance coefficient.23 If this model is linearized about an equilibrium (or cruise) velocity, 𝑣xe , then
the system can be modeled by a linear first order plant model and linear analysis tools can be used to study the PID control.
In this case, the PID controller is being used to drive the fully nonlinear model in simulation. As formulated, the model
will enable control scenarios to be simulated provided control gains are provided. In addition, the simulation will solve for
control torque 𝜏c , which would enable determining the required motor torque and power levels. In addition, the computed
no-slip traction force, F, can be compared to expected levels of maximum traction force so that the no-slip assumption can
be evaluated. A block diagram in Figure 7.82 illustrates PID controlled torque-driven wheel system.
The following pseudocode shows how a simulation function file integrates the torque-driven wheel state equations with
the a full PID control algorithm:

def torque_driven_wheel(x, t, vr, g, m, meff, Jt, cd, fr, Kp, Ki, Kd, tauf):
vx, Ie, zd = x[0], x[1], x[2]
r = vr # set reference command
e = r - vx # error
De = (-zd+e)/tauf # error derivative
tauc = Kp*e + Ki*Ie + Kd*De # PID control torque
Fd = (specify a disturbance function)
Fa = cd*vx*abs(vx) # aerodynamic force
Fr = fr*m*g*sign(vx) # rolling resistance force
# state equations
vxdot = (tau_c/rw - Fa - Fr + Fd)/meff
Iedot = e
zddot = (-zd+e)/tauf
return [vxdot, Iedot, zddot]
Note that the vdot equation depends on generally nonlinear forces. The force Fc is determined by the PID controller,
which depends on the error, error integral, and error derivative terms. Within this function, the PID controller output is

23 In this form, this simple model can be taken as a longitudinal performance model for a vehicle, where m is the total mass and J is the total
inertia of “rotating parts.” As such, the model is a reasonable basis for studying cruise control.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
7.6 Simulation of Controlled Systems 459

computed in synchronized time with the state equations. It would also be possible to integrate actuator and sensor models
within this function.

7.6.2 Continuous-Time State Equations and Discrete Feedback Control

As many systems are implemented using discrete-time control, a controlled system simulation can take advantage of a
different reticulation that takes advantage of the system boundaries clarified in Figure 7.83. A discrete-time (DT) controller
will process measured sensor feedback and generate a control signal over some finite loop time, Δt. Thus, controlled input(s)
to the plant may follow a zero-order hold (as from an analog-to-digital, or A/D, converter) over this finite time. Further,
any sensor inputs may also generally be sampled every Δt. Over this time period, the digital controller and measurement
and processing systems will prepare the next command. Thus, rather than simulate a controller using a framework meant
for continuous-time model simulation, as in Section 7.6.1, an alternative approach is to construct a “big loop” that more
closely emulates the realization conveyed in Figure 7.83.

7.6.2.1 Modeling Discrete PID

We can adopt some of the ways that digital control methods are practically implemented to more closely simulate expected
behavior. Extensive coverage of these methods can be found in Ogata [136] and in Astrom and Hagglund [153].
In this approach, the PID algorithm is discretized as done for implementation in a digital controller. Each of the controller
terms (P, I, and D) is considered in turn to compose a discretized PID output. The discrete proportional term is simply
computed by up (k) = Kp ek = Kp (rk − ysk ), where rk and ysk are the values of the reference (command) and the measurement
output sampled at tk . The integral term can be approximated simply by a summing algorithm based on a basic trapezoidal
rule. The form of the integral of error at any discrete time is then,
[ ]
Ie (k) = Ie (k − 1) + Δt e(k) + e(k − 1) ∕2, (7.91)

which effectively gives a term similar to Tustin’s approximation as shown in [153]. Finally, a simple backward difference
is taken here to estimate the derivative term, De (k) = (e(k) − e(k − 1))∕Δt. As such a basic discrete PID algorithm is com-
posed by,

u(k) = Kp e(k) + Ki Ie (k) + Kd De (k), (7.92)

Note that Δt in the I and D terms is absorbed into the controller gains for practical implementation. There are many ways
to implement a PID algorithm discretely for simulation or in a practical digital controller. Descriptions of the various ways
this can be done are presented more completely by Astrom and Hagglund [153].

7.6.2.2 Implementing Discrete PID in Simulation

The use of a discrete PID in a simulation of a continuous-time plant is less complex than the version for a “continuous
PID” simulation. The discrete approximations and calculations are directly implementable. Note also that the discrete PID
algorithm provides a control update every Δt for input to the continuous-time plant. In a simulation implementation, this
control input is then used as a constant input to an ODE simulation of the continuous-time parts of the system. This is not
unlike what will happen in a real system. Outputs from the ODE simulation can be “sampled” in a suitable form for use by

Mixed-signal Continuous-time hardware/plant

hardware and
Power source Disturbances
software

Digital Power
D/A modulation/
controller Actuator Plant Sensor(s)
amplification

Measurement
and processing Feedback

Figure 7.83 Basic single-input–single-output description of a discrete-time control of continuous-time systems.

 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
460 7 Modeling Feedback Control Systems

Table 7.8 Steps in a discrete-type PID algorithm for simulation with system ODEs.

Steps in discrete-time PID control simulation

1 Initialize system parameters and variables

and simulation variables
2 Initialize plant system states
3 Initialize controller states
4 Set controller properties/gains
5 Define the controller time step, Δt
6 Start the “big loop,” simulate until end time, tend
Set iteration value, k
7 Update reference (command) input, rk
8 Determine plant output, yk (function of state)
9 Determine the feedback signal, ysk (account for sensor dynamics)
10 Compute error, ek = rk − ysk
Update sum of error, Ie (k) = Ie (k − 1) + (ek + ek−1 )∕2
(note, using error sum formula per [153]; also, may include anti-windup algorithm)
Estimate error derivative as, De (k) = ek − ek−1
(note: error derivative as simple difference per [153])
11 Compute the controller output, uk = Kp ek + Ki Ie (k) + Kd De (k)
Use controller output, uk , to determine actuation output, uc (k)
(note: this could involve a model of an actuator, look-up table, etc.)
15 Use ODE solver to simulate ẋ = f(x, u) from tk to tk + Δt
– Use states at end of previous step as initial conditions
– Select a smaller time step for ODE simulation than Δt
– Use constant uc (k) over Δt
18 Compute any outputs at end of time period
19 Save states and outputs at end of time, tk + Δt for next step
19 Save error value, ek
Go to step 6, repeat
(end of big loop)
Process/display/save results

the discrete-time implementation. These steps are summarized in Table 7.8 for a discrete PID. It should be mentioned the
concept could be extended to integrate a different discrete-time control algorithm for simulation.
Another advantage of this form of the simulation of a control system with a plant is that it becomes possible to very closely
emulate the overall system. As such, the form of the control algorithm can be more easily ported over to an actual digital
control platform. Indeed, this has been the practice in contemporary virtual control prototyping environments.

7.7 Summary
The need to analyze and design a system with physical feedback effects or feedback control is a very common motivation for
physical dynamic system modeling. Many of the methods introduced in Chapters 5 and 6 for linear system analysis in the
time and frequency domain find extensive use for these purposes, but in addition it can be important to assess the models
used in feedback control carefully. This has been an emphasis of this chapter, to give an essential review or introduction to
how physical system modeling can be helpful in the system control task. The topics and examples have been selected either
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
7.8 Problems 461

to reinforce linear system analysis techniques or to show how some systems can benefit from a careful consideration of
the physical model. It is a careful balance, of course, since control analysis and design often calls for simplifying a physical
plant to be controlled in order to make the design task possible. On the other hand, there can be many cases where it is
necessary to explore how a given controller will behave when confronted with more realistic effects. These are a few of the
issues that have been explored in this chapter.
The chapter also attempts to touch on a broad range of topics that are typically covered in an introductory course in
feedback control. It is emphasized, however, that some of the detailed derivations or descriptions of concepts are necessarily
brief and the reader is expected to conduct more in-depth study into any topics by exploring the many cited references.
It is also expected that the methods introduced in this chapter can be used to return to earlier chapters to explore how
many of the “open-loop” systems studied in a modeling context can now be studied using classical and/or state-variable con-
trol. Some of the problems included at the end of this chapter demonstrate both the model and control system development,
while some also simply review basic control system analysis techniques.

7.8 Problems
Problem B-7.1 Manual automobile steering system Develop a feedback control system block diagram for a vehicle’s
heading direction that includes the manual automobile steering system. Identify elements that perform the function
of actuator, process, sensor, and controller and identify the controller output signal, process output signal, actuator
output signal, controller output signal, reference signal, and error signal. Many of these elements and signals will be
intrinsic to the human driver.

Problem B-7.2 Water level control with float valve Develop a feedback control diagram for a basic water level control
problem (such as in a storage tank and a toilet). An on/off float valve should be assumed as the sensing and actuation
mechanism. On/off controllers commonly use a differential gap characteristic. Research this concept and describe how
it applies to this system. Identify elements that perform the function of actuator, process, sensor, and controller and
identify the controller output signal, process output signal, actuator output signal, controller output signal, reference
signal, and error signal.

Problem B-7.3 Drebbel’s incubator Mayr [134] describes Cornelius Drebbel as “the inventor of the first feedback
mechanism of the West” for a temperature-regulating system devised for an egg incubator. A schematic is shown in
Figure B-7.3 .

Damper

Water Push-rod
jacket
Temperature-controlled
region
Alcohol
Mercury

Heated gases Fire

Figure B-7.3 Schematic of Drebbel’s incubator as described in Mayr.

Source: [134]/Massachusetts Institute of Technology.

Drebbel’s furnace is formed by a fire box (furnace) with a flue at the top and a damper to allow passage of air (hot
gases, smoke). Within the fire box is a double-walled incubator, and the hollow walls are filled with water that transfers
heat evenly to the incubator. The water jacket and incubator can be assumed to have the same temperature. The sys-
tem exhibits feedback control as follows. The water temperature is monitored by a temperature sensor formed using
a glass vessel filled with alcohol and mercury. As fire heats the water, the alcohol expands and forces the mercury
(which remains a fluid) to push the float-push-rod/linkage up, lowering the damper and thus causing the fire to burn
less. As the water temperature falls, the alcohol contracts, causing the damper to open and inducing the fire to grow.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
462 7 Modeling Feedback Control Systems

A desired temperature level can be set by adjusting the length of the riser, which sets the opening of the damper for a
given expansion of the alcohol.
Given this description, a feedback control system block diagram can be developed taking water jacket temperature
as the controlled variable. Assume the water jacket has a net input heat as the difference between the furnace heat and
heat that leaves via the walls of the incubator/jacket. The furnace is thus an actuator element with inputs from the
fire and the valve opening, which is adjusted by the riser/pushrod/lever mechanism. Elements in this diagram can be
modeled as nonlinear functions of input variables. For example, the water jacket has net heat in with water temperate
as output.
Identify elements that perform the function of actuator, process, sensor, and controller and identify the controller
output signal, process output signal, actuator output signal, controller output signal, reference signal, and error signal.

Problem B-7.4 Watt’s flyball governor Assume a linear transfer function has been developed that relates the output
lever adjustment, zs , from a spring-loaded governor to an engine’s shaft speed, as indicated in Figure B-7.4.24 Research
the classical configuration in Watt’s governor-based speed control of a steam engine, where the governor output zs
adjusts a valve lever to regulate fuel supply and thus speed [134]. Complete the feedback control diagram initiated in
Figure B-7.4 by making suitable assumptions and explain how an error function is formed. Identify elements that per-
form the function of actuator, process, sensor, and controller and identify the controller output signal, process output
signal, actuator output signal, controller output signal, reference signal, and error signal.

Flyballs

Sleeve
Engine dynamics
Horizontal lever
motion

Valve Valve lever Driven by rope

To engine and pulley
inlet Steam from engine
Rope and
Exhaust
pulley,

(a) (b)

Figure B-7.4 (a) Schematic of Watt’s flyball governor. (b) Preliminary layout for closed-loop feedback control
diagram for engine speed control per Watt’s approach.

Problem B-7.5 Position control of inertial mass using potentiometers A simple feedback control is to be employed
to position a large rotational inertia in response to a reference input position, 𝜃i as shown in Figure B-7.5 using a
permanent-magnet DC (PMDC) motor. The load inertia combines the motor rotational inertia and an effective load
inertia in Jm , with the connecting shaft very stiff. Damping in the motor is considered negligible. An electronic (ideal)
amplifier with gain provides an output to drive the motor armature circuit as Va = Ka (Vo − Vi ), regardless of the current
drawn by the motor, here indicated by im .

Input Output
potentiometer potentiometer
+

Electronic
amplifier
gain =
–
= –

Figure B-7.5 Basic PMDC motor feedback servomechanism using resistive

potentiometers.

24 See Problem B-4-40.

 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
7.8 Problems 463

The voltages Vo and Vi are from the output and input rotational (resistive) potentiometers, respectively. These
potentiometers are used to detect the output angle, 𝜃o , and the input (command) angular position, 𝜃i . Assume
the potentiometers are identical and each has total resistance R. The angle can vary from a minimum, 𝜃min , to a
maximum, 𝜃max , with the total angular displacement less than one revolution. It is desired to have the rotational
inertia rotate back and forth within this range of motion. Assume the potentiometer has resistance Rmin = 0 Ω at 𝜃min
and Rmax = R Ω at 𝜃max .
a) Determine a potentiometer constant, Kp , which depends on the total “swipe” angle (𝜃max − 𝜃min ) and the total
resistance, R, and which relates Vi − Vo to the angle difference, 𝜃i − 𝜃o . This constant has units of volts/angle.
b) Show that the transfer function relating 𝜔m to Va can be expressed, 𝜔m ∕Va = Km ∕(𝜏m s + 1).
c) Find the close-loop transfer function relating 𝜃o to 𝜃i in terms of Kp , Ka , Km , and 𝜏m .
d) Given Ka = 200 V/V, potentiometer constant, Kp = 1 V/deg, motor time constant, 𝜏m = 1 sec, motor steady-state
gain, Km = 0.01 (∘ /sec)/V, what is the value of the steady-state gain for the closed-loop transfer function?

Problem B-7.6 Integrating motor torque–speed curves into linear feedback control models Example 7.1 showed
how the steady-state torque–speed curves for a permanent-magnet DC (PMDC) motor, as in Figure B-7.6 (a), can
be integrated into a motor transfer function model. Subsequently, a feedback control model as in Figure B-7.6 (b)
can be constructed.
Describe how a similar approach can be followed for motor models that have nonlinear torque–speed characteristics,
such as for the two-phase AC motor shown in Figure B-7.6 (a). Explain how the key block diagram elements and/or
constants would be defined in such cases.

PMDC motor 2-phase AC motor

+ + 1
+
–

(a) (b)

Figure B-7.6 (a) General motor torque–speed curves. (b) Block diagram motor torque–speed curve with
rotational inertia load.

Problem B-7.7 dc tachometer bridge The schematic in Figure B-7.7 depicts a dc tachometer bridge, which is con-
structed by connecting two tachometers back-to-back. Differences (or errors) in the angular velocities can be detected
by the output (error) voltage [137].

Tachometer 1 Tachometer 2
+ +

1 – – 2

“Error voltage”

Figure B-7.7 Schematic for a dc tachometer bridge.

Assume that each tachometer can be modeled by an ideal gyrator with modulus rt . For the ideal case where there
are no losses, build a bond graph and assume both 𝜔1 and 𝜔2 are specified inputs and that there is no current drawn
at the output voltage connection. Show that the output voltage Vo is proportional to the velocity difference.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
464 7 Modeling Feedback Control Systems

Problem B-7.8 Control system with unity feedback Find the response of y when r is a unit step (Figure B-7.8 ).

+ 10
+1 2
–

Figure B-7.8 Closed-loop control with unity feedback.

Problem B-7.9 Position control with velocity feedback For the position control system shown in Figure B-7.9 , deter-
mine the step response of 𝜃o to a step input 𝜃i .

+ 100
+2
–

0.1 + 1

Figure B-7.9 Closed-loop control

diagram for position control system.

Problem B-7.10 Performance characteristics with velocity feedback For the system in Figure B-7.10 , determine the
value of k such that the damping ratio of the closed-loop system is 𝜁 = 0.5. Then obtain the rise time, tr , peak time, tp ,
maximum overshoot, Mp , and the settling time, ts , for the response in y to an unit step in r.

+ 16
+ 0.8
–

1+

Figure B-7.10 Closed-loop control

with velocity feedback.

Problem B-7.11 Positive feedback The effect of “positive” feedback is commonly associated with systems “going unsta-
ble.” For the system shown Figure B-7.11 , determine if there is a condition that ensures stability.

+
+1
+

Figure B-7.11 Closed-loop control

diagram with positive feedback.

Problem B-7.12 Depth regulation using ballast control system A depth control system for a submersible vehicle is
proposed as in Figure B-7.12 . The controller proposed is a PD controller, and a second-order system plant model
is suggested.
a) Find the open-loop transfer function, and, for 𝜏d = 1∕3, determine the open-loop poles and zeros.
b) Neatly sketch or plot (using software) a root locus for this system.
c) Find a value for Kc that will set one of the closed-loop poles at s = −3.5
d) You are asked to consider a PID controller in comparison with the PD controller. Explain which controller is better
for this system. Make sure to be clear in what way it is “better” (that is, be specific about the criteria you use). You
may use “rough” root locus sketch(s) or use software to explore options.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
7.8 Problems 465

=0
+ 5
1+ 2 +4 +5

1.5

Figure B-7.12 PD control of a submersible.

Problem B-7.13 Buoyant force control for underwater vehicle By regulating the volume of air space, a buoyant force,
Fb , acting on an underwater vehicle can be kept proportional to its depth, zs ; that is, the buoyant force is Fb = Kb zs ,
where Kb is a constant.
a) Formulate a simple model for purely vertical motion of the submerged vehicle of mass m. Assume it can be modeled
as a point mass and ignore any friction/losses in vertical motion. Discuss how this depth control works and suggest
any changes to the control as necessary.
b) Consider integrating a system that manages the filling and emptying of the ballast so that the buoyant force is now
given by Fb = 𝜌gV, where V is the displaced volume (filled with air). Formulate a closed-loop control system that
integrates this form of buoyant force generation given the system model as in (a) and proposed a proper control
structure. Assume feedback is available from direct measurement of the depth, zs .

Problem B-7.14 Active suspension for vehicle ride control In Figure B-7.14 , a force actuator is connected in parallel
with spring and damping model elements that represent the passive suspension. The forces on the suspended mass (m)
are those due to a spring (k), a damper (b), and the controlled force from the actuator, Fc . Refer to solved Problem A-2-3
which examines the passively-suspended mass moving over a surface. Similarly, Figure B-7.14 shows a mass translat-
ing horizontally in the x-direction while vertically supported by the spring–damper–actuator suspension. The surface
elevation varies with the position traveled, x, according to a known ground surface profile, zg (x).
a) Develop a bond graph of the open-loop system plant and show that the states are the vertical velocity of the sus-
pended mass, 𝑣m , and the spring deflection, xk , with inputs 𝑣g = ż g and Fc (controlled force).
b) Write an output equation for y = am and then derive the transfer functions am ∕𝑣g and am ∕Fc .
c) Use the results from (b) to create an open-loop block diagram with input Fc and including 𝑣g as an input disturbance.
d) Develop a closed-loop transfer function incorporating the feedback from the accelerometer measurement as shown
in Figure B-7.14 .
e) Derive the transfer function am ∕𝑣g for the closed-loop system.
f) For m = 1, b = 2, k = 2, and K𝑣 = 2, identify the closed-loop poles and zeros and conduct a root locus analysis
treating Ka as a gain of interest.
g) For the values in (f), determine a value for Ka that will give a damping of 𝜁 = 0.707.
h) The actual force actuator is found to have first-order system dynamic behavior, Fc ∕u = K∕(𝜏s + 1), with gain K and
time constant, 𝜏, and so, u = (Ka s2 + K𝑣 s)zm . Rework steps (d)-(g) in this case.

1/

=
+
+
Ideal force actuator

Figure B-7.14 Active suspension for vehicle ride control.

 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
466 7 Modeling Feedback Control Systems

Problem B-7.15 Vehicle heading angle control The system shown in Figure B-7.15 is a proposed controller for the
heading angle of a self-guided vehicle. The reference, r, is the desired heading angle, K is a gain, Gm = 5∕(s2 + 2s + 5), is
a motor/controller system model that drives 𝛿, the steering angle of a front wheel, Gp = 1∕(s(s + 3)) represents a model
for the (heading) vehicle dynamics, and H = 1 (ideal, unity feedback). The output is controlled heading angle, 𝜓.
a) Determine the open-loop transfer function, GH.
b) If K = 2, what is the static velocity error constant, K𝑣 ?
c) It has been proposed that a steady-state error to a ramp input of 5% can be achieved by increasing K by a factor of
30 from the case in (b). Explain why this would work.
d) Neatly sketch the Bode plots for GH for the case in (c). Indicate basic factors and a composite sketch for both
magnitude and phase.
e) Show that the phase margin indicates the system is not closed-loop stable? Find the gain margin and gain crossover
frequency as well.
f) A lag compensator is proposed to maintain good steady-state response but to improve stability. Find a new gain
crossover frequency where you can achieve a desired phase margin of about 65∘ . What is this new gain crossover
frequency?
g) The next step in a lag compensator design requires finding the change in gain required at the new crossover fre-
quency (to make the magnitude 0 dB). What is that gain change?
h) Analysis and design: complete the final steps in a lag compensator design and use a step simulation to indicate the
results.

Figure B-7.15 Block diagram for vehicle heading

angle control.

Problem B-7.16 Inverted pendulum on a cart – PD control A PD controller was recommended for stabilizing an inverted
(simple) pendulum on a cart system in Problem A-7-7.
Consider the system with the following parameter values: mc = 1000 kg, m = 200 kg, and l = 10 m.
a) Sketch the closed-loop block diagram.
b) Derive the closed-loop transfer function.
c) Determine values of Kp and 𝜏d assuming the values provided for m, mc , and l.
d) For the case of r = 0 (regulating the angle at 𝜃 = 0∘ C), solve for the initial condition response assuming 𝜃(0) = 10∘ C.
e) Extended analysis: compare the efficacy of the PD controller using the same Kp and 𝜏d when controlling the linear
model versus the fully nonlinear model case. This simulation will need to be conducted using a Runge–Kutta algo-
rithm. Study various values of initial pendulum angle. It may also be worth examining the control force required to
achieve stabilization.

Problem B-7.17 Modeling for temperature control using embedded heating element A system is designed to maintain
a block of material (e.g., aluminum) at a specified temperature using an embedded heater that has resistance, Rh . When
current flows through the resistive heater, it generates heat, qh = 𝑣 ⋅ i = Rh i2 (Figure B-7.17 ).
a) Begin with the constitutive relation for a solid (see Chapters 2 and 3 discussions on thermal element modeling) and
explain the assumptions you’d make to model the solid as a lumped linear thermal capacitance, Cm .
b) Develop a bond graph that models all the significant power flows influencing the rate of change of internal energy
in the block. Assume that heat transfer to the ambient can be modeled by, qa = Ka (Tm − Ta ). Explain your under-
standing of all these terms. What is Ka ? (conceptually) Apply causality to this bond graph and show that the entropy
state for the thermal capacitance of the material, Sm , is a state of the system.
c) Derive a state equation first in terms of the entropy of the block material, Sm , and then show that you can convert
this equation into a state equation in terms of the temperature, Tm . You should find that this equation is linear in
Tm , but the current is a nonlinear term:
Km Ṫ m + Ka Tm = Rh i2 + Ka Ta .
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
7.8 Problems 467

Uniform block
temperature Conductive walls

Ambient
temperature
Controlled
electrical
source

Figure B-7.17 Thermal block temperature control.

d) It is desired to regulate Tm at an equilibrium value of Tmo . Since the heat generating function, qh = Rh i2 , let the
equilibrium current, io , be the value required to maintain the material at Tmo . Find a relation for io . Assume that
the following parameters are provided for this problem: Rh = 2Ω, Ka = 1.46 × 10−6 W/K, Cm = 0.00146 J/K (K =
Kelvin). What is io if Tmo = 100 ∘ C and Ta = 50 ∘ C?
e) To obtain a linear model, linearize the nonlinear current relation in the temperature ODE of part (d) about the
quiescent (equilibrium) point (Tmo , io ). The resulting linear equation should take the form:

Km Ṫ m + Ka Tm = qho + Kh Δi + Ka Ta ,

where Δi = i − io and qho = Rh i2o (the equilibrium heat required to maintain at Tmo ).
f) What is the time constant for the linearized system?

Problem B-7.18 Temperature control using embedded heating element An automatic temperature control system is to
be used for the system studied in Problem B-7.17. Assume material temperature is being measured using a thermocou-
ple which has a transfer function,

T̂ m 1
Ht (s) = = ,
Tm 𝜏t s + 1

where 𝜏t is the thermocouple time constant, and T̂ m is the (calibrated) measured temperature. The measured temper-
ature is compared with a set point temperature, Tr , and then amplified to produce a drive current, i = K𝑣 (Tr − T̂ m ).
a) Draw the closed-loop feedback control diagram and label all the blocks and signals.
b) Derive the closed-loop transfer function relating the controlled temperature, Tm , to the set point temperature,
Tr . Include the ambient temperature, Ta , as a disturbance input (as well as the equilibrium heat, qho ). Use only
P-control.
c) The closed-loop transfer function should indicate that the system is second order. Find the natural frequency and
damping for the closed-loop system.
d) Assume the system has been regulating temperature at 100 ∘ C (Tmo ). Develop a simulation for the following sce-
narios. Specific plots will be requested here in an updated form of this document. You will need to assume some
parameter values for this step. Work this step for the linear case and consider the following: Case 1, At time t = 0,
the ambient temperature increases by 20 ∘ C; Case 2, At time t = 0, the reference temperature is increased by 20 ∘ C.
e) Now implement the controller in step (d) on with the nonlinear plant to determine if the linear controller works
effectively.
f) Would there be any benefit in using a PI, PD, or PID controller for this system?

Problem B-7.19 Belt-drive control system The belt-drive system shown in Figure B-7.18 (a) is driven by a dc motor with
a torque–speed curve as shown. The indicated stall torque is given by, 𝜏s = rm Vs ∕Rm − 𝜅m 𝜔m , where rm is the torque
constant, Rm is the motor armature resistance, and 𝜅m is the slope of the torque–speed curve. A steady-state model of
the motor (see Chapter 4) shows that 𝜅m = rm 2 ∕R + B , where B is a linear effective motor damping. Assume the
m m m
combined motor and shaft/pulley inertia is Jm .
a) Adopt a steady-state model for the dc motor as conveyed by the bond graph model below and consider the pulleys/-
belt can be modeled by the equivalent schematic of Figure B-7.18 (b). Show that the indicated causality allows the
motor to be modeled using the given (steady-state) torque–speed model.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
468 7 Modeling Feedback Control Systems

+
– +

(a) (b)

Figure B-7.18 (a) A tape (or belt) drive system and (b) equivalent schematic.

R: R:

S 1 G 1

b) Use the bond graph to show that the two state variables are the effective belt (rotational) displacement, 𝜃b , and the
load inertia rotational velocity. Derive the state equations.
c) Find a transfer function that relates the load velocity, 𝜔L , to the input voltage, Vs (t).
d) Configure a closed-loop control for the plant transfer function derived in (c) and considering a controller transfer
function, Gc = Vs ∕eV = K, where K is an amplifier gain, and eV = Vr − Vb , with Vr a voltage reference input and
feedback voltage Vb = Kf ⋅ 𝜔L , where Kf is the gain of a speed sensor. Derive the closed-loop transfer function 𝜔L ∕Vr .
e) For the parameter values Jm = 1 × 10−4 kg m2 , JL = 0.05 kg m2 , KL = 201.8 N m/rad, BL = 0 N m sec/rad, rm ∕Rm =
0.25 N m/A, Ke = 0.05 N m/(rad/sec), and Kf = 0.01, show that the control system in (d) results in an oscillatory
response for 𝜔L for an positive value of K.
f) Propose and evaluate compensator designs that may resolve the oscillation problem.

Problem B-7.20 Boat stability gyro A large flywheel is used to create an anti-roll gyro for installation on a boat as shown
in Figure B-7.20 . An onboard IMU (not shown) provides a measure of the (boat) roll angle, 𝜙, and is used as feedback to
regulate roll induced by disturbance loads. This is accomplished by using a precession motor as shown, which pitches
the main gyro fore and aft (about its y or pitch axis) according to a proportional control, 𝜔y = K𝜃 ⋅ 𝜃, where K𝜃 is a
constant gain and 𝜔y is the pitch velocity. Assume that the moment of inertia of the boat about the pitch axis, Iy , is very
large compared to that about the other axes. Refer to Section 4.7.3 for insight into gyroscopic coupling.
a) Sketch a block diagram of the proposed roll control system which should incorporate the external roll torques as
disturbance inputs.
b) Determine the transfer function, G = 𝜔x ∕𝜏x , where 𝜔x is the roll velocity and 𝜏x is the roll torque (disturbance) on
the boat.
c) Make any observations about how well the system can be designed to regulate roll, particularly the relativity stability.
Inner gimbal Boat body-fixed
(rotates relatives axes
to chassis)

Spinning
rotor (roll angle)

Precession motor
External roll (fixed to boat chassis)
(disturbance) torques

Figure B-7.20 Ship anti-roll control.

 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
7.8 Problems 469

Problem B-7.21 Magnetic levitation control The schematic in Figure B-7.21 depicts an electromagnetic coil used to
suspend a mass (that may be a permanent magnet) as studied in Problem B-4-31.
a) As in Problem B-4-31, assume the forces on the mass are given by Fem = Gi2 ∕xg2 + Fpm (xg ), as indicated in
Figure B-7.21 , and derive the state equations, assuming the current, i, is specified as an input. The state equations
should be linearized about an equilibrium gap, xgo , and assumed equilibrium current, io , to show that a transfer
function relating the gap, xg , to the drive current, i, is given by,
xg Ko
= ,
−i s2
− a2
where Ko and a are defined for the equilibrium values of xg and i.
b) The block diagram below shows a compensator Gc in the forward path to specify the drive current, −i, to regulate
xg about the equilibrium.

+ –
+ 2 2
–

=1

c) Derive the closed-loop transfer function, xg ∕r.

d) Determine conditions on the system parameters (a, K, z, p, Ko ) to guarantee closed-loop system stability.
e) For the regulation case where r = 0, assume Ko = 2, and a = 1. Specify values for z and p based on the work in part
(d).
f) Find a value of K that will give a steady-state error of 5% for a step input in r.

due to if is
current PM

Air gap FBD:

Suspended
mass
Magnetic
flux lines

Figure B-7.21 Mass suspended by electromagnet, with inset showing

electromagnetic force due to current and effect of permanent magnet.

Problem B-7.22 Magnetic levitation control with disturbance response Reconsider the magnetic levitation system in
Problem B-7.21. The control system described in the block diagram does not account for external (uncontrolled) dis-
turbance forces.
a) Show how the control system model can be reconfigured to account for disturbance forces, Fd . Make sure to begin
with the original physical system model.
b) Redraw the block diagram with disturbance forces indicated.
c) For a regulation case where r = 0, assume Ko = 2, a = 1, z = 2, p = 3. Using the K as found in Problem B-7.21,
determine the steady-state value of xg for a unit step response in the disturbance, Fd .
d) Conduct a root locus analysis to refine the K value and to improve the system relative stability.
e) Simulate the system for a unit step disturbance input.

Problem B-7.23 Model of overhead crane (or gantry) A schematic for a simplified model of an overhead crane (or
gantry) system is shown in Figure B-7.23. The slider of negligible mass is controlled to move with velocity 𝑣c (t) along
a rail, which has negligible friction. The crane is assumed to have a point mass, mb , at the end of a rigid link of link L
having negligible mass/inertia. There is no damping in the pivot. The intent is to control the position of the mass, mb ,
using the velocity 𝑣c (t).
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
470 7 Modeling Feedback Control Systems

Slider
Rail

Figure B-7.23 Overhead crane or gantry system.

The nonlinear state equations are as follows. For the angular motion of the crane:
1 𝑣 (t)
𝜃̇ = 𝜔 = h − c cos 𝜃.
mb L2 L
The momentum equation is,
ḣ = −mb 𝑣c (t)L𝜔 sin 𝜃 − mb gL sin 𝜃.
For tracking the translational position of the mass, we introduce,
ẋ m = 𝑣mx = L𝜔 cos 𝜃 + 𝑣c (t).

a) Develop a set of linearized equations for small 𝜃 and for zero nominal velocity of the slider.
b) Use the linear state-space equations to find the transfer function xm ∕𝑣c .
c) For mb = 10 000 kg and L = 50 m, plot the poles and any zeros of the transfer function, xm ∕𝑣c .

Problem B-7.24 Control of overhead crane (or gantry) For the overhead crane, do the following.
a) Draw a closed-loop control block diagram assuming xm can be directly (and ideally) measured and the input is a
reference position xmr . Label all the blocks and signals.
b) Derive the closed-loop transfer function for the general case of PID control of xm for a controller of the form,
( )
1
Gc (s) = K 1 + TD s + .
TI s
c) Consider the case for PD control. Can this controller stabilize the mass motion (that is, would you be able to adjust
the gains to control how the mass is brought to a “stop”)? Study how the poles of the system change for PD control
(use the system parameters given in the original problem; set a value of TD = 1, look at effect of K). If this controller
does not work, then consider PID.
d) Develop a simulation of the LINEAR system with a controller (any PID form). Test with a reference step input of
5 m. Try to tune the controller to achieve minimal overshoot and a settling time of about 5 seconds. Report the
controller gain(s).
e) As a final test, implement the linear controller you used in step (d) to control the nonlinear overhead crane system
(as given in previous problem). Don’t use a step input. Instead, shape the reference input in the form, xmr (t) =
xmro tanh(t∕a), to achieve a relatively smooth step change over 2 seconds (the value of a can effectively “stretch”
the hyperbolic tangent function over time). Let xmro = 5 m. Try to tune the controller to achieve minimal overshoot
and a settling time of about 5 seconds.
f) If you completed both (d) and (e), compare/discuss briefly any observations.
NOTE: the next three problems have analogous physical plant models.

Problem B-7.25 Two-tank sewer system control Consider the two-tank sewer system with controlled pumping studied
in Example 7.13. This system is also studied in Chapter 5 as Problem B-5-23.
[ ]T
a) Consider the model given in Example 7.13, with states ordered as x = V 1 Qp V 2 . As in Example 7.13, set all
parameter values to unity.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
7.8 Problems 471

b) Show that the system is controllable.

c) Study observability assuming only V 1 is measured; that is, C = [1 0 0].
d) Show that the TF relating state V 1 to the control input Qc (t) is,
V1 −1
= 2 .
Qc s(s + 2)
e) Use root locus to show that the system cannot be stabilized for any gain, K.
f) Design a full-state feedback (regulator) control assuming all states. Assume that you want to place the three poles
for the controlled system at: −2.12 ± j2.12 and −2.12. Find the gain matrix, K.
g) Use MATLAB to solve for the initial condition response given assuming V 1 (0) = 1, Qp (0) = 0, and V 2 (0) = 0.5.
Compare the “uncontrolled” initial condition response of the system to the response obtained using the regulator
state-space feedback control design from (e). For the controlled case, also plot the actuator output, Qc = −Kx.
h) Assume the system is at equilibrium with V 1 = V 2 = 0 and the pipe flow is zero, Qp = 0. If the disturbance flow Qd
varies sinusoidally (minimum of zero to peak of 1, so there is a mean value), simulate the response to show how
well the state-space feedback can regulate the system at equilibrium.
i) Independent case study: investigate whether a compensator design can stabilize this system using a SISO frequency
domain approach (loop shaping). Compare to the state-space approach.

Problem B-7.26 Two-car train The two-car train system in Figure B-7.26 is also studied in Chapter 5 as Problem B-5-21.
[ ]T
a) Complete all the steps in the state-space model as in Problem B-5-21 and order the states as x = 𝑣1 xk 𝑣2 . For
the remainder of the steps, set all parameter values to unity.

1 2

1 2

Figure B-7.26 A two-car train

system with driving force, Fc (t), and
load force, Fd (t).

Repeat steps (b)–(h) as in Problem B-5-23; however, now the controlled variable is 𝑣1 , the controlled force is Fc , and
the disturbance force is Fd .

Problem B-7.27 Two-disk spring coupled system with control For the two-disk spring-coupled system in Figure B-7.27
(also studied in Problem B-5-22), complete the steps as in Problem B-5-23. In this case, the variable to be controlled
is 𝜔1 .
[ ]T
a) Complete all the steps in the state-space model as in Problem B-5-22, and order the states as x = 𝜔1 𝜃k 𝑣𝜔2 .
For the remainder of the steps, set all parameter values to unity.

1 2

1 2

Figure B-7.27 Two

rotational inertias (disks)
coupled by a spring with
driving torque, 𝜏c (t), and
load torque, 𝜏d (t).
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
472 7 Modeling Feedback Control Systems

Repeat steps (b)–(h) as in Problem B-5-23; however, now the controlled variable is 𝜔1 , the controlled torque is 𝜏c , and
the disturbance torque is 𝜏d .

Problem B-7.28 Pressure-compensated variable-displacement pump driving hydraulic motor with load.25 An
open-loop model for a variable-displacement hydraulic pump and motor drive is described in Problem A-7-5.
The system in Figure B-7.28 shows that system with a feedback mechanism to automatically adjust the yoke angle,
𝜙i , in order to regulate the system pressure (due to transient behavior, leakage, disturbances, etc.). If the manual control
shuts off the supply to the load, the pressure will increase, adjusting the yoke so that the pump flow decreases to a very
low level (matching leakage),26 but the pressure is regulated at a desired level. If the manual control is opened, there
will be flow to the motor and the pressure drops, but then the yoke is adjusted to increase pump flow and bring the
pressure back up. There will be a transient period whenever pressure regulation occurs that depends on the system
characteristics and the yoke mechanism. Assume the yoke mechanism has rotational inertia Ji , effective rotational
stiffness, Ki , and damping, Bi . In addition, the effective torque arm has length Li and the compensation piston has area
Ai , as shown in the detail of Figure B-7.28 (b).

Variable-displacement Regulated
Manual Gear ratio
pump with pressure
pressure feedback control

Pressure-
Load inertia and
adjusting
damping
screw
Yoke angle,

(a) (b)

Figure B-7.28 (a) Pressure-compensated variable-displacement pump driving a rotational

hydraulic motor and load. (b) Detail of the yoke mechanism with pressure-compensation feedback.

The physical feedback mechanism imposes a torque on the yoke due to a net pressure acting on the piston due
feedback pressure, Pf , the set pressure, Pset , and equilibrium adjustment, Pza ; that is, 𝜏ci = Ai ⋅ Li ⋅ Pnet , where Pnet =
Pset + Pza + Pf .
a) Extend the model of the system (referring to the bond graph from Problem A-7-5) but now introduce the physical
effects of the pressure-compensation feedback. In addition to the previous dynamic states, those due to the dynamic
effects of the yoke mechanism (inertia, spring, damping) should be accounted for, as well as the feedback effect.
Identify states using causality on the extended model.
b) Derive system state equations.
c) Formulate a feedback control diagram using the extended model.
d) Conduct a simulation of the baseline model using the parameters provided in solved Problem A-7-5, but also the
following:
Ji = 0.1 load mass moment of inertia, lb-in2
Ki = 100 rotational stiffness, lb-in/rad
Bi = 0.1 rotational damping, lb-in-sec/rad
Ai Li = 0.5 effective combined piston area, length, in3
Pset = 1000 control pressure, in2
Pza = 1.11 equilibrium adjustment pressure, in2 (per [114])
Initial conditions for all states should be determined based on: (a) pump input is at 𝜔p = 180 rad/sec, (b) pressure
settings are as specified above, and (c) the yoke is not moving, so it’s velocity is initially zero (but actual angle is not
known).

25 This problem is adapted from Doeblin [114].

26 These systems have replenishing pumps that maintain the contained system volume.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
7.8 Problems 473

Simulate from the initial state assuming the manual vale is closed and then opened at t = 0.003 seconds. Simu-
lation should be run until a steady state is reached with the valve opened, solving for the regulated pressure, the
motor load inertia angular speed, and the position of the yoke mechanism, 𝜙i over time. The accuracy of the pressure
regulation should be assessed, as well as time to settle.
e) Re-evaluate the line assuming the fluid lines have been increases so the compliance increases by a factor of 10 and
the damping in the yoke mechanism decreases by about 50%.
f) Are there any conditions under which this system may go unstable? Explain how you would go about making this
assessment.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
475

Multiport Modeling and Energy Methods

This chapter introduces modeling with multiport energy-storing and dissipative elements. Multiports are necessary when
an energy-storing or dissipation effect in a physical system needs to be modeled using more than one variable or state.
This means, for example, that C and I elements require more than one power port because there are n > 1 distinct ways to
“pump” energy into a device or process. This concept was introduced in Chapter 3 using a simple cantilever beam, such as
the one shown in Figure 8.1(a). In this case, it is possible to store elastic energy in the beam by independently deflecting its
tip along one axis, say x, as well as by twisting the tip about y-axis (out of the page) by angle 𝜃 (slope at beam tip). Indeed,
one could also twist the beam along its length (z-axis) as shown in Figure 8.1(b). In this case, there could be stored energy
that depends on three displacement variables, (x, 𝜃y , 𝜃z ), requiring a 3-port C.
It is important to distinguish how this type of model abstraction is different from single-port elements. While a linear
translational spring, for example, has two physical connections, the connections do not define distinct power flow ports.
The velocity of each end is distinct, but the force is the same at each end (see Chapter 3, Figure 3.9).
In addition, multiports require an n-dimensional constitutive relation that relates the key variables. Hence, a multiport
I-element is defined by f = Φ( p), where it should be recognized that each flow, fi , depends on n distinct momentum states,
pi , or fi = Φi (p1 , p2 , … , pn ). If the device or process under study does not call for such a description, it is not a multiport.
Similarly, a multiport R element would require identifying a dissipative function, e = Φd (f ). Note that there can be some
situations where a collection of single-port elements and an interconnecting junction structure can be combined into a
multiport. Karnopp et al. [38] refer to these as implicit fields, distinguishing from those cases where the field is explicit. The
term field is commonly used in bond graph literature to refer to either of these multiport representations.
In addition to properly identify the need for a multiport, it is necessary to formulate the constitutive behavior. In some
cases, the system lends itself to direct formulation of the constitutive relations based on direct application of fundamen-
tal physical laws and relations for a given case. Often, however, energy methods can be especially useful when modeling
certain types of systems with coupled energy storage and dissipation. As such, this chapter introduces a more systematic
use of energy methods in modeling and analyzing physical systems. A key part of modeling with multiport systems is for-
mulating relations to model the constitutive behavior. These methods will be illustrated using a broad range of examples.
The approach is shown to be especially useful for understanding electromechanical (EM) and magnetic devices and sys-
tems. In addition, as Lagrangian methods can be a very effective and efficient energy-based approach for a wide class of
systems, we introduce how they can be integrated with a bond graph approach. Lagrange methods can be especially use-
ful when dealing with complex mechanical systems, notably those that have dependent kinetic energy storage elements.
Finally, variational and minimization principles are reviewed to provide an additional approach for arriving at estimates
for constitutive properties, such as when elements have nonideal geometric characteristics.

8.1 Multiport Concept and Usage

An energetic basis was used in Chapter 1 to illustrate modeling power flow into system elements with multiple ports. This
concept was the basis for defining the basic bond graph elements in Chapter 3. Up to this point, however, model elements
with multiple ports (i.e., multiports) have primarily been used to represent the ideal 0 and 1 junctions shown in Figure 8.2.
Indeed, while junction elements as well as two-port transformers and gyrators are strictly multiports, they are used to
represent power-conserving effects in systems that neither store or dissipate energy.

Modeling of Physical Systems: Simulation and Control, First Edition. Raul G. Longoria and Joseph J. Beaman.
© 2025 John Wiley & Sons Ltd. Published 2025 by John Wiley & Sons Ltd.
Companion Website: www.wiley.com/go/Modeling_of_Physical_Systems/1e
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
476 8 Multiport Modeling and Energy Methods

(a) (b)

Figure 8.1 (a) A 2-port C to represent coupled elastic (potential) energy storage in a beam due to tip loads. (b) Combined loading on
a beam that would require a three port C.

Figure 8.2 The common (a) effort and (b) flow junctions are multiports that
neither store nor dissipate energy via the power flows, i = ei fi , at the n ports.

(a) (b)

Table 8.1 Summary of multiport C, I, and R element bond graph representations and key relations,
with constitutive relations for the i port indicated by function, Φs,i (), where the subscript “s”
indicates these are single-valued functions (see Chapter 4).

Potential storage Kinetic storage Dissipation

Out of necessity, two-port C and R elements were introduced in Chapter 3 in order to model coupled thermodynamic
effects. This necessity arises for a wide range of practical problems in all energy domains, especially when energy is stored
or dissipated as a result of multiple power/energy interactions. A more complete introduction and application of multiport
modeling will be given in this chapter. The definitions of multiport C, I, and R elements are merely extensions of those
already given in Chapter 3 and summarized in Table 8.1. Here bold notation is used to represent vectors in functional
relations.

8.1.1 Energy-Storing Multiports

The need for using a multiport model element as opposed to a single-port form was briefly addressed in Chapter 3.
For energy-storing elements, a key first step is to determine the number of independent states required to quantify the
stored energy. For a multiport C element, the stored potential energy can be written as,

(q) = Uq = e ⋅ dq, (8.1)

∫
where this integral represents an n-dimensional line integral over the n generalized displacements of the ports of a C
element. A potential coenergy can also be defined as,
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
8.1 Multiport Concept and Usage 477

Ue = q ⋅ de. (8.2)
∫
The constitutive relation, e = Φs (q), must be known to evaluate the energy, Uq , or the inverse relation, q = Φ−1
s (e), for the
coenergy, Ue .
Recall the cantilevered beam from Chapter 3, shown again if Figure 8.3(a). The physical “port” connections are at the tip
of the beam in the form of translational and rotational variables. Considering only the potential energy-storing effects of
the beam, the constitutive relations for the two-port C model shown in Figure 8.3(b) would take the form,
[ ] [ ][ ]
F 2EI 6 −3L x
= 3 .
𝜏 L −3L 2L2 𝜃
These relations convey how the force and torque each depend on both states: F = F(x, 𝜃) and 𝜏 = 𝜏(x, 𝜃). The potential
energy can be shown to be, within an arbitrary constant,

Ux𝜃 = Fdx + 𝜏d𝜃 + constant =

6EI 2 6EI
x − 2 x𝜃 +
2EI 2
𝜃 + 
constant.
∫ ∫ L3 L L
The inverse relation is,
[ ] [ ][ ]
x L 2L2 3L F
= ,
𝜃 6EI 3L 6 𝜏
and the coenergy to within an arbitrary constant is,
L3 2 L2 L 2
UF𝜏 = F + F𝜏 + 𝜏 .
6EI 2EI 2EI
Recall the two-port C is used to represent internal energy storage for a thermodynamic substance in Chapter 3. In that
case, the constitutive relations were the equations of state and related temperature and pressure to volume and entropy in
the form, T = T(S, –V ) and P = P(S, –V ), respectively.
The cantilever beam and thermodynamic substance represent system elements that would be difficult to model using
an interconnection of single-port elements and junction structure. A multiport representation is essential.1 In some cases,
it can be convenient to condense multiple single-port C elements connected by lossless junction structure into a multiport
form. Systems in the former case are sometimes referred to as implicit “fields” [38]. A common example is for the system
shown in Figure 8.4(a), which can be modeled either as in Figure 8.4(b) using single-port elements (implicit field) or as an
explicit multiport as in (c).
Note that the states for the two-port in Figure 8.4(c) would be x1 and x2 , which would not be the same as the states xa
and xb for the model in (b). This should be evident since ẋ b = ẋ 2 − ẋ 1 . The compact form of the model in part (c) might
be desirable in some cases, but it does obscure the true nature of the system. In any case, this example serves mostly to
illustrate how multiports can be formed from interconnected assemblages of basic elements.

Figure 8.3 (a) Cantilever beam with two physical ports. (b) Two-port
representation.

(a) (b)

(a) (b) (c)

Figure 8.4 (a) Two-springs with multiple inputs. (b) Bond graph (implicit field) (c) explicit multiport form.

1 These cases are also referred to as explicit fields or multiports in the bond graph literature [38].
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
478 8 Multiport Modeling and Energy Methods

Example 8.1 Deriving two-port constitutive relations

Find the constitutive relations for the two-port of Figure 8.4(c).
Solution
The constitutive relations can be derived by considering the form of the bond graph in Figure 8.4(b). From port 1,
F1 = Fa − Fb = ka xa + kb xb = ka x1 + kb (x2 − x1 ),
since ẋ 1 = ẋ a , ẋ 2 = ẋ b + ẋ 1 , and taking integration constants to zero. Further, F2 = Fb = kb xb = kb (x2 − x1 ). In summary,
[ ] [ ][ ]
F1 k + kb −kb x1
= a ,
F2 −kb kb x2

(a) (b) (c)

Figure 8.5 (a) Three capacitors in Π-configuration: (b) implicit bond graph and (c) multiport form.

The two models for Figure 8.4 had the same number of states, but this may not always be the case when converting to
a multiport. Indeed, systems may have energy-storing elements interconnected so there is dependent causality. In these
cases, it can be convenient to eliminate a derivative causality by converting to an explicit multiport form. Consider the
Π-configuration of capacitors in Figure 8.5(a) with the bond graph in (b). Depending on the causality at bonds 1 and 2,
the bond graph in (b) may give 3 or 2 states. The equivalent two-port in Figure 8.4(b), however, will at most have only two
states depending on causality applied. Solved Problem A-8-1 provides details on how to form the two-port model in Figure
Figure 8.5(c).
While converting implicit fields to multiport forms is possible and useful in some cases, it is not strictly necessary.
Multiports for kinetic energy (I form) in mechanical systems typically are derived from implicit fields for rigid bodies [38].
An example is the multiport model of a rigid beam shown in Figure 8.6, which for purposes of this example is assumed to
have planar motion with small rotation. The explicit multiport I can be formulated as shown in Figure 8.6(b). In this case,
the kinetic energy takes the form,

( p) = Tp = f ⋅ dp, (8.3)
∫
where this is a line integral over the n port momenta and the kinetic coenergy is,

Tf = p ⋅ df, (8.4)
∫
where the use of “f” in the subscript is used to distinguish coenergy.
For such a uniform and rigid beam constrained to move in a plane, the constitutive relations are given as,
[ ] [ ][ ]
𝑣1 2 2 −1 p1
= ,
𝑣2 m −1 2 p2
and the kinetic energy is,
2 ( 2 )
Tp1 p2 = p − p1 p2 + p22 .
m 1

(a) (b)

Figure 8.6 (a) Planar rigid beam with applied forces. (b) Two-port I representation.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
8.1 Multiport Concept and Usage 479

The inverse constitutive relation is,

[ ] [ ][ ]
p1 m 2 1 𝑣1
= ,
p2 6 1 2 𝑣2
and the kinetic coenergy is,
m( 2 )
T𝑣1 𝑣2 = 𝑣1 + 𝑣1 𝑣2 + 𝑣22 .
6
The origin of these relations is from an implicit model of the rigid body in planar motion with assumed small angle motion
[38].
The only other common multiport I model may be for the electrical transformer with mutual inductance. That case is a
direct extension of the electrical inductor model which has a constitutive relation of the form, 𝜆 = Li. For a multiport, the
flux linkage and current variables form vectors and the inductance matrix has cross-terms defined by the mutual inductance
values. The reader is referred to Karnopp et al. [38] for further discussion as well as specific use cases in [32, 154].

8.1.2 Dissipative Multiports

The concept of a two-port R was introduced in Chapter 3 in the context of showing how any single-port R element also has
an inherent thermal port. Heat transfer processes can be modeled as 2-port R elements. The multiport R shown in Table 8.1
shows that a multiport form can be used to model physical processes that: (i) are inherently dissipative and (ii) require a
constitutive relation in the form of a vector algebraic relation of the form e = 𝚽R (f). It is emphasized that the multiport
R is one where, for example, the effort at one port would depend on multiple (independent) flow and/or effort variables
(depending on the causal from). For example, for a basic two-port with flows as inputs,
e1 = Φ1 (f1 , f2 ),
e2 = Φ2 (f1 , f2 ).
These relations can arise for either implicit or explicit multiports, and both forms may couple dissipative processes from
different energy domains. The use of a multiport R may be suggested on a first examination of a system when there appears
to be dissipative coupling in a device or process. The multiport form can help identify possible relations that may need to
be determined experimentally. A deeper examination of the problem may reveal that one or more variables does not need
to considered. In any case, the model basis provides guidance for the development and or design of experiments.
For example, consider a controlled ablation process that uses an electrically powered plasma [155]. These processes are
used in tissue surgery, where a pump is used to draw tissue and byproducts through a gap. Optimizing the design of this type
of device and improving the controlled operation can be aided by the simplified model shown in Figure 8.7. The multiport
R suggests a coupling between the electrical, fluid, and the thermal power flows. The mechanical motion of the device
would impact the effective gap between the device and targeted tissue, 𝛿, which is modeled here by modulation of the R.
Constitutive relations for this 3-port R might take the form,
i = Φ1 (Vs , ΔP, T, 𝛿)
Q = Φ2 (Vs , ΔP, T, 𝛿)
fs = Φ3 (Vs , ΔP, T, 𝛿),
where ΔP is the hydraulic pressure drop across the effective gap/wand/tissue, and T is the working temperature.

Tube Pump
Pulse
generator

Peristaltic
pump

(a) (b)

Figure 8.7 (a) Schematic of a controlled ablation concept powered by a pulsed generator with voltage Vs (t) and controlled peristaltic
pumping. (b) Proposed 3-port R for modeling the coupled dissipative processes.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
480 8 Multiport Modeling and Energy Methods

Implicit R fields can be used to more compactly represent an interconnection of single-port R elements by junction struc-
ture [38]. The constitutive relations for the related explicit field or R multiport will follow from the individual single-port
relations, and these will inherently require resolving algebraic loops. As long as the single-port elements in the implicit
field have proper dissipative constitutive relations (see Chapter 4), the multiport form will be dissipative. This follows since
the power dissipated at a single port is equal to the heat generated: d = e ⋅ f = T ⋅ fs ≥ 0. The second law of thermody-
namics imposes the restriction that the dissipated power is non-negative, and a corresponding restriction is placed on the
constitutive relation as well (see Section 4.6). It should follow that as long as any single-port R element models are intercon-
nected properly with ideal junction structure, then any implicit R field should lead to a proper dissipative set of constitutive
relations.
Only for the case where all the single-port R elements in an implicit field have linear constitutive relations can linear
resistive or conductive matrices be derived for the explicit form. That is, these would assume specific inputs on all the ports
as follows:
e = R ⋅ f (resistive) or, f = G−1
c ⋅ e (conductive),

where Gc = R−1 . If these matrices are given, they provide a way to check that the multiport is dissipative since, for example,
the R matrix must be positive definite. This follows since the dissipated power would be, d = fT e and for the linear case,
e = Rf. Thus, d = fT Rf > 0 requires R is positive definite (or positive semidefinite for cases where a system includes
gyrators; see [38]). Refer to Karnopp et al. [38] for a discussion on the former case (implicit) and relation to Onsager and
Casimir forms [156, 157].

8.1.3 Summary
A wide range of physical processes and devices can be modeled with multiports. Identifying whether a multiport is essential
begins by making proper assumptions, reticulation of a system under study, and using any provided physical informa-
tion or given relations. As done for single-port elements, lumped-parameter multiports extend how we can reticulate a
system into separate energy-storing and dissipative elements. Proper junction structure can then be used to interconnect
additional effects. The port variables for a multiport may be from different energy domains, as the multiport forms an
energetic-coupling. The form of the multiport will imply the need for proper constitutive relations. These may not be explic-
itly known at the time the model is formed but the key point is that the port variables that need to be related are thus
identified. This knowledge is a key initial step in the modeling process.
The use of multiports and formation of models with multiports will be discussed in subsequent examples. In addition,
this chapter will introduce and/or review methods from discipline specific areas commonly used in particular to derive
constitutive relations. In particular, we will make use of energy methods for these purposes.

8.2 Causality and Constitutive Relations for Multiports

The causality assignment on single-port C, I, and R elements interconnected as implicit fields follows the successive causal-
ity assignment procedure (SCAP). There may be cases where imposed causality leads to some energy-storing elements
having all integral, all derivative, or a mixed combination of causality. And in some cases, this may motivate creation of
an explicit form to simplify repeated use, especially if derivative causality can be subsequently eliminated. This section
deals primarily with illustrating causality assignment on explicit multiports and identification of states. Some particulars
on derivation of constitutive relations for multiports are also discussed and then examples follow to illustrate the use of
multiports in bond graph modeling.

8.2.1 Causality on Multiports

The successive causality assignment procedure should continue to be applied even for systems that include multiport ele-
ments. For multiport energy-storing elements, each port should be treated independently of the others since each has a
distinct state variable. Thus, it possible to complete causality on energy-storing multiports so that the n ports have com-
plete integral, complete derivative, or a mixed causality. This is summarized for the C multiport in Table 8.2. A similar table
to that for the C multiport is constructed as Table 8.3 for the I multiport.
If the interconnection of a multiport in a system results in any of the ports having derivative causality (for example, for
effort specified into a C element port), then the state associated with those ports will not be states of the system. This requires
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
8.2 Causality and Constitutive Relations for Multiports 481

Table 8.2 Complete integral, complete derivative, and mixed causality options for C multiports.

Complete integral Complete derivative Mixed

Table 8.3 Complete integral, complete derivative, and mixed causality options for I multiports.

Complete integral Complete derivative Mixed

inverting the constitutive relations, which may not necessarily be an algebraically tractable task. For this reason, it is often
preferred to enforce complete integral causality on energy-storing multiports when possible. This means that the causality
assignment procedure may initially focus on multiport assignment when possible.
Dissipative R multiports follow similar assignment as for single-port R elements. Again, some algebraic manipulation
may be required depending on final form of the causality on each port. It is most common (and physically meaningful) that
the temperature-entropy port has effort imposed, as shown in Table 8.4. The other bonds with general effort-flow pairs will
have causality imposed based on the system causality assignment.

Table 8.4 Resistive, conductive,

∑ and mixed causality options for R multiports, where total
power dissipated d = ei fi = T ⋅ fs .

Complete resistive Complete conductive Mixed

 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
482 8 Multiport Modeling and Energy Methods

Figure 8.8 An elastic cantilevered beam with attached barbell-like tip load.

Example 8.2 Barbell-loaded cantilever beam

A cantilevered beam loaded by a “barbell-like” load at its tip is as shown in Figure 8.8. The barbell can both translate and
rotate, thus inducing a combined force-torque load on the beam. This beam can be modeled as a 2-port C, and linear beam
theory shows that the constitutive relation between the force (F) and the torque (𝜏) at the tip can be related to displacements
x and angle 𝜃 by,
[ ] [ ][ ]
F 2EI 6 −3L x
= 3 , (8.5)
𝜏 L −3L 2L 2 𝜃

where E is the elastic modulus, L is the length, and I is the area moment of the beam [15].

a) Develop a bond graph that integrates the beam with the barbell, including the effect of gravity. Assume planar motion
and account for linear losses in the rotational motion of the barbell. Assign causality to the final bond graph and identify
system states.
b) Derive the system state equations.
Solution (a)
The bond graph for the system is shown in Figure 8.9. The barbell is modeled essentially by two ideal pendula of mass m
[ ]T
with rigid links of length l. With causality applied as shown, the system is found to have four states: x = pm , x, 𝜃, h .

Figure 8.9 Bond graph for cantilever beam as 2-port C with barbell
tip load.

Solution (b)
With causality as assigned, the four state equations are derived as follows. For the pm ,

ṗ m = −F(x, 𝜃) + 2mg,

where F(x, 𝜃) is from 2-port beam constitutive equation 8.5. The beam deflection state, x, equation is,

ẋ = 𝑣m = pm ∕(2m).

Then, from the rotational side of the multiport, the barbell momentum equation is,

ḣ = −𝜏(x, 𝜃) − 𝜏B = −𝜏(x, 𝜃) − Bh∕(2ml2 ),

and for the beam tip slope, 𝜃,

𝜃̇ = 𝜔 = h∕(2ml2 ).

It is helpful to note how multiports are integrated with single-port elements. The same methods introduced in Chapter 3
to structure a bond graph are essential, here making note of the fact that the distinct translational and rotational velocities
are represented by 1-junctions that can then be used to interconnect the elements on each side of the 2-port C. Note also
the convenience of having integral causality on both ports of the 2-port, enabling direct use of the constitutive relations in
functional form.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
8.2 Causality and Constitutive Relations for Multiports 483

8.2.2 Multiport Constitutive Relations

Some physical devices or processes may seem difficult to model without using a multiport element. For example, many
elastic beams such as the cantilever in Example 8.2 would be difficult to model using single-port elements. As we’ll see in
Section 8.3, many electromechanical devices have physical coupling that requires a multiport form. There are some simpli-
fied forms that have been used in the past in some cases, but these do have inherent limitations (see, e.g., Busch-Vishniac
and Paynter [158]). Often, selection of a model is dictated more by a legacy formulation and/or analysis approach, or the
availability of parameter data that fits a representation. In any case, a multiport form does require properly identifying its
need in the system model reticulation. As mentioned earlier, it is important to identify that the multiport assumes each port
has independent variables and, in the case of energy-storage, an associated state. Those cases presented up to this point have
satisfied this basic requirement and the constitutive relations have been arrived at using “first principles.” For example, for
basic beam models with two ports, the two coupled constitutive relations can often be derived using classical superposition
principles (e.g., see Burr [42]). In many cases, however, it can be much more effective, if not necessary, to rely on energy
methods.
Generalizing energy storage elements to a multiport form dictates that the energy function now depends on a state vector,
x, or  = (x) = (x1 , … , xn ). This implies that we can effect a unique change in the energy Δi by work at the ith port by,
say, the product, yi ⋅ Δxi . This is one way to interpret the fundamental derivative relations (first introduced in Chapter 4),
yi = 𝜕(x)∕𝜕xi , as defining the constitutive relation at port i of an energy-storing multiport.2 Indeed, the energetic basis
provides a way to use derivative relations to determine constitutive relations from either an energy or coenergy function,
which more fundamentally follow directly from the energy expressions 8.1–8.4; that is:
𝜕Uq
ei = , i = 1, … , n (efforts from potential energy), (8.6)
𝜕qi
and,
𝜕Ue
qi = , i = 1, … , n (displacements from potential coenergy), (8.7)
𝜕ei
and, for kinetic energy and coenergy, we have,
𝜕Tp
fi = , i = 1, … , n (flows from kinetic energy), (8.8)
𝜕pi
and,
𝜕Tf
pi = , i = 1, … , n (momenta from kinetic coenergy). (8.9)
𝜕fi
These energy-based relations form the foundation for methods in deriving constitutive relations. Examples of their use
were first shown in Chapter 4 (e.g., see Solved Problem A-4-5) and their use in discipline specific methodologies will be
illustrated in Sections 8.3 and 8.4. Before demonstrating their use in the following examples, it is important to relate this
discussion to Maxwell reciprocity, which provides a second restriction to impose on constitutive relations for energy-storing
elements (also in Section 4.6.1).
While the derivative relations 8.6–8.9 are used to help derive constitutive relations, Maxwell reciprocity provides a basis
for “checking” constitutive relations derived for energy-storing relations. This holds for either implicit or explicit forms.
Maxwell reciprocity takes the form,
𝜕yi 𝜕yj
= . (8.10)
𝜕xj 𝜕xi
The Maxwell reciprocity provides a quick check on the correctness of a derived set of constitutive relations. For the linear
case, the capacitive C or inertive I matrices in,

q = Ce (capacitive), (8.11)

f = Ip (inertive), (8.12)

2 See Section 4.6.1 in Chapter 4.

 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
484 8 Multiport Modeling and Energy Methods

respectively, will be symmetric. This is evident in Example 8.2, where the stiffness matrix for the 2-port cantilever in
equation 8.5 is symmetric. Maxwell reciprocity provides a quick way to check either derived linear or nonlinear consti-
tutive relations, helping find algebraic errors, sign errors, etc. If the reciprocity condition is not satisfied, this implies that
the constitutive relations are not conservative, from an energy standpoint. This means the implied energetic coupling may
add or dissipate energy, thus violating the model assumptions. Corresponding restrictions for dissipative multiports were
discussed in Section 8.1.2 and previously in Chapter 4. The following examples illustrate the use of multiports in basic mod-
els and how energy methods can be used to complete constitutive relations. Additional ways to use energy methods will be
described in Sections 8.3 and 8.4 for electromechanical applications and when applying Lagrange equations, respectively.

Example 8.3 Two-dimensional spring–mass system

Consider the two-dimensional spring–mass system shown in Figure 8.10(a) with a linear spring. Derive the x and y forces
imposed on the mass by a two-port C, shown in the bond graph of Figure 8.10(b). Assume the spring has initial length lo .
Then derive the state equations for this system.

(a) (b)

Figure 8.10 (a) Two-dimensional spring pendulum system (dashed line tube used to indicated constrained motion along length of
spring). (b) Bond graph with 2-port C element.

Solution
The potential energy stored in the 2-port C can be expressed as,
[√ ]
1 1
Uxy = k1 (r − lo )2 = k1 x2 + y2 − l2o ,
2 2
where lo is the free length of the spring. Using the derivative relations yields,
[ ]
𝜕Uxy lo
Fx = = k1 x 1 − √ ,
𝜕x x2 + y2 ]
[
𝜕Uxy l
Fy = = k1 y 1 − √ o .
𝜕y x + y2
2

Note that,
𝜕Fx 𝜕Fy [ ]−3∕2
= = k1 xy x2 + y2 ,
𝜕y 𝜕x
which is a direct consequence of Maxwell reciprocity. Given these constitutive relations derived from potential energy, the
state equations can be derived using standard bond graph techniques. From the causality indicated, there are four states:
[ ]T
x = px , x, py , y . The corresponding state equations can be found as,
( )
l
ṗ x = −Fx = −k1 x 1 − √ 2o 2 ,
x +y
ẋ = px ∕m, ( )
l
ṗ y = mg − Fy = mg − k1 y 1 − √ o ,
x2 + y2
ẏ = py ∕m.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
8.2 Causality and Constitutive Relations for Multiports 485

Example 8.4 Tank with a float

Consider next a tank with a float depicted in Figure 8.11. In this system, the displacement of the mass is assumed to cause
significant changes in the tank’s water height, thus implying there is coupled energy storage. Use energy methods to derive
the constitutive relations for the tank and use these relations to obtain the state equations for the system with source
pressure Ps (t) as an input.

Long pipe

(a) (b)

Figure 8.11 (a) Hydraulic system, tank with float. (b) Bond graph with 2-port C element.

Solution
In order to use the derivative relations, it is first necessary to calculate the potential energy in the multiport C. Now energy
is a state function, so the value of the function depends only on the present state values (x and V – in this case) and not on
past history of the states. This concept can be used to great utility in the calculations of energy.
Starting with an empty tank, we could conceive of the mass raised a height x with a corresponding energy of mgx.
With a fixed x and mass m not totally submerged, we could then fill the tank with fluid which results in a pressure at
the tank bottom of,
⎧ 𝜌g
⎪ A2 V–, V– < A2 x
Pc = ⎨ 𝜌g ( ) .
– − A2 x + 𝜌gx, V
⎪ A2 − A1 V – ≥ A2 x
⎩
– > A2 x as,
This relation can be used to calculate the energy stored in the system for a fluid volume V
– [ ]
A2 x
𝜌g V
𝜌g ( )
UV– x = mgx + –d–
V V+ – − A2 x d–
V V,
∫0 A2 ∫A2 x A2 − A1

which yields
𝜌g ( 2 )
UV– x = mgx + –V − 2A1 –Vx + A1 A2 x2 .
2(A2 − A1 )
Since UV– x is a state function, any path, not just the particular one conceived above, will yield the same result for the
energy. The force F and the pressure P can then be calculated as,
𝜕UVx 𝜌gA1 ( )
F= = mg + A x−V ,
𝜕x A2 − A1 2
𝜕UVx 𝜌g ( )
P= = V − A2 x ,
𝜕V A2 − A1
and the state equations are,
𝜌gA1 ( )
ṗ = −F = −mg − A2 −A1
A2 x − –V ,
ẋ = px ∕m,
𝜌g ( )
Γ̇ = Ps − RΓ∕I − A2 −A1
–V − A2 x ,
–̇ = Γ∕I,
V
where I is the fluid inertia and R is the fluid resistance.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
486 8 Multiport Modeling and Energy Methods

Example 8.5 Equivalent inertia

The system in Figure 8.12(a) is found to have a dependent piston mass energy storage, since the piston velocity is related
to the fluid flow rate via the piston are, as shown by the bond graph of Figure 8.12(b). Use energy methods to define
an equivalent fluid inertia element that accounts for the piston translational mass, as represented by the bond graph in
Figure 8.12(c).

Piston ram

(a) (b) (c)

Figure 8.12 (a) Hydraulic system with dependence between fluid and piston mass. (b) Bond graph with transformer. (c) Bond graph
with equivalent inertia derived using energy methods.

Solution
The two inertias in this model – fluid inertia in the pipe and the mass of the ram – are not independent, which can be inferred
from the bond graph of Figure 8.12(b). They can be efficiently combined into one equivalent inertia using energy methods,
simplifying the effort needed to obtain explicit state equations. The first step is to calculate the total kinetic coenergy in the
fluid and the piston-ram,
1 2 1
TQV = IQ + mV 2 .
2 2
If the energy is to be referred to the fluid domain, then using the kinematic constraint, V = Q∕A, reduces the kinetic
coenergy to,
1
TQV = (I + m∕A2 )Q2 ,
2
This coenergy can be used to derive an equivalent momentum per unit area from a derivative of the energy,
𝜕TQ ( )
m
Γeq = = I + 2 Q,
𝜕Q A
and therefore an equivalent inertia Ieq ,

Ieq = I + m∕A2 .

The system model is reduced to that of Figure 8.12(c). Now, using Γeq as the state variable for the system results in the
simple state equation,

Γ̇ eq = Ps (t) − RΓeq ∕Ieq .

This derivation should be compared with the standard equation derivation using the bond graph of Figure 8.12(b) and
state Γ,

Γ̇ = Ps (t) − RΓ∕I − Pp
Pp = Fp ∕A = m𝑣∕Ȧ = mQ∕Ȧ 2 ̇
= mΓ∕(IA2
)
( )
1 + m∕(IA ) Γ̇ = Ps (t) − RΓ∕I
2
( )
Γ̇ = Ps (t) − RΓ∕I ∕Ieq ,

which is equivalent to the much simpler expression in terms of Γeq .

 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
8.2 Causality and Constitutive Relations for Multiports 487

Example 8.6 Two degree-of-freedom spring-mass system

Use energy methods to derive the state equations for the two degree-of-freedom (DOF) spring–mass system shown in
Figure 8.13(a).

-field -field

(a) (b)

Figure 8.13 (a) Two degree-of-freedom (dof) spring–mass system and (b) bond graph with indicated C- and I-fields.

Solution
The kinetic coenergy contained in the two I elements is
1 1
T𝑣1 𝑣2 = m1 𝑣21 + m2 𝑣22 ,
2 2
and the potential energy contained in the two C elements can be expressed in terms of the displacements of the two
masses as,
1 1
Ux1 x2 = k1 (x1 − x2 )2 + k2 x22 .
2 2
Then the momentum of mass 1, p1 is,
𝜕T𝑣1 𝑣2
p1 = = m1 ẋ 1 ,
𝜕𝑣1
and momentum 2 is,
𝜕T𝑣1 𝑣2
p2 = = m2 ẋ 2 .
𝜕𝑣2
The force due extension x1 is,
𝜕Ux1 x2 ( )
Fx1 = = k1 x1 − x2 ,
𝜕x1
and force due to extension x2 is,
𝜕Ux1 x2 ( )
Fx2 = = −k1 x1 − x2 + k2 x2 .
𝜕x2
Equating time rate of change of momentum of the two masses to the forces applied to these masses then yields the
following equations in terms of energies:
( )
d 𝜕T𝑣1 𝑣2 𝜕Ux1 x2
=− + F1 ,
dt 𝜕 ẋ 1 𝜕x1
( )
d 𝜕T𝑣1 𝑣2 𝜕Ux1 x2
=− + F2 ,
dt 𝜕 ẋ 2 𝜕x2
or
m1 ẍ 1 = −k1 (x1 − x2 ) + F1 ,
m2 ẍ 2 = +k1 (x1 − x2 ) − k2 x2 + F2 ,
which represents a second-order form for the state equations. This derivation represents a simple form of Lagrange
equations, which will be discussed more fully later in this chapter.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
488 8 Multiport Modeling and Energy Methods

8.3 Electromechanical Systems Modeling

An engineering application area that is particularly suited for study by energy methods and the use of multiport models is
EM systems and devices. Of particular interest are four types of EM forces of electric origin that are of practical importance
[31]. These are forces on: (a) a free charge in an electric field, (b) a polarizable medium in an electric field (c), a moving
charge (current) in a magnetic field, and (d) magnetizable material in a magnetic field.
The design of systems and materials to control and direct EM forces is a main concern in EM system modeling. Methods
that use field descriptions based on the Maxwell equations form a foundation for determining resulting force densities but,
in general, this requires a deeper level of specialization in this area. Commercially available software is available to analyze
these systems if required, especially when the intent is to optimize geometric and material parameters.
The emphasis in this section is on lumped-parameter descriptions of EM devices that can dynamically interact with
other system components. We assume slowly varying fields for which these forces are either dominated by electric fields
(electroquasistatic or EQS) or magnetic fields (magnetoquasistatic or MQS), but not both simultaneously [159]. We adopt
multiport elements to represent these two classes of EM interaction, as shown in Figure 8.14. This section will further
describe these model elements and show how energy functions can be used to derive expressions for EM mechanical forces
and torques. These relations rely on derived (or directly measured) electrical or magnetic characteristics of the device.
Practical devices and system can be modeled by integrating the EM multiport with electrical and equivalent magnetic
circuit models, as will be shown in this section.
The reader is expected to have familiarity with electromagnetic concepts as might be introduced in an undergraduate
engineering physics course. Key variables and concepts that are used in this chapter were previously reviewed in Chapter 2
(Kirchhoff Systems), Section 2.3. For a more thorough review, refer to [2, 31, 159]. Basic magnetic circuit modeling will be
introduced in this section.

8.3.1 EM Capacitive Multiports

8.3.1.1 Electroquasistatic (EQS) Multiports
An EM system or device that can be modeled by an electroquasistatic3 or capacitive multiport system can generally have n
multiple electrical ports and m mechanical ports, as shown in Figure 8.14(a). In order to use such a model, the voltage and

(a) (b)

– +
+ –
Iron plunger

N turns

(c) (d)

Figure 8.14 (a) Bond graph representation of electric coupling. (b) Bond graph representation of magnetic coupling, with gyrational
coupling between circuit variables (𝑣, i) and field variables (M, 𝜑).
̇ (c) Prototype electric coupling device. (d) Prototype magnetic
coupling device.

3 The terms electroquasistatic and magnetoquasistatic are adopted from Haus and Melcher [159] and convey the inherent approximation used
for lumped-parameter electromechanics.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
8.3 Electromechanical Systems Modeling 489

force constitutive relations must be determined. These depend on both charge and displacement states,

Vi = Vi (q1 , … , qn ; x1 , … , xm ), (8.13)

Fj = Fj (q1 , … , qn ; x1 , … , xm ). (8.14)

The voltage relations are commonly assumed to be electrically linear and are either known or can be derived from material
and geometric characteristics. The force functions, on the other hand, are not typically known, but they can be found if we
can determine the total energy stored in the EM multiport,
∑
n
∑
m
Uqx = V ⋅ dq + F ⋅ dx = Vi dqi + Fj dxj . (8.15)
∫ ∫ i
∫ j
∫

Given this (scalar) energy function, the EM force at port j is given by,
𝜕Uqx
Fj = , j = 1, … , m. (8.16)
𝜕xj
However, the energy function in 8.15 cannot be found since the Fj are required. Fortunately, it is possible to calculate Uqx
by taking advantage of the path independence of conservative (i.e., lossless) energy functions [31]. To illustrate, consider a
two-port capacitive-plate device as in Figure 8.14(c). The electrical energy at a particular value of charge q and position x
can be calculated by directly summing VΔq + FΔx over a general path, as prescribed by equation 8.15. But, for EM systems,
it is often much more desirable to choose a particular path of integration that first assembles the system geometrically (path
OA), as illustrated in Figure 8.15(a), and then imposes the electric field (path AB).
Since energy is a state function and therefore path independent, this path will produce an identical value of energy as
any general path. On path OA, the electric field is turned off (dq = 0) and therefore EM force is identically zero and no
contribution to the energy calculation is obtained along this portion of the total path. Once point A is reached, there is no
further change in geometry (dx = 0) and all subsequent energy flows through the electrical port.
This principle enables us to determine the energy Uqx as if the system were entirely electrical. The idea then is that any
state of energy can be reached and as such a path can be taken where the unknown force terms can be made identically zero,
thus allowing the energy function Uqx to be found. The following example demonstrates the application of this methodology,
which is useful for a wide range of similar electroquasistatic multiport models.
In some cases, we may prefer to express the energy as an electric coenergy function,4

UVx = V ⋅ q − Uqx = q ⋅ dV − F ⋅ dx, (8.17)

∫ ∫
if we choose to use voltages as states, rather than charge (note the energy variable subscripts are efforts and displacements).
In this case, we would then derive force relations from,
𝜕UVx
Fj = − , j = 1, … , m. (8.18)
𝜕xj

General General
path path

(a) (b)

Figure 8.15 Path OAB used to calculate energy at position B for (a) electric system and (b) magnetic system.

4 The expression, UVx = V ⋅ q − Uqx is an example of a Legendre transformation, commonly used in analytical mechanics and thermodynamics
[8, 21].
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
490 8 Multiport Modeling and Energy Methods

Further, we would require that the following hold:

𝜕UVx
qj = , j = 1, … , n. (8.19)
𝜕Vj
We can use coenergy the same as an energy function. In addition, it is not necessary to transform all of the variables; a
mixture of charge–voltage variables could be used on multiport elements. Any “untransformed” variables can be evalu-
ated from the coenergy using a derivative relation as above. Note the negative sign when using a force–coenergy relation
(equation 8.18), where displacements xj represent untransformed states (mechanical costate is force Fj ).

Example 8.7 Electrostatic voltmeter

Electrostatic voltmeters can be constructed in a fashion similar to that in Figure 8.16(a). Obtain state equations for the
voltmeter that can be used to describe motion of the rotor for a change in input voltage Vs (t).

+
Stator
–

Rotor
rotor inertia

(a) (b)

Figure 8.16 (a) Schematic of an electrostatic voltmeter. (b) Bond graph model.

Solution
State equations can be obtained using standard methods once voltage Vc and torque 𝜏c are obtained as functions of charge
q and angle 𝜃 (2-port constitutive relations). Assuming the device is constructed so that fringing fields can be ignored, the
area of the plates is large compared to the shaft, and electric field is uniform between the plates, then electric displacement
field across the gap is,
q q
D = = 2 , 0 < 𝜃 < 𝜋∕2.
A r 𝜃
For linear isotropic material between the plates, the electric field is then,
1 q
E= D= 2 0 < 𝜃 < 𝜋∕2,
𝜀 𝜀r 𝜃
where 𝜀 is the permittivity, and the voltage is found as,
d 1
Vc = Eds = q= q 0 < 𝜃 < 𝜋∕2,
∫ 𝜀r 2 𝜃 C(𝜃)
where C(𝜃) is the angle-dependent capacitance. Electric energy can then be calculated as,
d 1
Uq𝜃 = 𝑣c dq = q2 = q2 0 < 𝜃 < 𝜋∕2,
∫ 2𝜀r 2 𝜃 2C(𝜃)
and therefore the EM torque, 𝜏c , is,
𝜕Uq𝜃 d
𝜏c = =− q2 0 < 𝜃 < 𝜋∕2.
𝜕𝜃 2𝜀r 2 𝜃 2
Note that the device is operated in regions in which there is significant but not complete overlap.
Now, taking advantage of the assigned causality, the state equations can be derived as,
d
q̇ = (Vs (t) − Vc )∕R = Vs (t)∕R − ,q
𝜀Rr 2 𝜃
𝜃̇ = h∕J,
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
8.3 Electromechanical Systems Modeling 491

( )
d q 2
ḣ = −𝜏 − K𝜃s − bh∕J = − K𝜃s − bh∕J,
2𝜀r 2 𝜃
𝜃̇ k = h∕J.
The state equations for angular displacements 𝜃 and 𝜃k are identical. The rationale for tracking both states is that each
angular displacement defines a distinct energy storage: one for mechanical potential energy in the spring and the other
helps quantify the electrical potential energy in the variable capacitance. Each requires defining a unique initial condition
to quantify energy in the system.
In this example, the capacitance, C(𝜃), was found from first principles, and note that 𝑣 = q∕C(𝜃) takes the form of an
electrically linear voltage relation; that is, it is linear in relating the electrical variables, but obviously nonlinear due to 𝜃. In
many practical cases, it is common to find that the position-dependent capacitance will be provided, in this way indicating
that the device is electrically linear and suitable for modeling as a two-port C.

8.3.1.2 Devices with Polarizable or Piezoelectric Material

The basic parallel plate capacitor can be used to guide modeling of many multiport EM devices that have energy stored in an
electric field (uniform fields and negligible fringing is assumed in lumped-parameter modeling). As shown in the previous
example, the capacitance often follows a form, 𝜖A∕d, where A is plate area and d distance between plates, for example.
While 𝜖 is often constant, generally any EM transduction can be formed when there is variation in material or geometric
parameters. As such, we can extend our use of a multiport C as a suitable model for common and practical devices that rely
on either polarizable or piezoelectric materials. We can extend equation 2.31 from Chapter 2 to include the effect of electric
polarization P [159],
D = 𝜖o E + P, (8.20)
to emphasize that the displacement flux density, D depends not only on external sources and sources within the material
(E) but also on the material polarization itself.
Consider electrets, so named by Heaviside in 1885 as counterparts to magnets [160], which refer to materials that can retain
a polarization for long periods of time. Electrets are able to provide a contained store of energy in an electric configuration.
In this way, devices can be formed that take advantage of the inherent active effect in electrets, not unlike how permanent
magnets are used in magnetic devices (see Sections 8.3.2 and 8.3.3). Piezoelectrics differ in that they establish an electric
field when a strain is applied. Electrodes are typically attached directly to piezoelectric materials to take advantage of this
capability for sensing or actuation. A review of how multiport modeling for both piezo- and pyroelectrics using a bond
graph perspective is given by Busch-Vishniac and Paytner [158]. See also the broad discussion by Busch-Vischniac [161]
for more details on how these types of materials are used in transduction applications. We can illustrate how polarizable
materials and piezoelectrics can be accounted for by the following example.

Example 8.8 Electret microphone

Consider an electret microphone formed by binding a polarized sheet with uniform polarization density of magnitude Po
to a fixed electrode, as shown in Figure 8.17(a). The upper electrode is electrically grounded and serves as a diaphragm for
the microphone, induced to move with velocity 𝑣(t). Use the proposed bond graph model in Figure 8.17(b) to determine
the relations between the voltage V across the sensing resistor, R, as a function of the total gap, h.
Solution
It is necessary to relate the macroscopic variables in the bond graph model (Vc , qc , etc.) to the properties of the electret
which require approximations based on the electric field relations. The dynamic charge qc needs to be related to the voltage

(a) (b)

Figure 8.17 (a) Schematic of an electret microphone. (b) Bond graph model.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
492 8 Multiport Modeling and Energy Methods

Vc , using Gauss’s law (see Chapter 2, equation 2.30), qc = ∮S D ⋅ dA, where D = 𝜖o E. In this case, qc = 𝜖o Ea A, where A is the
area of the diaphragm/electrode. Across the microphone, however, the potential difference is,
h
Vc = Ex dx = dEb + (h − d)Ea ,
∫0
where the electric fields are taken as uniform, since we assume that the polarization surface charge density is uniform and
the equipotential boundaries are plane and parallel. Taking the electric displacement field, D, at the interface as equal:
𝜖o Ea = 𝜖o Eb + Po (assuming the same permittivity). Thus, Eb = Ea − Po ∕𝜖o and,

Vc = d(Ea − Po ∕𝜖o ) + (h − d)Ea = hEa − dPo ∕𝜖o .

Now we can use the relation for qc = 𝜖o AEa to find the voltage relation we seek,
h dP
Vc = Vc (qc , h) = q − o,
𝜖o A c 𝜖o
revealing one of the required constitutive relations that motivates the 2-port C element in Figure 8.17(a). Note how the first
term takes the form qc ∕C(h), where C(h) = 𝜖o A∕h is a capacitance that depends on h. Note also that the second term can
be rewritten so we get the form,
h d
Vc = Vc (qc , h, qpo ) = q − q ,
𝜖o A c 𝜖o A po
where qpo = APo (which has units of charge). In this case, the 2-port C in Figure 8.17(b) could be a 3-port to include a static
polarization port with an imposed effort source.
In any case, the bond graph and relations as given are sufficient to provide a nonlinear set of state equations,
q̇ c = −i = −Vc (qc , h)∕R,
ḣ = 𝑣(t),

with output equation y = Vc (q, h) that can be solved by simulation. Alternatively, this system can be linearized to find a
transfer function, G(s) = Vc ∕h, which would reveal that the electret microphone has high-pass characteristics.

A three-port C as in Figure 8.18 is also recommended by Busch-Vishniac and Paynter [158] for mod-
eling piezoelectric transducers. This form can be reduced to the two-port form as well, and this model
is contrasted with classical circuit analog models in [158]. Of note is the limitation with circuit analog
forms, which do not allow assessment of piezoelectric materials in the model without considering a
Figure 8.18 A specific load. The multiport model is preferred in cases where a more complete picture of the trans-
three port C for the duction process is desired and especially to understand nonlinear behavior, which can be essential
electret microphone in design. The piezoelectric multiport can also be readily extended to other applications, such as in
in Figure 8.17.
Wangcharoenrung and Longoria [162] and Wangcharoenrung [163].

8.3.2 EM Magnetoquasistatic Devices

Modeling devices that are magnetoquasistatic can typically adopt one of the two approaches. The first takes advantage of
those cases where the inductance is known to vary with a displacement variable, for example, L(x) or L(𝜃). In this case, an
ideal energy-storing multiport model can usually be used that connects with an electrical drive system. When it is necessary
to account for more detail about the magnetic design of a system or device under study, magnetic circuit modeling can be
adopted along with multiport magnetomechanical elements. This latter approach is discussed in Section 8.3.3.

8.3.2.1 Lossless EM Multiport

A lossless EM multiport can be used to model EM interactions when a known position-dependent inductance is specified
or can be readily measured. As for electroquasistatic cases, an assumption is that the device is electrically linear. In this case,
a relation of the from, i = 𝜆∕L(𝜉), with 𝜉 either a translational or rotational displacement, is assumed to hold. For example,
for a “pull-in” solenoid as in Figure 8.19(a), the inductance is L(x), where x is the displacement of the solenoid plunger
about a center position, where the measured values of inductance (as viewed at the electrical terminal) reach a peak value.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
8.3 Electromechanical Systems Modeling 493

0.4
0.3
0.2
0.1
0.0
1 2 3 4 5 6 7 8

(a) (b)

Figure 8.19 (a) Schematic of a solenoid with position-dependent inductance, L(x). (b) Two equivalent multiport models, upper case
explicitly showing magnetic variables.

The inductance will tend to decrease as the plunger moves away from center. When the solenoid is “energized,” an EM
force, Fem , will tend to pull the plunger back to the center position.
This type of device can be modeled by considering power interactions at the electrical and mechanical ports. Two equiv-
alent bond graph models are shown in Figure 8.19(b). One version explicitly shows that the magnetoquasistatic device
inherently stores magnetic and mechanical energy in a two-port C. This form is useful when it is necessary and/or possible
to use known magnetic material properties and device geometry, and will be discussed in more detail later in this section.
For many practical cases where the inductance is specified or will be measured at the electrical port, the two-port “IC” ele-
ment is very convenient and sufficient. This multiport directly couples the electrical and mechanical ports used to integrate
with a system. The electrical port presents inductance behavior to an electrical driving circuit while the mechanical port
presents an EM “spring-like” effect.
The two-port constitutive relations for the IC take the form,

i = i(𝜆, x) = 𝜆∕L(x) (as given), (8.21)

Fem = F(𝜆, x) (to be determined). (8.22)

Note that this model can have multiple electrical and/or mechanical ports, such as when modeling a two-phase alternator.
Formulating the multiport model equations follows steps similar to those used for the EM multiport C. First, the device is
typically assumed to be electrically linear, so for a single electrical port we could use, i = 𝜆∕L(x). A force equation can be
found from an energy function using the derivative relation,
𝜕𝛌x
F= , (8.23)
𝜕x
where now,

𝜆x (𝜆, x) = i ⋅ d𝜆 + F ⋅ dx, (8.24)

∫ ∫
and “()” is the total stored energy, with one of the energy terms “kinetic” (in terms of a momentum state, flux linkage, 𝜆)
and the other “potential” (in terms of displacement state, displacement, x). Because of this hybrid form, Karnopp et al. [38]
have referred to the IC as a mixed energy-storing element. As done for the multiport C energy function, an expression for
(𝜆, x) is found by assuming an integration path in state space chosen to force unknown terms to zero. The resulting scalar
energy function can be used to derive the force or torque relation(s) as in equation 8.23.

Example 8.9 Solenoid-suspended mass

A solenoid plunger of mass m is electromagnetically suspended against gravity as shown in Figure 8.20(a). The solenoid has
an inductance L(x) = 2𝜇o a2 N 2 ∕(b + x), where x is the air gap distance, 𝜇o is the permeability of free space, N is the number
of turns, and a and b define geometry of the core material as shown. The air gap, x, changes as the plunger moves up and
down (within low-friction guides not shown). Assume the input to the coil is a voltage, V, and the coil has resistance Rc .
Assume the plunger is suspended only by the solenoid electromagnetic force, Fem .

a) Develop a bond graph model of the system, apply causality, and identify system states
b) Derive the state equations
c) Derive the electromagnetic force, Fem , required for the equations in (b).
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
494 8 Multiport Modeling and Energy Methods

Air gap

Plunger

(a) (b)

Figure 8.20 (a) Solenoid with plunger mass suspended in gravity. (b) Bond graph with a two-port IC representing the
magnetoquasistatic coupling.

Solution (a)
Since L(x) is provided, it is convenient to use an IC element to model the magnetoquasistatic coupling, as shown in the
[ ]T
bond graph of Figure 8.20(b). From the applied causality, the system states are x = 𝜆, x, pm .
Solution (b)
The state equations follow from the bond graph:
𝜆̇ = −R𝜆∕L(x) + V,
ẋ = p∕m,
ṗ m = −Fem (𝜆, x) + mg,
where Fem is the EM force induced by the solenoid.
Solution (c)
To find Fem (𝜆, x), we write the energy from equation 8.24,
 𝜆
𝜆 𝜆2 (b + x)
𝜆x (𝜆, x) = i ⋅ d𝜆 + F
 ⋅ dx = ⋅ d𝜆 = ,
∫ ∫ ∫0 L(x) 4𝜇o a2 N 2

where the second term has been made zero by selection of a proper path. Then, from equation 8.23,
𝜕 𝜆2
Fem = 𝛌x = .
𝜕x 4𝜇o a2 N 2
Note that the force will always be attractive, working to reduce the air gap.

Although flux linkage 𝜆 is a natural state variable to express magnetic energy, it can also be convenient to use current i,
as the costate variable, to express magnetic energy. In this case, the energy takes the form i,x
∗
(i, x), where the ()∗ indicates
the coenergy form,

ix
∗
= i ⋅ 𝝀 − 𝛌x = 𝝀 ⋅ di − F ⋅ dx. (8.25)
∫ ∫
In this case, the force relations would be determined using,
𝜕ix
∗
Fj = − , j = 1, … , m, (8.26)
𝜕xj
and the electrical terminal relations must obey,
𝜕 ∗
𝜆j = ix , j = 1, … , n. (8.27)
𝜕ij
As discussed for electroquasistatic elements, it is not necessary to transform all of the variables, and here we could use a
mixture of flux-current variables. Also, note again the difference in the sign for the force–coenergy relations (equation 8.26).
It is very common in the electromechanics literature to use the coenergy form, since current is often used as the state rather
than flux linkage (e.g., see Woodson and Melcher [31], etc.).
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
8.3 Electromechanical Systems Modeling 495

Example 8.10 Salient-pole machine

A schematic of a simplified salient pole machine is depicted in Figure 8.21(a). For details on this type of machine, see
Woodson and Melcher (pp. 150–165) [31]. The device shown consists of a stator of highly permeable magnetic material
with two coils wound and connected in series as shown. The machine also has a high permeability rotor that can rotate
with angle 𝜃. This machine is electrically driven by a sinusoidal current, is (t) = I sin(𝜔r t + 𝜋∕4). It is desired to calculate the
input mechanical energy per cycle when the machine is running at a constant speed, 𝜔r = 𝜃, ̇ such that 𝜃 = 𝜔r t. All losses
are to be neglected, and it has been determined experimentally that the flux linkage at the electrical port can be expressed,
𝜆 = (Lo + L2 cos(2𝜃))is (t).

(a) (b)

Figure 8.21 (a) Salient pole machine. (b) Bond graph with a two-port C representing the magnetoquasistatic coupling.
Solution
Since electric current is (t) and generator angle 𝜃 are given as inputs, it is natural to express the EM torque 𝜏em in terms of
these variables. This torque can be calculated from the magnetic coenergy, i𝜃∗ as,
𝜕i𝜃∗
𝜏em = − ,
𝜕𝜃
where coenergy can be calculated from the electric port relation as,
1
i𝜃∗ = 𝜆di = (L + L2 cos(2𝜃))i2 .
∫ 2 o
The EM torque is then,
𝜏em = L2 i2 sin(2𝜃) = L2 i2 sin2 (𝜃 + 𝜋∕4) sin(2𝜃).
since 𝜔r t = 𝜃. The input energy per cycle can then be calculated by integrating over one cycle (𝜃 from 0 to 2𝜋) as,
2𝜋
𝜋
cycle = 𝜏d𝜃 = L i (t)2 .
∫0 2 2s
Note that because current is specified as an input on the electrical port, neither 𝜆 (nor 𝜑) are state variables of this system.
This example is examined again in Example 8.20 using a Lagrange approach. This type of electrical machine is also studied
in Solved Problem A-8-7.

One form of a generalized EM (GEM) device model for representing practical EM machine types can be depicted as in
Figure 8.22. This machine consists of two stator (a,b) and two rotor (1,2) coils with the following assumptions: (1) lossless,
(2) unsaturated (electrically linear), (3) smooth rotor and stator (radially symmetric), (4) unity coupled (no leakage), and
(5) equal turn coils (N1 = N2 = Na = Nb ).
Under assumption (1), the behavior of the primitive machine is governed entirely by the nature of stored energy which
can be expressed as,
𝜆𝜃 = 𝜆𝜃 (𝝀, 𝜽),
with flux linkage vector,
[ ]
𝝀T = 𝜆1 𝜆2 𝜆a 𝜆b ,
or conversely we could use coenergy,
i𝜃∗ = i𝜃∗ (i, 𝜽),
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
496 8 Multiport Modeling and Energy Methods

(a) (b)

Figure 8.22 (a) Schematic representation of generalized electromagnetic machine (GEM). (b) Bond graph representation.

with current vector,

[ ]
iT = i1 i2 ia ib .
All subsequent developments are derived from this basic relation.
For electrically linear systems, the absence of saturation (assumption (2)) implies flux linkage is linearly related to current,
𝝀 = L(𝜃)i,
where L(𝜃) is an angle-dependent inductance matrix. Using this constitutive relation, coenergy can be expressed in the
quadratic form as,
1 T
i𝜃∗ = i L(𝜃)i.
2
The only assumed nonlinearity in the machine lies in the sensitivity of the inductance matrix to variations in shaft angle.
For the primitive machine, this sensitivity is expressible as simple direction cosines as we next discuss.
Under the additional assumptions (3) and (4), the variation of L with shaft angle, 𝜃, is simply prescribed as follows:
⎡ 1 0 cos 𝜃 sin 𝜃 ⎤
⎢ ⎥
0 1 sin 𝜃 − cos 𝜃 ⎥
L = L⎢ ,
⎢cos 𝜃 sin 𝜃 1 0 ⎥
⎢ sin 𝜃 − cos 𝜃 0 1 ⎦ ⎥
⎣
where L is a material- and size-specific constant. We observe two particular properties of this matrix:
1) It is symmetric (as it ought to be); attesting to the passive reciprocity of the underlying magnetic field.
2) It is furthermore symmetric with respect to rotor (1,2) and stator (a,b); reflecting the relativistic invariance of the field
(either rotor or stator or both could rotate).
From the coenergy state function i,𝜃∗
, we may derive shaft torque (corresponding to relative angular position 𝜃 between
rotor and stator) by taking the derivative of Ti𝜃 with respect to 𝜃,
𝜕i𝜃∗
𝜏=− ,
𝜕𝜃
but we have already observed that, through an assumption of electric linearity, the coenergy function becomes i𝜃∗ = iT Li∕2.
Consequently, torque is given by,
1 𝜕L
𝜏 = − it i.
2 𝜕𝜃
The angular derivative matrix 𝜕L/𝜕𝜃 is sufficiently recurrent to deserve a symbol of its own,
𝜕L
K= ,
𝜕𝜃
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
8.3 Electromechanical Systems Modeling 497

so we may write,
1
𝜏 = − iT Ki.
2
Corresponding to the inductance matrix of the primitive machine, we may evaluate the K matrix as,
⎡ 0 0 − sin 𝜃 cos 𝜃 ⎤
⎢ ⎥
0 0 cos 𝜃 sin 𝜃 ⎥
K = L⎢ .
⎢− sin 𝜃 cos 𝜃 0 0 ⎥
⎢ cos 𝜃 sin 𝜃 0 0 ⎥⎦
⎣
The resultant torque is then given by,
[ ]
𝜏 = L (i1 ib + i2 ia ) cos 𝜃 + (i2 ib − i1 ia ) sin 𝜃 .
Given flux linkage vector 𝝀 for this lossless machine, the corresponding voltage vector follows by simple time differentiation,
d𝝀
V= .
dt
Substituting 𝝀 = Li and using the operator s = d()∕dt yields,
V = (sL)i + L(si),
but
d𝜃 𝜕L
sL = = 𝜔K,
dt 𝜕𝜃
thus we obtain the final expression,
V = 𝜔Ki + L(si), (8.28)
where the first term on the right-hand side is the generated or speed voltages and the second term represents the induced or
transformer voltages. It has long been customary to designate the two terms on the right by these names. A similar situation
arises in fluid mechanics: convective acceleration (corresponding to speed voltage) and local acceleration (corresponding
to transformer voltage). As in the case of fluid motion, induced voltages occur only if currents are time variable (i.e., no
transformer effects exist for strict DC). On the other hand, the convective analog, speed voltage, exists even for steady
currents so long as K ≠ 0. This discussion may prompt one to perceive of the analogy to the case of uniform fluid flow.
For our generalized electrical machine, we may, in fact, display all these effects explicitly, K and L having been evaluated
above:
⎡V1 ⎤ ⎧⎡ 0 0 − sin 𝜃 cos 𝜃 ⎤ ⎡ 1 0
⎫
cos 𝜃 sin 𝜃 ⎤ ⎪ ⎡i1 ⎤
⎢ ⎥ ⎪⎢ ⎪ ⎥ ⎢ ⎥ ⎢ ⎥
⎢V2 ⎥ = ⎨⎢ 0 0 cos 𝜃 sin 𝜃 ⎥ 0 1 sin 𝜃 − cos 𝜃 ⎥ ⎪ ⎢i2 ⎥
L𝜔 + ⎢ Ls⎬ .
⎢Va ⎥ ⎪⎢− sin 𝜃 cos 𝜃 0 0 ⎥ ⎢cos 𝜃 sin 𝜃 1 0 ⎥ ⎪ ⎢ia ⎥
⎢V ⎥ ⎪⎢ cos 𝜃 sin 𝜃 0 0 ⎥⎦ ⎢ sin 𝜃 − cos 𝜃 0 1 ⎥⎦ ⎪ ⎢⎣ib ⎥⎦
⎣ b ⎦ ⎩⎣ ⎣ ⎭
Moreover, these relations may be consolidated into a quasi-impedance matrix as follows:
⎡V1 ⎤ ⎡ Ls 0 − sin 𝜃L𝜔 + cos 𝜃Ls cos 𝜃L𝜔 + sin 𝜃Ls ⎤ ⎡i1 ⎤
⎢ ⎥ ⎢ ⎥⎢ ⎥
⎢V2 ⎥ = ⎢ 0 Ls cos 𝜃L𝜔 + sin 𝜃Ls sin 𝜃L𝜔 − cos 𝜃L𝜔⎥ ⎢i2 ⎥
.
⎢Va ⎥ ⎢− sin 𝜃L𝜔 + cos 𝜃Ls cos 𝜃L𝜔 + sin 𝜃Ls Ls 0 ⎥ ⎢ia ⎥
⎢V ⎥ ⎢ cos 𝜃L𝜔 + sin 𝜃Ls sin 𝜃L𝜔 − cos 𝜃L𝜔 0 Ls ⎥ ⎢i ⎥
⎣ b⎦ ⎣ ⎦ ⎣ b⎦
Now we observe that speed voltages appear only if mutual induction terms (off diagonal terms) appear in the matrix (main
diagonal terms correspond to self-inductance). Alternatively, systems in this form can be practically solved by simulation.
Rearrange equation 8.28 and express in the time domain,
di
L(𝜃) = −𝜔K(𝜃)i + V, (8.29)
dt
where the functional dependence on 𝜃 is explicitly shown. Equations in this form can be solved using ordinary-differential
equation solvers as described by Shampine and Reichelt [164]. It is assumed that these would be integrated with state
equations for 𝜔 and 𝜃.
For more discipline-specific details on machine models of the type described in the foregoing, the interested reader can
refer to Fitzgeral et al. [32] or Krause et al. [165, 166].
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
498 8 Multiport Modeling and Energy Methods

8.3.3 Magnetic Circuit Modeling and Devices

Many magnetic devices and systems can be practically modeled using the methods described in Section 8.3.2. In some
cases, however, functions describing how inductance varies with mechanical displacements or rotations, and essential
for deriving force or torque relations, respectively, may not be known. In those cases, information about the magnetic
design of a system such as the specific use of magnetic core materials, coils, and permanent magnets is used to formulate a
model in the magnetic domain. Building magnetic circuit models is common in magnetic design [154, 167, 168], and these
concepts are introduced in this section. By interpreting with a bond graph basis, a nonexpert is able to integrate magnetic
system characteristics into unified models of magnetic devices. This calls for proper selection of magnetic power conjugate
variables, as will be discussed in the following. While refined design and optimization of magnetic devices may require
using finite-element software, lumped-parameter models are useful in system analysis and design. Woodson and Melcher
[31], Haus and Melcher [159], Watson [167], or Chai [168] can be referred to for more insight into magnetic devices and
their design.

8.3.3.1 Properties of Magnetic Materials

Magnetic material properties influence how power flows through magnetic structures in getting to and from electrical and
mechanical energy domains. Before discussing the mechanisms for interaction between these energy domains, it is helpful
to review some of the underlying variables used to describe magnetic material properties and their behavior. The insight
gained will provide insight into the use of lumped-parameter magnetic circuit modeling and adoption of magnetomotive
force, M, as an effort variable and rate of change of total magnetic flux, 𝜑̇ = f𝜑 , as a flow variable. The product of these
variables is a measure of power flow within magnetic circuit models, that is, M ⋅ f𝜑 . Power conjugate variables are essential
for unifying models across energy domains, as we are apt to do when building system models with bond graphs. This
energetic basis helps define basic modeling elements, and fortunately we can follow definitions from classical magnetic
circuit modeling.
First, recall that magnetic field variables B (magnetic flux density) and H (magnetic field strength) are defined at a point
in a magnetic material, and both quantities are vectors, B and H. These vector quantities are in parallel and proportional
in magnitude in free space but not necessarily for a given magnetic material. Thus, it is common to express B as due to a
free space component and one due to material,
B = 𝜇o (H + ) = 𝜇o H + Bi (units of T, or Wb/m2 ). (8.30)
In this relation,  is magnetization (density) (same units as H; compare to electric polarization in equation 8.20) and Bi is
called intrinsic induction5 with same units as B [167]. Magnetic materials are those that enable large B to be achieved with
relatively low levels of H. If the material is not magnetic, Bi is zero. For a linear material, Bi = 𝜇o 𝜒m H, where 𝜒 is called the
magnetic susceptibility [159, 167], which allows us to write,
B = 𝜇o H + Bi = 𝜇o (1 + 𝜒m )H = 𝜇o 𝜇r H = 𝜇H, (8.31)
which includes the definition of relative permeability, 𝜇r = 𝜇∕𝜇o (dimensionless), a parameter useful in comparing material
“magnetizability.” Often, it is the relative permeability that is referred to when citing material properties in this context.
For a given magnetic material, the magnetization curve is a plot of B versus H, commonly provided by material man-
ufacturers, and also called the B − H curve. The curves can vary depending on how they are measured including initial
state of the material and the sample geometry. It is assumed that the measurement is made by varying H quasistatically
for a sample of the material in a closed flux path. Typically, this is done using a toroid, as illustrated in Figure 8.23(a). The
induced (azimuthal) magnetic field strength and induced magnetic flux are both measured over repeated cycling to form
this classic characterization that reveals any inherent hysteresis in the material as well.6 The resulting curve reveals other
features, such as the saturation limits, indicated in Figure 8.23(b) and (c), which are useful in conceptualizing how a simpli-
fied model is formed. Another element of such a model is the linear trend in Figure 8.23(c), which has near zero (dynamic)
hysteresis width, Hdyn . Materials having hysteresis widths are useful in various applications, distinguishing soft and hard
magnetic materials, the latter with very large widths useful for permanent magnets while soft are good for inductors and
transformers [167].

5 Alternatively, intrinsic flux density, or magnetic polarization.

6 Hysteresis itself refers to the history dependent property of a system, here for a magnetic material by the presence of the area in the B − H
curve.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
8.3 Electromechanical Systems Modeling 499

Saturation limits

Slope =

(a) (b) (c)

Figure 8.23 (a) Typical toroid-based measurement of B − H curve, and (b) B − H curve, (c) linear model approximations.

(a) (b)

Figure 8.24 (a) The integrated and inverted/flipped B − H curves help define M − 𝜑 relations for use in magnetic circuits and the
related; (b) current versus flux linkage form as well.

8.3.3.2 Stored Energy

Considering the basic toroid, with center length l and cross-sectional area A, composed of a uniform, linear magnetic
material, the total energy stored can be expressed, m = l ⋅ A ⋅ ∫ HdB = l ⋅ A ⋅ ∫ (M∕l)d(𝜑∕A) = ∫ Md𝜑. With reference to
Figure 8.24(b), we see this is of the form of a standard potential energy storage element, so m = U𝜑 , and taking M as effort
and 𝜑 as a displacement-type variable. In the approximate relations, we have made use of the definition of magnetic flux
through an area, 𝜑 = ∫ B ⃗ ⋅ dA,
⃗ and the commonly used approximate mmf relation, M = H ⋅ l. Also, note the axes have been
flipped to indicated M versus 𝜑. It is this basic result, M = 𝜑, that forms the foundation for understanding many basic
electromagnetic devices.
Now for a linear isotropic material with B = 𝜇H, M = l ⋅ H = lB∕𝜇 = (l∕𝜇A)𝜑, and we can define a basic relation for
reluctance  = l∕𝜇A. This relation reveals dependence on key geometric and material properties that can be taken advan-
tage of to design EM devices. We also define magnetic permeance as  = 1∕, which can thus be thought of as a magnetic
capacitance. Thus, highly permeable materials are those where 𝜇 → ∞, essentially then acting as efficient transmitters
of magnetic flux as they offer “no reluctance.” Magnetic materials are known to have 𝜇 = 𝜇r 𝜇o with 𝜇r on the order of
2000–80 000, where 𝜇o is the permeability of free space. It is common to find in classical texts on EM devices, notably
Woodson and Melcher [31], that devices with core materials indicated by “𝜇 → ∞” are meant to be neglected in the mod-
el/analysis, with due attention given to the “air gaps” which present higher reluctance and are the points at which sensing
and actuation occur.

8.3.3.3 Magnetic Circuit Variables and Elements

Magnetic circuit models often make an analogy of the relation between M and 𝜑 to a resistive element, M = 𝜑,
viewing reluctance, , as a resistance and M and 𝜑 as voltage and current, respectively [168, 169]. For the purposes of
lumped-parameter modeling in a bond graph approach this relation is taken as capacitive in nature, and thus foundational
in defining storage of magnetic energy. Thus, we have taken magnetomotive force, M (effort), and the time rate of change
of magnetic flux, 𝜑̇ = f𝜑 , as effort and flow variables, respectively. The magnetomotive force at points or nodes in magnetic
structures can be thought of as analogous to voltage potential at nodes in an electric circuit, while f𝜑 = 𝜑̇ is analogous to
current. Thus, magnetic power flow is the product,  = M ⋅ 𝜑̇ (units of Watts).
The magnetomotive force (M or mmf) between two points a and b is defined by Mab = ∫a H
b
⃗ ⋅ d⃗s, similar to how we define
a voltage potential in an electric field (or circuit). Another analogy is H to mechanical stress and B to strain. The magnetic
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
500 8 Multiport Modeling and Energy Methods

material description between magnetic flux density, B ⃗ , and the magnetic field intensity, H,⃗ B ⃗ = 𝜇 H,
⃗ helps describe the
magnetic field density, 𝜖m = ∫ H⃗ ⋅ dB⃗ . Note that 𝜖m has units of A⋅V⋅sec/m3 ; that is, energy/volume.
We can now define ideal junction elements for magnetic circuits, using 0- and 1-junctions that correspond to a flux law
and mmf law, respectively. The flux law states that the net flux going into or out of a node (here a common mmf node) is zero,
thus being analogous to Kirchhoff’s current law (KCL). The mmf laws states that the net mmf drop around a magnetic circuit
is equal to the net mmf rise and is thus analogous to Kirchhoff’s voltage law (KVL) [168]. These concepts are illustrated in
Figure 8.25(a) and (b), along with examples of each case in Figure 8.25(c) and (d).
Magnetic circuit models are typically comprised of the following basic elements:

● Soft magnetic materials. These provide low reluctance paths (help “duct” the flux), typically made from silicon steel or
low-carbon steel, with saturation limitations (about 2.1 Tesla). The relative permeability can range, 2000 < 𝜇r < 80 000.
● Coils that provide a connection to the electrical domain, form electromagnets, and usually made using copper and some-
times using aluminum (for weight).
● Permanent magnets that provide “dc” flux, typically at a gap, distinguished by parameters such as Br , Hc , and Br Hc (max.
energy product), as well as by demagnetization curves and temperature effect on flux density.

Useful schematics and corresponding bond graph elements are summarized in Figure 8.26. It is helpful to draw equivalent
magnetic circuits as well (e.g., see Nasar [169] and Chai [168]), prior to construction of a bond graph form.
One of the most useful basic relations is the estimated reluctance,  = l∕𝜇A, where l is a centerline length, 𝜇 is the
permeability, and A is the effective cross-sectional area. The inverse of reluctance is permeance, m = 1∕ (the subscript
used to distinguish this from power). It is common in the literature to find schematics of magnetic devices that indicate
𝜇 → ∞ for a section of core material. This implies that the reluctance of that section/component in a magnetic circuit model
may be very small compared to other parts of the magnetic circuit. The basic reluctance relation provides insight into how

Flux law mmf law

(a) (b)

Gap

(c) (d)

Figure 8.25 Magnetic circuit ideal junction elements: (a) flux law as 0-junction, (b) mmf law as 1-junction, (c) example use of flux
law, and (d) example use of mmf law.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
8.3 Electromechanical Systems Modeling 501

(a) (b)

(c) (d)

Figure 8.26 Key magnetic circuit elements in schematic and corresponding bond graph form: (a) coil element, (b) core,
(c) core junction, and (d) air gap.

a wide range of transducers (sensors or actuators) can be built by taking advantage of changes that can occur in either
geometric (l or A) or magnetic material (𝜇) properties.

8.3.3.4 Magnetic Circuit Examples

Consider the magnetic device in Figure 8.27(a). A model of this device is conveyed by the equivalent circuit in Figure 8.27(b),
where a circuit gyrator element is used to represent the transformation between electrical and magnetic circuits, with the
magnetomotive force, Ni, equivalent to voltage on the magnetic circuit. Note that capacitor elements are used to represent
the magnetic energy storage elements, which here have a common magnetic flux, 𝜑. A reluctance parameter is indicated
for each of the storage elements by . While an true analogy for capacitance would call for using permeance,  = 1∕,
either parameter is suitable in this context.
Since magnetic energy is taken as potential in form, that is, U𝜑 = ∫ Md𝜑, those basic elements in Figure 8.26 that include
magnetic energy storage in core material or in air are represented by C elements.7 Equivalent circuits can be constructed
based on magnetic device schematics and then those can be converted into bond graph form. Some reduction may be done
in bond graph form, followed by causality assignment and state variable determination. In the magnetic domain, only the
flux variable, 𝜑, for C elements are used as energy-storing states. No kinetic energy element is needed in the magnetic
energy domain.
In constructing the equivalent bond graph, the magnetic circuit can then be interpreted as an electrical circuit. As drawn
in Figure 8.28(a), a preliminary step is taken to emphasize that each distinct mmf (effort) node is a 0-junction. The rest of

+ +

– –

(a) (b)

Figure 8.27 (a) Electromagnet core with air gap, (b) Equivalent magnetic circuit with magnetic energy storage elements indicated by
reluctances, .

7 As previously mentioned, references such as Nasar [169] and Chai [168] use resistive elements in equivalent circuits to represent reluctance
elements.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
502 8 Multiport Modeling and Energy Methods

Using magnetic

Equivalent with

(a) (b)

Figure 8.28 (a) Bond graph for magnetic device in Figure 8.27; (b) equivalent reduced form.

(a) (b)

Figure 8.29 (a) Solenoid from Figure 8.20(a) now with relevant reluctances indicated in flux path. (b) Magnetic circuit with
reluctance elements, some having dependence on the air gap distance x.

Figure 8.30 Bond graph for magnetic circuit in Figure 8.29(b), including the plunger mass and force due to gravity.

the bond graph follows, and then in Figure 8.28(b) a reduced form is given where one C is used to represent the combined
reluctance values which add (since they are series connected). Since the reluctance values are constant, the series connected
gyrator and capacitive element in the reduced form can be interpreted as the effective inductance when viewed from the
electrical port. Indeed, for this case, L = N 2 ∕T , where T = ab + bc + cd .
Consider now the solenoid example from Figure 8.20(a), which is redrawn in Figure 8.29(a) now emphasizing key reluc-
tance elements in the flux path. The schematic in (a) is converted to an equivalent magnetic circuit form in Figure 8.29(b),
where it is assumed that a cylindrical magnetic structure has the five principal reluctances in series (common flux). The
plunger is shown to have a reluctance that may change due to the change in air gap, but it is only at the air gap where
a magnetomechanical force is induced as modeled by a two-port magnetomechanical C element. The model is shown in
bond graph form in Figure 8.30.

8.3.3.5 Magnetomechanical Force Interactions

Section 8.3.2 has described ways for estimating forces and torques induced by magneto-quasistatic devices. The following
examines how permanent magnets induce forces or torques due to stored magnetic energy where reluctance can vary due to
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
8.3 Electromechanical Systems Modeling 503

Air gap
Moving
core
+

Stator

(a) (b)

Figure 8.31 (a) Toroid with gap. (b) Magnetic device with moving core.

changes in moving mechanical components. Many transducers that make use of this mechanism are commonly categorized
as being of a variable-reluctance type. Assuming a magnetomechanical two-port C, the potential energy is given by,

U𝜑x (𝜑, x) = M ⋅ d𝜑 + F ⋅ dx, (8.32)

∫ ∫
assuming a translational mechanical displacement. The force can then be found from,
𝜕U𝜑x
F= . (8.33)
𝜕x
These relations can be extended to multiple ports in the magnetic and mechanical domain.
Consider the toroid shown in Figure 8.31(a) which has a gap of length xg and assuming uniform gap flux field the area is
taken as that of the core, A. The air gap reluctance is thus, g = xg ∕𝜇o A, where 𝜇o is the permeability of free space. Since the
core material is very high, core ≈ 0. Thus, the energy stored in this device is U𝜑xg = (xg )𝜑2 ∕2. Now, the force attracting
the pole faces can be determined as,
𝜕U𝜑xg 𝜑2
F= = . (8.34)
𝜕xg 2𝜇o A
The force will act to reduce the reluctance, thus tending to close the gap. The same force will act to move the core in
the device of Figure 8.31(b). In both cases, the force is independent of displacement since the field in the gap is assumed
uniform. The basic relation in equation 8.34 is useful for baseline force calculations with electromagnets.

Example 8.11 EM relay system

Use energy methods to obtain constitutive relations for the electromagnetic force Fa and for the electromagnetic relay
system shown in Figure 8.32. Express this relation as a function of displacement xa , at the gap, and flux linkage, that is,
𝜆 = NΦ, Fa = Φ(xa , 𝜆), where N is the number of turns of the coil.
Solution
If we assume a uniform field in the gap between the core piece and the pivot arm and no energy stored outside of this gap
(neglect fringing and energy stored in highly permeable core and pivot), then flux linkage is,
𝜆 = N𝜑 = NBA,
where A is the area of the core and B is the magnetic flux density. Since air is a linear isotropic medium,
B = 𝜇H,

Figure 8.32 Schematic of an electromagnetic relay. Pivot arm

Contacts

Spring
+
Magnetic
–
coil Input
voltage
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
504 8 Multiport Modeling and Energy Methods

where 𝜇 is the magnetic permeability and H is the magnetic field intensity. Magnetomotive force is found from Ampere’s
law,

M= H ⋅ ds = Hxa = Ni,
∮
where the path of integration is taken around the pivot arm, across the air gap, and through the coil with the only contri-
bution to the integration occurring across the gap. We then have,
x
M = a 𝜑,
𝜇A
or,
1 x xa 1
i= M= a 𝜑= 𝜆= 𝜆,
N 𝜇NA 𝜇N 2 A L(xa )
where L(xa ) = 𝜇N 2 A∕xa is a displacement-dependent inductance. The magnetic energy can then be calculated as,
xa 2
U𝜑xa = Md𝜑 = 𝜑,
∫ 2𝜇A
or the coenergy is,
xa 1
𝜆xa = id𝜆 = 𝜆2 = 𝜆2 .
∫ 2𝜇N 2 A 2L(xa )
Electromagnetic force is then derived using either,
𝜕U𝜑xa 𝜑2 𝜕𝜆xa 𝜆2
Fa = = or Fa = = .
𝜕xa 2𝜇A 𝜕xa 2𝜇N 2 A
Note that, for uniform fields, force is not a function of displacement.

Note also that being able to estimate the reluctance provides insight into electrical port variables as well. While the force
is independent of gap, the reluctance is not and recall that the inductance as viewed from the electrical port is L = L(xg ) =
N 2 ∕g (xg ). Further, one could measure L(xg ), say, for different values of xg . This highlights how ignoring the actual core
reluctance can be suitable for making estimates of the force at the gap but it is not sufficient for determining the electrical
inductance. As xg goes to zero this would indicate that L → ∞. Of course, this is not true since we know that the actual
device reluctance is,
l xg
 = core + g (xg ) = c + .
𝜇A 𝜇o A
So, when xg = 0, the reluctance is not zero and thus inductance is finite due to the effects of the core’s reluctance.

Example 8.12 Angular motion magnetic device

It has been proposed to construct an angular motion EM actuator in the form shown in Figure 8.33. This device features an
electromagnetic coil that induces magnetic flux 𝜑 in the core and gap. An iron blade positioned outside of the gap is attached
to a (non-ferrous) flywheel that constrained to rotated about a fixed axis. The intent is to generate rotational motion in the
flywheel by drawing the blade into the gap by powering the coil.
Assume the core has 𝜇 = ∞. When i = 0 in the coil, assume the blade is initially set at an angle that will enable it to be
“pulled-in.” Rationalize and estimate the trend in the total magnetic circuit reluctance as a function of 𝜃, and then sketch
the trend of how the inductance (looking into the coil input) varies with 𝜃 where 𝜃 = 0 corresponds to when the blade is
centered in the air gap.
Solution
Begin by setting upper and lower bounds on the expected reluctance. When the blade sits centered in the gap, the air gaps on
each side of the blade will dominate the reluctance. However, the gap reluctance values will be much smaller when the blade
is fully outside of the gap, and most of the flux must pass through the full air gap. Thus, at 𝜃 = 0 we have min ≈ 2t∕𝜇o Ac ,
where t is the gap on each side of the blade and Ac is the core cross-sectional area. When 𝜃 = 𝜃max , the reluctance will be
max = lg ∕𝜇o Ac , where lg is the full gap distance and, lg ≫ 2t. Note that 𝜃max is that angle at which the reluctance will be
influenced by the blade’s presence. For 𝜃 > 𝜃max , it is assumed that reluctance will no longer change with 𝜃.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
8.3 Electromechanical Systems Modeling 505

Side view Flywheel

Air gap Iron blade Iron blade

Core

Cross-section
of core

(a) (b)

Figure 8.33 (a) An electromagnet with rotationally pivoted blade/core; (b) side view to illustrate rotation of blade into air gap.

Figure 8.34 Trends in reluctance  and inductance L for

angular-motion device in Figure 8.33 as function of 𝜃.

Now, as 𝜃 decreases from 𝜃max to 𝜃min , the reluctance will vary according to the effective change in the area. Without a
specific blade/core shape and without a detailed field analysis, it is only possible to guess that the reluctance will decrease
until min is reached as 𝜃 ↓. In Figure 8.34, a linear trend is indicated for illustration. In this case, the trend in L(𝜃) ∼
N 2 ∕(𝜃), as shown. The slope in these trends reflect the electromagnetic torque induced within the range ±𝜃max , since
𝜏 = 𝜕U𝜑𝜃 ∕𝜕𝜃 ∼ 𝜕(𝜃)∕𝜕𝜃. Thus, when assembling such a device, care should be taken to place the blade within |𝜃| < 𝜃max
to ensure sufficient pull-in torque.

The previous example is similar to the more common rotational magnetic actuator shown in Figure 8.35. The torque
induced on the rotor will tend to align the rotor with the magnetic core (to reduce reluctance). The reluctance can be
estimated given the geometry indicated. The magnetic circuit shows that a total reluctance is the sum,
2lg
 = core + rotor + gap (𝜃) ≈ gap (𝜃) = , (8.35)
𝜇o A(𝜃)
where A(𝜃) is a function describing how the flux path cross-sectional area changes with angle 𝜃. Either the area and thus
reluctance may be approximated by the geometry or sometimes implied by observed trends in measured inductance [31].
Computational field analysis or field estimation methods (e.g., [170]) can also be used to provide a more exact determina-
tion, especially when a design is being optimized for size, shape, and material selection.
Given the reluctance, the stored magnetic energy can then be determined,

U𝜑𝜃 = Md𝜑 + 𝜏d𝜃,

∫ ∫

+
Rotor
Uniform gap,

Axial thickness, (into page)

Figure 8.35 A rotational magnetic actuator with rotor free to rotate, shown rotated by angle 𝜃 from alignment with pole faces where
𝜃 = 0.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
506 8 Multiport Modeling and Energy Methods

since M = gap (𝜃)𝜑, and so, assuming a convenient integration path making the second term zero,
1
U𝜑𝜃 =  (𝜃)𝜑2 ,
2 gap
The torque is then found by,
𝜕U𝜑𝜃 1 𝜕gap (𝜃) 2
𝜏em = 𝜏em (𝜑, 𝜃) = = 𝜑. (8.36)
𝜕𝜃 2 𝜕𝜃
This actuation mechanism relies on change in reluctance with position, illustrating well a type used in a wide range of
variable-reluctance actuators, such as stepper motors. Example calculations of the torque characteristics for the actuator in
Figure 8.35 can be found in Solved Problem A-8-9.
The device model can be integrated with mechanical inertia of the rotor and effective damping as shown in the bond
[ ]T
graph of Figure 8.36. Causality here indicates the independent states x = 𝜑 (or 𝜆), 𝜃, h .
While the discussion up to this point has emphasized magnetic devices designed for actuation, it should be made clear
that the same models apply to magnetic devices for sensing applications. In these cases, however, the electrical design
would be used for inferring changes in reluctance given a change in the position of a moving magnetic circuit element.
Variable-reluctance mechanisms of this type are used in a wide range of transducers that rely on detection of displacement
to measure different physical quantities.

8.3.3.6 Modeling Devices with Permanent Magnets

In addition to coils, magnetic devices make use of permanent magnets (hard magnetic materials) to provide flux in magnetic
circuits. As with electromagnetic coils, the flux induced by a permanent magnet is usually directed toward an air gap for
force or torque generation. Permanent magnets (PMs) also provide flux for inducing charges to move in coils for sensing
applications. For the purpose of modeling such effects, it helps to understand the role of “B” as provided by a magnet.
Consider for example a block of PM material with area Am and length lm , as in Figure 8.37(a). Assume the PM has 𝜇m , so
the reluctance is m = lm ∕𝜇m Am .
In static applications where the flux does not change significantly (𝜑̇ ≈ 0), a PM can be modeled by an ideal mmf (effort)
source, as shown in two forms in Figure 8.37. We might take,
lm l l
Mm = Rm 𝜑m = Bm Am = m Bm = m Br ,
𝜇m Am 𝜇m 𝜇m
where Br is the residual flux density or remanence, indicated in Figure 8.38(a), which shows a typical demagnetiztion curve,
as conventionally plotted in the second quadrant. The value of 𝜇m is called the incremental permeability and is usually close

Figure 8.36 Bond graph for system in Figure 8.35.

Figure 8.37 (a) Ideal effort (mmf) source model. (b) Ideal flux rate
(flow) source model.

(a) (b)
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
8.3 Electromechanical Systems Modeling 507

BH (energy) curve
12
Neodymium
Operating line Rare-earth 8
Alnico
cobalt
Ceramic 4

0
10 8 6 4 2 0
Demagnetization
curve
(a) (b)

Figure 8.38 (a) Typical PM demagnetization curve with overlaid operating line model. (b) Typical B − H trends for common PM
families (both conventionally plotted in second quadrant).

to a value of 1, but it can vary. Clearly, variation in flux will lead to changes in the mmf and 𝜇m values. This can be especially
true for different types of magnets, as reflected in the typical demagnetization curve shown in Figure 8.38(b). A dashed line
on the graph in Figure 8.38(b) shows (qualitatively) how the trend in the energy product BH, say, for the neodymium
curve might plot. A maximum BH line will identify the maximum energy product, (BH)max , values of which are given in
Table 8.5 for different magnet types. The table also lists values for Br (a measure of how much flux can be produced) as
well as for coercivity, Hc (a measure of how hard to demagnetize with an external field). These characteristics can be highly
temperature dependent, a common problem to deal with since magnets are often enclosed in many practical designs. The
reader is referred to specialized books such as by Campbell [171] for more insight into permanent magnet materials and
their applications.
In general, a PM induces a magnetic field that will depend on load properties. Consequently, if the flux does change, say,
due to a moving element and thus changing air gap, the mmf changes according to the B − H curve. Thus, a better approx-
imate model for a permanent magnet (PM) in a magnetic circuit is as a mmf source in series with a reluctance (Thevenin
form), or using a flux source (Norton form) [168]. These common simplified source models are shown in Figure 8.39. Note
that traditional Thevenin and Norton equivalents are defined using ideal effort and flow sources in combination with resis-
tive elements. The analogy in conventional magnetic circuits as in Chai [168] is to take reluctance as analogous to resistance.
Here, we use a more strict analogy since reluctance is the inverse of permeance which is related to magnetic potential energy
storage. Thus, the source models in Figure 8.39 use capacitive elements and indicated as reluctance (which is the inverse
of permeance).
An example use of the Thevenin form from Figure 8.39(a) is illustrated using the magnetic circuit in Figure 8.40(a).
This circuit shows the permanent magnet with some leakage flux, indicated by reluctance l , and a total effective reluctance
that accounts for the core material as well as the air gap, T . A bond graph equivalent is shown in Figure 8.40(b).

Table 8.5 Table of properties for common magnet types [168].

Parameter (Units) Ceramic Alnico Rare Earth Neodymium

(BH)max (GOe) 1–4 1.3–9 5–28 25–50+

Br (kGa) 2.3–4.1 6.7–13.5 4.7–10.9 10.5–13.5
Hc (kOe) 1.86–3.4 0.47–1.9 4.5–9.2 8.5–11.5

Oe = 1000∕4𝜋 A/m, Gauss, Ga = 10−4 Wb/m2 (or Tesla, T)

Figure 8.39 (a) Thevenin (mmf) source model. (b) Norton (flow) source model.
+ +

+
–
– –
(a) (b)
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
508 8 Multiport Modeling and Energy Methods

+
–
–

(a) (b)

Figure 8.40 (a) schematic and circuit. (b) bond graph.

Air gap

(a) (b) (c)

Figure 8.41 (a) Permanent magnet sourcing flux in an air gap. (b) Bond graph model. (c) Graphical depiction of the static analysis
with PM B − H curve transformed into mmf-𝜑 curve with overlaid linear air gap reluctance.

In design applications where a magnetic circuit is “driven” by a PM, consideration may need to be made for the region
of the B − H curve over which operation will take place. Often, this operating region is chosen to minimize the volume of
magnet needed by maximizing the amount of energy delivered to an air gap, for example. This type of design is discussed in
texts on magnetic device design [167, 168, 171]. The figure given below uses a bond graph to illustrate this common design
example in Figure 8.41(a), modeled for purposes of this discussion by two interacting C elements. The PM is strictly not a
“basic” (magnetic) energy storing element. The constitutive behavior for a magnet is here conveyed on an mmf-𝜑 plot along
with an overlaid linear reluctance model for the air gap in Figure 8.41(c). The intersection of these curves determines an
operating point (a type of source–load analysis as in Chapter 4). Note that if the air gap changes the operating (intersection)
point will move along the magnet characteristic curve, Mm − 𝜑m . Again, even though the permanent magnet is treated
as a nonlinear C element, the fact that the constitutive law involves hysteresis implies the element is not basic.8 If the
system response remains within a small region, it is preferred to adopt the source models discussed previously. Only if it
was necessary to, say, “track” the state of the magnet throughout its hysteresis loop might it be necessary to reconsider this
model. In such cases, flux state of the magnet could be used to keep track of the system state on a B − H loop. The hysteresis
model could be approximated by a nonlinear R–C combination for these purposes (e.g., see Paynter [88] and/or Karnopp
[89]).
For more insight into analysis and design with permanent magnets and interactions with coils, the reader is referred
to specialized texts by Chai [168], Campbell [171], and Furlani [172]. Some technical articles addressing practical devices
that may be of interest by Kamerbeek [173] and Timmerman [174] have also been addressed by Karnopp [89], and later
summarized in Karnopp et al. [38]. Unique challenges in estimating flux-induced torques in devices with coils and moving
magnets are also discussed by Campbell [175] and Furlani et al. [176], the latter of which provides insight into how a
combined use of magnetic circuit approximations and finite element analysis (FEA) analysis are often essential for more
effective predictions.

8.4 Lagrange’s Equations in System Modeling

This section shows how Lagrange’s equations, which use kinetic coenergy, Tf , and potential energy, Uq , functions, can be
integrated with bond graph models. The classical Lagrange approach defines the Lagrangian,  = Tf − Uq , to derive the
Lagrange’s equations,
( )
d 𝜕 𝜕
− = Qj .
dt 𝜕 q̇ j 𝜕qj

8 For definition of a “basic” element, see Section 4.6.1.

 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
8.4 Lagrange’s Equations in System Modeling 509

These are n second-order differential equations in the generalized coordinates, qj , j = 1, … , n. The method scales effectively
as the number of modeling elements and coordinates increases, so the steps used in applying Lagrange’s equations to a
simple single degree-of-freedom system are similar to those needed to describe systems of higher complexity with many
more generalized coordinates.
This section describes how a Lagrangian subsystem can be defined for those parts of a system that may be more effectively
modeled with the Lagrange equations. Lagrange subsystems are particularly suited for modeling those parts of a system that
have position-dependent kinetic energy storage. This includes the wide class of nonlinear mechanical systems, particularly
with holonomic kinematic constraints. A systematic step-by-step methodology is described for formulating the Lagrange
subsystem model and integrating with a bond graph of the rest of a system.
For completeness, a derivation of Lagrange’s equations is presented at the end of this section, although it is not needed to
apply the methodology presented. In addition, Hamiltonian and Co-Lagrange formulations are also discussed to reinforce
related energy methods.

8.4.1 Application of Classical Lagrange’s Equations

The classical form of the Lagrange’s equations provides an analytical approach to deriving system mathematical models
by using energy relationships. Through proper selection of variables for generalized coordinates, equations equivalent to
Newton’s laws, Kirchhoff’s laws, and other foundational physical laws can be formulated. A thorough introduction to the
foundational principles and derivations of Lagrange’s equations can be found in many classical texts (e.g., Lanczos [8],
Goldstein [68]), and a good summary and resource of problems is found in Wells [177]. The following discussion lays out
terminology and basic definitions used in a classical Lagrange equation approach. First, the Lagrangian for such systems is
̇ and Uq is the potential
defined by  = Tf − Uq , where Tf is the kinetic coenergy (expressed in terms of flow variables, f = q)
energy. Application of the Lagrange’s equations,
( )
d 𝜕L 𝜕L
− = 0, (8.37)
dt 𝜕 q̇ j 𝜕qj

leads to second-order ordinary differential equations for the n qj generalized coordinates. The generalized coordinates are
the set of independent coordinates necessary to describe the system completely. The generalized flows would be fi = q̇ i .
Note that the kinetic coenergy is expressed in terms of flow type variables. Equation 8.37 is sufficient for systems with no
dissipation or inputs (nonzero right-hand side). Many systems often have more coordinates than the minimum set qj , so
constraint relations need to be identified. However, for those systems where the qj are easily identified, the application of
Lagrange’s equations provides a systematic way to derive system equations.

Example 8.13 Lagrange’s equations for simple pendulum

The simple pendulum shown in Figure 8.42 is a single degree-of-freedom system requiring only one generalized coordinate,
q = 𝜃.
In this case, the kinetic coenergy is, T𝜃̇ = ml𝜃̇ ∕2. Using the downward vertical axis as the datum,
2

the potential energy is, U𝜃 = mgl(1 − cos 𝜃). With no losses, the Lagrangian is,
1 ̇2
 = T𝜃̇ − U𝜃 = ml𝜃 + mgl(1 − cos 𝜃),
2
so that,
( )
d 𝜕T𝜃̇ 𝜕T𝜃̇ 𝜕U𝜃
− + = 0,
dt 𝜕 𝜃̇ 𝜕𝜃 𝜕𝜃
Figure 8.42
gives the expected second-order ODE for the simple pendulum, Simple pendulum.

ml2 𝜃̈ + mgl sin 𝜃 = 0.

Practical applications require: (i) incorporating nonconservative effects such as inputs and dissipation, and (ii) dealing
with systems that have more coordinates and constraints that define the generalized coordinates for a system.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
510 8 Multiport Modeling and Energy Methods

8.4.1.1 Nonconservative Effects

Any nonconservative effects can be integrated into a model based on Lagrange’s equations by including generalized efforts
on the right-hand-side, commonly symbolized by Qi ’s, but here we will use Eqi to represent any nonconservative effort
associated with the generalized displacement, qi . Associating these efforts with the generalized coordinates implies work is
done in accord with energy conservation principles. This allows us to account for net power flowing into a system either by
sources or dissipation (the latter of course being negative in sign). It is convenient to think of any nonconservative effects as
being efforts applied to the conservative Lagrange subsystem, being distinct from those effects that arise from conservative
effects (such as springs, etc.), which would be accounted for by derivatives of the Lagrangian energy function.
Applied or dissipative efforts are included “on the right hand side” of the Lagrange’s equations, appropriately associated
with the equation for a given generalized coordinate. For example, if a torque, 𝜏a , from a motor was applied at the pivot about
𝜃 defining motion of the simple pendulum in Example 8.13, then that effort would be readily associated with the Lagrange
equation for 𝜃. Similarly, dissipation effects may be readily associated with the generalized coordinates. By observation,
linear damping at the pivot would impose a damping torque, 𝜏b = b𝜃.̇ Note that we’d need to make sure to account properly
for the sign of these efforts. This is why it is useful to think of these in terms of net power into the conservative system and
to formulate a power function for each generalized coordinate q such as,
∑ ∑
q = + (applied efforts into system) ⋅ q̇ − ̇
(dissipative efforts) ⋅ q.
̇ which is often the case. The total
Note that this allows for the applied and dissipative efforts to generally be functions of q,
nonconservative efforts for a particular generalized coordinate can then be found from: Eq = 𝜕q ∕𝜕 q. ̇ For example, consider
the torque supplying power to the pendulum about 𝜃 and the damper dissipating power also about 𝜃, then the net power
into the conservative Lagrange system is: 𝜃 = +𝜏a ⋅ 𝜃̇ − (b𝜃) ̇ ⋅ 𝜃,
̇ so E𝜃 = 𝜕𝜃 ∕𝜕 𝜃̇ = +𝜏a − b𝜃,
̇ and the Lagrange equation
for the simple pendulum is now,
̇
ml2 𝜃̈ + mgl sin 𝜃 = +𝜏a − b𝜃.
The specific power function described above for linear dissipative effects is typically referred to as a Rayleigh dissipation
function; i.e., the dissipated power being, Df = (1∕2)𝛼f 2 , where 𝛼 is a proportionality constant and f an associated flow
variable, which is either a q̇ or related to the q̇ variables of a system. Different types of power functions for dissipative effects
have a similar form (see, for example, [177]). Consider one more basic example in adding damping for the mass-spring
system.

Example 8.14 Lagrange’s equations for mass-spring-damper

The mass-spring-damper system in Figure 8.43 requires only one generalized coordinate, q = x, here x being the displace-
ment of the mass.

Figure 8.43 Mass–spring–damper system.

Thus, the system kinetic coenergy is, Tẋ = 12 mẋ 2 , and the potential energy, Ux = 12 kx2 . There is a nonconservative effort
associated with x, Ex , which is typically derived from a dissipation function, commonly taking the form of a Rayleigh dis-
sipation function, which was originally defined for velocity-dependent forces. For example, we might take Dẋ = bẋ 2 for a
linear viscous force and Ex = −𝜕Dẋ ∕𝜕 ẋ = −bx.̇
In this way, applying Lagrange’s equations directly gives,
( )
d 𝜕Tẋ 𝜕T ̇ 𝜕Ux
− x + = Ex = −bx, ̇
dt 𝜕 ẋ 𝜕x 𝜕x
so,
m̈x + bẋ + kx = 0,
as expected for this system. Input forces could also be added using the Ex .
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
8.4 Lagrange’s Equations in System Modeling 511

8.4.1.2 Constraint Relations

Sometimes it is relatively easy to associate a particular nonconservative effort with a generalized coordinate, as the previous
examples have illustrated. However, most problems for which a Lagrange approach would be advised feature m-dependent
coordinates where m > n, n being the number of generalized coordinates required to describe a system configuration. In
such cases, defined relationships between the dependent coordinates and a minimum set of independent generalized coor-
dinates are defined from the physical system structure and are commonly referred to as constraints. Constraints were also
discussed in relation to modulated two-port elements (transformers and gyrators) in Chapter 4. While constraints arise in
many ways in all types of systems, the case of complex mechanical systems with more body coordinates than actual degrees
of freedom (which is n, i.e., the number of generalized coordinates) is a use case of particular interest here.
Holonomic constraints are those that take the form, 𝜙j (q1 , q2 , … , qn , t) = 0 (j = 1, 2, … , r), allowing to solve for m coor-
∑n
dinates in terms of remaining n and possibly time. Non-holonomic constraints are those expressed as i=1 aij dqi + aji dt
= 0 (j = 1, 2, … , m), with the a’s generally functions of the q’s and time, and not integrable, so more coordinates than DOF
are needed to describe system. A more practical way to distinguish these constraints and their influence on the model-
ing process was discussed in Chapter 4 in relation to the modulation structure of transformers and gyrators. The following
example illustrates these concepts.

Example 8.15 Rolling disks and constraints

The disk rolling in a plane as shown in Figure 8.44(a) requires consideration of the translational mass, m, and the rotational
inertia about the axis of rotation, J. If we assume the disk does not leave the surface, we need at most two co-ordinates (x, 𝜃)
to define the system motion. If we now constrain to rolling without slip then ẋ = rd 𝜃, ̇ which is a holonomic constraint, as
x = rd 𝜃.
Figure 8.44 Rolling disk constraints.

(a) (b)

In contrast, the motion of a disk rolling on a 2D plane as shown in Figure 8.44(b), again constrained to stay on the plane
and with no tilt (about the contact point and along an axis in the plane of the disk), requires coordinates (x, y, 𝜃, 𝜓), where
𝜓 is the angle about the disks vertical axis. A rolling constraint would allow us to relate ẋ and ẏ to 𝜃̇ and 𝜓, ẋ = rd 𝜃̇ sin 𝜓
and ẏ = rd 𝜃̇ cos 𝜓, but we find that,
dx − rd sin 𝜓 ⋅ d𝜃 = 0,
dy − rd cos 𝜓 ⋅ d𝜃 = 0,
which are non-holonomic constraints. We would need to independently find 𝜓 in order to complete this constraint, which
is reflected in the bond graph of Figure 8.45(b).
Figure 8.45 Rolling disks: holonomic versus nonholonomic.

(a) (b)

One way to interpret the implication of the difference between constraints illustrated by this example is that
non-holonomic constraints, being represented by multiport Ts or Gs, require information from an external system
(e.g., for yaw, 𝜓) to complete a model description. Holonomic constraints are those that are not modulated, such as
the rolling wheel, or those that are modulated by coordinates (i.e., information or energetic states) from an attached
power bond. The basic slider-crank is a good example of a modulated transformer that relates system variables through a
holonomic constraint.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
512 8 Multiport Modeling and Energy Methods

While such an insight may not be applicable to all cases, for many problems of practical interest, this way of distinguishing
holonomic and non-holonomic constraints can very useful. The presence of a non-holonomic constraint can complicate the
application of Lagrange’s equations by requiring Lagrange multipliers. The use of Lagrange’s equations with Lagrange mul-
tipliers is very broadly used, of course, and provides a way to model complex, constrained multibody systems, as discussed
in Haug [72].

8.4.1.3 Example Applications

Additional application examples of the classical form of Lagrange’s equations are now reviewed. Emphasis is given to
examples that illustrate nonconservative effects as well as holonomic constraint relations.

Example 8.16 Roller mass–spring–damper

The roller shown in Figure 8.46 has mass, m, and rotational inertia, J, about its axis of rotation. It is assumed rolling occurs
without slip. The following steps are taken to formulate the Lagrange’s equations.
Key elements: mass/roller (m, J), damper (b), and spring (k)
Dependent coordinates to describe system: x, 𝜃
Kinematics and constrained motion: 𝜃̇ = x∕r
̇ (no slip)
Generalized coordinate(s): q = x
Total kinetic coenergy: Tẋ 𝜃̇ = mẋ 2 ∕2 + J 𝜃̇ ∕2
2

Kinetic coenergy relation: Tẋ = (1∕2)(m + J∕r 2 )ẋ 2

Potential energy: Ux = kx2 ∕2
Dissipation function: Dẋ = bẋ 2 ∕2
Lagrangian:
 = Tẋ − Ux = (1∕2)(m + J∕r 2 )ẋ 2 + kx2 ∕2.
Lagrange equation:
( )
d 𝜕Tẋ 𝜕T ̇ 𝜕Ux 𝜕D ̇
− x + = − x.
dt 𝜕 ẋ 𝜕x 𝜕x 𝜕 ẋ
Final form:
̇
(m + J∕r 2 )̈x + kx = −bx.

Figure 8.46 Roller mass–spring–damper.

Example 8.17 Inverted spring pendulum

The inverted pendulum in Figure 8.47 is stabilized by the linear spring, k. The mass position is defined by xm and ym , and
the spring deflection about the vertical is xk . The pendulum can also be described using the angular displacement 𝜃. These
are a set of dependent coordinates for describing the system, as only 𝜃 is required.
Key elements: point mass (m), link (L, massless), spring (k), and a = constant
Dependent coordinates to describe system: xm , ym , xk , 𝜃
Generalized coordinate(s): q = 𝜃 can describe all motions
Kinematics and constrained motion:
⎡ ẋ m ⎤ ⎡ L cos 𝜃 ⎤
⎢ ̇ ⎥ ⎢ ⎥
⎢ ym ⎥ = ⎢ L sin 𝜃 ⎥ ⋅ 𝜃,
̇
⎢ ẋ k ⎥ ⎢ a cos 𝜃 ⎥
⎢ 𝜃̇ ⎥ ⎢ 1 ⎥
⎣ ⎦ ⎣ ⎦
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
8.4 Lagrange’s Equations in System Modeling 513

Figure 8.47 Inverted spring pendulum.

Total kinetic coenergy: T𝜃̇ = mL̇𝜃̇ ∕2

2

Potential energy: U𝜃 = mgL cos 𝜃 + k(a sin 𝜃)2 ∕2

Dissipation function: n/a
Lagrangian:
 = T𝜃̇ − U𝜃 = mL𝜃̇ ∕2 + mgL cos 𝜃 + k(a sin 𝜃)2 ∕2,
2

Lagrange equation:
( )
d 𝜕T𝜃̇ 𝜕T ̇ 𝜕U𝜃
− 𝜃 + = 0.
dt 𝜕 𝜃̇ 𝜕𝜃 𝜕𝜃
Final form:
ml2 𝜃̈ − mgL sin 𝜃 + ka2 sin 𝜃 cos 𝜃 = 0.
For small motion:
ka2 − mgL
𝜃̈ sin 𝜃 + 𝜃 = 0,
mL2
revealing stability condition, ka2 > mgL

Example 8.18 Roller pendulum

A planar roller (M, Jo ) has an attached simple pendulum as shown in Figure 8.48 point mass (m) formed with massless
rod OA rigidly attached at O. A spring (k) and damper (b) system restrains the roller. This system has m = 3 dependent
coordinates to describe motion: x, y, 𝜃, with n = 1 generalized coordinate.

Figure 8.48 Roller pendulum.

Rigidly attached
to roller

Kinematics and constrained motion: x = r𝜃, y = l(1 − cos 𝜃), 𝑣A

find 𝑣A : ⃗r A = (x − l sin 𝜃)î − l cos 𝜃̂j, so ⃗ṙ A = (ẋ − l𝜃̇ cos 𝜃)î + l𝜃̇ sin 𝜃̂j, and 𝑣2A = ⃗ṙ A ⋅ ⃗ṙ A ⇒ 𝑣2A = ẋ 2 − 2xl𝜃̇ cos 𝜃 + l2 𝜃̇
2

Generalized coordinate(s): q = 𝜃

̇ A = mẋ ∕2 + Jo 𝜃̇ ∕2 + m𝑣A ∕2
2 2
2
Total kinetic coenergy: Tẋ 𝜃𝑣
⇒ T𝜃̇ = mr 𝜃 ∕2 + Jo 𝜃̇ ∕2 − mrl𝜃 𝜃̇ cos 𝜃 + ml 𝜃̇ ∕2
2 ̇2 2 2 2

Potential energy: U𝜃 = mgy + (1∕2)kx2 = mgl(1 − cos 𝜃) + (1∕2)k(r𝜃)2

Dissipation function: D = −bẋ 2 ∕2 = −br 2 𝜃̇ ∕2
2

Lagrangian:  = T𝜃̇ − U𝜃 = mr 2 𝜃̇ + Jo 𝜃̇ ∕2 − mrl𝜃 𝜃̇ cos 𝜃 + ml2 𝜃̇ ∕2 − mgl(1 − cos 𝜃) − (1∕2)k(r𝜃)2

2 2 2
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
514 8 Multiport Modeling and Energy Methods

Figure 8.49 Solenoid-cart-pendulum system with dashed portion partitioned for Lagrange subsystem modeling.

Lagrange equation:
( )
d 𝜕T𝜃̇ 𝜕T ̇ 𝜕U𝜃 𝜕D
− 𝜃 + =− .
dt 𝜕 𝜃 ̇ 𝜕𝜃 𝜕𝜃 𝜕 𝜃̇
Final form:
(mr 2 + ml2 + Jo − mrl𝜃 cos 𝜃)𝜃̈ + mrl(cos 𝜃 − 𝜃 sin 𝜃)𝜃̇ + mgl sin 𝜃 + kr 2 𝜃 = −br 2 𝜃.
̇

For small motion:

(mr 2 + ml2 + Jo )𝜃̈ + br 2 𝜃̇ + (mgl + kr 2 )𝜃 = 0.

8.4.1.4 Subsystem Partitioning

The application of Lagrange’s equations is well suited for a large class of systems and provides a very systematic way for
formulating system equations for those parts considered conservative (no dissipation), while any nonconservative effects
(applied efforts, dissipation) can be introduced as needed. It is also possible to leave any aspects of a system that can be
readily modeled by other approaches as external effects, and only use the Lagrange subsystem to handle those systems that
have highly coupled dependent energy storage. As shown in the examples above, those cases with holonomic constraints
lead to a highly formulaic approach. As such, we can partition such Lagrange subsystems in the initial stages of a model
formulation, leaving any other parts of a system to be modeled using a direct bond graph approach. For example, the
system in Figure 8.49 shows an EM solenoid interacting with a coupled-mass-pendulum system. The dashed box indicates
a possible partition for a Lagrange subsystem.
In this case, those parts of the system outside the Lagrange subsystem could be modeled with conventional bond graph
methods. The partitioning should be done so that there is a preferred causal relation between the systems. Since Lagrange
equations are formed with effort type inputs, it is preferred in Figure 8.49 to leave the translational spring outside of the
Lagrange system even though the spring itself could simply add to the Lagrange potential energy function. It is more effi-
cient in the Lagrange subsystem method to “accept” the spring force as a “nonconservative” input. Of course, there are
other alternatives: (i) the solenoid mass and the spring could both be lumped into the Lagrange subsystem, and (ii) the
conservative aspects of the solenoid, plus mass, plus spring, all could be lumped into the Lagrange subsystem. Again, the
decision on where to partition is a matter of preference and convenience. The initial case lends itself to a study where an
actuating system is under study, thus allowing for quicker remodeling. Such an approach will become even more apparent
in Section 8.4.2 when we convert the Lagrange subsystem itself into bond graph form.
Another motivation for partitioning a system in this way is to avoid the need for Lagrange multipliers. In such cases,
a partition could be established to remove the need for non-holonomic constraints. Building overall system models that
include partitioned Lagrange subsystems is discussed further in the Section 8.4.2 and illustrated in worked examples at the
end of this chapter.

8.4.2 Lagrange Subsystem in a Bond Graph

We now consider a multiport kinetic energy field with m ports represented by an I element in Figure 8.50(a), equivalently
represented by a gyrationally coupled C element in Figure 8.50(b).9 Alternatively, we can consolidate representation of the

9 These gyrators have a unity modulus and are referred to as symplectic gyrators, introduced here so the generalized case will coupled energy is
succinctly represented by a single multiport C.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
8.4 Lagrange’s Equations in System Modeling 515

(a) (b)

̇ and
Figure 8.50 Bond graph representation of Lagrange equations using (a) I and C multiports divided, dependent flows x,
̇ and dependent momenta p; the
dependent momenta p, and (b) I represented as a gyrationally coupled C multiport, dependent flows x,
two multiport C are subsequently merged (see Figure 8.51).

Figure 8.51 Bond graph representation of Lagrange equations using a Lagrange subsystem representation with combined C
̇ and momenta p.
multiports, independent flows q, ̇

potential energy stored in the m-port multiport C in Figure 8.50(a) and (b) with the kinetic energy all into a single, collective
multiport C as shown in Figure 8.51.
̇ are independent (integral causality can
Assume that due to constraints in the system, only n of the I field’s m flows, x,
only be assigned on n of the ports). This implies that the m dependent flows ẋ can be expressed in terms of n independent
flows q̇ by,
ẋ 1 = T11 q̇ 1 + · · · + T1n q̇ n
⋮ ⋮ (8.38)
ẋ m = Tn1 q̇ 1 + · · · + Tmn q̇ n
which can be written in compact form as,
ẋ = Tq̇ (8.39)
where it should be noted that:
a) The T matrix is m × n and not generally invertible.
b) T can be and often is modulated by q, that is, T(q).
c) If equation 8.38 is integrable such that Tij = 𝜕xi ∕𝜕qj then x = x(q) and x can be eliminated in terms of q and the con-
straint is called holonomic.10
In the following, we provide steps in formulating a Lagrange subsystem model and then illustrate its use with some
examples.
Step 1. Express all kinetic coenergy of the system in terms of (dependent) flows ẋ as Tẋ (x)
̇ (distinguish this “T” as kinetic
coenergy by using associated flow variables as subscripts).
Step 2. Replace the dependent flows ẋ with independent flows q̇ using the transformation ẋ = Tq, ̇ and then express the
kinetic coenergy by,
̇ = TT(q)q̇ (q,
Tẋ (x) ̇ q) = Tqq ̇ q).
̇ (q,

10 We will limit ourselves to holonomic systems in this text. For a discussion of non-holonomic systems, see Neı̆mark and Fufaev [69].
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
516 8 Multiport Modeling and Energy Methods

Note that the transformation is generally modulated. This implies that the coenergy is a function of both generalized
̇ and displacement, q. This is reminiscent of magnetic coenergy discussed earlier in Section 8.3 in which the
flow, q,
coenergy is a function of both current (electrical flow variable) and mechanical displacement.
Step 3. Define generalized momenta as,
𝜕Tqq
̇
p= . (8.40)
𝜕 q̇
Note that this definition provides a relationship between each momentum variable, pi , and its associated flow vari-
̇
able, q̇ i , and generally with additional independent flows, q.
Step 4. Express the potential energy of the system as a function of generalized displacements q as Ux (x) = Uq (x(q)) =
Uq (q). The ability to express potential energy as a function of independent coordinates (q) greatly facilitates the use
of Lagrange equations. There are systems for which it is not possible to express the potential energy in terms of the
generalized displacements of the kinetic energy elements. In such cases, the use of a Lagrange formulation is less
desirable.
Step 5. Define a generalized conservative effort for each generalized displacement,
𝜕Tqq
̇ 𝜕Uq
eq = − + . (8.41)
𝜕q 𝜕q
This generalized effort consists of two terms, one due to potential energy storage and one due to the displacement
dependence of the kinetic energy (coenergy in this case). Both of these terms result from derivative energy rela-
tions. The sign difference between the two terms is simply a result of the difference between energy and coenergy
derivative relations. Now, the astute reader might recall that our major justification for introducing two forms of
energy, potential and kinetic, and two types of storage elements, I and C, was to have one form solely a func-
tion of displacement (potential energy) and the other solely a function of momentum (kinetic energy). With the
inclusion of displacement-dependent kinetic energy (e.g., due to displacement-modulated constraints, T(q), or
displacement-dependent kinetic energy), it begs the question:
When is it useful (or not) to delineate between two types of energy?
The answer appears to be pedagogic – It is a most efficient way to initiate an understanding of the concept of energy.
When we take up thermodynamic systems in Chapter 9, we sometimes find it useful to treat energy without regard
to a division into separate elements.11
Step 6. Express the net power into the (Lagrange) subsystem due to nonconservative effort sources and dissipation as,

n = enc ⋅ ẋ = enc ⋅ T(q)q,

̇

from which we can define a generalized nonconservative effort as,

E = enc ⋅ T(q). (8.42)

The six steps provide all the elements necessary to formulate a Lagrange subsystem model. If the nonconservative efforts
are specified as inputs to the subsystem, then we have n generalized momenta and n generalized displacement states. The
state equations for the momenta in vector form are defined by,

ṗ = −e + E, (8.43)

or in the expanded form as,

ṗ j = −ej + Ej j = 1, … , n. (8.44)

In the process of setting out the key elements in the six steps above, we also find the n state equations for the generalized
displacements, since generally from equation 8.40,
𝜕Tqq
̇
p= ̇ q) = M(q) ⋅ q,
= 𝚽(q, ̇
𝜕 q̇

11 This is the point of view represented in Figure 8.51 where storage of potential and kinetic energy has been combined into a single C multiport
element.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
8.4 Lagrange’s Equations in System Modeling 517

which can be inverted to find,

q̇ = M−1 (q)p. (8.45)

Classically, the result in equation 8.44 has been presented in an equivalent form as12
( )
d 𝜕Tqq ̇ 𝜕Tqq
̇ 𝜕Uq
− + = Ej j = 1, … , n, (8.46)
dt 𝜕 q̇ j 𝜕qj 𝜕qj

or compactly as,
( )
d 𝜕 𝜕
− = Ej j = 1, … , n Lagrange Equations, (8.47)
dt 𝜕 q̇ j 𝜕qj

where  is the Lagrangian defined as,

̇ − Uq .
 = Tqq (8.48)

This general method is now applied to specific application problems.

8.4.3 Mechanical and Electromechanical System Examples

The following examples illustrate how a Lagrange subsystem can be integrated into a system model. Many mechanical sys-
tems that have position-dependent inertia effects are particularly suited for these methods. EM systems can also be modeled
this way, although the methods in Section 8.3 can be used to deal most cases where either capacitance or inductance vary
with mechanical displacements (x or 𝜃).
To illustrate the method, consider the system studied in Example 8.6, repeated here in Figure 8.52. The system kinetic
coenergy is,
1 1
m ẋ 2 + m ẋ 2 .
Tẋ 1 ẋ 2 =
2 1 1 2 2 2
Now, for this particular problem, both flows, ẋ 1 and ẋ 2 , used to define the kinetic coenergy of the system are independent
of each other and therefore these flows can be taken as an acceptable set of generalized flows13 (generalized flows must be
independent).
In this case, the transformation matrix T reduces to the identity matrix and the generalized displacements are simply x1
and x2 . Generalized momenta are then,
𝜕Tẋ 1 ẋ 2
p1 = = m1 ẋ 1 , (8.49)
𝜕 ẋ 1
𝜕Tẋ 1 ẋ 2
p2 = = m2 ẋ 2 , (8.50)
𝜕 ẋ 2
which are simply the linear momenta of masses 1 and 2, respectively.
The potential energy stored in the two springs in terms of the displacements x1 and x2 is,
1 1
Ux1 x2 = k (x − x2 )2 + k2 x22 .
2 1 1 2
Note that these displacements are different than the ones that would be used for a standard bond graph set of displacement
states (standard states would be q1 = x1 − x2 and q2 = x2 ).

Figure 8.52 Two degree-of-freedom (dof) spring–mass system.

12 See Goldstein [68] or Lanczos [8].

13 Although an obvious pair, these two flows are not the only pair that could be used. As in any other system problem, the number of states
needed to define the system energy content is fixed but their exact form is not.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
518 8 Multiport Modeling and Energy Methods

To use Lagrange equations in their most direct form, the generalized displacements used to define potential energy are
the integrated flows used to define the kinetic coenergy. The generalized conservative efforts are then,
𝜕Ux1 x2 ( )
e1 = = k1 x1 − x2 ,
𝜕x1
𝜕Ux1 x2 ( )
e2 = = −k1 x1 − x2 + k2 x2 .
𝜕x2
The net power into the system due to sources and dissipation (dissipation is zero for this system) is,

Pn = F1 ẋ 1 + F2 ẋ 2 ,

from which we can identify that the nonconservative efforts are, E1 = F1 and E2 = F2 . Now we can write the 2n state
equations for this Lagrange system for j = 1, 2:
ṗ 1 = −k1 (x1 − x2 ) + F1 ,
ṗ 2 = k1 (x1 − x2 ) − k2 x2 + F2 ,
ẋ 1 = p1 ∕m1 ,
ẋ 2 = p1 ∕m2 .

These equations are equivalent to the second-order state equations derived in Example 8.6. As a Lagrange subsystem, the
bond graph appears as shown in Figure 8.53. Note that there is no need to include the transformation since it is just an
identity relation for this system.

Example 8.19 Revisiting two-dimensional spring–mass system

Reconsider the two-dimensional spring-mass system from Example 8.3, and repeated in Figure 8.54(a). Now, rather than
cartesian coordinates, use cylindrical coordinates, (r, 𝜃), and develop a Lagrange subsystem model.
Solution
Only two generalized coordinates are needed to model this system. This system has only a point mass, so the total kinetic
coenergy can be easily expressed in cartesian coordinates as,
1 2 1 2
Tẋ ẏ = mẋ + mẏ .
2 2

Figure 8.53 Lagrange subsystem bond graph for two degree-of-freedom spring–mass system. The T in this case is an identity matrix,
and thus not needed, but is included here for illustrative purposes.

(a) (b)

Figure 8.54 (a) Two-dimensional spring pendulum system and (b) transformation structure.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
8.4 Lagrange’s Equations in System Modeling 519

The transformation between cartesian coordinates and cylindrical coordinates is,

̇
ẋ = sin 𝜃 ⋅ ṙ + r cos 𝜃 𝜃,
̇
ẏ = cos 𝜃 ⋅ ṙ − r sin 𝜃 𝜃,
from which, we can form the following matrix relation between the m = 4 dependent flows and the n = 2 independent
flows,
⎡ ẋ ⎤ ⎡ sin 𝜃 r cos 𝜃 ⎤
⎢ ̇ ⎥ ⎢ ⎥[ ]
y cos 𝜃 −r sin 𝜃 ⎥ ṙ
ẋ = ⎢ ⎥ = Tq̇ = ⎢ ,
⎢ ṙ ⎥ ⎢ 1 0 ⎥ 𝜃̇
⎢ 𝜃̇ ⎥ ⎢ 0 1 ⎥
⎣ ⎦ ⎣ ⎦
and the holonomic result,
x = r sin 𝜃,
y = r cos 𝜃.
These relations can be represented by a multiport transformer as in Figure 8.54(b). Flow junctions are included for all
the dependent flows as an example of how the Lagrange subsystem interacts with non-conservative effects associated with
any of those variables.
Given the transformation relations, the kinetic coenergy can be expressed in terms of the independent flows and dis-
placements,
1 [ ]
̇ 2 .
Tṙ 𝜃ṙ = m ṙ 2 + (r 𝜃)
2
Generalized momenta are then defined by,
𝜕Tṙ 𝜃ṙ
p= ̇
= mr,
𝜕 ṙ
𝜕Tṙ 𝜃ṙ
h= ̇
= mr 2 𝜃,
𝜕 𝜃̇
where p is a translational momentum and h is an angular momentum. Note how the dependence of h on r is properly
accounted for through this energetically consistent relation.
The potential energy in terms of generalized displacements r and 𝜃 is,14
1
Ur𝜃 = k1 (r − lo )2 ,
2
where lo is the unstressed spring length, so conservative efforts are given by,
𝜕T ̇ ̇ 𝜕Ur𝜃
er = − r𝜃r + = −mr 𝜃̇ + k1 (r − lo ),
2
𝜕r 𝜕r
e𝜃 = 0.
Up to this point, we have composed the conservative effects of the Lagrange subsystem model. Any contributions to the
model that can be more readily or effectively modeled on the dependent flows side of the T contribute to the net power
into the Lagrange subsystem. These are effects that arise from sources and dissipation. A complete bond graph is shown in
Figure 8.55.
In this model, we consider the gravity effort source mg acting along the y-axis. There is also some linear pivot friction
that acts about 𝜃. For these external effects, the net power into the Lagrange system is,
n = 0 ⋅ ẋ + (mg) ⋅ ẏ + 0 ⋅ ṙ + (−𝜏B ) ⋅ 𝜃̇ = mg(cos 𝜃 ṙ − r sin 𝜃 𝜃)
̇ − 𝜏B 𝜃.
̇

From this relation, the nonconservative efforts are determined by collecting terms on each independent flow (or taking
partial derivatives with respect to each independent flow),
Fr = mg cos 𝜃,
𝜏𝜃 = −mgr sin 𝜃 − 𝜏B .

14 We could also have included gravity with an additional term −mgr cos 𝜃, but for illustration purposes, we are accounting for gravity here as a
source via the ẏ -dependent flow.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
520 8 Multiport Modeling and Energy Methods

Figure 8.55 Complete bond graph with Lagrange subsystem for spring-pendulum, with gravity force and pivot friction shown on the
“nonconservative” (left) side of the transformer.

Note that causality on the complete bond graph indicates four states as expected: p, h, r, and 𝜃. The state equations are then
derived as in equation 8.43 for momenta and equation 8.45 for displacements, representing the corresponding Lagrange
equations and kinematics, respectively. For this system, the state equations are:
ṗ = −er + Fr = −k1 (r − lo ) + mr 𝜃̇ + mg cos 𝜃,
2

ḣ = −e𝜃 + 𝜏𝜃 = −mgr sin 𝜃 − B𝜃, ̇

ṙ = p∕m,
𝜃̇ = h∕(mr 2 ),
where 𝜃̇ would substitute into the ḣ equation for completion.

Up to now, our examples of Lagrange equations have dealt with strictly mechanical systems. Although often only dis-
cussed in this context, these equations are not limited to mechanical systems. The following examples illustrate how
position-dependent inductance can be dealt with using the Lagrange subsystem approach.

Example 8.20 Lagrange of salient-pole machine

An EM system that can be effectively analyzed with Lagrange equations is the salient pole machine studied in Example 8.10.
Besides the magnetic energy stored in the machine, we will include in our analysis kinetic energy and dissipation of the
rotor with inertia J and linear bearing resistance b and electrical resistance in the coils of linear resistance R. It is desired
to find a set of state equations that describe the rotation of this system when we apply a voltage V = Vo sin 𝜔t.
Solution
The kinetic coenergy of the system, which includes magnetic coenergy is,
1 1
Ti𝜔r 𝜃 = (Lo + L2 cos 2𝜃)i2 + J𝜔2r ,
2 2
where 𝜔r = 𝜃̇ (rotor angular velocity, to distinguish from electrical voltage frequency, 𝜔). The potential energy is identically
zero for this problem. Generalized momenta are then derived as,
𝜕T ( )
𝜆 = 𝜕ii𝜔r 𝜃 = Lo + L2 cos 2𝜃 i,
𝜕T (8.51)
h = 𝜕𝜔i𝜔r 𝜃 = J𝜔r ,
r

and conservative efforts are

𝜕Ti𝜔r 𝜃
e𝜃 = − = L2 sin 2𝜃,
𝜕𝜃
eq = 0,
where q̇ = i. The net power into the system is,
n = (𝑣 − Ri)i + (−b𝜔r )𝜔r ,
thus nonconservative efforts are,
Ei = 𝑣 − Ri,
E𝜔r = −b𝜔r .
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
8.4 Lagrange’s Equations in System Modeling 521

Now, the Lagrange state equations yield,

𝜆̇ = 𝑣o sin 𝜔t − Ri,
ḣ = −L2 i2 sin 2𝜃 − b𝜔r .

Using expressions for generalized linkage 𝜆 and angular momentum h relations above, current i and angular velocity 𝜔r
can be eliminated from these expressions to obtain a complete set of state equations as,

𝜆̇ = 𝑣o sin 𝜔t − 𝜆
L2 sin 2𝜃 b
ḣ = − ( )2 𝜆 − J h
2

Lo + L2 sin 2𝜃
̇𝜃 = h∕J.

Example 8.21 EM solenoid system

Consider the EM system shown in Figure 8.56. Develop a model using a Lagrange approach, with the Lagrange subsystem
incorporating both the EM and the mechanical-coupled energy.
Solution
This example includes not only electromechanics but also dependent energy storage which occurs due to a kinematic
̇ z,
constraint. Note there are three dependent flows (q, ̇ indicated on the graph, but ż and ẏ are dependent due to the
̇ y)
transformer constraint,

y2 + z 2 = r 2 ,

or

yẏ + zż = 0.

Choosing independent coordinates as charge q and vertical position y, we have the following form relating dependent
and independent flows:

⎡ q̇ ⎤ ⎡1 0 ⎤[ ]
q̇
ẋ = ż = Tq̇ = ⎢ 0 n( y) ⎥
⎢ ⎥ ,
⎢ ⎥ ⎢ ⎥ ẏ
⎣ ẏ ⎦ ⎣0 1 ⎦
√
where n(y) = −y∕ r 2 − y2 . We can now anticipate the bond graph will take the form shown in Figure 8.57.

Figure 8.56 Electromechanical solenoid-linkage system schematic.

Centered in
solenoid

Massless, rigid
bar
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
522 8 Multiport Modeling and Energy Methods

Figure 8.57 Bond graph for the solenoid-linkage system. Note that the nonconservative forces are Eq = Vs (t) − Rq̇ and
Ey = −Fb n(y) − mg.

The kinetic coenergy Tq̇ yy

̇ can then be formulated as,

1 2 1 1
Tq̇ yy
̇ = mẏ + Mn( y)2 ẏ 2 + L( y)q̇ 2 ,
2 2 2
and the potential energy is,
1
Uy = K( y − yo )2 ,
2
where yo is the free length position of the spring.
First, find the generalized momenta,
𝜕Tq̇ yy
̇
𝜆= = L( y)q̇ (as expected),
𝜕 q̇
𝜕Tq̇ yy
̇
p= ̇
= (m + Mn2 )y.
𝜕 ẏ
Then, we see that the conservative efforts are,
𝜕Tq̇ yy
̇ 𝜕Uq̇ ẏ
eq = − + = 0,
𝜕q 𝜕q
𝜕Tq̇ yy
̇ 𝜕Uq̇ ẏ dn q̇ dL
2
ey = − + = −Mnẏ 2 − + K( y − yo ).
𝜕y 𝜕y dy 2 dy
The net power into the Lagrange subsystem is (by summing the net powers on the left side of the T):

 = (𝑣 − Rq) ̇ ⋅ q̇ + (−bn2 ẏ − mg) ⋅ y,

̇ ⋅ q̇ + (−Fb ) ⋅ ż + (−mg) ⋅ ẏ = (𝑣 − Rq) ̇

̇ and remember n = n(y). The nonconservative efforts are thus,

where we used Fb = bż = bny,

̇
Eq = Vs (t) − Rq,
Ey = −bn2 ẏ − mg.

Now we can write the state equations:

𝜆̇ = −eq + Eq = Vs (t) − Ri = −R𝜆∕L( y) + Vs (t)

dn q̇ dL
2
ṗ = −ey + Ey = Mnẏ 2 + − K( y − yo ) − bn2 ẏ − mg
dy 2 dy
q̇ = 𝜆∕L( y)
ẏ = p∕(m + Mn2 ),

where it should be noted that n = n(y) and a final form would replace q̇ and ẏ with their final forms above. Also, the q state
is strictly not needed and the number of state equations could be reduced to three.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
8.4 Lagrange’s Equations in System Modeling 523

8.4.4 Dealing with Variable-Mass Effects

It was shown in Chapter 4 that systems with variable mass can be modeled using controlled sources. This is one way to
account for the momentum flux term that arises in Newton’s law for open systems. Whether for a closed (constant mass) or
open (variable mass) system, Newton’s law dictates that for a translating mass [178],

ṗ = m𝑣̇ = F − 𝜙, (8.52)

where p = m𝑣 is the momentum of the changing system mass, m, and 𝑣 is velocity. In this form, F represents the net external
forces and 𝜙 is the momentum flux out of the system. For closed systems, 𝜙 = 0. See discussions and example applications
in Tiersten [83] and Siegel [84]. In Chapter 4, for example, 𝜙 took the form 𝜙 = ṁ ⋅ u, where ṁ was the rate of the mass
leaving a system and u the (relative) velocity of mass flowing out of a system. Clearly momentum flux has units of force, and
thus one way to account for this in a bond graph is by attaching an effort at the same velocity (1-junction) as a particular
system mass in question (e.g., see Section 4.7.6). Using a controlled-source enables equation 8.52 to be represented within
a bond graph model. While the approach may be suitable for some cases, it may not scale and can lead to errors if not
energetically consistent. It may also be difficult to see how to combine multiple components and/or additional effects in
more complex systems. For example, not all system components with variable-mass effects have constant volume. Physical
components may have changes in size (length, volume, etc.) that also need to be considered in a systematic manner. Lastly,
there could be thermodynamic considerations within these contexts.
This section describes how variable-mass effects can be accounted for by using a Lagrange subsystem approach. A
Lagrange subsystem can be effective when kinetic energy is influenced by variations in displacement variables associated
with volume or length changes. Fortunately, these effects can usually be modeled by a conservative effort within a Lagrange
subsystem approach. What remains is to show how the variable-mass effect can be modeled by a nonconservative effort to
account for momentum flux. Introducing these two key efforts can provide a basis for modeling many practical situations
where variable-mass effects may arise. Each problem may also require modeling interconnections to other physical effects.
In some situations, all that may be known is that mass (or inertia) of an element varies as a function of time, say, m(t).
If the means by which this change is occurring is not influenced by the system under study, then the controlled source
model approach in Chapter 4 may be suitable. As with any ideal source model, it is assumed that there is no “back effect”
on how the mass will vary, but there will be an influence on the system. In those cases, a momentum flux is defined by
̇ with u the relative velocity between the “main” mass 𝑣, and the velocity of the added mass. Without any additional
𝜙 = mu,
information about how the mass is changing, a controlled source method would be a practical way to account for the
effect. It is also necessary to assign a proper sign convention for the momentum flux. In a Lagrange approach, this extra
term can be introduced as a nonconservative effort. In Step 6 for formulating a Lagrange subsystem model in Section 8.4.2,
nonconservative efforts are identified by a net input power relation, associating specific efforts (forces or torques) with the
independent flows (e.g., velocities).
To illustrate this approach, recall the simple case shown in Figure 8.58(a), where a mass slides with friction while mass is
added at a known rate ṁ with velocity 𝑣o . This example was considered in Chapter 4 (see Figure 4.74). The kinetic coenergy
for this simple case is,

1
Tẋ = ̇ ẋ 2 ,
(m + mt)
2 o

Friction
Convective interface
(a) (b)

Figure 8.58 (a) System with variable-mass (b) Model using a Lagrange subsystem form.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
524 8 Multiport Modeling and Energy Methods

so px = 𝜕Tẋ ∕𝜕 ẋ = (mo + mt)

̇ x.̇ With potential energy Ux = 0, the conservative force is ex = −𝜕Tẋ ∕𝜕x + 𝜕Ux ∕𝜕x = 0. Finally,
the nonconservative forces are found by examining net input power,
n = −Fb ẋ + 𝜙ẋ + F(t)ẋ = (−Fb + 𝜙 + F(t))ẋ = Ex x,
̇
̇ o , 𝑣o is the velocity of the input mass, and Fb = b𝑣. Now the state equations are,
where 𝜙 = m𝑣
ṗ x = −ex + Ex = −Fb + 𝜙 + F(t) = −b𝑣 + m𝑣
̇ o + F(t),
ẋ = px ∕(mo + mt).
̇
This model is conveyed using a Lagrange subsystem in Figure 8.58(b). Note that the convective interface shows momentum
effectively taken away as mass is added. To show this model aligns with the conventional form typically derived for these
problems, express as a single model equation in terms of the velocity, 𝑣. First differentiate the generalized momentum,
d [ ]
ṗ x = ̇ ẋ = m(t)𝑣̇ + m𝑣,
(mo + mt) ̇
dt
where m(t) = mo + mt. ̇ Then,
ṗ x = m(t)𝑣̇ + m𝑣
̇ = −b𝑣 + m𝑣
̇ o + F(t),
and a model equation can then be expressed,
̇ − 𝑣o ) − b𝑣.
m(t)𝑣̇ = −m(𝑣
This approach can be applied to variable-mass problems with specified mass rates, as described in dynamics textbooks [12].
These problems do not typically specify how the mass changes relate to changes in system characteristics. Some problems
with variable-mass will depend or cause changes on a displacement state. In those cases, the kinetic coenergy and potential
energy functions will be able to model these effects in the Lagrange subsystem. Typical examples include when a mass is
attached to and takes-up increasing lengths of rope or chain with specified mass per unit length. In these types of situa-
tions, there are two key changes that should arise in a Lagrange subsystem model: (i) the conservative effort will include
components due to how potential energy and/or kinetic conenergy vary with a given displacement (step 5 in Section 8.4.2),
and (ii) the effect of convected kinetic energy into or out of the system should be included. The latter enables modeling
power and convected energy interactions with other system components and/or the environment. The convective interface
warrants further discussion.
A convective interface accounts for kinetic energy being conveyed between system elements. A positive sign can be asso-
ciated with kinetic energy convected into a system. On a power bond, the effort variable is defined by the kinetic energy
per unit displacement variable, while the flow variable is taken as the time rate of change of the displacement variable.
For example, consider a translational system with a lumped mass and an attached cable or rope. As the mass moves, the
amount of mass added to (or released from) the system mass will have an associated convected kinetic energy per unit
length as the effort variable and velocity as the flow variable. These variables will form a power conjugate pair. Similar sets
of variable pairs can be defined to model convected energy in rotational, fluid, and thermal–fluid systems. The following
example shows how this approach can be used to model the variable-mass effect in the classical Taite-Steele falling chain.

Example 8.22 Tait-Steele falling chain

A classical dynamics problem considers a rope or chain that falls from a table as shown in Figure 8.59(a). The chain could
be piled so the length that remains on the table is not initially moving or the section extended along Region 1 could be
initially moving with the part that falls (Region 2). In the idealized case, Region 2 is isolated as in part (b) of the figure,
with no friction (e.g., no links rubbing or in contact with the table; no aerodynamic friction). It is also assumed that the
chain does not pile up vertically, so there is no potential energy in Region 1. For this case, the classical model equation is
[179, 180],
z2 z̈ 2 = gz2 − ż 22 , (8.53)
where z2 is the length extending from the top of the table (Region 2), and thus z1 + z2 = l. The total mass is taken as m so
the mass per unit length is 𝜌l = m∕l.
The falling chain problem is often used as an example of a variable-mass system. Show how a bond graph approach can be
used to model the classical falling chain problem by accounting for the change in length and by using a convective interface
to model the effect of variable mass.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
8.4 Lagrange’s Equations in System Modeling 525

Region 1

Region 2

(a) (b)

Figure 8.59 (a) Schematic of a falling chain/rope of the Tait-Steele type. (b) Model of falling section with tension and momentum
flux.

Derivative causality

Region 1 Region 2

Sign change to Convective interface

enforce
(fixed length)

Figure 8.60 A bond graph model for the falling chain problem, including representation of both the falling region 2 and the feed
section region 1.

Solution
Wong [58] and Denny [180, 181] describe classical Lagrange approaches to the falling chain problem. In the following, a
bond graph and Lagrange subsystem approach are used to illustrate modeling the interaction of two elements with variable
mass. Each region of the changing length, z1 and z2 , will define a Lagrange subsystem. The key relations (following steps in
Section 8.4.2) are summarized in Table 8.6. A bond graph can be constructed by using a Lagrange subsystem to represent
each size/mass variable region of the falling chain as shown in Figure 8.60. Each subsystem represents the energy stored
[ ]T
for each pair of state variables (z1 , p1 ) and (z2 , p2 ); however, causality indicates the states are x = z1 , z2 , p2 since there is
a derivative causality and p1 is dependent. There are two paths that connect the regions: (i) the tension force, with cable
assumed inextensible, and (ii) the convective interface, to represent exchange of momentum between regions 1 and 2.
Note that the momentum flux here is modeled using a modulated gyrator (see similar forms in fluid–structure interaction,
Section 4.7.5). Another point about the bond graph is the change in sign on ż 1 to enforce the constant chain/rope constraint,
z1 + z2 = l, which requires, ż 1 = −ż 2 .
The bond graph reveals that the tension force right at the input to region 2 can be determined by,
Ft = ṗ 1 + e1 + 𝜙1 ,
where the defined generalized momentum,
ṗ 1 = 𝜌l 𝑣21 + 𝜌l z1 𝑣̇1 ,
and from the bond graph,
1 2
𝜙1 = 𝜌𝑣 .
2 l 2
Thus, the tension force for the case where the chain is piled up and 𝑣̇ 1 ≈ 0 is given by, and since 𝑣1 = 𝑣2 ,
1 1
Ft = 𝜌l 𝑣22 − 𝜌l 𝑣22 + 𝜌l 𝑣22 = 𝜌l 𝑣22 , (8.54)
2 2
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
526 8 Multiport Modeling and Energy Methods

Table 8.6 Summary of Lagrange subsystem equations for falling chain, regions 1 and 2.

Region 1 Region 2

Kinematic variables z1 , 𝑣1 z2 , 𝑣2
1
Kinetic coenergy T1∗ = 𝜌 z 𝑣2
2 l 1 1
T2∗ = 12 𝜌l z2 𝑣22
𝜕T1∗ 𝜕T2∗
Generalized momentum p1 = 𝜕𝑣
= 𝜌l z1 𝑣1 p2 = 𝜕𝑣2
= 𝜌l z2 𝑣2
1
−z2
Potential energy U1 = 0 (stays on table) U2 = ∫0 (𝜌l dz2 )gz2 = − 12 𝜌l gz22
𝜕T1∗ 𝜕U1 𝜕T2∗ 𝜕U2
Conservative efforts e1 = − 𝜕z + 𝜕z1
e2 = − 𝜕z + 𝜕z2
1 2

e1 = − 21 𝜌l 𝑣21 +0 e2 = − 21 𝜌l 𝑣22 − 𝜌l z2 g
Nonconservative efforts E1 𝑣1 = Ft 𝑣1 − 𝜙1 𝑣1 E2 𝑣2 = −Ft 𝑣2 + 𝜙2 𝑣2
(power in/out) ⇒ E1 = F t − 𝜙1 ⇒ E2 = −Ft + 𝜙2
Momentum equation ṗ 1 = −e1 + E1 ṗ 2 = −e2 + E2
Motion equation ż 1 = p1 ∕(𝜌l z1 ) ż 2 = p2 ∕(𝜌l z2 )

Total Total
volume, volume,

Region 2 Region 2
Region 1 Region 2
partially filled
(a) (b) (c)

Figure 8.61 (a) Piston 1 drives fluid from region 1 to region 2, which is completely filled and exits to atmosphere. (b) A basic bond
graph to model the system in (a). (c) Detail of region 2 partially filled with a changing volume and including piston 2.

agreeing with the results in Wong et al. [182] and Denny [180].
The state equations can now be completed as,
ż 1 = −𝑣2
ż 2 = p2 ∕𝜌l z2
ṗ 1 = −e2 + E2
1 1
= + 𝜌l 𝑣22 + 𝜌l gz2 − 𝜌l 𝑣22 + 𝜌l 𝑣22
2 2
= +𝜌l gz2 .
At the end of the day, these equations combine to give the classical result in nth order form,
ṗ 2 = 𝜌l 𝑣22 + 𝜌l z2 𝑣̇ 2 = 𝜌l gz2 ,
or,
z2 z̈ 2 = gz2 − ż 22 ,
as in equation 8.53.
In applying the Lagrange subsystem approach to this known variable-mass problem, we can show how to arrive at the
same known solution. A useful pattern is now provided for guiding how other more complex problems that have interacting
elements of variable-mass elements might be modeled with Lagrange subsystems.

While modeling of variable-mass effects date back to Cayley [183] and Tait and Steele [179] in the late 1800s, there has
been ongoing and practical interest, especially for more complex systems. Using a Lagrange formulation has typically
proved effective, particularly for dealing with how stored kinetic energy (or coenergy for convenience) may depend on
flow (or momentum) as well as displacement variables. Beaman and Breedveld [184, 185] used a system such as shown
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
8.4 Lagrange’s Equations in System Modeling 527

in Figure 8.61(a) to demonstrate an approach and also to illustrate a way to account for open system effects. For a closed
system case, first consider Region 2 being filled with an incompressible fluid as shown, for which a basic bond graph as
in Figure 8.61(b) may suffice to represent the total inertial and loss effects, albeit for a limited range of operation. Beaman
and Breedveld introduced a case where the volume of fluid in region 2 may be changing, as shown in Figure 8.61(c) where
the fluid is contained by a second moving piston. Rather than a fixed fluid inertia as in (b), this case has a fluid inertia that
depends on the changing displacement variable –V2 .
Assume the kinetic coenergy in region 1 is very small compared to that in region 2 (area 1 is very large compared to area
2), the kinetic coenergy in region 2 can be written,
( )2
1 1 Q
T(Q, –V2 ) = TQ–V2 = m𝑣22 = 𝜌–V2 .
2 2 A2
The generalized (fluid) momentum can now be determined using the derivative relation,
𝜕TQ–V2 𝜌–V2 Q
Γ= = = Γ(–V2 , Q),
𝜕Q A22
showing either Γ or Q can be chosen as a system state, and that both depend on V
– 2 . Choosing Γ, the stored kinetic energy
is, TΓ–V2 = A22 Γ2 ∕(2𝜌–V2 ).
Considering now a closed system as in Figure 8.62, where external forces F1 and F2 are applied, the rate of change of
stored kinetic energy is,
𝜕TQ–V2 𝜕TQ–V2 𝜕TQ–V2
= Γ̇ + –̇ 2 = Γ̇ ⋅ Q + e2 ⋅ –
V V̇ 2 = F1 𝑣1 − F2 𝑣2 − R ,
𝜕t 𝜕Γ 𝜕 –V2
where power losses have also been included for the converging section as R . Figure 8.62(b) shows how power flows into
the two ports of a Lagrange subsystem C (coupled energy storage).15 Since the flow rate is common in this case, a 1-junction
structure interconnects all the power flows. The two state equations for this system are found as,

Γ̇ = −e2 + F1 ∕A1 + F2 ∕A2 − PR (Γ), (8.55)

–̇ 2 = Q = ΓA22 ∕𝜌–V2 ,
V (8.56)

where PR (Γ) is a pressure drop that depends on state Γ. Note the introduction of e2 as a conservative effort associated with
the changing volume and given by, e2 = 𝜕TΓ–V2 ∕𝜕 – V2 . This result would also follow from e2 = 𝜕TQ–V2 ∕𝜕 –
V2 , as in step 5 of the
procedure given in Section 8.4.2.
Beaman and Breedveld proposed a change to an open system by removing piston 2 as in Figure 8.63. In this case, region
2 is filled to volume state V
– 2 by flow, Q. There is no exit flow, Qo , until –V2 reaches –VT ; that is,
{
0, if –V2 < –VT
Qo = . (8.57)
Q, if –V2 = –VT

Changing Energy storage

volume,

Region 2

Region 1
(a) (b)

Figure 8.62 (a) A closed fluid system where an incompressible fluid flows in a converging section. The volume of fluid within
region 2, –
V2 , changes. (a) A bond graph with Lagrange subsystem to represent kinetic energy storage with volume dependence.

15 Note that this multiport can also be represented by a mixed-energy IC multiport as in Karnopp et al. [38].
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
528 8 Multiport Modeling and Energy Methods

Total volume, Energy storage

Region 2

Region 2
partially filled Convective interface

(a) (b)

Figure 8.63 (a) The system of Figure 8.62 as an open system. The volume of fluid within region 2, – V2 , changes and can exit.
(a) A bond graph with Lagrange subsystem to represent kinetic energy storage with volume dependence, including a convective
interface to model expelled kinetic energy at the outlet. Resistive effects in the converging nozzle are not included here.

When flow exits, the model should also account for convected kinetic energy at the outlet by a conditional relation,
{ 2
𝛾 ∕2𝜌, Qo > 0
𝜅= , (8.58)
0, Qo ≤ 0
where 𝛾 = A2 Γ∕–V2 is the fluid momentum per unit volume. The bond graph in Figure 8.63(a) shows one way to construct
this conditional effect [185]. The modulation variable 𝜈 is given by,
{
1, –V2 ∕–VT = 1 (filled)
𝜈= . (8.59)
0, –V2 ∕–VT < 1
With this construction, the state equations now become,

Γ̇ = F1 ∕A1 − Po − 𝜈 ⋅ (𝜅 − e2 ), (8.60)

–̇ 2 = Q − Qo .
V (8.61)

A simpler open system can be formed simply by having a leaking piston, but in that case, there would be no need for the
convective interface. Similar applications of these methods can be found in Willson and Traver [77] as well as Redfield
[186]. Beaman and Breadveld also showed how these methods could be extended to an open and moving control volume
system. Further discussion on open system effects for thermodynamic systems is given in Chapter 9.

8.4.5 Derivation of Lagrange Equations

A derivation of Lagrange equations is presented here that is independent of physical domain, showing that these equations
are a direct consequence of energy in the form of a power balance (action–reaction in Newton’s terminology). Therefore,
starting with a power balance for the input bonds on the multiport I element in Figure 8.50(a) or the gyrationally coupled
C element in Figure 8.50(b), we have
∑
m
∑
m
ei ẋ i = ṗ i ẋ i ,
i=1 i=1

which can be combined with the kinematic constraint (insert equation number) to yield
∑
m
∑
n
∑
m
∑
n
ei Tij q̇ j = ṗ i Tij q̇ j . (8.62)
i=1 j=1 i=1 j=1

For holonomic systems, this becomes

∑m
∑
n
𝜕xi ∑m
∑
n
𝜕xi
ei q̇ j = ṗ i q̇ j . (8.63)
i=1 j=1
𝜕q j i=1 j=1
𝜕q j
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
8.4 Lagrange’s Equations in System Modeling 529

Using kinetic coenergy, which is defined as

∑
m
Tẋ = p ⋅ dẋ = pi dẋ i ,
∫ ∫ i=1

the momentum can be derived as

𝜕Tẋ
pi = .
𝜕 ẋ i
Using this result in 8.63 gives
[ ] n
∑m
∑
n
𝜕xi ∑m
d 𝜕Tẋ ∑ 𝜕xi
ei q̇ j = q̇ ,
i=1 j=1
𝜕qj i=1
dt 𝜕 ẋ i j=1 𝜕qj j

or equivalently
[ ( ) ( )]
∑m
∑
n
𝜕xi ∑∑ d 𝜕Tẋ 𝜕xi
m n
𝜕Tẋ d 𝜕xi
ei q̇ = ⋅ − ⋅ q̇ j .
i=1 j=1
𝜕qj j i=1 j=1 dt 𝜕 ẋ i 𝜕qj 𝜕 ẋ i dt 𝜕qj

Noting that
( )
𝜕xi 𝜕 ẋ d 𝜕xi 𝜕 ẋ i
= i and = ,
𝜕qj 𝜕 q̇ j dt 𝜕qj 𝜕qj

and reversing the order of summation yields the result

[m ] [ ( ) ]
∑ n
∑ 𝜕xi ∑n
d 𝜕Tqq ̇ 𝜕Tqq ̇
ei q̇ j = − q̇ j ,
i=j i=1
𝜕qj j=1
dt 𝜕 q̇ j 𝜕 q̇ j

where now the coenergy Tẋ = Tqq ̇ is expressed as a function of both flow q̇ and displacement q rather than x,
̇ i.e., Tẋ (x)
̇ =
̇
̇ (T(q)q).
Tqq
Now since all the q̇ j are independent, this implies that,
( )
∑m
𝜕xi d 𝜕Tqq ̇ 𝜕Tqq ̇
ei = − j = 1, … , n. (8.64)
i=1
𝜕q j dt 𝜕 ̇
q j 𝜕 ̇
q j

∑
m
As indicated in Figure 8.50(a) and (b), the effort variable in the expression ei 𝜕xi ∕𝜕qj can be split into two terms, one due
i=1
to potential energy deviation and one due to nonconservative efforts as,
m ( )
∑m
𝜕xi ∑ 𝜕Ux 𝜕xi ∑ m
𝜕Ux 𝜕xi ∑
m
𝜕x
ei = − + enci =− + enci i ,
i=1
𝜕q j i=1
𝜕x i 𝜕q j i=1
𝜕x i 𝜕q j i=1
𝜕qj
where Ux is potential energy expressed in terms of x. We can then express the first term as,
∑m
𝜕Ux 𝜕xi 𝜕Uq
= ,
i=1
𝜕xi 𝜕qj 𝜕qj
where Uq is potential energy expressed in terms of independent displacement q. Using this result one form of Lagrange
equations, can then be expressed as,
( )
𝜕Uq ∑ m
𝜕xi d 𝜕Tqq ̇ 𝜕Tqq ̇
− + enci = − j = 1, … , n. (8.65)
𝜕qj i=1
𝜕q j dt 𝜕 ̇
q j 𝜕 ̇
q j

Defining p̃ j as a generalized momentum p̃ j = 𝜕Tqq

̇ ∕𝜕 q̇ j , ej a conservative effort ej = −𝜕Tqq
̇ ∕𝜕qj + 𝜕Uq ∕𝜕qj , and Ej as a non-
∑
m
∑m
conservative effort Ej = enci 𝜕xi ∕𝜕qj = enci Tij , Lagrange equations can also be expressed as,
i=1 i=1

p̃̇ j = −ej + Ej j = 1, … , n.

which is identical to the previously stated equation 8.44.

 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
530 8 Multiport Modeling and Energy Methods

This equation is equivalent to the bond graph representation Figure 8.51 in which all energy is combined into a single
multiport C element. The gyrationally coupled bonds attached to this element pertain to generalized momenta while the
remaining attached bonds yield conservative efforts. There is no strict division of kinetic and potential energy into separate
elements.
The conservative effort term ej results from change in the difference between kinetic coenergy and potential energy with
displacement deviation as

ej = 𝜕(−Tqq
̇ + Uq )∕𝜕qj .

We will sometimes find it useful to use kinetic energy rather than coenergy. Kinetic energy Tpq
̃ can be related to coenergy
̇ by the Legendre transformation,
Tqq

Tpq ̃ q,
̇ =p
̃ + Tqq ̇

and the conservative effort term becomes,

ej = 𝜕(+Tpq
̃ + Uq )∕𝜕qj .

Note the similarity between this effort and the electromagnetic forces and torques that are derived by the derivative relations
on energy. In the following, we will generalize this observation.

8.4.6 Hamiltonian and CoLagrange Formulations

In the discussion of Lagrange equations, kinetic energy (I elements) was the basis for the derivation with potential energy
(C elements) taking a subordinate role. In this case, the potential energy stored in the C elements is expressed as a function
of the integrated flow variables of the I elements, q. This classical Lagrangian method was historically the first introduced
to deal with generalized state determined systems. However, it was not realized until comparatively recently that there also
exists a completely dual form16 of Lagrange equations in which potential energy takes the prominent role, with the kinetic
energy stored in the I elements expressed as a function of the integrated effort variables of the C elements, that is, e [187].
Rather than use kinetic coenergy and potential energy as in the classical method, the dual formulation uses potential coenergy
and kinetic energy to define the motion of systems. The two formulations are expressed side-by-side in Table 8.7.

Table 8.7 Table of Lagrange and Hamilton equation forms.

Classical Lagrange form Dual Lagrange form

( ) ( )
d 𝜕 𝜕 d 𝜕∗ 𝜕∗
− =E − =F
dt 𝜕f 𝜕q dt 𝜕e 𝜕p
 = Tf − Uq ∗ = Ue − Tp
E = nonconservative efforts F = nonconservative flows
Classical Hamiltonian form Dual Hamiltonian form
𝜕H 𝜕H
q̇ = q̃̇ = − +F
𝜕 p̃ 𝜕p
𝜕H 𝜕H
p̃̇ = − +E ṗ =
𝜕q 𝜕 q̃
H = Tp̃ + Uq H = Tp + Uq̃
Intrinsic Hamiltonian form
𝜕H
p̃̇ = − +E
𝜕 q̃
𝜕H
q̃̇ = − +F
𝜕 p̃
H = Tp̃ + Uq̃

16 In system modeling, a dual form relates to another by interchanging pairs of terms. For example, we might refer to a 0-junction as the dual of
the 1-junction.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
8.4 Lagrange’s Equations in System Modeling 531

Figure 8.64 A series spring, mass, and damper system and associated bond
graph.

Example 8.23 Spring–mass–damper model using dual Lagrange formulation

Use the dual form of Lagrange equations to derive a set of state equations for the mechanical system having a spring,
damper, and mass connected in series, as shown in Figure 8.64.
Solution
In this system, we find that the velocity ẋ m used to define kinetic coenergy of the mass, Tẋ m = mẋ 2m ∕2, is not directly related
to spring displacement, xs , which is used to define potential energy, Uxs = Kxs2 ∕2. Classical Lagrange methods are more
preferred when the generalized coordinates are directly related, including through a holonomic constraint.
In this case, however, we find that the force, Fs , that defines spring potential coenergy, UFs = Fs2 ∕2K, is directly related
to mass momentum, p, by ṗ = Fs . This is true because the forces on the spring and damper are common, as shown in the
bond graph of Figure 8.64. This momentum defines mass kinetic energy, Tp = p2 ∕2m. In this case, the dual Lagrange form
is more applicable for deriving the system model. From Table 8.7,
∗ = Fs2 ∕2K − p2 ∕2m.
Thus
( ) ( )
d 𝜕∗ 𝜕∗ d Fs
− = + p∕m = 𝑣nc .
dt 𝜕Fs 𝜕p dt k
The nonconservative flow, 𝑣nc , associated with the force Fs is identified here at the 0-junction, thus 𝑣nc = Vb = −Fs ∕b.
This allows us to write now the state equations,
Ḟ s = −kp∕m − Fs ∕b,

ṗ = Fs .
This example demonstrates how the use of the dual Lagrange (or CoLagrange) formulation can lead to first-order state
equations with effort and momentum variables as states.

Lagrange equations (both classical and dual forms) can also be expressed in classical Hamiltonian form by using a Legen-
̇ and kinetic coenergy
dre transformation. For the classical Lagrangian, this consists of changing from flow variables, f = q,
to generalized momentum variables, p̃ = 𝜕Tf ∕𝜕f, and kinetic energy as,
H = p̃ ⋅ f −  = p̃ ⋅ f − Tf + Uq = Tp̃ + Uq ,
where H, the Hamiltonian, is the total stored system energy, that is, H = .
Now with this substitution, state equations can be written in terms of the generalized momentum and displacement
variables as follows:
𝜕H
p̃̇ = − + E, (8.66)
𝜕q
𝜕H
q̇ = , (8.67)
𝜕 p̃
where we have used the fact that,
𝜕H 𝜕 𝜕T 𝜕Uq
=− =− f + ,
𝜕q 𝜕q 𝜕q 𝜕q
𝜕H
= f.
𝜕 p̃
This form is also summarized in Table 8.7.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
532 8 Multiport Modeling and Energy Methods

̇ and potential coenergy

From the dual Lagrange (or CoLagrange) equations, we can change from effort variables, e = p,
to displacement variables, q̃ = 𝜕Ue ∕𝜕e, and potential energy as,
H = e ⋅ q̃ − ∗ = e ⋅ q̃ − Ue + Tp = Tp + Uq̃ .
With this substitution, we have
𝜕H
ṗ = 𝜕 q̃
, (8.68)

q̃̇ = − 𝜕H
𝜕p
+ F. (8.69)
Note that in general p ≠ p̃ and q ≠ q̃ for the same system.

Example 8.24 Spring–mass–damper model using Hamiltonian formulations

Use Hamiltonian methods to obtain state equations for the system depicted in Figure 8.65, a spring–mass–damper system
with a prescribed platform motion, Vo (t), and a prescribed force on the mass, Fo (t).
Solution
We will use three different sets of state variables by which to obtain the Hamiltonian in order to show the variety of forms
Hamilton’s equations might take for the same system. Starting from a classical Lagrangian method, we would use momen-
tum and displacement of the mass as states. Using these states to express the Hamiltonian yields,
H(pm , xm , t) = Tpm + Uxm = p2m ∕2m + K(xm − xo (t))2 ∕2,
where xo (t) = ∫ Vo (t)dt. Note the Hamiltonian is an explicit function of time due to the presence of xo (t) in the expression.
State equations become,
𝜕H
ṗ m = − + Fo (t) = −K(xm − xo (t)) + Fo (t),
𝜕xm
𝜕H
ẋ m = = pm ∕m.
𝜕pm
For the dual Lagrange, we would use momentum and displacement of the spring as states resulting in,
H(ps , xs , t) = Tps + Uxs = (ps − po (t))2 ∕2m + Kxs2 ∕2,
where po (t) = ∫ Fo (t)dt. Now the state equations are,
𝜕H
ṗ s = − = Kxs ,
𝜕xs
𝜕H
ẋ s = + Vo (t) = −(ps − po (t)) + Vo (t).
𝜕ps
Finally, we could use the intrinsic energy states, mass momentum, and spring displacement, resulting in Hamiltonian,
H(pm , xs ) = Tpm + Uxs = p2m ∕2m + Kxs2 ∕2,
and state equations,
𝜕H
ṗ m = − + Fo (t) = Kxs + Fo (t),
𝜕xs
𝜕H
ẋ s = − + Vo (t) = −pm ∕m + Vo (t).
𝜕pm

Figure 8.65 Spring–mass–damper with moving foundation. (a)

Schematic. (b) Bond graph.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
8.5 Variational and Minimum Principles 533

Even though all three Hamiltonians are numerically equal, note that only the use of intrinsic Hamiltonian variables
renders a Hamiltonian (and therefore an energy function) free from explicit time dependency.

8.4.7 Summary
This section reviewed the use of Lagrange equations in different forms:

● Classical form, no input: equation 8.37

● Lagrange subsystem form: 8.43 (including 8.44)
● Extended classical forms: 8.46 and 8.47.

All of these methods provide a systematic approach that includes several key steps:

1) Expressing kinetic coenergy in terms of independent flow variables.

2) Expressing potential energy in terms of integrated independent flow variables (displacements).
3) Obtaining generalized nonconservative efforts as E = enc ⋅ T, where T is a constraint transformation matrix.

A consistent way for dealing with variable-mass effects was described, building on discussion about this class of problems
from Chapter 4. In addition, a derivation of the Lagrange equations using effort-flow variables was provided to reinforce the
relation with a bond graph model, and both dual Lagrange (or CoLagrange) and Hamiltonian formulations were reviewed
and shown to be related to Lagrange forms by a Legendre transformation.

8.5 Variational and Minimum Principles

8.5.1 Introduction to Calculus of Variations

In analysis, design, and modeling in engineering and physics, we are often interested in functions that minimize integral
forms. The branch of mathematics that deals with this problem is called Calculus of Variations. A typical problem of Cal-
culus of Variations, and one of the earliest to be analyzed, is the brachistochrone problem which was first formulated and
solved by John Bernoulli in 1696.17 In this problem, a trajectory of minimum time is to be found for a particle falling in a
uniform gravity field in going from a point A to a point B as shown in Figure 8.66.
The time for the particle to transverse the distance from A to B can be shown to be,
√
b
1 1 + (dy∕dz)2
t= √ √ dz.
2g ∫a yi − y
The trajectory, y(z), has to be chosen to minimize t and also intersect the points A and B.
For the present, we will be interested in problems similar to the brachistochrone; problems in which, given a functional
 of the form,
tf
= ̇ t)dt,
(x, x,
∫ti
and also given boundary values xi = x(ti ) and xf = x(tf ), we wish to find the curve, x(t) which minimizes  .

Figure 8.66 Brachistochrone problem: mass particle constrained to a frictionless path from A to
B in a gravity field. What is minimum time path?

17 Lanczos [8].
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
534 8 Multiport Modeling and Energy Methods

Figure 8.67 Optimal path x(t) and comparison path x∗ (t) used to derive necessary conditions for
optimal path.

8.5.1.1 Derivation of Euler–Lagrange Equations

We start with a problem for which the brachistochrone is a prototype. The class of problem we will initially study is problems
in which  depends on three scalar variables, x, x(t), and t.18 In addition, we assume that  is twice differentiable with
respect to its arguments. The condition for  [x(t)] to be minimum can be stated as:

Given x(t), the path which minimizes  , and given any other path x∗ (t), then  [x(t)] ≤  [x∗ (t)].

This implies that to find an extremal path, one method would be to compare it with other possible paths as shown in
Figure 8.67.
Theoretically, we should allow x∗ (t) to vary over all possible trajectories and then pick x(t) as the trajectory that minimizes
 . Since this is not mathematically an easy task, we further restrict x∗ (t) to be in some set M. The main problem of Calculus
of Variations can then be stated as the following.

Given some set M of functions, find the curve(s) for which the integral F[x(t)] has a least value.19

In Calculus of Variations, this set is usually called the functions admissible for comparison. The set M for which we will
focus our attention are those functions x∗ (t) which are:
1) Differentiable over (ti , tf )
2) Pass through the points (xi , ti ), (xf , tf )
Otherwise x∗ (t) is completely arbitrary. Admissible comparison functions of this type can be represented as,
x∗ (t) = x(t) + 𝜀z(t),
where 𝜀 is a numerical parameter and z(t) is any differentiable function that is zero at ti and tf
z(ti ) = z(tf ) = 0.
Placing the comparison function into the functional  yields
tf
 [x∗ (t)] = (x + 𝜀, ẋ + 𝜀z)dt.
̇
∫tI
For a particular arbitrary function z(t), then F[x∗ (t)] becomes just a function of 𝜀, that is, F[x∗ ] = F(𝜀). From ordinary
calculus, we know that a necessary condition for a differentiable function to have a minimum at point is for its derivative
to vanish at that point. Also, we know from the construction of x∗ (t) that a minimum of F(𝜀) occurs at 𝜀 = 0. This implies
that
tf ( )
d 𝜕 𝜕
(0) = z+ ż dt = 0,
d𝜀 ∫ti 𝜕x 𝜕 ẋ
is a necessary condition for a minimum. This condition must be true for any admissible z(t).
In order to obtain a necessary condition in a more useful form, it is convenient to integrate the second term in this
equation, by parts, which yields,
tf [ ( )] ]tf
𝜕 d 𝜕 𝜕
− z(t)dt + z(t) = 0.
∫ti 𝜕x dt 𝜕 ẋ 𝜕 ẋ ti

18 The reason for the symbol  will be presently explained. Also, restricting ourselves to scalar arguments is strictly for ease of discussion. This
restriction will be lifted in the following.
19 Aleksandrov et al. [188].
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
8.5 Variational and Minimum Principles 535

Since z(ti ) = z(tf ) = 0, this gives,

tf [ ( )]
𝜕 d 𝜕
− z(t)dt = 0.
∫ti 𝜕x dt 𝜕 ẋ
This equation must hold for any admissible z(t). The only way for this to be true for all possible z(t) is to have,
( )
d 𝜕 𝜕
− = 0. (8.70)
dt 𝜕 ẋ 𝜕x
This is the famous Euler–Lagrange equation. It represents a differential equation which a differentiable minimizing path,
x(t), must satisfy.

Example 8.25 Shortest distance between two points

The distance, S, between two fixed points A (yA , xA ) and B (yB , xB ) in the x − y plane can be expressed as
√
S= xB 1 + y ′ 2 dx,
∫xA
where y′ = dy∕dx. Using the Euler–Lagrange equation to minimize the functional S yields the following differential
equation:
[ √ ]
d 𝜕 ′ 2
( 1 + y ) = 0,
dx 𝜕y′
which can be solved in the closed form as
[ ]
yB − yA y x − yB xA
y= x+ A B .
xB − xA xB − xA
As expected, the solution is a straight line between A and B.

8.5.2 Hamilton’s Principle

The similarity between the resulting Euler–Lagrange equation in equation 8.70 and the form of equation 8.47 will be readily
apparent. In fact for systems without nonconservative efforts, that is, those arising from external inputs or from dissipative
effects, they are identical. This implies that the dynamics of physical systems can be viewed in a radically different way.20
Rather than the dynamics of physical systems being determined by a set of differential equations, the dynamics may be
determined by minimization of a functional form.
This method of viewing system dynamics is termed Hamilton’s principle,21 which we state in two dual forms. But before
we state these principles, we need to define more precisely what we mean by a variation. Given a comparison function
x∗ (t, 𝜀) = x(t) + 𝜀z(t), the variation 𝛿x of a function x(t) is defined to be
𝜕x∗
𝛿x = 𝜀 = 𝜀z(t) = x∗ (t, 𝜀) − x(t),
𝜕𝜀
or simply the difference between the comparison function and x(t). Note that with this definition, the variation of a deriva-
tive is
𝜕 ẋ ∗ d
𝛿 ẋ = 𝜀 = 𝜀ż = (𝛿x),
𝜕𝜀 dt
the variation of a function f (x) is
𝜕f 𝜕x∗ 𝜕f
𝛿f (x) = 𝜀= 𝛿x,
𝜕x 𝜕𝜀 𝜕x
and the variation of a functional form is
𝜕
tf
𝜕(x + 𝜀z, ẋ + 𝜀z,
̇ t) tf
𝛿 = 𝜀= dt = 𝛿dt.
𝜕𝜀 ∫ti 𝜕𝜀 ∫ti

20 Feynmann et al., [189].

21 Lanczos [8]. Also see Crandall et al. [16].
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
536 8 Multiport Modeling and Energy Methods

Hamilton’s principle can then be stated as:

For a system in which the potential energy, Uq , is expressible in terms of the independent integrated flows q = ∫ fdt which define the
kinetic coenergy, Tf , a comparison path q∗ (t, 𝜀) of a dynamic system is the true path if, and only if, the variation
t [ ]
̇ dt = 0.
𝛿 = ∫ti f 𝛿 + enc 𝛿q

In this expression,  = Tf − Uq , ti and tf are initial and final times for the path, and enc are the nonconservative (nc)
efforts.
The dual form for Hamilton’s principle can be stated as:

For a system in which the kinetic energy, Tp , is expressible in terms of the independent integrated efforts p = ∫ edt which define the
potential coenergy, Ue , a comparison path p∗ (t, 𝜀) of a dynamic system is the true path if, and only if, the variation
t [ ]
̇ dt = 0..
𝛿 = ∫ti f 𝛿∗ + fnc 𝛿p

In this expression, ∗ = Ue − Tp , and fnc are the nonconservative (nc) flows.

Example 8.26 Mass–spring system model using Hamilton’s principle

For the mechanical system of Figure 8.65, use Hamilton’s principle to obtain the state equations.
Solution
The kinetic coenergy for this system can be expressed Tẋ m = mẋ 2m ∕2 and potential energy as Uxm = K(xm − xo )2 ∕2. Hamil-
ton’s principle then states that,
tf [ ]
𝛿 = ̇ m dt = 0.
𝛿§̇ − § + Fo 𝛿x
∫ti
The variation of kinetic coenergy is,
𝛿Tẋ m = mẋ m 𝛿xm ,
and the variation of the potential energy is,
𝛿Uxm = K(xm − xo )𝛿xm ,
which results in,
tf [ ]
𝛿 = mẋ m 𝛿 ẋ m + (Fo − K(xm − xo ))𝛿xm dt = 0.
∫ti
Integrating the first term in this integral by parts gives,
tf [ ]
𝛿 = mẋ m 𝛿 ẋ m + (Fo − K(xm − xo )) 𝛿xm dt = 0.
∫ti
For arbitrary variations 𝛿xm , this implies,
d
Fo − K(xm − xo ) − (mẋ m ) = 0,
dt
which is a second-order state equation for the system.
We could obtain an equivalent state equation by expressing the potential coenergy as,
Uṗ s = ṗ 2s ∕2K,
and the kinetic energy as,
Tps = (ps − po )2 ∕2m,
and then using the dual form for Hamilton’s principle as,
tf [ ]
𝛿 = 𝛿(Uṗ s − Tps ) + Vo 𝛿ps
∫ti
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
8.7 Problems 537

[ ]
tf
ṗ s ps − p o
= 𝛿 ṗ + (Vo − )𝛿ps dt
∫ti K s m
[ ]
d ṗ s
tf
p − po
= Vo − s − 𝛿ps dt = 0,
∫ti m dt K
where we have integrated the potential coenergy term by parts. For arbitrary variation, 𝛿ps , we then have
p − po d ṗ s
Vo − s − = 0,
m dt K
which represents another form for the state equation.

The use of Hamilton’s principle, at least for lumped systems, is a somewhat tedious method to obtain state equations.
The real value in Hamilton’s principle is the different view it presents of dynamical systems. This view has led to advances
in finite-element techniques [56, 190], and many other branches of engineering and physics.

8.6 Chapter Summary

This chapter has reviewed how derivatives of energy and coenergy are remarkably helpful for obtaining constitutive rela-
tions of energy storing elements. Energy methods are particularly well suited to key engineering applications, such as EM
system modeling. The use of Lagrange equations arises from a straightforward application of derivatives of energy and coen-
ergy, and a formulation compatible with a bond graph approach was described. Developing Lagrange subsystem models
that can be coupled with bond graphs of other subsystems provides a very powerful modeling platform. A Lagrange subsys-
tem is particularly helpful when dealing with displacement-dependent effects that arise in complex mechanical systems,
as well as for EM devices.
As shown in this chapter, energy methods find many practical uses in modeling and analysis of physical systems.
Although these techniques could be viewed as distinct from methods Chapters 3 and 4, energy methods are a natural
progression in showing how a bond graph basis can be used to understand a wide range of system types. This philosophy
continues in the Chapter 9 as we revisit thermodynamic system modeling.

8.7 Problems

Modeling systems with multiport elements

Problem B-8.1 Barbell-loaded cantilever with twist The cantilever beam in Example 8.2 was modeled by a two-port
C-element which coupled the power flow into the translational and rotational motion at the beam’s tip. The example
showed how to model the case where a barbell was attached at the beam’s end and could move with (small) planar
translational and rotational motion. Propose a model for the case where now the barbell can also rotate “out of the
plane” as shown in Figure B-8.1. Propose constitutive relations for the beam in this case (not necessary to derive in
exact form) and integrate with the barbell inertial motions. Apply causality, identify states, and derive system state
equations.

Figure B-8.1 Instrument

support structure.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
538 8 Multiport Modeling and Energy Methods

Problem B-8.2 Simulation of two-dimensional spring–mass system Revisit the two-dimensional mass–spring system
in Example 8.3, shown in Figure 8.10(a). Consider the main spring has stiffness k1 = 10 N/m with an initial (unstressed)
length, lo = 1 m, and let mass m = 1 kg. Introduce a rotational spring at the support pivot, k2 , along with rotational
pivot damping, b2 . Consider also that there are some linear damping forces on the mass decomposed into the x and y
directions with damping coefficients bx and by , respectively.
a) Build the bond graph for the upgraded system, apply causality, and derive the state equations
b) For the parameters given below, assume spring k1 is initially stretched out so that ro = lo + 0.5 m and the angle is
set to 45∘ . With the mass then released at rest from this point, simulate until the mass comes to rest.
System parameters: m1 = 1 kg, k1 = 10 N/m, k2 = 5 N⋅m/rad, b2 = 1 N⋅m⋅sec/rad, bx = by = 1 N⋅s/m

Problem B-8.3 Tank-float system with feedback to valve The tank-float system in Example 8.4 is reconfigured with a
feedback linkage as shown in Figure B-8.3 meant to resupply the tank as the pump operates to deliver flow Qp (t). The
valve is controlled so that flow is directed either into the tank or back to the reservoir (via Q𝑣 ). Develop a model of this
system, assign causality, and derive state equations. Assume that the linkage has negligible inertia and pivot friction
and that the force to change the valve is also very small. Propose a proper model for the valve component.

Low-friction
guides
Long pipe

Valve Pump

Figure B-8.3 Tank-float system with feedback and valve.

Problem B-8.4 Ktesibios In his book on the origins of feedback control, Mayr refers to the water clock of Ktesibios,
which is considered the first recorded system with true feedback control [134]. This water clock relies on a float valve/-
supply tank that provides a controlled flow of fluid, illustrated schematically in Figure B-8.4 (based on description
from Diels [191] per Mayr [134]). This key element features a float valve meant to maintain the tank fluid height, and
thus the controlled output flow Q, which is essential for the waterclock timekeeping. Develop a bond graph that could
help describe the operation of this float valve/tank system in sufficient detail to enable causality assignment and state
variable determination. Similarity to the tank-float system in Example 8.4 may be useful in this model study.

Supply of
fluid

Float
valve Controlled
flow

Figure B-8.4 Float valve as used by

Ktesibios [134, 191].

Problem B-8.5 Triton’s Wall A concept is proposed for an energy-generating device that uses a pivoted sea-wall as
shown in Figure B-8.5. The idea is to use the natural wave forcing on the wall to pump fluid via a fluid line into a
turbine generator. You have been asked to consult on this proposed project and to determine whether your company
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
8.7 Problems 539

(or government) should fund whether significant power can be generated in this manner. You are asked to develop a
preliminary model for Triton’s Wall.

Parallel spring-damper
combination
Waves
Assume channel has a known depth
Pivoted
sea-wall
Water
Fluid line
Water turbine
Ocean

Generator
Pivot
To load

Figure B-8.5 Energy generation from waves.

Present your analysis, including a description of all elements of the model. Derive or present all required constitutive
equations for key elements that would be sufficient to make preliminary estimates of performance. Present a bond
graph of the entire (interconnected) system, including a resistive load, and derive if possible a mathematical model in
any form (system of ODEs, frequency response) suitable for answering questions about the design. Assume you would
be provided sufficient information about the waves in a particular installation site.

Electromechanical Multiport Systems

Problem B-8.6 Capacitive position detection Develop a bond graph for the system shown in Figure B-8.6. Assume
that C(x) = 𝜖A∕(d + xc ), where d is the distance between the plate attached to mass, m, and the fixed ground electrode.

Figure B-8.6 Capacitive position

detection.

a) Build a bond graph, assign causality, and identify state variables

b) Derive the complete state equations
c) Explain whether and/or how position, xc , the motion relative to equilibrium position d, can be inferred from a
measurement of Vout .

Problem B-8.7 Butterfly capacitor The capacitance of the “butterfly capacitor” in Figure B-8.7 varies with rotor posi-
tion according to the relation C(𝜃) = C0 + C1 cos(2𝜃), where C0 and C1 are constants. The rotor has axial moment of
inertia J and turns on low-friction fixed bearings.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
540 8 Multiport Modeling and Energy Methods

Figure B-8.7 Butterfly condenser.

a) Develop a bond graph model of this system and derive the state equations. The need is to model the rotor dynamics,
especially when an ideal constant-voltage source Vo is connected across the electrical terminals of the capacitor.
Assume there is negligible electrical resistance.
b) Use your equations to determine the equilibrium positions of the rotor and indicate the stability.
c) What is the natural frequency of small oscillations in the neighborhood of a given stable equilibrium position?
d) What is the maximum electrical torque available if: C0 = 15 × 10−12 F, C1 = 10 × 10−12 F, and Vo = 1000 V?
e) If the rotor shaft is 1 millimeter in diameter and the rotor mass is 100 grams, estimate the minimum coefficient of
friction required to prevent motion. Assume gravity acts in the direction indicated.

Problem B-8.8 Capacitive weighing system Develop a complete mathematical model of the system shown in
Figure B-8.8 using a bond graph approach. Express the model in terms of first order differential equations. Use or
derive constitutive equations implied in the diagram by the parameters and/or functions expressed. For example, the
electrical capacitive element shown has electrically linear behavior described by the capacitance,
Co
C(x) = ,
1 + (x∕do )2
where x is the displacement of the movable (top) plate about a “zero” position, defined by when the distance between
the two plates is do . The movable plate is mounted on a flexible guide having stiffness, kf . Assume the mass of the
movable plate is lumped into the rotational inertia of the rod, J, and that the rod rotational displacement about the
pivot is small. Friction about the pivot is negligible.

Rigid rod

Figure B-8.8 Capacitive weighing system.

Problem B-8.9 EM balance In the EM balance shown in Figure B-8.9, a rigid beam pivots about a fulcrum. The beam
is balanced by two passive springs and by forces induced by the movable plate capacitor and solenoid devices shown.
Assume small motions xc and xs about a balanced position and that the mass of the movable plate and solenoid plunger
are relatively small compared to the beam. Let
𝜖A
C(xc ) = and, L(xs ) = Lo − axs ,
d + xc
where d is the distance between the plates when xc = 0, A is the plate area, and a is a known constant (and xs <
(Lo − Lmin )∕a, Lmin ≥ 0).
a) Model the system and develop a bond graph. Assume the input to the circuit is voltage Vs (t)
b) Find the required constitutive relations for any multiport elements
c) Derive the complete state equations for the case where L1 = L2 = L∕2, and L is the length of the entire beam.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
8.7 Problems 541

Rigid beam

Figure B-8.9 Electromechanical balance.

Problem B-8.10 Capacitive pendulum The pendulum in Figure B-8.10 is formed by a particle of mass m having electric
charge +q. It swings in a plane on fixed-length (massless) support l coming close to a flat plate of mass M having charge
−q. It is known that the electrical (potential) energy stored by this particle-plate configuration is,
q2
U(q, d) = ,
16𝜋𝜖d
when the point charge is a distance d from the plate, and 𝜖 is the dielectric constant of the intervening medium. With
the pendulum resting vertically straight down and the spring (stiffness k) unstretched, the spacing between the particle
and the plate is a.

Figure B-8.10 Capacitive

pendulum.

a) Begin by developing a bond graph model of the pendulum system. Explain two different models: (i) modeling the
restoring torque due to gravity by using a modulated transformer, and (ii) modeling the restoring torque due to
gravity by using a capacitive element. Show how the restoring torque is derived in each case.
b) Develop a model to account for the torque induced by the presence of the charged plate. You will need to account for
the relative motion of the particle and plate. Derive the force generated by the particle-plate interaction and show
how this gives you the torque on the pendulum.
c) Complete an integrated model of the pendulum-plate system in bond graph form, apply all implied inputs, apply
causality, and identify the state variables.
d) Derive a complete set of state equations.

Problem B-8.11 Three-plate capacitor Consider the three-plate capacitor which has a central plate that can move only
in the x-direction as shown in Figure B-8.11. When the central plate is in the equilibrium position (x = 0), the springs
are unstressed and the distance between the three plates are equal to a. The plates have area A and the permittivity of
the capacitor material is 𝜖o .
a) Assume a is very small compared to the other plate dimensions so there is negligible field fringing. Propose expres-
sions for each C(x) of the two capacitors formed by the three plates. Sketch a schematic of the electrical circuit
formed.
b) Find an energy function for the energy stored in each capacitor, Uc (qi , x), where i = 1, 2 (1 represents the left capac-
itor, 2 the right).
c) Determine expressions for the EM force induced by each capacitor.
d) Develop a bond graph that integrates the three-plate capacitors with the mass and spring elements. Assume Vo is
an ideal voltage source and consider the switch to be closed. Assume each spring has stiffness k∕4.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
542 8 Multiport Modeling and Energy Methods

e) Assign causality, identify state variables, and derive state equations for this system.
f) With the plate x = 0 and voltage Vo applied, show that the system achieves a stable equilibrium.
g) Assume that the system has been at equilibrium for some time and then at t = 0 the switch is opened. Explain what
will happen. It can be helpful to assume an ideal switch to the new configuration and to draw a new system bond
graph, taking the initial conditions from part (f).

Figure B-8.11 Three-plate

capacitor system.

Problem B-8.12 Differential capacitive accelerometer One type of differential capacitive sensor is formed by two
capacitive plate elements that have changing gaps and controlled voltages, Vs , formed into an electrical circuit as shown
in Figure B-8.12(a). This schematic shows an accelerometer configuration meant to detect the motion of the base struc-
ture, 𝑣g (t). The seismic mass, m, is supported by the spring–damper combinations (kb , bb ), with each dynamic gap gi
(i = 1, 2) defining a capacitance Ci (x). The output voltage is used to infer 𝑣g (t).

– + – +

– + – +

(a) (b)

Figure B-8.12 (a) Simplified schematic of a differential capacitive

accelerometer and (b) circuit schematic.

a) Model the two capacitors as fixed-gap parallel capacitor plates as in the electrical circuit model of Figure B-8.12(b).
In this case, capacitance, Ci = 𝜖A∕gi , where 𝜖 is the permittivity, A is the plate area, and gi is the respective gap.
Develop a bond graph and derive system equations. Show that an output voltage expression can be given by [192],
C1 − C2 g − g1
Vo = ⋅ 𝑣s = 2 ⋅ Vs ,
C1 + C2 g1 + g2
where g1 and g2 are the gap distances for each capacitor.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
8.7 Problems 543

b) For an accelerometer that uses the differential-capacitance concept, the gaps are dynamic: g1 = g1o − x and
g2 = g2o + x, with g1o and g2o the gaps when the system is at rest and x is the center plate displacement. Build a
complete bond graph model for the accelerometer.
c) Derive state equations and an output voltage equation to explain how or whether this sensor can provide a useful
measure of the motion, 𝑣g (t).

Problem B-8.13 Multiply-excited electric-field system The system with two moving plates shown in Figure B-8.13
has two electrical terminals and two mechanical “ports.” Plates 1 and 2 only move along the x1 and x2 directions,
respectively, and are restrained by spring–damper elements. Each electrical terminal has a current source and shunt
resistor as shown.

Fixed base
Moving plate
mass Moving plate
mass
+
–
Fixed ground

Plates have width

into page
– +
Fixed base

Figure B-8.13 A system with two electrical and two

mechanical ports coupled by an electric field.

a) Assume the space between the plates is filled with air so permittivity is 𝜖o . Show that the electrical constitutive
relations for a four-port C can be expressed by [31],

q1 = C1 (x1 , x2 )V1 − Cm (x1 , x2 )V2 ,

q2 = −Cm (x1 , x2 )V1 + C2 (x1 , x2 )V2 ,

where
[ ]
𝜖o 𝑤 l1 + (lm − x2 )
C1 (x1 , x2 ) =
x1
[ ]
l2 lm − x2
C2 (x1 , x2 ) = 𝜖o 𝑤 + x1
x2 x1
𝜖 𝑤(l − x2 )
Cm (x1 , x2 ) = o m
x1
b) Develop a bond graph model that shows the interconnection between the electrical and mechanical systems using
a four-port C element.
c) Derive an expression for the stored potential energy, Uq1 q2 x1 x2 (q1 , q2 , x1 , x2 ).
d) Use the energy function from (c) to derive the EM forces induced on each plate, F1 and F2 .
e) Apply causality on the bond graph and derive system state equations,

Problem B-8.14 Electromagnetic relay The EM relay shown in Figure B-8.14 is to be modeled using lumped-parameter
elements. Consider that as the mass m slides, there is some linear friction b and the effective inductance changes
according to L(x) = muo AN 2 ∕(l1 − x). There is resistance R in the coil which has N turns. Assume the spring has a
linear stiffness k.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
544 8 Multiport Modeling and Energy Methods

= N
–
Time, s

(a) (b)

Figure B-8.14 (a) Electromechanical relay with dc voltage input.

(b) Transient current due to step in voltage.

Assume that m = 0.01 kg, b = 0.1 N s/m, k = 100 N/m, lo = 20 mm, l1 = 30 mm, A = 100 mm2 , and R = 1 Ω.
a) Develop a bond graph model of this system and derive state equations. Write an output equation for the current, i.
b) Assume the initial states implied in Figure B-8.14(a), with V = 0 at t = 0. Assume that V = 6 volts at t = 10 ms, hold
until a steady-state current is reached, then step V back to zero. The current transient response (over the relatively
small transient periods) will take the characteristics shape shown in Figure B-8.14(b). Make sure to consider special
conditions and considerations for changes in the model, especially when x → l1 and when contact is made. Make
plots of the following variables from time t = 0 until a steady state is reached: input voltage, V, mass velocity, 𝑣m ,
mass position, x, current, i.

Problem B-8.15 Angular EM actuator An angular motion EM actuator is illustrated Figure B-8.15(a). The “driving
blade” is made of iron and moves into the air gap when the electromagnet is powered by voltage source V(t). The
blade is shaped so that the inductance of the driving coil varies linearly with the angle relative to the core. Specifically,
L(𝜃b ) = a + b|𝜃|, where a and b are constants and 𝜃b is the angle relative to the core as shown in Figure B-8.15(b). Other
system parameters are: R = coil resistance, J = moment of inertia of the flywheel/blade, B = coefficient of friction
between torsion rod and bearing, and K = torsional stiffness of the torsion rod. Assume that voltage input Vs (t) is an
ideal voltage source.

Bearing Flywheel Side view Rod unstressed

Torsion rod,
Flywheel at

Iron blade Iron blade

Air gap
Core

Driving coil Cross-section

of core

Iron blade
angle relative
to core

(a) (b)

Figure B-8.15 (a) Angular motion electromechanical actuator and (b) side view illustrating rest (or
unstressed) position of rod relative to core.

a) Develop a bond graph model of this system and derive an explicit expression for the EM torque on the blade.
Show that the nonlinear state equations are:
⎡ V(t) − R 𝜆 ⎤
⎡ 𝜆̇ ⎤ ⎢ L(𝜃) ⎥
⎢𝜔̇ ⎥ = ⎢ b𝜆 2
− B 𝜔 − K 𝜃⎥ ,
⎢ ̇ ⎥ ⎢ 2JL(𝜃)2 J J ⎥
⎣𝜃 ⎦ ⎢ ⎥
⎣ 𝜔 ⎦
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
8.7 Problems 545

but clearly define the definition of 𝜃 as used in these equations with reference to the unique definitions of 𝜃k and
𝜃b in Figure B-8.15(b)
b) Develop a simulation model to study the response behavior of the system when a “square wave” is applied to the
drive coil input to move the blade in and out of the air gap. Assign proper initial conditions for the elements in the
system. Specifically, assume that the iron blade initially rests outside of the air gap (choose an angle) when there is
no EM torque applied. The parameter values listed below should be used.
J = flywheel inertia = 0.001 kg m2 , K = 1.0 N m/rad = torsion rod stiffness, B = 0.01 N sec/rad = bearing resistance,
R = 100 Ω, L(𝜃) = 1 − 2 |𝜃| ∕𝜋 H (a = 1, b = −2∕𝜋).
NOTE: The value of L cannot be negative so it needs to be assumed that L takes on a constant value, L when the
blade moves beyond the core.
c) Investigate the frequency response of this system using the simulation model. Develop both magnitude and phase
relationships over an appropriate frequency range.

Problem B-8.16 Electromechanically suspended mass A mass M is to be suspended against the force of gravity by an
EM actuator as shown in Figure B-8.16. With z indicating the distance of the mass from the ground, it is determined
that L(z) = Lo ∕(1 − z∕a)4 , where Lo and a are known constants for the EM actuator being used. The system has two
inputs, a constant dc voltage Vo and a varying voltage Vs (t).

+
–
+
–
–

Figure B-8.16 An
electromechanically
suspended mass.

a) Develop a bond graph model and show that the system states can be taken as 𝜆 (or i), p (or 𝑣), and z.
b) For Vs (t) = 0, Vo is set to induce a current that will hold the mass at equilibrium. Determine the equilibrium
conditions.
c) Linearize the state equations for the equilibrium in part (b).

Problem B-8.17 Electromechanically balanced inverted pendulum An inverted simple pendulum is to be stabilized by
an electromagnet as shown in Figure B-8.17. The mass is made of a ferromagnetic material so as the pendulum angle
𝜃 varies the effective inductance of the circuit changes in a known way, L(𝜃) = Lo (1 + a sin 𝜃 + b cos 2𝜃), where Lo , a,
and b are positive constants. The current source provides a constant current, i(t) = io , such that i2o Lo = 6Mgl with no
other forces exerted except those due to gravity.
a) Develop a bond graph of this system, apply causality and identify the state variables.
b) Derive the state equations.
c) Solve for all the static equilibria.
d) Linearize the state equations.
e) Express the model as a linear second-order equation in standard form and find expressions for the system damping
ratio, 𝜁, and the undamped natural frequency, 𝜔n .
f) Describe the stability criteria for this system.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
546 8 Multiport Modeling and Energy Methods

Figure B-8.17
Electromechanically balanced
inverted pendulum.

Problem B-8.18 EM coupling network – 1 A system shown in Figure B-8.18 has known relations,

𝜆1 = ax1 i31 + bx1 x2 i2

𝜆2 = bx1 x2 i1 + cx2 i32 ,

where a, b, and c are known constants. Write an expression for the total coenergy stored in the coupling network?

Generic magnetic
field coupling

Figure B-8.18 Word bond graph for a generic

electromechanical coupling.

Problem B-8.19 Rotating machine with superconducting coils A rotating machine concept that uses superconducting
coils is illustrated schematically in Figure B-8.19. A stator (fixed) coil is driven by a current source, is (t), while the
rotor is to be fashioned with a shorted superconducting coil. The prototype design has been tested revealing electrical
relations,

𝜆1 = L1 i1 + Lm cos 𝜃 ⋅ i2
𝜆2 = Lm cos 𝜃 ⋅ i1 + L2 i2 ,

where 𝜃 is the angular displacement of the superconducting rotor.

The machine designer proposes that to take advantage of the superconducting properties of the rotor, the machine
must be put into operation in the following steps:
i) With the rotor terminals in an open circuit, the shaft is set at 𝜃 = 0, and then the current source is (t) is set to io .
ii) The rotor terminals are then shorted, so the initial flux imposed on the superconducting coils, 𝜆2 , will be conserved.
iii) The machine is then operated by driving the current, is (t), in a desired way.
Develop a model that can explain this system as follows:
a) Develop an EM system model using the given relations and integrate using a bond graph.
b) On a complete model, assign causality for the operational state of the machine, and identify any state variables of
the system.
c) Express the system as one equation in the variable 𝜃, assuming no damping.
d) Optional: Is it relevant to include mechanical inertia? How do things change if you drive with voltage instead of
current?
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
8.7 Problems 547

Normal conducting
rotor

Superconducting
rotor

Shorted

Figure B-8.19 A rotating

superconducting rotating machine
concept.

Magnetic Circuits and Devices

Problem B-8.20 Magnetic circuit modeling – 1 For the magnetic device shown in Figure B-8.20:
a) Draw an magnetic circuit model, accounting for reluctance effects from all the core material and the air gap.
b) Use the results from (a) to formulate a bond graph model.
c) Simplify the bond graph and find the equivalent total reluctance; sketch a simplified bond graph.
d) Write an expression for the effective inductance of the electromagnet and core as viewed from the electrical port.

Figure B-8.20 Electromagnet-powered core.

Problem B-8.21 Magnetic circuit modeling – 2 The magnetic device shown in Figure B-8.21 includes a permanent
magnet. Do the following:
a) Draw an magnetic circuit model, accounting for reluctance effects from all the core material and the air gap, and
include a model for the permanent magnet.
b) Use the results from (a) to formulate a bond graph model.
c) Simplify the bond graph and find the equivalent total reluctance; sketch a simplified bond graph.

+ +

– –

Figure B-8.21 Electromagnet-powered core with permanent magnet.

Problem B-8.22 Magnetic device with movable element and two electrical ports A magnetic device is used to mag-
netically constrain a movable element of mass m between two pole faces of area A1 and A2 , as shown in Figure B-8.22.
Assume the device has a rectangular overall structure. When the movable element is centered, the air gap lengths are
equal to a, which is the desired equilibrium gap distance. As the mass moves from the centered position by x, the air
gaps change as indicated. Assume the movable element has centerline length b and that the poles have equal length,
c. The poles have width 𝑤i = Ai ∕𝑤 (i = 1, 2), where 𝑤 is the thickness into the page. Assume the core paths have
centerline lengths shown (d and e).
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
548 8 Multiport Modeling and Energy Methods

Thickness into page,

(core)

Area

Area

+ – + –

Figure B-8.22 A magnetic device with two

electrical ports and a single movable element
with variable cross-sectional area.

a) Sketch an equivalent magnetic circuit for this device, and note that the core material has permeability, 𝜇. Include
all magnetic and electrical elements deemed necessary in the magnetic circuit (the mechanical elements can be left
off in this step). Assume that V1 and V2 are input voltage sources.
b) Develop a bond graph model based on the circuit in part (a), incorporating the moving mass element, including an
force, F(t), to represent next forces imposed on the mass along the x-axis. Use a two-port magnetomechanical C to
represent energy stored in each gap.
c) If 𝜇 → ∞, explain how the model can be simplified and why; redraw the bond graph of the simplified system.
Explain how the two mechanical ports can be combined into a single port, with velocity 𝑣 = ẋ and a net EM force.
d) Further simplify the system of part (c), assuming currents are specified as inputs, and then let the electrical terminal
relations take the form, 𝜆1 (i1 , i2 , x) and 𝜆2 (i1 , i2 , x). Specify the form of the implied inductance, L(x), in terms of any
parameters indicated in the diagram.
e) Determine the coenergy (i1 , i2 , x) stored in the EM coupling for the model of part (d).
f) Derive the EM force using the coenergy in part (c).
g) For the bond graph in part (d), derive the system state equations, taking i1 , i2 , and the unknown forces F(t), as
inputs.

Problem B-8.23 Electromagnetically suspended plate In Figure B-8.23, a rigid plate is suspended by an electromagnet.
The plate is free to move in the x and y directions as shown.
a) Assume that a current source is connected to the coil terminals, i = i(t). Develop a bond graph model and identify
state variables using causality. Assume that the core and plate materials have 𝜇 → ∞. Make use of the geometric
information, notably that the plate has a depth (into page) of d.
b) Show that the force equation (ignoring fringing fields) is,
[ ]
𝜇o N 2 di(t)2 a − 2x x(a − x) ̂
F= î− j ,
2a y y2
and then write in terms of states of the system model.
c) Develop state equations for the system.
d) Consider the case where a voltage source V(t) is connected to the coil terminals and that the coil wire has resistance,
Rc . Repeat steps (a)–(c) for this case.
e) Repeat part (d) when core and plate permeability must be taken into account and a complete magnetic circuit model
must be developed.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
8.7 Problems 549

+ –

Depth into,
page,

Mass,

Plate length, , Thickness,

Figure B-8.23 Electromagnetically suspended plate in x − y.

Problem B-8.24 Diaphragm pressure sensor Figure B-8.24 shows a schematic of a diaphragm-based pressure sensor
that uses permanent magnets and coils to induce a measurable output at an electrical port.

Fluid input pressure, Circular diaphragm

Sensor volume,
Air gap
Permanent magnets
and coils
Core structure
Electrical output terminals

Figure B-8.24 Permanent-magnet diaphragm pressure sensor.

a) Draw a magnetic circuit that includes the permanent magnets and coils and reluctance effects from the permeable
diaphragm, the core structure, and the air gap.
b) Develop a bond graph that integrates a pressure source with the sensor fluid volume, interaction with the flexible
diaphragm, and a model of the magnetic system elements, including the magnetomechanical multiport coupling
the diaphragm motion and air gap to the total effective air gap magnetic flux; include a sensing resistor, Rs , across
the electrical output terminal.
c) Derive the constitutive relation for the magnetomechanical multiport.
d) Apply causality, identify system variables, and derive system state equations.
e) Write an output equation for the voltage across a resistor, Rs .

Lagrange Equations

Problem B-8.25 Simulation of the block-strut-block system Set up a simulation of the block-strut-block system studied
in Solved Problem A-8-12 which includes an input force on the mass along the x direction, F(t).
a) Begin by solving for the equilibrium configuration; i.e., what are x, y, and 𝜃 before turning on force?
b) Then simulate by turning on the force F(t) at t = 1 second, letting F(t) = Fax sin(2𝜋fa t), where fa is the frequency
in Hz. Simulate for 10 seconds. From the simulation, plot the x, y, and 𝜃, as well as the spring force. Use the same
parameter values from Problem B-4-37, experimenting with Fax as necessary.
c) Compare with the baseline results in Problem B-4-37 and comment on the results.

Problem B-8.26 Solenoid-driven cart pendulum The schematic shown in Figure B-8.26 models a system with a
cart-pendulum device driven by a solenoid actuator.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
550 8 Multiport Modeling and Energy Methods

Figure B-8.26 Solenoid-driven cart pendulum.

a) Develop a Lagrange subsystem model for the cart-pendulum system. Formulate this subsystem model so that you
can incorporate friction at the pendulum pivot, as well as external forces and friction on the cart. Derive the state
equations for this subsystem.
b) Assume the solenoid is electrically linear with L(x1 ) = Lo + 𝛼x1 , where x1 is the position of m1 relative to the fixed
coil and 𝛼 is a constant. Derive the constitutive equations for the solenoid actuator, and indicate how you would
model this device in bond graph form.
c) Develop a complete bond graph model for the solenoid actuator system coupled with the cart-pendulum subsystem
model. Assign causality and list the states for this system.
d) Derive and/or summarize the complete system state equations. If any terms/variables in these equations have been
derived in previous steps, you do not have to substitute them into the final equations, but make sure to indicate
(and box) those key results.

Problem B-8.27 Solenoid-spring piston pendulum Figure B-8.27 shows a solenoid configured to induce motion in the
piston as well as in the pendulum. Let the solenoid have an inductance, L(zs ) = A − Bzs , where A and B are constants

Seal
friction

Pivot
friction

Figure B-8.27 Solenoid-spring piston

pendulum.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
8.7 Problems 551

and zs is zero at the midpoint as shown. The return spring, ks is unstressed when the solenoid plunger of mass, ms ,
is at the edge of the solenoid (so when the solenoid is turned off, it pushes the solenoid back out. When “energized,”
the solenoid will pull in the plunger, and then the spring will force the solenoid mass back out when the solenoid is
turned off.
Develop a bond graph model of the complete system, apply causality, and derive complete state equations. Note that
the piston mass and solenoid mass in this system are rigidly attached (separated by distance d).

Problem B-8.28 Crank pendulum A simple pendulum of length l is constrained to rotate by angle 𝜃 about prismatic
̇ about the z axis, as shown
joint P which to a rotating hub with diameter R. The hub rotates with angular velocity angle
in Figure B-8.28.

(a) (b)

Figure B-8.28 (a) Crank-driven pendulum in side view. (b) Isometric

schematic.

a) Write an expression for the total kinetic energy of the pendulum mass, m, and the hub moment of inertia, Jz , in
terms of dependent coordinates.
b) Express the kinetic coenergy in terms of independent coordinates (𝛼, 𝜃), and write a transformation matrix relation-
ship between all five variables (x, y, z, 𝛼, 𝜃) and the independent variables.
c) Assume the crank-pendulum system operates in gravity and develop a complete system model accounting for an
input crank torque about the z axis, 𝜏(t). Assume the only significant losses are due to drag forces on the pendulum
mass. Assume the drag forces can be modeled as Fi = Φi (𝑣i ), where i = x, y, z and each 𝑣i can be related to state
variables through the results of part (a).
d) Apply causality to the complete system bond graph, identify states, and derive the state equations.

Problem B-8.29 Crane-belt system The system in Figure B-8.29 is a type used to move heavy loads in a mechanical
shop or warehouse. Assume the cart is driven using a stiff drive belt using motor M1 . The load with mass m hangs by

Idler pulley Drive belt Drive pulley

Crane pulley
Motor,
Crane cart

Crane support Note: the crane

Crane wheels
pulley does not
Cable touch crane
support
Cargo

Figure B-8.29 A crane-belt system for transporting loads.

 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
552 8 Multiport Modeling and Energy Methods

inelastic cable. Assume the bridge or crane support is stiff. Develop a model including elements you deem necessary
to represent this system for predicting necessary drive requirements, predicting response, and implementing and/or
testing controlled behavior. Develop a bond graph to guide state equation derivation and identify parameters and con-
stitutive relations that would be needed to begin formulating a complete model. Assign all parameter values needed
for your model formulation.

Problem B-8.30 Overhead crane or gantry Consider the overhead crane (or gantry) system in Solved Problem A-8-13.
Let mb = 10 000 kg and L = 50 m, and develop a simulation of the nonlinear equations. Use a velocity input 𝑣c (t) with
̇ as well as the position xm and xc . Find
a smooth transition to a final velocity of about 2 m/s. Plot transient values of 𝜃, 𝜃,
a value of B that will cause the mass to come to a stop after about 6 cycles of oscillation.

Problem B-8.31 Quartz yaw rate sensor modeling You’ve secured a job with a sensor company that builds quartz
rate sensors. These sensors measure angular velocities and have become very common in inertial measurement units
used on many types of vehicles. Your company makes a sensor that relies on a tuning fork design to detect angular
rotation rate. In this device, the tuning fork “tines” are actively driven to move in the plane of the fork as shown in
Figure B-8.31(a) (see, e.g., Madni et al. [193, 194]). Motion of the tines out of the plane is induced by rotation and can be
sensed, and thus related to the angular rotation rate, 𝜔i (t) (an input). You’re asked to help scale the device for different
applications, so you want to build a model to study how the device works.

Sensed Lumped
motion approximation
Driven
Driven motion
motion Turntable

Rotation Input
shaft

(a) (b)

Figure B-8.31 (a) Tuning fork concept for yaw rate sensor indicating
directions of driven and sensed motion; rotation is to be sensed.
(b) A lumped parameter approximation of the sensor device.

A simplified lumped parameter model shown in Figure B-8.31(b) is to be used for preliminary modeling. The effective
masses, m, slide without friction on a turntable with moment of inertia, J. Each mass is supported by spring–damper
elements that model the tuning fork tines (kr , br ). During operation, one set of tines is driven by a known piezoelectric
force actuation along r, Fp (t), on each mass. Build a Lagrange subsystem model to represent the turntable and mass
dynamics. You may find it better to keep the input shaft model shown (stiffness, ks , damping, bs ) outside of the Lagrange
subsystem.
a) Use r and 𝜃 as the generalized coordinates, and write the total kinetic co-energy in terms of the independent flows,
ṙ = 𝑣r and 𝜃̇ = 𝜔. Make sure to include the turntable inertia, J, the stiffness model for the tuning for tines (use kt ),
and both effective masses.
b) Complete and clearly identify all steps of a Lagrange subsystem model. Include effective damping in the tuning fork
tine material.
c) Sketch a complete system bond graph by integrating with the shaft elements and input angular velocity, 𝜔i (t). The
forcing, Fp (t), should be included, along with any other nonconservative effects. Apply causality and identify the
state variables.
d) Derive the state equations for this model.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
8.7 Problems 553

Problem B-8.32 Magic man A popular child’s toy shown in Figure B-8.32(a) is referred to as “magic man” because the
body of the clown straddles the rolling ball and is stabilized by a counter-mass, ms . The conceptual schematic provided
in Figure B-8.32(b) illustrates how this system may be realized. For the purposes of this exercise, assume all motion
is planar as in Figure B-8.32(c) (constrained to the x − z plane), even though in reality there can be significant yaw
(about z) and roll (about x), depending on the initial conditions. The angular motion of the body about the y-axis is
quantified by angle 𝜃.

Figure B-8.32 (a) Magic man. (b) Model schematic. (c) Motion of clown body relative to rolling ball as
balanced pendula.

When the ball rolls on the ground, there can be some induced rolling resistance modeled by the relation, Fr = fr W,
where fr is a dimensionless rolling resistance coefficient and W is the effective contact force. This force is applied in the
x direction and always opposes motion of the body. It is also necessary to model sliding between the ball and the ground,
since there may either be pure rolling or sliding depending on the particular conditions. In addition, the bearing that
supports the body on the ball also introduces friction between these two bodies.
a) Develop a bond graph of this system. Note that this system does not have any inputs. Motion is usually induced
by giving the whole system an initial translational velocity. For this reason, your model should not have any input
effort or flow sources.
b) Assign causality to your system, identify state variables and derive state equations for this system.
c) Simulation: Take the total mass of this system as 0.2 kg, and the ball outer radius as 90 mm. Assume the ball
mass alone is 0.1 kg. Estimate the other parameters needed to simulate the behavior of your system model. For
example, you need to estimate the value of Jb for the ball (it is a hollow sphere, thin walled). Assume that the
friction coefficient between the ball and ground is about 0.3. Consider the case were at t = 0 the system is released
moving in the x direction at a known velocity, 0.25 m/sec with the body pitch angle is zero. Assume that the ball
is not initially rotating (𝜔b (t = 0) = 0). Develop a simulation model to: (i) fine tune unknown parameters, and (ii)
solve for the response of the system from the time the ball touches the ground to the point that all motion ceases.

Problem B-8.33 Directional windmill The windmill system shown in Figure B-8.33 features an automatic turning gear
mechanism whose invention in 1750 is credited to Meikle [134]. The basic idea is to use a fantail-driven circular rack
and pinion to drive the housing so that the main sail will directly face the oncoming wind. The fan tail gear ratio is on
the order of 3000 to 1 (total of worm gear and rack and pinion units).
a) This system has coupled inertia elements if there is negligible compliance in the drive shafts. Study the system
and identify the rigid bodies needed to model the system, and develop a bond graph by inspection (using rigid
body models, transformers, etc.) Assume that all the shafts and gear matings are rigid. Include inertia of the turret,
fan tail, main sail, and all gearing. Include a simple model to represent the interaction between the wind and the
turbine blades for generating torque, including the effect of the wind direction. Also include the load as a known
torque input, 𝜏m (t) as shown. Apply causality to identify the coupling between the inertial elements, and the level
of derivative causality.
b) Rebuild the bond graph of part (a) including any compliance you deem important. Reassign causality and identify
the state variables. It is not necessary to derive state equations.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
554 8 Multiport Modeling and Energy Methods

c) Remodel the system as in part (a) using a Lagrange subsystem formulation. For this model, develop full state
equations and identify whether the feedback operation that maintains the main sail direction (axial) into the wind
could be quantified (i.e., do you think it is possible to develop a model that could be used to design and/or predict
performance, such as by simulation).

Top view
Housing Housing

Fan tail
Fan tail
Worm Drive Worm
gear pinion Support wheel
Main Main
structure sail Support
Drive rack sail
structure

(a) (b)

Figure B-8.33 Directional windmill (a) side view schematic and (b) top view.

Problem B-8.34 Mass sliding down an incline with heavy chain A mass sliding down a fixed-angle incline shown
in Figure B-8.34(a) is attached to a heavy chain that is fed without friction and from rest from point S. This system
was studied as an example in Chapter 4 with relation to Figure 4.75. Develop a model using a Lagrange subsystem and
incorporate a representation for the variable-mass effect using the convective interface, as suggested by the preliminary
bond graph in Figure B-8.34(b).

Heavy chain
(mass/length)

Friction

Convective interface

(a) (b)

Figure B-8.34 (a) Mass sliding down an incline, with heavy chain. (b)
Preliminary bond graph using Lagrange subsystem.

Problem B-8.35 Ballon with a heavy rope load A balloon rises, carrying a heavy rope that increases in length with
the balloon’s height as shown in Figure B-8.35. Develop a bond graph model that treats the heavy rope as a separate
system, clearly identifying the effective tension in the rope at the intersection of the rope and balloon. Derive the state
equations for this system.

Air friction

Heavy rope
(mass/length)

Figure B-8.35 Balloon rising

with attached heavy rope.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
8.7 Problems 555

Problem B-8.36 Moving and open tank The moving and spring-restrained tank shown in Figure B-8.36(a) was used by
Beaman and Breedveld [185] to demonstrate how moving and open system effects can be represented within a bond
graph basis. Study the referenced article and build on the methods discussed therein as well as in Section 8.4.4. Recreate
the models as formulated by Beaman and Breedveld.

Region 1
Region 2

Figure B-8.36 A moving tank that

loses fluid.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
557

Thermodynamic Systems

In this chapter, we extend ways for modeling thermodynamic systems and the use of internal energy and its relation
to thermal systems. Chapters 2 and 3 laid the groundwork for how thermodynamic effects can be accounted for when
modeling physical systems. Those discussions were limited to heat transfer processes and ways for modeling closed
thermodynamic systems; that is, those in which the amount of matter within the system does not change. Engineering
thermodynamic processes may also involve open systems subjected to significant irreversible heat and work transfers, and
these concepts are introduced in Chapters 2 and 3. Principles from classical thermodynamics will continue to be reviewed,
in a fashion compatible with our discussion in previous chapters. In addition, we will draw upon concepts from the field
of irreversible thermodynamics in order to include dynamic change. Emphasis will be placed on how system network
concepts based on bond graphs can be used effectively with both thermodynamic and open systems in order to develop
overall system models. In addition, we introduce how bond graphs can be used to understand and incorporate chemical
system effects in our models. Some introductory concepts are repeated in this chapter, but the reader may also find it
helpful to review sections in Chapters 2 and 3 that introduce thermodynamic systems modeling as well.

9.1 Thermodynamic Systems and Relations

Thermodynamics is often defined as the study of energy, its forms and its transformations, and so conforms with the theme
of this book. A reader familiar with thermodynamics should find similarities with the bond graph modeling approach taken
throughout this book. For example, our definition of system and environment in Chapter 1 is identical to that usually given
in thermodynamic texts. Thermal systems must be treated differently, however, because of consideration must be given not
only to internal energy but also often to entropy changes.
To understand the need for internal energy, first recall how we considered the energy content of a system of particles,1
which can be viewed as the molecules of a rigid body. If the rigid body constraint is removed, we can still account for energy
due to net molecular motion (in terms of total linear momentum and angular momentum). This leaves unaccounted for
the energy due to vibratory motion of individual particles and relative displacement between particles. Since the number
of molecules is usually prohibitively high (>1023 for typical systems), and since Heisenberg’s uncertainty principle forbids
complete and simultaneous knowledge of particle motion and position, we are forced to average over the extent of the system
to reduce the states to a manageable number. These states show up as system momentum and displacements in mechanical
systems, system charge and magnetic flux in electrical systems, and system volume and momentum in hydraulic systems.
In contrast, the energy of thermodynamics is formulated to account for all the “internal” states of the system and thus
ensure the constancy of energy. Callen [21] clearly states this concept:

The study of mechanics (including elasticity) is the study of one set of surviving coordinates. The subject of electricity
(including electrostatics, magnetostatics, and ferromagnetism) is the study of another set of surviving coordinates.
Thermodynamics, in contrast, is concerned with the macroscopic consequences of the myriads of atomic coordinates
that, by virtue of the coarseness of macroscopic observations, do not appear explicitly in a macroscopic description
of a system.

1 For example, see Section 2.1.2, Figure 2.7.

Modeling of Physical Systems: Simulation and Control, First Edition. Raul G. Longoria and Joseph J. Beaman.
© 2025 John Wiley & Sons Ltd. Published 2025 by John Wiley & Sons Ltd.
Companion Website: www.wiley.com/go/Modeling_of_Physical_Systems/1e
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
558 9 Thermodynamic Systems

These principles are the underlying reason why we choose a certain set of variables for thermodynamic system modeling.
Section 9.1.1 expands on why certain variables are useful and suitable for system modeling.

9.1.1 Equilibrium and States

As stated above, thermodynamic states do not uniquely determine the configuration of each internal element of the system,
but rather they are averages over all of the possible internal states of the system. For example, constant values of macroscopic
observables such as pressure and temperature do not imply a constant configuration of the thermodynamic system but
rather the average of a number of internal states which are continually changing. Now this averaging cannot be effectively
done unless we invoke another condition on the thermodynamic system – equilibrium. In classical thermodynamics, the
states are only determined if the system satisfies this condition. It is necessary to recognize that classical thermodynamics
bears a misleading name and is in point of fact concerned only with static, near-equilibrium conditions. Indeed, as we shall
see, only under further restrictive assumptions can the principles of such thermostatics be applied to real processes and
open systems with steady flows.
This fact brings with it the necessary implication that one of the conjugate power variables in thermodynamic systems
must generally have near-uniform value over the extent of the system. So as previously shown we consider,

m = P V̇ ̇
and T = T S, (9.1)

as appropriate variables representing the energy exchange in a simple thermodynamic system with its environment, where
m is the mechanical power, P is the pressure, V is the volume, T is the thermal power, T is the temperature, and S is the
entropy.
Under static equilibrium, the variables P and T must have uniform values throughout a homogeneous system (i.e., control
volume). Such variables have been recognized over the history of thermodynamics by several distinguishing names, perhaps
the most universal being intensive variable (others are intrinsic variables, potentials, and affinities). The intensities are
then assumed not to vary among the parts when we divide a system into subsystems. On the other hand, the conjugated
̇ must vanish under static conditions, yielding the integrals, V and S, constant. So too has this property
variables, V̇ and S,
been recognized by nomenclature, whereby such variables have been called extensive variables (others are capacities or
extrinsic variables), since if the system is divided into subsystems, the extensive variables distribute among the parts.
In summary, we may state that for a homogeneous thermodynamic system in equilibrium, considered as separated into
parts, the intensive variables are common to the parts and the extensive variables sum over the parts. For example, pressure
is common to a thermodynamic system in equilibrium while the volume of the individual parts of the system sum to the
total system volume. Intensive and extensive variables of simple thermodynamic systems are listed in Table 9.1.
Since equilibrium plays such a vital role in thermodynamics, it is important that we have a firm grasp of this concept.
This understanding is complicated by the fact that thermodynamic equilibrium and its relation to energy and entropy is
not as precise or simple as we might desire. Conservation of energy, as we have used throughout the entirety of this text,
is equivalent to the 1st law of thermodynamics, and, as we discussed in Chapter 3, physical system state is determined by
state variables which uniquely define energy content – both amount and form.
For example, a spring–mass system with 100 J (amount) of energy is not completely specified until we know the dis-
tribution (form) of this energy between spring and mass. This is determined by knowing spring displacement, q, mass
momentum, p, and fundamental energy equation,

 = (q, p) = Uq (q) + Tp (p). (9.2)

Table 9.1 Extensive and intensive variables

of simple thermodynamic systems.

Extensive Intensive

Entropy, S Temperature, T
Volume, V Pressure, P
Mole, Ni Chemical potential, 𝜇i
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
9.1 Thermodynamic Systems and Relations 559

From this fundamental equation, we can derive functional relations for spring force F and mass velocity 𝑣 as,
𝜕 𝜕Uq
F = F(q) = = , (9.3)
𝜕q 𝜕q
𝜕 𝜕Tp
𝑣 = 𝑣(p) = = . (9.4)
𝜕p 𝜕p
Also associated with each state is a rate law:
∑ ∑
ṗ = Fj ⇒ Δp = Fj Δt (9.5)
∑ ∑
q̇ = 𝑣j ⇒ Δq = 𝑣j Δt
which can be interpreted as conservation laws – momentum in the case of mass; displacement in the case of a spring.
In order to obtain states of thermodynamics systems, we need to define variables that specify internal energy distribution.
For the most basic thermodynamic system, this can be done with just one variable – entropy. The fundamental energy
equation for a simple thermodynamic system is then,
 = US (S), (9.6)
where US is the internal energy and S is the entropy. Thus, a functional equation of state (or constitutive relation) for system
temperature can be obtained from this equation using the derivative relation,
𝜕 𝜕US
T= = , (9.7)
𝜕S 𝜕S
where T is the thermodynamic temperature [21]. It remains to briefly elucidate the concept of entropy and its relation to
thermodynamic equilibrium.
Some things occur naturally while others do not. For example, as illustrated in Figure 9.1, if a gas is initially constrained
to lie in half of an available volume, it will spontaneously expand to fill this volume if the constraint is removed. In this
case, energy is dispersed from a more ordered distribution in (a) to a less ordered distribution in (b). It can also be easily
shown that case (b), when gas fills the total volume, is much more probable than any case when gas fills only a fraction of
the available volume [195, 196]. This is a general result for thermodynamic systems that can be stated:

The direction of spontaneous change is from an ordered state of low probability to a random state of high probability.
The equilibrium state corresponds to the most probable state of all possible internal states of the system.

For the particular example of Figure 9.1, the number of ways Ω of finding N atoms in a volume V is proportional to the
volume raised to the Nth power,
Ω(V, N) ∝ V N .
Entropy can be statistically defined in terms of Ω as,
S ≡ k ln Ω = k ln V N ,
where k is a constant and Ω and V are appropriately normalized. This definition of S gives a maximum for the most probable
configuration of the system. It also defines S as an extensive property; that is, if the amount of matter doubles from N to 2N,
the entropy also doubles as,
S = k ln V 2N = 2k ln V N .

(a) (b)

Figure 9.1 Dilute gas in a vessel. (a) Low probability of occurrence. (b) Most probable state (equilibrium).
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
560 9 Thermodynamic Systems

The above description of entropy is statistical. Classically, entropy was first developed in terms of heat transfer by Clausius
[197]. The change in internal energy is related to heat transfer by the first law of thermodynamics,
ΔU = ΔQ + ΔW,
where U is the internal energy, Q is the heat, and W is the work. Heat directly increases the random molecular motion of
internal energy while work involves organized correlated motion which can subsequently but not necessarily degrade to
thermal motion. Since heat is a randomizing influence, the entropy, which is a measure of dispersal, should increase with
heat input. Heat, by itself, is not a state of a system but rather an energy transfer to a system. In order to obtain a system
state, we divide randomizing influence by an amount of randomization already in a system. Since temperature is a measure
of system randomization, this leads to the following definition for change of entropy due to heat transfer,
dQrev
dS = ,
T
where heat transfer is restricted to reversible (near-equilibrium) transfer in order that S satisfy state conditions.2 This is the
classical definition of entropy attributed to Clausius and it can be shown to be equivalent to the statistical definition.3
Regardless of definition or interpretation, entropy has the following properties important to our subsequent discussion:
1) Entropy achieves a maximum at the most probable system state (equilibrium state).
2) Entropy is an extensive property.
3) Entropy is a thermodynamic state variable.
Rate laws for entropy or entropy production are defined by dissipative processes, as we have seen in earlier discussions
on heat transfer and resistive elements (which intrinsically generate entropy).

9.1.2 Gibbs Internal Energy

All state variables are created equal – but some are more equal than others.
– H.M. Paynter, with acknowledgment to George Orwell.

… the relation between the volume, pressure, and temperature affords a less complete knowledge of the properties
of the body than the relation between the volume, entropy, and energy.
– J.W. Gibbs

Beginning with the seminal work of J. Willard Gibbs [199], it has been useful to represent the state of a simple ther-
modynamic system with internal energy as a function of entropy, volume, and mass or moles of material constituents.
The fundamental energy relation may then be expressed as,
U = US VN1 N2 …Nn (S, V, N1 , N2 , … , Nn ) (9.8)
where entropy S quantifies dispersal of system energy, volume V quantifies system size, and moles Ni quantifies system
matter content. From this fundamental energy relation, we can define thermodynamic temperature, T, pressure, P, and
chemical potentials, 𝜇i , as,
𝜕USVNi
T = 𝜕S
, (9.9)

𝜕USVNi
−P = 𝜕V
, (9.10)
𝜕USVNi
𝜇i = 𝜕Ni
, (9.11)
Equating power flow into such a thermodynamic system to rate of change of energy gives,
𝜕USVNi 𝜕USVNi 𝜕USVNi
 = U̇ SVNi = ⋅ Ṡ + ⋅ V̇ + ̇
⋅ N. (9.12)
𝜕S 𝜕V 𝜕Ni
This can be represented graphically by a multiport C element as shown in Figure 9.2(a).

2 Review the discussion of state variables and state determined systems in Chapter 1.
3 Still another interpretation of entropy can be given in terms of information theory. See Tribus [198].
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
9.1 Thermodynamic Systems and Relations 561

μ1 Ṅ1
··
−P · μn
C
V˙
– Ṅn

T Ṡ

(a) (b)

Figure 9.2 (a) Multiport C element representing storage of internal energy. Note that since pressure decreases as volume increases
(i.e., 𝜕USV Ni ∕𝜕V is negative), pressure is indicated on the mechanical port with −P. This follows the sign convention of having all
power flow positive into the C element. Alternatively, one could indicate P as positive and show the half arrow pointing away from the
C element. (b) Piston-cylinder system containing a gas.

Table 9.2 Efforts, flows, and displacements (states) of simple

thermodynamic systems.

Generalized Thermal Mechanical Chemical

Effort, e T −P 𝜇i
Flow, f ̇ f
S, ̇ Q
V, Ṅ i , fNi
S
Displacement, q S V Ni

On this bond graph, efforts are T, −P, and 𝜇i , and flows are Ṡ = fS , V̇ = Q, and Ṅ i = fNi , and stored energy is given by
equation 9.8 in terms of states S, V, and Ni .
States of internal energy have a special property.4 Noting Tables 9.1 and 9.2, we see that thermodynamic states are a set
of extensive variables and efforts are conjugate intensive variables. This result leads to a unique mathematical property of
internal energy – Internal energy is a homogeneous first-order function of its extensive state variables [21]. That is, if all states
of an internal energy function are multiplied by a parameter 𝜆, the resulting energy is equal to the original internal energy
multiplied by this same parameter, or,
USVN (𝜆S, 𝜆V, 𝜆Ni ) = 𝜆 USVN (S, V, Ni ). (9.13)
This is a mathematical consequence of the fact that internal energy of simple thermodynamic systems is additive. This
results in a special form that energy density takes. Dividing internal energy by total system mole number N yields molar
specific energy u,
1
u= U (S, V, N1 , N2 , … , Nn ),
N SVN
∑
n
N= Ni ,
i

but, due to the first-order nature of internal energy, we have,

( )
S V N1 Nn
u = USVN , , ,…, .
N N N N
Defining s, 𝑣, xi as molar-specific entropy, molar-specific volume, and as mole fraction, xi = Ni ∕N, respectively,5 then
yields,
u = u(s, 𝑣, x1 , ..., xn ). (9.14)

4 The proper definition of these states requires at least local thermodynamic equilibrium.
5 Equivalently, mass- or volume-specific quantities can be defined, but we will find molar quantities more useful in our present discussion.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
562 9 Thermodynamic Systems

In particular, for a single component system, molar-specific energy can be expressed in terms of molar-specific
quantities as,
u = u(s, 𝑣, 1) = u(s, 𝑣). (9.15)
Intensive variables T and P can also be expressed in terms of molar specific quantities as,
𝜕u
T= , (9.16)
𝜕s
𝜕u
−P = . (9.17)
𝜕𝑣
9.1.2.1 Thermal Inertance
For thermodynamic system modeling at ordinary temperatures, total thermal energy is given by u = u(s, 𝑣). Hence, defining
a thermal inertance of the form, It = 𝜕u∕𝜕f , where f = ds∕dt, an entropy flow rate (analogous to a velocity), is not known
to exist. A quantum mechanical phenomena arising in condensed matter and referred to as phonons suggest wave-like
behavior at near zero absolute temperatures. Since waves often suggest an exchange of energy between potential and
kinetic forms, there could be some thermal inertia in this context. These phenomena are out of the scope and interest
of this book.

9.1.2.2 Legendre Transformations

Although entropy, volume, and moles are natural state variables to express thermodynamic energy, these variables are
difficult to measure and control in actual processes. In this case, it is desirable to obtain thermodynamic functions (his-
torically these have been chosen as negative coenergies, presumably because as temperature and volume are increased,
entropy and pressure decrease) that have intensive rather than extensive parameters as independent variables. This can be
achieved via Legendre transformations of internal energy. For example, in order to change from entropy to temperature,
this involves changing from internal energy to Helmholtz free energy, F, as,
F(T, V, Ni ) = −UTVNi = USVNi − TS, (9.18)
where UTVNi is the coenergy with S transformed to T.
Helmholtz free energy is most useful in isothermal systems and it has derivative relations,
𝜕F 𝜕F 𝜕F
−S = , −P = , and 𝜇i = .
𝜕T 𝜕V 𝜕Ni
The two other common thermodynamic Legendre transforms and associated relations are:

Enthalpy, H:
H(S, P, Ni ) = −US(−P)Ni = USVNi + PV (9.19)

𝜕H 𝜕H 𝜕H
T= , V= , and 𝜇i = .
𝜕S 𝜕P 𝜕Ni
Gibbs free energy, G:
G(T, P, Ni ) = −UT(−P)Ni = USVNi − TS + PV (9.20)

𝜕G 𝜕G 𝜕G
−S = , V= , and 𝜇i = .
𝜕T 𝜕P 𝜕Ni
Enthalpy is useful for isobaric and flow systems and Gibbs free energy is most useful for isothermal, isobaric systems
(this occurs most frequently in chemical systems).
Thus, we see that common Legendre transforms of internal energy correspond to the four possible causalities of the
thermal and mechanical ports of a thermodynamic C element, as shown in Figure 9.3.
The (total) coenergy with respect to all states (including now the matter ports) is,6
∑
n
UT(−P)𝜇i = USVNi − TS + PV − 𝜇i Ni , (9.21)
i

6 See discussion by Breedveld [74] on this and additional topics covered in this chapter.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
9.1 Thermodynamic Systems and Relations 563

−P T −P T
C C
V˙
– S˙ V˙
– S˙
(a) Fundamental (b) Helmholtz free energy

−P T −P T
C C
V˙
– S˙ V˙
– S˙
(c) Enthalpy (d) Gibbs free energy

Figure 9.3 Possible causalities on thermal and mechanical ports.

which can be explicitly calculated by noting that,

USVNi (𝜆S, 𝜆V, 𝜆Ni ) = 𝜆USVNi (S, V, Ni ),
differentiating with respect to 𝜆 and setting 𝜆 = 1 to give the Euler equation,
∑
n
USVNi = TS − PV + 𝜇i Ni . (9.22)
i

Combining this result with equation 9.21, we see that total coenergy is identically zero,
UT(−P)𝜇Ni (T, P, 𝜇i ) = 0. (9.23)
From this derivation, we see that one intensity can be eliminated in terms of the others. This implies that a thermody-
namic C element cannot have all-derivative causality [74], that is, T, P, 𝜇i cannot simultaneously be inputs, since the efforts
are related. This also implies that the equations of state,
T = T(S, V, Ni ),
P = P(S, V, Ni ),
𝜇i = 𝜇i (S, V, Ni ),
are not uniquely invertible to obtain S,V, and Ni .
A differential form of this constraint can be derived from the Euler equation 9.22 as
∑
n
−SdT + VdP − Ni d𝜇i = 0, (9.24)
i

which is known as the Gibbs–Duhem relation. This relation finds use when modeling open systems in Section 9.4.3.

9.1.2.3 Maxwell Reciprocity

Internal energy and its Legendre transformations all satisfy Maxwell reciprocity and therefore qualify as state functions.7
Due to the number of thermodynamic state functions that are used, there are many Maxwell relations.8 As an example,
we have,
𝜕F
−P = , (9.25)
𝜕V
and therefore,
𝜕P 𝜕2 F
− = . (9.26)
𝜕T 𝜕T𝜕V
Conversely, we have,
𝜕F
−S = , (9.27)
𝜕T
and,
𝜕S 𝜕2 F
− = . (9.28)
𝜕V 𝜕V𝜕T

7 See Section 4.6 for discussion on Maxwell Reciprocity.

8 Maxwell relations are notably useful in experimental determination of thermodynamic properties.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
564 9 Thermodynamic Systems

Since Helmholtz free energy satisfies Maxwell reciprocity, we can equate 9.26 and 9.28 as,
𝜕S 𝜕P
= . (9.29)
𝜕V 𝜕T
For example, for an ideal gas, we know that,

PV = NRT, (9.30)

where R is the universal gas constant. From this relation, we have,

𝜕P NR
= , (9.31)
𝜕T V
and using 9.29 yields,
𝜕S NR
= , (9.32)
𝜕V V
which says that entropy of an ideal gas in a system changes inversely with volume and directly with amount of matter.

Example 9.1 Vessel with photonic or electromagnetic internal energy

An “empty” vessel with interior walls held at temperature T𝑤 is assumed to contain photonic or electromagnetic internal
energy. The fundamental energy equation for this system is given as,

USV = 𝛼S4∕3 V −1∕3 , (9.33)

where 𝛼 is a constant. Obtain equations of state T(S, V) and P(S, V), Helmholtz free energy, total coenergy UT(−P) , and
internal energy and pressure in a rigid vessel of volume V a = 1 m3 having walls at a room temperature of T𝑤 = 298 K.
Solution
This example has an advantage in demonstrating most concepts of the previous discussion but with a reduced set of variables
(i.e., N = 0 in an empty vessel and can be eliminated from consideration). For instance, USV can be easily demonstrated to
be 1st order homogeneous as,

USV (𝜆S, 𝜆V) = 𝛼(𝜆S)4∕3 (𝜆V)−1∕3 = 𝜆(𝛼S4∕3 V −1∕3 ) = 𝜆USV (S, V).

and internal energy per volume (rather than per mole) u can be expressed in terms of entropy per volume s as,
U ( )4∕3
S
u = SV = 𝛼 = 𝛼s4∕3 .
V V
Equations of state can be derived from the fundamental relation 9.33 as,
𝜕USV ( )
4𝛼 S 1∕3 4𝛼 1∕3
T= = = s , (9.34)
𝜕S 3 V 3
𝜕USV ( )
𝛼 S 4∕3 𝛼
−P = =− = − s4∕3 , (9.35)
𝜕V 3 V 3
and then Helmholtz free energy F can be found via a Legendre transform as,

F = −UTV (T, V) = USV (S(T, V), V) − TS(T, V). (9.36)

Entropy can be found from the temperature equation of state 9.34 as

27
S= VT 3 , (9.37)
64𝛼 3
which can be used to express internal energy USV in terms of temperature and volume as

USV (T, V) = bVT 4 , (9.38)

where b = 81∕256 𝛼 3 . Equation 9.38 is called Stefan–Boltzmann’s law for radiation and for a vacuum b = 7.56 ×
10−16 J/m3 K 4 . Using 9.36, 9.37, and 9.38, Helmholtz free energy is,
b
F = − VT 4 . (9.39)
3
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
9.2 Equations of State for Ideal Gases 565

The total coenergy can be found from a Legendre transform as

UT(−P) = TS − PV − USV = −F − PV. (9.40)
From the pressure equation of state 9.35 and entropy relation 9.37, we have
b 4
P= T . (9.41)
3
Substituting this result in 9.40, coenergy is
b b
UT(−P) = VT 4 − VT 4 = 0, (9.42)
3 3
which is the expected result for a first order homogeneous energy function. Note that 9.41 demonstrates that intensive
variables P and T are not independent.
Internal energy and pressure at T = 298 K in a 1 m3 vessel can be calculated from 9.38 and 9.41 as
J
USV = 7.56x10−16 3 4 ⋅ 1m3 (298 K)4 = 5.96 × 10−6 J,
mK
7.56 × 10−16 J∕m3 K 4
P= ⋅ (298 K)4 = 1.99 × 10−6 N∕m2 .
3
From these numbers, we see that radiation pressure is 11 orders of magnitude lower than atmospheric pressure
(≈0.1 × 106 N/m2 ) and therefore negligible in any earthly application.

9.2 Equations of State for Ideal Gases

Before we can proceed with development of predictive, quantitative models for thermodynamic systems, we must obtain
equations of state that can adequately represent internal energy storage in a system. These are not to be confused with system
state equations which are differential equations describing change in system state variables with time. Equations of state are
simply the constitutive relations for a thermodynamic C element and, as such, represent static, equilibrium energy storage
functions. Like all element constitutive relations, thermodynamic equations of state are based on underlying microphysical
structure. However, prediction of accurate equations of state from microphysical statistical analysis for most engineering
systems remains an unfulfilled dream. We thus resort to empirical techniques. This section is limited to a discussion of
simple thermodynamic systems that have pure (one material state N), compressible (volume state V), thermodynamic
(entropy state S) substances.

9.2.1 Ideal Gas

The ideal gas relation can provide an accurate approximation for gases over a suitable range of temperatures and pressures.
It is usually characterized by two relations, a mechanical equation of state relating pressure to specific volume and temper-
ature and experimental determination of constant volume-specific heat. These equations can be expressed as,
RT
P= Mechanical equation of state, (9.43)
𝑣
du
= c𝑣 Constant volume specific heat, (9.44)
dT
where R is the universal gas constant and c𝑣 is the molar-specific heat capacity. For an ideal gas, internal energy and
therefore c𝑣 are solely functions of temperature and in many engineering cases c𝑣 can be considered a constant.
These two equations are equivalent to a single energy relation, u(s, 𝑣). This equivalence can be deduced by noting that
the change in specific internal energy with respect to specific entropy and volume is,
𝜕u 𝜕u
du = ds + d𝑣 = Tds − Pd𝑣. Gibbs equation. (9.45)
𝜕s 𝜕𝑣
Solving for entropy change yields,
du P
ds = + d𝑣
T T
c𝑣 R
= dT + d𝑣,
T 𝑣
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
566 9 Thermodynamic Systems

which can be integrated directly as,

T ( )
c𝑣 𝑣
s − so = dT + R ln , (9.46)
∫To T 𝑣o
where 𝑣o is a standard specific volume found at a reference pressure Po and temperature To , commonly chosen as To = 298 K
and Po = 0.1 MPa, as,
RTo
𝑣o = ,
Po
and so is a reference entropy,
so = s(To , Po ).
Using 9.44, we can obtain an energy relation as,
T
u(T) = uo + c𝑣 dT, (9.47)
∫To
where uo is the reference energy. Equations 9.46 and 9.47 represent a parametric form for the energy relation u(s, 𝑣) with T
a parameter.
If c𝑣 is a constant, T can be eliminated explicitly. For constant c𝑣 , we have,
( ) ( )
T 𝑣
s − so = c𝑣 ln + R ln ,
To 𝑣o
which yields,
( 𝑣 )R∕c𝑣 ( )
o s − so
T = To exp . (9.48)
𝑣 c𝑣
where R∕c𝑣 = (𝛾 − 1). It is sometimes convenient to express this result in the form,
( 𝑣 )𝛾−1 ( )
s − so
T = To o exp . (9.49)
𝑣 c𝑣
where 𝛾 = (c𝑣 + R)∕c𝑣 = cp ∕c𝑣 , and R = cp − c𝑣 (Mayer’s relation). Recall that these quantities are defined in terms of ther-
modynamic properties [20],
( )
𝜕u
c𝑣 = , (9.50)
𝜕T 𝑣
( )
𝜕h
cp = . (9.51)
𝜕T P
The fundamental internal energy equation is then,
UTN = Nu(T) = N(uo + c𝑣 (T − To ))
[ (( )𝛾−1 ( ) )]
𝑣o s − so
USVN = Nu(T(s, 𝑣)) = N uo + c𝑣 To exp −1 , (9.52)
𝑣 c𝑣
where we note that s = S∕N and 𝑣 = V∕N.
From this fundamental relation, we can then derive equations of state (constitutive relations for a thermodynamic C
element) for an ideal gas with constant specific heat as in Table 9.3. Note that by substituting 9.53 into 9.55, 𝜇 simplifies

Table 9.3 Equations of state for ideal gas with constant specific heat.

( 𝑣 )𝛾−1 ) (
𝜕USVN 𝜕u s − so
T= = = To o exp (9.53)
𝜕S 𝜕s 𝑣 c𝑣
( 𝑣 )𝛾 ( )
𝜕USVN 𝜕u s − so
−P = = = −Po o exp (9.54)
𝜕V 𝜕𝑣 𝑣 c𝑣
( )
𝜕USVN ( ) ( 𝑣 )𝛾−1 s − so
𝜇= = u − Ts + P𝑣 = uo − c𝑣 To + c𝑣 + R − s To o exp (9.55)
𝜕N 𝑣 c𝑣
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
9.2 Equations of State for Ideal Gases 567

( ) ( )
to 𝜇 = uo − c𝑣 To + c𝑣 + R − s T. Further, if uo is taken as c𝑣 To , then 𝜇 further simplifies to 𝜇 = c𝑣 + R − s T. Since
( )
c𝑣 + R = cp , 𝜇 is also equivalent to cp − s T, which can be convenient to use in computations.9
The relation 9.55 is a direct result of the Euler equation 9.22. Since the use of chemical potential may be less familiar to
some engineers, it is worth noting that this equation can also be expressed in a more conventional form as,
P
𝜇 = 𝜇 o (T) + RT ln , (9.56)
Po
where
T
𝜇 o (T) = 𝜇o + (cp − so )(T − To ) − cp T ln ,
To
and,
𝜇o = uo − To so + RTo .
If we again substitute c𝑣 To for uo , then 𝜇o = (c𝑣 + R − so )To = (cp − so )To , which is the same form as the simplified expres-
sion of 𝜇 in equation 9.55. Now, 𝜇 o (T) becomes,
T T
𝜇 o (T) = (cp − so )To + (cp − so )(T − To ) − cp T ln = (cp − so )T − cp T ln .
To To
Substituting this expression for 𝜇 o (T) into equation 9.56 and making use of the Gibb’s equation,
T P
s − so = cp ln − R ln ,
To Po
it can be shown that equation 9.56 is equivalent to equation 9.55 in its simplified form,
𝜇 = (cp − s)T. (9.57)

Example 9.2 Mechanical compression of argon gas

Consider a laboratory apparatus in which it is desired to subject a specimen to a relatively large change in temperature and
pressure over a short time duration. The apparatus consists of a chamber that can be filled with a gas through an inlet valve,
while a piston/mass is suspended, as shown in Figure 9.4(a). The walls of the chamber are insulated. The upper wall of the

Release

Area, A

Ar

Valve
Specimen

(a) (b)

Figure 9.4 Apparatus for short duration temperature and pressure change. (a) Schematic of laboratory apparatus and (b) bond graph
model of adiabatic closed system model.

9 In simulations, it is good practice to use tabulated values of cp and c𝑣 and to then define R = cp − c𝑣 and 𝛾 as cp ∕c𝑣 .
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
568 9 Thermodynamic Systems

chamber formed by the suspended piston, which can be dropped by the force of gravity upon release. The mass is chosen
such that its subsequent drop into the chamber results in a desired temperature and pressure rise in the chamber.
If the chamber is filled with argon (Ar) at standard temperature and pressure, and a mass mp = 50 kg is dropped from a
height relative to the chamber floor of 10 cm, predict the pressure, temperature, and volume of the chamber as a function
of time. The piston has area A = 0.001 m2 , so the initial volume of contained Ar is 0.0001 m3 .
Solution
For a closed, adiabatic chamber, Ṅ = Ṡ = 0 and therefore thermal and material ports have zero power. A bond graph model
for the system is depicted in Figure 9.4(b). The I element represents the mass kinetic energy, and the C element represents
the argon internal energy. The transformer couples mechanical translation to the fluid domain with transformer modulus
A. Applying standard bond graph equation techniques, we have state equations:
𝑣̇ p = A(P − PA )∕mp − g, (9.58)
V̇ = A𝑣p , (9.59)
where 𝑣p = pp ∕mp is the mass velocity (define positive upward), V is the volume of contained gas, and PA is the atmospheric
pressure. It remains to specify the constitutive relation for chamber pressure, P and temperature, T. Assuming argon can
be adequately modeled as an ideal gas with constant specific heat, the chamber pressure is found by equation 9.54 as,
( 𝑣 )𝛾 ( ) ( ) ( )
o s − so No 𝑣o 𝛾 s − so
P = Po exp = Po exp , (9.60)
𝑣 c𝑣 V c𝑣
where so and No are values of specific entropy and number of moles, respectively, at an initial time. These variables both
remain constant throughout the process due to adiabatic and closed system constraints. Thus pressure can be expressed,
( )𝛾
Vo
P = Po , (9.61)
V
which is the usual expression for pressure during adiabatic compression. Chamber temperature can be found in similar
fashion as,
( )𝛾−1
Vo
T = To . (9.62)
V
where Po and To are the initial pressure and temperature. Numerical parameters to be used in a simulation include initial
pressure Po = PA = 0.101 MPa, To = 298 K, initial momentum 𝑣po = 0, and, for argon, 𝛾 = 1.667. Using these quantities
in equation 9.61 and state equations 9.58 and 9.59 yields a set of nonlinear differential equations. These equations can
be solved numerically to obtain chamber pressure, temperature, and volume as functions of time. Numerically integrating
these equations yields the results in Figure 9.5. The pressure and temperature have been normalized by critical pressure and
temperature, respectively,10 to obtain reduced quantities as Pr = P∕Pcr and Tr = T∕Tcr . For argon, Pcr = 4.86 MPa (≈48 atm)
and Tcr = 151 K. From the plot, we see that the peak reduced pressure is approximately 0.9 and reduced temperature is 8.9
with volume ratio (V∕V o ) reaching a minimum of approximately 0.1. The entire cycle occurs in approximately 0.3 s.
As a general rule, if Pr < 1 and Tr > 0.4 for a particular gas, then an ideal gas relation will be within 10% of experimental
values. As pressure is lowered, temperature raised accuracy will be better. Since both of these criteria are easily met for the
entire range of process temperature and pressure, we expect the prediction in Figure 9.5 to be accurate.

9.2.2 Van der Waals Gas

As pressure is raised and temperature lowered, the ideal gas relation will increasingly deviate from experimental values for
real thermodynamic substances. If we plot compressibility, Z = P𝑣∕RT, as a function of reduced pressure and temperature,
we can quantify this deviation. From Figure 9.6, which is a compressibility plot for argon, we see that argon can deviate
substantially from an ideal gas for appropriate values of Pr and Tr (Z = 1 for an ideal gas).
An improvement on the mechanical equation of state was suggested by J.D. van der Waals in 1873 as,
RT a
P= − Mechanical equation of state. (9.63)
𝑣 − b 𝑣2
10 Recall that critical temperature, Tcr , is the lowest temperature for which an isotherm shows no phase change (no liquid-vapor two-phase
region) on a P − 𝑣 diagram. The critical point is the point where the critical isotherm is tangent to the saturated vapor line. The critical pressure,
Pcr , is the pressure at this tangent point.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
9.2 Equations of State for Ideal Gases 569

Piston height Temperature, pressure, and volume ratios

10
9

8
, cm

5
Tr
7
Chamber
bottom 6
0
0 0.1 0.2 0.3
5
Time, sec
Piston velocity 4
1
3
0.5
, m/s

0 2
Vr Pr
–0.5 1

–1 0
0 0.1 0.2 0.3 0 0.1 0.2 0.3
Time, sec Time, sec

Figure 9.5 Simulation results for closed, adiabatic argon system modeled as an ideal gas with Pk = 0.101 MPa and Tk = 298 K.
Motion of the piston shows it falling from 10 cm height, reaching a peak minimum height, then rising back up to original position.

1.0 Tr = 2

0.8
Tr = 1.5

0.6 Tr = 1.2
Z
0.4
Tr = 1

0.2 Lee and Kesler

Van der Waals
0
0 0.5 1 1.5 2 2.5 3
Pr

Figure 9.6 Compressibility plot for simple gases (Ar) and Van der Waals gas, including correlations of Lee and Kesler [200].

The correction due to the b parameter accounts for the small but finite volume of the molecules in the fluid while the a
parameter accounts for proper internal pressure due to molecular attractions.
Unlike an ideal gas, internal energy for a Van der Waals gas is a function of both temperature (or entropy) and volume.
The relation between internal energy and c𝑣 becomes a partial derivative rather than a total derivative as in equation 9.44,
thus
( )
𝜕u
= c𝑣 Constant volume specific heat, (9.64)
𝜕T 𝑣
where specific internal energy is considered as a function of temperature and specific volume,
u(s, 𝑣) = u(s(T, 𝑣), 𝑣). (9.65)
Like an ideal gas, equations 9.64 and 9.65 are equivalent to a single energy relation, u(s, 𝑣). This relation can be found by
( ) ( )
𝜕u 𝜕u
du = dT + d𝑣
𝜕T 𝑣 𝜕𝑣 T
( )
𝜕u
= c𝑣 dT + d𝑣.
𝜕𝑣 T
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
570 9 Thermodynamic Systems

Now (𝜕u∕𝜕𝑣)T is not the same as (𝜕u∕𝜕𝑣)s = −P, but rather

( ) ( ) ( )
𝜕u 𝜕u 𝜕s 𝜕u
= +
𝜕𝑣 T 𝜕s 𝜕𝑣 T 𝜕𝑣 s
( )
𝜕s
=T − P.
𝜕𝑣 T
From our discussion of Maxwell reciprocity (equation 9.29), we have (𝜕s∕𝜕𝑣)T = (𝜕P∕𝜕T)𝑣 and therefore,
( ( ) )
𝜕P
du = c𝑣 dT + T − P d𝑣. (9.66)
𝜕T 𝑣
This form for energy change is particularly useful for computing internal energy from a mechanical equation of state and
c𝑣 data. To find the change in u from a reference state, we then have,
T 𝑣( ( ) )
𝜕P
u − uo = c𝑣 dT + T − P d𝑣. (9.67)
∫To ∫𝑣o 𝜕T 𝑣
Note that the second integrand is zero for an ideal gas relation but not for a Van der Waals gas which gives,
a
du = c𝑣 dT + 2 d𝑣,
𝑣
or,
T
a a
u = uo + c𝑣 dT − + . (9.68)
∫To 𝑣 𝑣o
Entropy change as a function of T and 𝑣 can be calculated directly from Gibbs equations 9.45 and 9.66 as
( )
c 𝜕P
ds = 𝑣 dT + d𝑣. (9.69)
T 𝜕T 𝑣
This equation is a general expression for entropy change for any mechanical equation of state and c𝑣 data. Specifically
for a Van der Waals fluid, we have,
c ( )
R
ds = 𝑣 dT + d𝑣, (9.70)
T 𝑣−b
which can be integrated directly as,
T ( )
c𝑣 𝑣−b
s − so = dT + R ln , (9.71)
∫To T 𝑣o − b
where reference conditions are related by11
RTo a
Po = − .
𝑣o − b 𝑣2
Equations 9.68 and 9.71 represent a parametric (in T) set of equations for the internal energy function u(s, 𝑣). For constant
c𝑣 , equation 9.71 gives,
( ) ( )
𝑣o − b R∕c𝑣 s − so
T = To exp . (9.72)
𝑣−b c𝑣
Recognizing that R∕c𝑣 = (𝛾 − 1) results in equation 9.72 being identical in form to the ideal gas equation 9.53, and 9.68
gives,
a a
u = uo + c𝑣 (T − To ) − + ,
𝑣 𝑣o
which upon substitution of 9.72 yields specific internal energy as,
(( ) ( ) )
𝑣o − b R∕c𝑣 s − so a a
u(s, 𝑣) = uo + c𝑣 To − exp −1 − + . (9.73)
𝑣−b c𝑣 𝑣 𝑣o
The fundamental internal energy expression is then,
USVN = Nu(s, 𝑣).
From this fundamental relation, we can then derive equations of state as in Table 9.4.

11 Van der Waals equation can be multivalued in 𝑣 for given P and T. These regions represent a two-phase fluid. The reference conditions, Po
and To , should be chosen outside of this region.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
9.2 Equations of State for Ideal Gases 571

Table 9.4 Equations of state for Van der Waals gas with constant specific heat. Note that since R∕c𝑣 = (𝛾 − 1) and (c𝑣 + R)∕c𝑣 = 𝛾),
equation 9.75 becomes identical to 9.53 and equation 9.75 is identical to 9.54 (in form), the difference being the a and b corrections.

( ) ( )
𝜕USVN 𝜕u 𝑣 − b R∕c𝑣 s − so
T= = = To o exp (9.74)
𝜕S 𝜕s 𝑣−b c𝑣
( )(c𝑣 +R)∕c𝑣 ( )
𝜕USVN 𝜕u RTo 𝑣o − b s − so a
−P = = =− exp + (9.75)
𝜕V 𝜕𝑣 𝑣o − b 𝑣−b c𝑣 𝑣2
𝜕USVN
𝜇= = u − Ts + P𝑣 (9.76)
𝜕N

Example 9.3 Mechanical compression of argon gas revisited

The laboratory apparatus of Example 9.2, appropriately redesigned for lower temperature and higher-pressure compression,
is now filled with argon at To = 151 K and Po = 2 MPa. If a mass of 500 kg is dropped from a height of 10 cm, predict the
pressure, temperature, and volume as a function of time.
Solution
The bond graph model remains unchanged from that described in the last example. The only change in the state equations
is in the constitutive relation for chamber pressure P. For the expected temperature and pressure range, an ideal gas relation
will no longer be a good approximation for argon. Instead we will assume that argon can be adequately modeled as a Van
der Waals gas with constant specific heat.
In order to use Van der Waals equation of state, we must determine the coefficients a and b for argon. Determination of
Van der Waals constants is usually done by matching at the critical point (sometimes an empirical match for a particular
pressure and temperature range is used instead). The critical point of a Van der Waals mechanical equation of state 9.63 is
the inflection point of the relation determined by,
RTcr a
Pcr = −
𝑣cr − b 𝑣2cr
( )
𝜕P RTcr 2a
=− + =0
𝜕𝑣 cr (𝑣cr − b)2 𝑣3cr
( 2 )
𝜕 P 2RTcr 6a
= − = 0.
𝜕𝑣2 cr (𝑣cr − b)3 𝑣4cr
Using these three equations to solve for a, b, and 𝑣cr in terms of Pcr and Tcr yields,
3RTcr
𝑣cr = , (9.77)
8Pcr
a = 3Pcr 𝑣2cr , (9.78)
b = 𝑣cr ∕3. (9.79)
Equation 9.77 gives a critical compressibility of,
Pcr 𝑣cr 3
Zcr = = = 0.375 ⇒ Van der Waals critical compressibility. (9.80)
RTcr 8
Although real gases have a critical compressibility of more like Zcr = 0.29, Van der Waals equation is quantitatively
accurate for low pressure and high temperature (but for higher pressures and lower temperatures than an ideal gas), and it
is qualitatively correct for much higher pressure and lower temperature.
In reduced form, Van der Waals equation is,
1 Tr a
Pr = − 2r , (9.81)
Zcr 𝑣r − br 𝑣r
where ar = a∕(Pcr 𝑣2cr ) = 3 and br = b∕𝑣cr = 1∕3 and compressibility is,
8 𝑣r a
Z = Zcr − r . (9.82)
3 𝑣r − br 𝑣r Tr
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
572 9 Thermodynamic Systems

Plotted in Figure 9.6 is a 12 parameter compressibility correlation due to Lee and Kesler [200] for simple fluids (principally
Ar, Kr, and methane) and the corresponding Van der Waals compressibility.
As can be seen, a Van der Waals relation, although still limited in accuracy to relatively low pressure and high temperature,
represents an improvement over an ideal gas (which has a constant compressibility of Z = 1).
Argon has a critical pressure and temperature of Pcr = 4.86 MPa and Tcr = 151 K, respectively, which results in Van der
Waals constants of a = 1.37 × 105 Jm3 /kmol2 and b = 0.0323 m3 /kmol. Using these constants in 9.75 and 9.74, the chamber
pressure and temperature can be found as,
( )
RTk 𝑣k − b (c𝑣 +R)∕c𝑣 a
P= − 2,
𝑣k − b 𝑣 − b 𝑣
( )
𝑣 − b R∕c𝑣
T = Tk k ,
𝑣−b
where initial conditions at time step k are related as,
RTk a
Pk = − ,
𝑣k − b 𝑣2
k
with 𝑣k = V k ∕Nk . Numerical simulation of the state equations results in a peak reduced pressure of approximately 2.8 and
a reduced temperature of 2.2 with volume ratio reaching a minimum of approximately 0.35.

9.2.3 Summary
We reviewed the concept of thermodynamic extensive and intensive variables and introduced entropy as a fundamen-
tal thermodynamic extensive state variable. The thermodynamic intensive variables are temperature, pressure, chemical
potentials, which are derivatives of a fundamental internal energy function. We introduced a multiport C element for
representing internal energy storage in our thermodynamic system models. We also discussed the properties of homoge-
neous first-order energy functions and showed how Legendre transformation is related to internal energy, Helmholtz free
energy, enthalpy, and Gibbs free energy. These relations can be interpreted through applied causality on the thermodynamic
multiport C element. Finally, we discussed Maxwell reciprocity for thermodynamic functions.
Section 9.3 introduces how these concepts can be extended to general thermodynamic systems that encompass multiple
energy domains. After discussing thermomechanical systems, we take up open systems and multicomponent systems.

9.3 Applications in Thermomechanical Systems

Besides simple thermodynamic systems as discussed in the foregoing, thermodynamic systems can encompass, along with
the thermal domain, general energy domains. Later we examine multiple material phases and constituents, while the
following touches on elastic and kinetic systems. The examples discussed in this section are meant to demonstrate how
a bond graph modeling formalism can be applied to these general systems. A more complete discussion of general thermo-
dynamic systems can be found in a number of advanced thermodynamic texts [201, 202].

9.3.1 Thermal Systems in Motion

In order to energetically include macroscopic motion of a thermal system, we need only include linear and angular
velocity–momentum quantities to our thermodynamic variables. Including extensive variables linear momentum, pi , and
angular momentum, hi , in our energy expression yields,
U = USVNpi hi (S, V, N, pi , hi ),
where i represents one of the three spatial coordinates. Velocities can be found from this expression as,
𝜕USVNpi hi 𝜕Uu
𝑣i = = ,
𝜕pi 𝜕 p̂ i
𝜕USVNpi hi 𝜕Uu
𝜔i = = ,
𝜕hi 𝜕 ĥ i
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
9.3 Applications in Thermomechanical Systems 573

and temperature is,

𝜕USVNpi hi 𝜕Uu
T= = ,
𝜕S 𝜕s
where p̂ i and ĥ i are matter-specific momenta (i.e., p̂ i = pi ∕N). Note that we are no longer making any distinction between
two forms of energy (kinetic and potential), but we are treating all energy in one fundamental function [74, 203]. The engi-
neering importance of macroscopic motion of thermodynamic systems is primarily related to open systems in which matter
can enter and leave a system. We thus delay further discussion of this topic until we take up open systems.

9.3.2 Thermoelastic Systems

We extend here methods for dealing with thermodynamic effects in mechanical systems and thermoelasticity. An introduc-
tion to thermoelastic effects is given in Section 3.7 of Chapter 3. That previous discussion and the one taken here adopt a
conventional thermodynamic approach with volume as a single extensive mechanical parameter (pressure is the conjugate
intensive variable). A more detailed analysis of thermoelasticity requires additional states in terms of elastic strains. This
additional information augments but does not invalidate our basic pressure–volume description [204]. In the full theory,
extensive variables include strain components with conjugate stress intensive variables.12 Thermoelastic modeling using
bond graphs is also described in recent work by Zanj et al. [206, 207].
Appending the additional strain components to our extensive variables and neglecting matter change (N) yields an inter-
nal energy of the form,
U = USV o 𝛾i j (S, V o 𝛾i j ),
where 𝛾i j is a Lagrangian strain tensor13 of a deformable body, i, j represent cartesian coordinates that take on values of
x, y, z or 1, 2, 3, and V o is unstrained volume.14 A stress tensor 𝜎i j is related to this energy expression as,
𝜕USV o 𝛾i j 𝜕u
𝜎i j = = ,
𝜕V o 𝛾i j 𝜕𝛾i j
and temperature can be obtained in the usual fashion as,
𝜕USV o 𝛾i j 𝜕u
T= = ,
𝜕S 𝜕s
where u is volume specific energy and s is volume specific entropy. The intensive–extensive pair 𝜎i j − V o 𝛾i j is then added
to other thermodynamic quantities to form a complete thermodynamic description of a thermoelastic system.
For linear elasticity in one dimension, stress and strain are related by Young’s modulus E as,
𝜎 = E𝛾,
and, if we include linear thermal stress this becomes
𝜎 = E(𝛾 − 𝛼(T − To )) Mechanical equation of state, (9.83)
where 𝛼 is a coefficient of thermal expansion and To is a reference temperature. This mechanical equation of state along
with determination of a constant strain volume-specific heat15
( )
𝜕u
= c𝛾 Constant strain-specific heat (9.84)
𝜕T 𝛾
can be used to characterize thermoelastic properties of a system. For one-dimensional thermoelasticity, the fundamental
internal energy expression can be obtained as a function of temperature and strain, u = u(s(T, 𝛾), 𝛾), as,
( )
𝜕u
du = c𝛾 dT + d𝛾
𝜕𝛾
( (T ) )
𝜕u
= c𝛾 dT + −T + 𝜎 d𝛾, (9.85)
𝜕T 𝛾

12 For a more general approach to modeling thermoelasticity with bond graphs, see Ingrim [205].
13 Make sure to distinguish this 𝛾i j from the 𝛾 as ratio of specific heats.
14 The constant V o is included to insure consistency of units, that is, 𝜎i j ⋅ V o 𝛾i j has units of energy.
15 Equal to constant volume-specific heat for isotropic materials.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
574 9 Thermodynamic Systems

where we have used the relations,

( )
𝜕u
c𝛾 =
𝜕T 𝛾
( ) ( ) ( )
𝜕u 𝜕u 𝜕s 𝜕u
= +
𝜕𝛾 T 𝜕s 𝜕𝛾 T 𝜕𝛾 s
𝜕u
T=
( ) 𝜕s
𝜕s ( )
𝜕𝜎
=−
𝜕𝛾 T 𝜕T 𝛾
( )
𝜕u
𝜎= .
𝜕𝛾 s
in a completely analogous fashion to similar expressions used to derive equation 9.66.16
For linear thermoelasticity (equation 9.83), with constant specific heat, the energy is then,
1
u = uo + c𝛾 (T − To ) + E𝛾 2 − E𝛼To 𝛾. (9.86)
2
Entropy change as a function of T and 𝛾 can be calculated directly from a Gibbs relation and 9.85 as,
du 𝜎
ds = − d𝛾
T T
c𝛾 ( )
𝜕𝜎
ds = dT − d𝛾. (9.87)
T 𝜕T 𝛾
For linear thermoelasticity with constant specific heat, we have,
s = so + c𝛾 ln(T∕To ) + E𝛼𝛾,
and temperature is then found as,
( )
s − so − E𝛼𝛾
T = To exp . (9.88)
c𝛾
Upon substituting 9.88 into 9.86, we obtain the fundamental form for specific internal energy,
( ( ) )
s − so − E𝛼𝛾 1
u = uo + c𝛾 To exp − 1 + E𝛾 2 − E𝛼To 𝛾. (9.89)
c𝛾 2
From this fundamental relation, we then derive equations of state for one-dimensional thermoelasticity (in entropy
form) as,
( ( ) )
𝜕u s − so − E𝛼𝛾
𝜎= = E𝛾 − E𝛼To exp −1 , (9.90)
𝜕𝛾i j c𝛾
( )
𝜕u s − so − E𝛼𝛾
T= = To exp . (9.91)
𝜕s c𝛾

Table 9.5 Equations of state of for linear thermoelastic body.

[ ( ∑ ) ]
𝜕u ∑ s − so − i j 𝛽i j 𝛾i j
𝜎i j = = Ei jkm 𝛾km − 𝛽i j To exp −1 , (9.92)
𝜕𝛾i j km
c𝛾
( ∑ )
𝜕u 𝜕u s − so − i j 𝛽i j 𝛾 i j
T= = = To exp . (9.93)
𝜕s 𝜕s c𝛾

16 The only difference is compressive pressure is assumed positive while compressive stress is negative.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
9.3 Applications in Thermomechanical Systems 575

For a more general three-dimensional linear thermoelastic configuration, but still with constant specific heat, we have a
fundamental internal energy of the form,
[ ( ∑ ) ]
s − so − i j 𝛽i j 𝛾i j 1∑ ∑
u = uo + c𝛾 To exp −1 + Ei jkm 𝛾i j 𝛾km − To 𝛽i j 𝛾i j , (9.94)
c𝛾 2 i jkm ij

where Ei jkm is an elastic modulus tensor and 𝛽i j is a thermoelastic modulus tensor which relate strain and temperature to
stress as,
∑
𝜎i j = Ei jkm 𝛾km − 𝛽i j (T − To ). Mechanical equation of state. (9.95)
km

Equations of state in entropy form are given in Table 9.5. Application of these concepts is illustrated in Example 9.4.

Example 9.4 Linear thermoelastic beam with attached mass

A linear thermoelastic beam with a thin flat cross-sectional and isotropic properties is vertically attached to a mass, as
shown in Figure 9.7. Derive a bond graph description and state equations for adiabatic and isothermal conditions.

Se

Thin flat beam σ22 = 0 γ̇22 –

Vo
g
2
Mg Aσ11 σ11 Tg
3 Se 1 T C Se
1 v γ̇11 –
Vo Ṡg
ṗ v σ12 = 0 γ̇12 –
Vo
M
I:M Se
(a) (b)

Figure 9.7 (a) Schematic of a mass hanging from a beam with thermoelastic effects and (b) bond graph model.

Solution
First, for isotropic material, the elastic and thermoelastic modulus tensors reduce to effectively just three parameters as,
Ei jkm = 𝜆𝛿i j 𝛿km + 𝜇(𝛿ik 𝛿jm + 𝛿im 𝛿kj )
𝛽i j = 𝛽
𝜈E
𝜆=
(1 + 𝜈)(1 − 2𝜈)
E
𝜇=
2(1 + 𝜈)
E𝛼
𝛽= ,
1 − 2𝜈
where 𝜆, 𝜇 are called Lamé constants and E, 𝜈, 𝛼 are, respectively, Young’s modulus, Poisson’s ratio, and coefficient of
thermal expansion, and 𝛿i j is a delta function, that is, 𝛿i j = 1, i = j; 𝛿i j = 0, i ≠ j. Second, for a thin flat plate beam, one
dimension is extremely large in comparison with the others (let us assume x3 ). In this case, the strain will usually be
negligible in the x3 direction. This is called plain strain and analysis can be performed in just two dimensions (x1 − x2 ) with
unit thickness along the x3 axis. In this case, the stress–strain relation reduces to,

⎡ 𝜎11 ⎤ ⎡ 𝜆 + 2𝜇 𝜆 + 2𝜇 0 ⎤ ⎡ 𝛾11 ⎤ ⎡1⎤

⎢𝜎 ⎥ =⎢ 𝜆 𝜆 0 ⎥ ⎢ 𝛾 ⎥ − 𝛽(T − To ) ⎢ 1 ⎥ . (9.96)
⎢ 22 ⎥ ⎢ ⎥ ⎢ 22 ⎥ ⎢ ⎥
⎣ 𝜎12 ⎦ ⎣ 0 0 2𝜇 ⎦ ⎣ 𝛾12 ⎦ ⎣0⎦
For an adiabatic system, the thermal bond on the bond graph of Figure 9.7(b) has a flow source input (Ṡ = 0) and standard
bond graph causality assignment indicates three states which can be taken as mass momentum, p, tensile strain in the
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
576 9 Thermodynamic Systems

1 direction, 𝛾11 , and volume-specific entropy s. Using 9.92, 9.93, and 9.96, and standard bond graph equation derivation
techniques, we obtain the following differential-algebraic set of state equations:
[ ( ( ) )]
s − so − 𝛽(𝛾11 + 𝛾22 )
ṗ = −A (𝜆 + 2𝜇)𝛾11 + 𝜆𝛾22 − 𝛽To exp − 1 + Mg,
c𝛾
1
𝛾̇ 11 = p,
ML
ṡ = 0,
and the following algebraic derivative causality equations:
𝜎12 = 0 = 𝛾12 ,
[ ( ) ]
s − so − 𝛽(𝛾11 + 𝛾22 )
𝜎22 = 0 = 𝜆𝛾11 + (𝜆 + 2𝜇)𝛾22 − 𝛽To exp −1 ,
c𝛾
where L is the beam length and A is the beam area (V o = AL). These equations can be solved numerically for p, 𝛾11 , and 𝛾22
(s is constant).
For the isothermal case, the thermal bond has an imposed temperature (effort source) so there are only two independent
states that can be taken as p and 𝛾11 . For this case, isothermal differential-algebraic state equations are:
[ ]
ṗ = −A (𝜆 + 2𝜇)𝛾11 + 𝜆𝛾22 − 𝛽(T − To ) + Mg,
1
𝛾̇ 11 = p,
ML
and the algebraic derivative causality equations:
𝜎12 = 0 = 𝛾12 ,
𝜎22 = 0 = 𝜆𝛾11 + (𝜆 + 2𝜇)𝛾22 − 𝛽(T − To ),
s = so + c𝛾 ln(T∕To ) + 𝛽(𝛾11 + 𝛾22 ),
which represent a set of linear differential-algebraic equations for p, 𝛾11 , and 𝛾22 and a simple nonlinear substitution for s.

9.3.3 Summary
In this section, we derived equations of state (constitutive relations for thermodynamic C element) for some simple
thermodynamic systems and showed how these are used to derive state equations. In the remaining sections of this
chapter, we will review ways to incorporate processes and rates. “Thermodynamic” dissipation processes arising from heat
transfer induced by mechanical friction and fluid and electrical resistance were introduced in Chapters 2 and 3. We will
subsequently examine ways for modeling chemical reactions within the same framework. We first discuss in the how open
thermodynamic system effects can be handled using a bond graph approach.

9.4 Modeling Open Thermo-Fluid Systems

Earlier discussions in this book about the effects of variable mass17 did not consider that matter convected into or out of
a system (or system element) in open systems might also introduce a significant thermodynamic change in the system.
Such inherent changes in thermodynamic state can arise in practical cases of fluid flow, and of particular interest in this
section those that may be compressible. It can be helpful in such cases to track the total “energy flux” conveyed due to both
mechanical and thermal interactions in these types of systems. If the flow through a port has a uniform velocity profile over
the flow area, the total flux via multiple (parallel) “modes” [1] is abstracted thus,
 = flow + thermal + power due to elastic, magnetic, etc., (9.97)

17 See Chapter 4, Section 4.7.6 and Chapter 8, Section 8.4.4.

 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
9.4 Modeling Open Thermo-Fluid Systems 577

In the following, we confine the discussion to the first two types. As such, for an assumed uniform flow across a
cross-sectional area, the power of interest in thermo-fluid open systems is composed of terms such as,
[ ]
P 𝑣2
thermofluid = +u+ + gz ⋅ fm . (9.98)
𝜌 2
where fm = dm∕dt = ṁ is mass flow rate. The key point is that there is now a way to distinguish how mechanical (pressure,
velocity, etc.) and thermal (internal energy, density) effects can be quantified as needed. The term in brackets can be taken
to represent an effective “effort” variable, which helps rationalize use of enthalpy, where h = u + P𝑣 and 𝑣 = 1∕𝜌), in this
context. Depending on the application, the power conjugate flow variable is taken as fm or, alternatively, using molar flow
rate, fN = dN∕dt.
This section introduces how open thermo-fluid systems can be modeled, and how to represent the flow of power when
conveyed by a substance that requires thermodynamic state description. A multiport thermodynamic C representing inter-
nal energy storage can be used to model coupled energetic interactions. Examples are introduced to illustrate what is
required when constructing these models, what can be gained in doing so, and to begin building insight into when certain
effects can be neglected or should be accounted for in dynamic systems modeling. In certain cases, there can be a preference
shown for a pseudo-bond graph approach as described in the literature [38], particularly for gas dynamics. Thus, a brief
discussion is given at the end of this section to contrast with pseudo-bond graphs, but mostly adhere to the use of “true”
bond graphs here as in the rest of this book.

9.4.1 One-Dimensional Compressible Flow in Systems

Many of the past chapters have described ways and provided examples for modeling incompressible flow within systems of
interest. Alternative methods are needed for modeling how air and other gases are used in practical devices and systems,
flowing internally in conduits (pipes, tubes, ducts, etc.) and through restrictions (e.g., valves, nozzles). It is necessary to
have ways for deciding whether any compressibility effects need to be considered. Fluid density changes in gas flows, for
example, can be much greater in comparison to those for liquids. Further, the density and viscosity of gases are typically
much less. These property characteristics for gases also result in a relatively lower value for the speed of sound. This can
require considering changes in the way gases flow, particularly through regions where the cross-sectional area changes
significantly, since it is more possible in gases that the flow velocity can approach that of sound. In such cases, there can
also be a significant change in how the mass flow rate depends on pressure. As such, thermodynamic effects more often need
to be properly accounted for in these situations. Some background is provided in the following to help describe how bond
graph models can account for these types of compressible flow effects. Naturally, more extensive and in-depth discussions
should be referred to as needed in textbooks on fluid mechanics and thermodynamics (e.g., [14, 208, 209]), as the intent
here is to provide guidance for system modeling applications.

9.4.1.1 Flowrate and Velocity Descriptions

In hydraulic systems, we often take the flowrate Q as the flow variable, since the fluid remains uncompressed and the
flow regime is clearly incompressible. Further, it is typically assumed in system modeling that the flowrate and velocity are
related by Q = A ⋅ 𝑣, where 𝑣 is taken as approximately constant across the flow cross-section (as in “plug flow”), and 𝑣
designates a mean fluid velocity. Corrections can be made if necessary by a factor 𝛼 that depends on the actual flow profile.
For example, for a common relation such as 𝛼 = ∫ 𝑣3 dA∕(𝑣3 A), 𝛼 = 1 for a flat velocity profile (e.g., turbulent flow) and
𝛼 = 2 for a parabolic flow profile (laminar flow). This correction factor could be used to better estimate the kinetic energy
2
term, say, 𝛼𝜌𝑣 ∕2. In subsequent discussions, the “(⋅)”̄ designation for mean velocity is typically dropped. More attention is
also paid to the velocity, although mass (or molar) flowrate is usually adopted as the “flow” variable in a bond graph context
when modeling compressible flows in systems.

9.4.1.2 Speed of Sound and Mach Number

The speed of sound, or sonic speed, is considered an important characteristic parameter when considering the flow of
compressible fluids. Sound can be modeled in systems as a plane wave propagating through a fluid (gas or liquid). As the
fluid oscillates perpendicular to the wave front, the local density oscillates about the value when the fluid is not disturbed.
The density is considered to increase under infinitesimally small compressions and to decrease under rarefactions. For many
purposes, these processes can be assumed to be adiabatic and reversible, and in such cases the transmission of sound is
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
578 9 Thermodynamic Systems

considered an isentropic process. This leads to a definition of the speed of sound for a gas, c, as [14],
√ |
dP |
c= | (9.99)
d𝜌 ||s=constant
For an ideal gas, P∕𝜌𝛾 = constant, so
dP d𝜌
ln P − 𝛾 ln 𝜌 = ln c ⇒ −𝛾 = 0.
P 𝜌
This can be used to write,
√ |
dP | P
| =𝛾 ,
d𝜌 ||s 𝜌
and thus, since P∕𝜌 = RT,
√
c = 𝛾RT. (9.100)

We can thus estimate the speed of sound from known fluid properties and temperature. For example, for standard
temperature (20 ∘ C) and pressure (1 atm), the speed of sound is 342 m/s (approximately 1120 ft/s).
The Mach number, M , introduced by Ernst Mach in the 1870s, is a non-dimensional number, M = 𝑣∕c, where 𝑣 is the
flow speed and c is the local speed of sound. This number can also be interpreted as the ratio of inertia forces in the fluid
to forces due to compressibility. Uncompressed fluids have c = ∞ and M = 0.

9.4.1.3 Reference State of Fluid

The thermodynamic state of a fluid in a flow field requires specifying two independent (intensive) properties, such as
pressure and temperature, as well as the fluid velocity. When describing compressible flows of ideal gases, it is convenient
to make use of isentropic stagnation properties, which are those defined when the velocity goes to zero. Thus there are
stagnation temperatures, To , pressures, Po , etc. These properties are defined by considering the process by which the fluid
velocity would be brought to zero. For an incompressible flow, for example, its dynamic state is defined by integration of
the momentum equation along a streamline, which is how the Bernoulli equation is defined. Recall, for the steady flow
case, this gives us,
P 𝑣2
+ + gz = constant.
𝜌 2
By decelerating the flow in a “lossless” manner to zero velocity and neglecting gravity, a stagnation pressure is defined as,

Po = P + 𝜌𝑣2 ∕2. (9.101)

For a compressible flow, a frictionless and adiabatic or isentropic deceleration process is used to define local stagnation
properties. For an ideal gas, the following relations can then be derived [14]:
[ ]𝛾∕(𝛾−1)
Po 𝛾 −1 2
= 1+ M , (9.102)
P 2
To 𝛾 −1 2
=1+ M , (9.103)
T 2
[ ]1∕(𝛾−1)
𝜌o 𝛾 −1 2
= 1+ M . (9.104)
𝜌 2
These relations allow us then to model the input and output state conditions (effort and flow sources) on systems using
thermodynamic state as well as the prevailing flow velocity. One way to judge whether compressible effects may be signif-
icant is by comparing the ratios Po ∕P at a given Mach number. For ideal gases, the difference between compressible and
incompressible flow pressure ratio is about 2% for M < 0.3 and 5% for M < 0.45 [14].
The stagnation properties for isentropic flow enable defining critical conditions when the Mach number is unity. For this
case, equations 9.102–9.104 become:
( ) [ ]𝛾∕(𝛾−1)
Po 𝛾 −1
= 1+ , (Po ∕P)c = 1.893, air, (9.105)
P c 2
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
9.4 Modeling Open Thermo-Fluid Systems 579

( )
To 𝛾 −1
=1+ , (To ∕T)c = 1.20, air, (9.106)
T c 2
( ) [ ]1∕(𝛾−1)
𝜌o 𝛾 −1
= 1+ , (𝜌o ∕𝜌)c = 1.577, air. (9.107)
𝜌 c 2
It is useful to remember the values for inverse of the temperature and pressure relations for air, which are ratios that arise
in practical analysis: (P∕Po )c = 0.5283, and (T∕To )c = 0.8333.

9.4.1.4 Enthalpy, Matter Flow Rate, and Power Convection Bond

By substituting enthalpy, h, in equation 9.98, we can now represent power flow in thermofluid systems by (removing the
effect of gravity),

 = h ⋅ fm , (9.108)

where h would be taken as the effort variable. For ideal gases, h is only a function of temperature.18 Brown [7] points out a
need to avoid ambiguity when using enthalpy alone as an effort in this context and recommends specifying both pressure
and enthalpy in defining a convection bond, which might be drawn with two bonds, such as,
P, h
(convection bond)
fm

In this book, it will be assumed that a pressure is implied when a bond’s effort variable is indicated by an “h,” as
h = u + P∕𝜌. As such, the additional dashed line will not be used to maintain a minimal form. Further, the pressure and
matter flow rate will dictate assignment of causality, as in the conventions used for incompressible flow [7]. This bond will
thus be used to represent ideal transmission of a thermodynamic substance, where it is assumed that there is no stored
energy, no heat transfer, no frictional losses, etc.
The distinct nature of the convection bond influences how junction elements are defined and used when constructing
bond graph models of interconnected elements. For example, an ideal 0-junction can be used to model thermo-fluid flow
that splits into different flow paths. Each subsequent bond would have the same pressure, as in incompressible flows, and
in that case also the same temperature and enthalpy. The matter flows, of course, would split as in any 0-junction, as shown
in Figure 9.8(a). The same is not true when two or more distinct flows merge into a single conduit (bond), as an irreversible
mixing would occur. As such, there may be a need to account for mismatched enthalpy, temperatures and/or pressures
by using R elements. If the flows are expected to be changing directions, then it would be expected that the irreversible
effects would be integrated into the junctions. Brown suggests irreversible 0- and 1-junctions designated using “0S”- and
“1S”-junctions [7]. Alternatively, dissipative R elements can be used to explicitly indicate such cases, as in Figure 9.8(b).
Naturally, the R elements would be specified in such a way that efforts on all bonds at the 0-junction are common.19

h1 R R
fm
1
1
1
fm
h h3
0 0 1
fm1 + fm2 + fm3 fm2 fm3
fm
3 1
h2
2
fm R
(a) (b)

Figure 9.8 Possible ways to model the (a) ideal (lossless) splitting and (b) irreversible (lossy) mixing of flows, with the latter using R
elements to indicate dissipation into heat.

18 This requires dh = cp (T)dT; that is, that cp is only a function of T [14].

19 Related discussion on forming models with “non ideal” effort junctions can be found in Katz [210], Breedveld [211], and Willson and Traver
[77].
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
580 9 Thermodynamic Systems

Isentropic flow
constant constant

2
Initial state Kinetic
energy
1 Total
energy

(a) (b)

Figure 9.9 (a) The stagnation properties are the same for all points in an isentropic flow, states 
1,
2,
3 (b) For a flowing fluid, the
kinetic energy per unit mass of the flow is the difference ho − h.

9.4.1.5 Steady Isentropic Flow and Stagnation Enthalpy

A steady, one-dimensional compressible flow can be used to model flow in some sections of systems of interest. In these
situations, there may be changes in the cross-sectional area due to nozzles, valves, and orifices, for example. It is useful
to consider the definition for (specific) stagnation enthalpy, ho = h + 𝑣2 ∕2, which is often made by thinking about steady
isentropic flow in a channel of arbitrary area. In the h − s diagram of Figure 9.9(a), points along the constant s line have the
same stagnation enthalpy. That is, when considering no losses or inputs to a defined control volume, the enthalpy at the
input and output “ports” are equal, and combining continuity and the first law gives,

h1 + 𝑣21 ∕2 = h2 + 𝑣2 ∕2 = h + 𝑣2 ∕2 = constant = ho .

Thus, stagnation enthalpy, ho , can be considered the total energy (per unit mass) of the fluid flowing under these
conditions. Alternatively, stagnation enthalpy is also considered the enthalpy reached when the fluid is decelerated
adiabatically (no heat transfer) to zero velocity. In these cases, ho is assumed constant for all points in an adiabatic flow
field. In Figure 9.9(b), the kinetic energy difference between the two pressures Po and P, or ho − h, represents the change
in the kinetic energy (per unit mass) for an isentropic flow.

9.4.1.6 Flow Through Orifices and Nozzles

When modeling incompressible flow through a “hydraulic” valve or restriction of any type, it is typically sufficient to model
the flow-pressure relation using a nonlinear function, Q = f (ΔP). For example, this functions tends toward a square root as
the restriction takes the form of a sharp edged orifice. Parameterization of this function requires finding some effective flow
coefficient which may also depend on fluid properties, and which may have some temperature dependence. For gases and
any cases of compressible flow, a model for the flow restriction due to valves or nozzles is a bit more complicated. A good
starting point in establishing this model is the case of isentropic flow through a conduit that has area variation, a classical
problem in compressible flow.
While the ability to quantify the flow through a valve for incompressible flow is possible given just the pressure (although
it is possible coefficients may vary slightly with temperature), more information is needed in the compressible flow case.
Aside from making use of mass continuity, the momentum equation, the first law, the second law, and two equations of
state (relating h and 𝜌 to entropy and pressure), the assumption of isentropic flow through a specified “region” of the device
(say with known area variation) enables building relations that can be used in a system modeling context.
To illustrate how to model compressible flow in problems of interest, consider the converging nozzle shown in
Figure 9.10(a). Air at known stagnation conditions is drawn in from a plenum chamber as a valve connected to a vacuum
pump is gradually opened. Assume the system is operating at steady state so only the nozzle effects are considered (no
energy storage). Not unlike the case for incompressible flow, it is expected that the mass flow rate through the nozzle will
be a function of pressures and temperature at  1 and  2 ,
( )
P
ṁ = Φ P2 , T2 , 2 , (9.109)
P1
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
9.4 Modeling Open Thermo-Fluid Systems 581

P1, T1 Valve
v1 ≈ 0 P2
Vacuum
ṁ pump P1 , h1 Isentropic P2 , h2
Throat Se Se
ṁ1 nozzle ṁ2
Plenum Nozzle
Upstream 
1 
2 Downstream

(a) (b)
Figure 9.10 (a) Air drawn through a converging nozzle by a vacuum pump. (b) Word bond graph representation.

and ṁ 1 = ṁ 2 . Here, we see that the ratio, P2 ∕P1 , will play a key role. Assuming isentropic flow, the stagnation properties
in the flow region are constant and the ratio (from 9.102),
[ ]𝛾∕(𝛾−1)
Po 𝛾 −1 2
= 1+ M2 ,
P2 2
can be used to determine the Mach number, M2 , at the throat,20 since Pt ≈ P2 . This is implied through the causality in the
word bond graph of Figure 9.10(b).
Estimating M2 allows determination of whether the flow may become compressible (for M ≲ 0.45). To begin with, it
is possible to estimate the mass flowrate using continuity,
ṁ = 𝜌2 𝑣2 A2 = 𝜌2 ⋅ M2 ⋅ c2 ⋅ A2 ,
√
where we see the need to know T2 since c2 = 𝛾RT2 and 𝜌2 = P2 ∕RT2 (assuming ideal gas, and 𝛾 = 1.4). Fortunately, we
again take advantage of stagnation property invariance for the isentropic flow assumption, so,
To 𝛾 −1 2
=1+ M2 ,
T2 2
taking To = T1 , and this provides T2 . These static relations illustrate a pathway toward modeling the flow through a typical
nozzle/restriction in a way that can be adapted for dynamic system modeling purposes. Clearly the constitutive behavior
is much more complex than for the incompressible flow case. Note that if P2 ∕P1 < 0.528 (for 𝛾 = 1.4), the flow becomes
choked (critical condition where M → 1), and so P2 will no longer affect the mass flow rate. Decreasing P2 further would
not increase the mass flowrate, as it can be shown that ṁ would depend only on the upstream conditions (at plenum in this
case).
Since it is assumed that there is no accumulated mass in the nozzle/orifice region, ṁ = 𝜌2 A2 𝑣2 . Assume the flow is
adiabatic (no heat transfer, negligible friction),
𝑣21 𝑣22
h1 + = h2 +, (9.110)
2 2
so an expression for the flow velocity at the orifice (taking 𝑣1 ≈ 0) is,
√
𝑣22 − 
𝑣21 = 2(h1 − h2 ) ⇒ 𝑣2 = 2(h1 − h2 ). (9.111)
Now, since Δh = cp ΔT, and for ideal gas P𝑣 = RT, cp = 𝛾R∕(𝛾 − 1),
√
2𝛾
𝑣2 = (P 𝑣 − P2 𝑣2 )
𝛾 −1 1 1
where recall 𝑣 = 1∕𝜌 is the specific volume. By taking P𝑣𝛾 = constant (isentropic process), it is possible for the case of the
nozzle/orifice flow in Figure 9.10(a) to show that (e.g., see Chapter 8 of [75]),
√ √
( )1∕𝛾 ( )(𝛾−1)∕𝛾
2𝛾 P1 P2 P2
ṁ = Cd A2 √ 1− , (P2 ∕P1 ) < (P2 ∕P1 )c , (9.112)
R(𝛾 − 1) T1 P1 P1

20 See Fox and McDonald [14] for distinctions between orifice, nozzle, and venturi.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
582 9 Thermodynamic Systems

Figure 9.11 Basic bond graph for modeling the converging nozzle and vacuum pump
system in Figure 9.10, with plenum and pump/valve taken as ideal effort sources.

and, for choked flow,

√
( )(𝛾+1)∕(𝛾−1)
A2 P1 𝛾 2
ṁ = Cd √ , (P2 ∕P1 ) ≥ (P2 ∕P1 )c , (9.113)
T1 R 𝛾 + 1
where P1 is taken as the stagnation pressure, and Cd is a discharge coefficient that varies with orifice geometry and flowrate.
While somewhat complicated in this explicit form, this well-known result clearly conveys how the flow depends on both
pressure and temperature. This rationalizes modeling devices of this type using a R element, as shown in Figure 9.11. It is
important to point out that in the case of reversing flow, the constitutive behavior may need to be adjusted to account for
which side is taken as the “upstream” versus “downstream.”

9.4.1.7 Flow in Conduits Such as Tubes, Pipes, and Ducts

Aside from flow through large area variations, such as in valves and nozzles, systems of interest include flow between
devices via various types of conduits. As shown in Figure 9.12, flow may enter and/or leave from each end with mass flow
rates fm1 and fm2 . We consider the cases where the flow may be assumed either isothermal or adiabatic for the purpose of
analysis [14, 76]. It is also assumed that there is steady flow, so ṁ = 0 within the control volume, implying fm1 = fm2 . Two
different conduit element models can be formulated by then taking the following (steady-flow) relations:
dP dx 𝑣2
− =f + 𝑣d𝑣 (momentum), (9.114)
𝜌 D 2
𝑣21 𝑣22
h1 + = h2 + (energy), (9.115)
2 2

Isothermal Flow in a Constant Area Conduit Assume the environment is such that there is isothermal compressible flow with
friction at the walls of the conduit. Let the friction induce a pressure drop that can be modeled by the Darcy–Weisbach
equation or Fanning equation (when using a hydraulic diameter), as in equation 9.114. It is thus necessary to determine
the friction factor, f , which depends on Reynolds number.
The basic result to discuss here provides mass flow rate as a function of pressure difference for a known temperature (see
[76], eq. 9.38, p. 269). For steady isothermal flow in a conduit of uniform diameter, the mass flow rate is given by,
√
[ ( )]−1
A L P1
fm = √ f + 2 ln (P12 − P22 )
RT D P2
√
√[ ( )]−1 [ ( )2 ]
AP1 √ √ L P1 P2
= √ f + 2 ln 1− , (9.116)
RT D P2 P 1

Q̇

P, T

P1, T1 P2, T2
1 2

Figure 9.12 ̇
Flow of air in a uniform area conduit, such as a tube, pipe, or duct, which may have both friction and heat transfer, Q.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
9.4 Modeling Open Thermo-Fluid Systems 583

where A is the constant conduit area and T is the assumed known temperature. There is a choking limit on this relation
that needs to be considered. This limit can be used to determine the maximum length of pipe for which isothermal flow
can hold without choking (see [14]).

Adiabatic Flow in Constant Area Conduit Deriving a relation for steady-state adiabatic mass flow rate requires considering
equation 9.115 and ideal gas conditions in the conduit, from which is found,
2𝛾
𝑣22 − 𝑣21 = (P 𝑣 − P2 𝑣2 ).
𝛾 −1 1 1
where 𝑣 = 1∕𝜌 is the specific volume. This allows defining an invariant quantity (as in [76]),
( ) ( )
fm1 2 2𝛾 fm2 2 2𝛾
𝜒= + P1 𝑣1 = + P 𝑣, (9.117)
𝜌1 A 𝛾 −1 𝜌2 A 𝛾 −1 2 2
so the value of 𝜒 can be determined given the conditions at an inlet or outlet.
Now, in equation 9.114, replace velocity, 𝑣, by fm ∕𝜌A,
( )2 ( )2
f fm f d𝜌
0 = 𝜌 ⋅ dP + dx − m .
2D A A 𝜌
This equation is integrated along the length of the conduit of length L, giving,
( )2 ( )2 ( )
2
fL fm f 𝜌1
0= 𝜌 ⋅ dP + + m ln . (9.118)
∫1 2D A A 𝜌2
2
An expression for ∫1 𝜌 ⋅ dP can be found using relation 9.117 to define,
[ ( )2 ]
𝛾 −1 1 fm
P= 𝜒𝜌 − ,
2𝛾 𝜌 A

which provides and expression for dP; namely,

[ ( )2 ]
𝛾 −1 1 fm
dP = 𝜒𝜌 + 2 d𝜌,
2𝛾 𝜌 A
2
This term can be used to determine ∫1 𝜌 ⋅ dP. With 𝜒 defined at  1 , insert in equation 9.118, and solve for the mass
flow rate,
√
√[ ( ) ( ) ]−1 [ ]
√ √
√ 𝛾 − 1 𝜌22 − 𝜌21 𝛾 + 1 𝜌1 f L 𝜌21 − 𝜌22
fm = A P1 ⋅ √ + ln +
4𝛾 𝜌21 2𝛾 𝜌2 2D 2𝜌1
√
√[ (( ) ) ( ) ]−1 [ ( )2 ]
√
AP1 √ 𝛾 −1 P2 2 𝛾 +1 P1 f L P2
= √ ⋅ √ −1 + ln + 1− . (9.119)
2RT1 4𝛾 P1 2𝛾 P2 2D P1

Just like equation 9.116, this relation is limited for M < 1 (non-choked flow). Recall that choked flow will occur when
(P2 ∕P1 ) ≥ (P2 ∕P1 )c , a case for which this ratio can then be replaced by equation 9.105. In these cases, the mass flow rate
becomes independent of the downstream pressure, P2 .
Equations 9.116 and 9.119 assume steady flow but can provide approximate component models for flow in a conduit.
In a bond graph, these component models can be represented by an R element like the converging nozzle in Figure 9.11.
As such, an explicit thermal port can provide a connection to thermal effects modeling, as will be shown in Example 9.5.

9.4.1.8 Summary
Flow of compressible fluids requires consideration of thermodynamic variables and properties. An example was presented
of isentropic flow through restrictions and nozzles, as commonly introduced in introductory treatments of compressible
flow. This enabled review of concepts and terminology, but also provided a basis for modeling components that arise in
many practical types of pneumatic devices and systems. The steady-flow conduit models, based on basic models presented
in introductory compressible flow treatments (such as in [14, 76]), can be used to represent components in devices and
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
584 9 Thermodynamic Systems

machines. While inertial effects are not typically as significant for gases as they may be for liquids (though they can be
in some flow systems), their effect can be considered using the same form in both. For the most part, storage of potential
(internal) energy forms the most significant effect to consider and it is for this reason that the general multiport thermo-
dynamic C becomes essential. This element also forms the basis for accounting for open system effects as discussed in the
following. Lastly, it should be made clear that open system effects arise in many types of systems. Further, gases and liquids
can both undergo compressibility, and in many cases gases can be modeled as uncompressed fluids.

9.4.2 Modeling Storage of Energy in Thermo-Fluid Systems

The multiport thermodynamic C is reviewed to reinforce its use in modeling systems where there can also be a change
in the stored matter. For completeness, an approach using a pseudo-bond graph “accumulator” that has been commonly
adopted in the literature is also reviewed in this section.

9.4.2.1 The Multiport Thermodynamic C

We adopt the general thermodynamic C introduced earlier in this chapter for tracking internal energy storage. As shown
in Figure 9.13, a (modulated) transformer is used to represent how mass fm or molar flow rate, fN = dN∕dt, contributes to a
convected entropy flow rate. In the latter case, for example, fs = s ⋅ fN , where s = S∕N. This model element makes it possible
to represent energy that is convected (i.e., transferred) as matter enters and/or leaves part of a system that stores energy and
matter. In some cases it may be preferred (or customary) to use mass flow rate, m, ̇ rather than molar flow rate. In this form,
integral causality will dictate three state equations,
Ṡ = f1 + fs ,
V̇ = Q,
Ṅ = fN , or ṁ = fm .

With the addition of the “matter” port, it is important to distinguish in the modeling that the volume state is often here
dictated by mechanical expansion/compression of the contained thermodynamic substance. This is in contrast to previous
fluid (hydraulic) system models where volume state and its rate of change was also tracking the amount of substance, given
assumed incompressible behavior. This role is now adopted by mass or molar flow rates. Further, we now have the ability
to track dynamic changes in effective density, 𝜌 = m∕V = 1∕𝑣.
The chemical potential, 𝜇, and molar flow rate, variables have not been extensively used up to this point. However, in
some cases, such as when dealing with gas systems, it may be preferred when tracking different “species” using moles, N,
rather than mass, m. By rearranging the specific internal energy equation 9.55, we get,

u = 𝜇 + Ts − P𝑣, (9.120)

Recalling that specific enthalpy is defined as h = u + P𝑣 and substituting equation 9.120 for u, specific enthalpy can be
expressed as,

h = 𝜇 + T ⋅ s. (9.121)

Ṡ = f1 + fs –˙
−P V

T T µ h=µ+T ·s
f1
0 C 1
fN
Ṡ Ṅ
“net heat
transfer”
s
T ·· T ·s
T
fs = s · fN fN
“convected entropy”

Figure 9.13 Open thermodynamic C, with a bond included in the thermal domain to show how the net entropy flowrate would be
f1 + fs .
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
9.4 Modeling Open Thermo-Fluid Systems 585

Equation 9.121 illustrates how the enthalpy from a convection bond forms the effort summation in Figure 9.13. In that
context, a power product can be formed with either molar flow rate, fN , or mass flow rate, fm . The basic definitions discussed
here are sufficient to begin representing some key elements when dealing with open thermodynamic system effects.

Example 9.5 Flow in a tube or conduit

Air flows in a rigid tube subjected to known pressure and temperature conditions at ends  1 and 2 , as shown in Figure 9.12.
The tube has constant cross-sectional area, A. Consider friction at the fluid–wall interface, and that there may be a net
imposed heat transfer, Q̇ = T ⋅ fsh . Formulate a lumped-parameter bond graph model that can account for stored energy
due to mechanical, thermal, and mass interactions. The schematic in Figure 9.14 represents such a model, where half of
the tube friction is partitioned into equal elements on each side of a thermodynamic C element. Identify states and derive
state equations. Discuss the following cases: (a) adiabatic flow with friction, and (b) frictionless flow with external heat
transfer into the stored air.

R R
C
U(S, –V, m)

(a) (b)

Figure 9.14 Representing flow of an ideal gas in a rigid conduit with friction and heat transfer as in (a) with a lumped-parameter
model in (b) that reticulates the flow with friction into two equivalent R elements on each side of a thermodynamic C that represents
stored energy and mass with heat transfer.

Solution
We consider the lumped model in Figure 9.14(b), with inertial effects assumed negligible for this example. This
“T-junction”-type lumped model has effective friction in the conduit partitioned into two identical R elements on each
side of a thermodynamic C. The bond graph for this flow system can be constructed by combining two flow resistance
models, as in Figure 9.11, with the open thermodynamic C model of Figure 9.13. This results in a bond graph for the
tube model element under the stated conditions as shown in Figure 9.15. The imposed inlet conditions are given as

External
heat transfer Sf
Q̇ = T fsh
Heat transfer Heat transfer
T fsh
from friction fsf 1 fsf 2 from friction

R s1 : T 0 T :s R
fs1 fs2

Δh1 fm1 T s1 T Ṡ Ts Δh2 fm2

h1 µ µ + Ts h2
Se 1 1 0 1 1 Se
fm1 fm1 fm2 fm2 fm2
µ ṁ

C
“Rigid −P
Sf
conduit” V˙ = 0
–
Figure 9.15 Bond graph for compressible flow in a uniform conduit with friction. All heat generated by the R elements (friction) is
assumed to flow into the tube control volume. Also note the different modulations by entropy values on the convection via
transformers, where s1 is the specific entropy at the inlet conditions and s = S∕m. If flow reverses direction, these modulation variables
would change to reflect flow from  2.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
586 9 Thermodynamic Systems

effort sources for the flow ports, and an imposed heat transfer is also indicated. For the rigid conduit, any mechanical
(volumetric) expansion is set to zero with a flow source. Unlike the conduit models for the mass flow rate given in
equations 9.116 and 9.119, however, it is desired to account for changes in stored mass in the tube.
Note that there is heat transfer from the friction R elements into the lumped mass in the tube control volume (at
0-junction), and there is also thermal energy convected in and out of the control volume via entropy flow rates from the
inlet and outlet mass flows.
From the assigned causality, there are three independent states: x = [V, S, m]T , with state equations:
V̇ = 0, (9.122)
Ṡ = fs1 − fs2 + fsf + fsh , (9.123)

ṁ = fm1 − fm2 , (9.124)

where
fs1 = s1 ⋅ fm1 ,

fs2 = (S∕m) ⋅ fm2 ,

fsf = fsf 1 + fsf 2 (total heat from friction),
̇
fsh = Q∕T(S, V, m),
and s1 is the specific entropy of the air at  ̇ 1 and ṁ 2 , which
1 . It is necessary to find expressions for the mass flow rates, m
are causally dependent on the frictional R elements, that is,
fm1 = Φ1 (P1 − Pc𝑣 ), (9.125)
fm2 = Φ2 (Pc𝑣 − P2 ), (9.126)
these nonlinear functions can be defined similar to pipe friction using the Darcy–Wiesbach equation,
ΔP f 𝑣2
= L ,
𝜌 D 2
where here L would be taken as half the tube length (for partitioning the friction in two). Also, Δh1 = h1 − (𝜇 + Ts1 ) and
Δh2 = (𝜇 + Ts) − h2 , with 𝜇 given by equation 9.57. It can then be shown that,
√
2𝜌D
fm = A ⋅ ⋅ |ΔP|. (9.127)
f (L∕2)
To test this model, consider air entering at 1 and being drawn in by conditions at  2 . Let the pressure P2 begin at P1
but is then dropped by about 0.5%, with a tube diameter, Dtube = 7.16 mm, a length Ltube = 10Dtube . The temperatures are
known at T1 = 296 K and T2 = 287 K, and initially P1 = P2 = 101.325 kPa. After 0.1 seconds, P2 is linearly decreased until
it reaches 0.995P1 at 2 seconds, then the pressure is held.
Simulation results are shown in Figure 9.16. The plot in (a) shows how the pressure and the temperature change in
proportion as P2 is dropped. In part (b), mass flow rates at 1 and 
2 are shown to converge quickly to give m ̇ → 0. There is
a very short time period where fm1 ≠ fm2 for this case. For comparison, the mass flow rates using the steady-flow relations for
isothermal and adiabatic flow in a tube are also plotted. These are computed using equations 9.116 and 9.119, respectively.
Finally, the plot in (c) shows how the total mass contained with the tube changes mass and reaches a steady-state value
fairly quickly, settling to about an 18% decrease. It should be noted that the relatively small drop in P2 of 0.5% is chosen
to keep the flow from choking. For such conditions, the mass flow rate relations would need to adopt a functional change
based on that condition.

9.4.2.2 Pseudo-Bond Graph Approach for Gas Dynamics

An alternative approach for modeling thermo-fluid gas dynamic systems makes use of pseudo-bond graph formulations.
A pseudo-bond relaxes the definition and selection of bond graph variables to a set that can be more familiar and thus
convenient in some applications. As such, the bonds do not represent power flow using a defined set of conjugate variables
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
9.4 Modeling Open Thermo-Fluid Systems 587

4 14 0
Pcv and Tcv
12
–10
3
10 Adiabatic
–20

% Mass change
2 8
% change

gm/sec
6 –30
2
1 4 Isothermal
P1 –40
2
0 –50
0
P2
–1 –2 –60
0 1 2 3 0 1 2 3 0 1 2 3
Time, Sec Time, Sec Time, Sec
(a) (b) (c)

Figure 9.16 Results from simulation of air flow in a rigid tube: (a) plots of pressure and temperature within the tube at the inlets  1
and  ̇ also results from
2 ; (b) plots of the mass flow rates at the inlets, fm1 and fm2 , and the rate of change of mass within the tube, m;
using steady-state mass flow rates for adiabatic and isothermal flow in the tube; (c) the dynamic change in the stored tube mass
showing a relatively fast transient settling to a steady-state value about 18% of original mass.

Net heat
transfer

Mass flow enthalpy

inlet area
Uniform thermodynamic variables

Mass flow enthalpy States: or and

inlet area
Net effective mechanical
F expansion

Figure 9.17 Schematic of a generic thermodynamic control volume with in/out flows, mechanical interactions, and heat transfer.

as in true bond graphs. Consequently, there is a mismatch with other systems that may use true bond graphs. Many users,
however, find the approach useful, especially when only the thermo-fluid aspects are to be considered. The approach
is rooted in some classical approaches that utilize a lumped-parameter approach (Eulerian viewpoint) with the energy
(rate-of-change) equation to describe the system dynamics. To contrast this approach with the use of a true bond graph
approach that uses the multiport thermodynamic C, consider the system in Figure 9.17.
For this system, assume that the contained gas has negligible kinetic energy, and the thermodynamic variables are uni-
form as indicated. The potential (internal) energy is here a function of temperature, volume, and the contained mass,
 = (T, V, m). The rate of change of the stored energy can be determined by,
( )
∑ 𝑣2i ∑
̇ = hi + ṁ i + T ⋅ Q̇ i + (−P) ⋅ V̇
i
2 i ⏟⏞⏟⏞⏟
⏟⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏟⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏟ ⏟⏞⏟⏞⏟ Mechanical
power term
Net convected Net heat
energy transfer

Assume kinetic energy in the input and output flows is negligible. Thus, only enthalpy is considered in the convection
terms, 𝜙1 = h1 ṁ 1 , 𝜙2 = h2 ṁ 2 , etc.
Consider a system as in Figure 9.17 with one inlet and one outlet flow. A pseudo-bond graph in a form proposed by
Karnopp et al. [38] is shown in Figure 9.18. This pseudo-bond graph C element indicates three state equations (with full
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
588 9 Thermodynamic Systems

0 Figure 9.18 Example pseudo-bond graph for thermodynamic accumulator element

φ1 φ2 with integral causality form. This model requires power modulation of a “flow source”
with m . This deviates from the strict rules discussed in Chapter 4 on modulation
0 structure, but it is an essential part of this accumulator model as formed by Karnopp
ṁ1 ṁ2 et al. [38].
φ1 − φ2 P ṁ

T T
0
˙
C
Q̇
m

Sf –˙
−P V

m V˙
= −P –
◦
T :A

F v

integral causality),
̇
̇ = 𝜙1 − 𝜙2 + m + Q, (9.128)
ṁ = ṁ 1 − ṁ 2 , (9.129)

V̇ = A𝑣. (9.130)
For an ideal gas,
U = mc𝑣 T,
PV = mRT.
Because the choice of bond variables deviates from those used in a “true” bond graph approach, Karnopp et al. [38] refer
to this pseudo-bond graph model as a thermodynamic accumulator. This can be a useful approach when modeling systems
with compressible fluid flowing between and within system components.
The accumulator model described in Figure 9.18 uses an unusual modulation of a “power flow source” to achieve in
the bond graph what is strictly achieved in the energy equation 9.128. This equation can be considered foundational
and is the basis for many classical approaches to modeling systems of this type. For this reason, the accumulator model
finds favor among some system modelers. Another reason is that the pseudo-bond variables adopted, namely, T − Q̇ or
̇ may be more familiar than those required for a true bond graph approach. It should be understood that both meth-
P − m,
ods, executed properly, will produce the same results. While the choice can come down to “user preference,” it will also
depend on whether a system under study is solely in the thermal domain or involves interaction with other energy domains.
A true bond graph approach is more effective in the latter case. Nevertheless, familiarity with both representations makes
it possible to understand and interpret work reported in the literature.

9.4.3 Models of Open Thermo-Fluid Systems

There are a large variety of ways that open thermo-fluid system effects can arise. Only a few ways have been described here
as needed for modeling devices and components in practical engineering systems. Chapters 4 and 8 have discussed ways
to handle systems with variable mass. In the present discussion, thermodynamic considerations have become necessary
as fluid properties, particularly for gases or liquid–gas mixtures, are influenced by temperature and pressure. Often, it is
necessary to incorporate open system effects and explore computationally whether it is significant. As such, the basic model
elements described in Sections 9.4.1 and 9.4.2 should provide a systematic way for a non expert to integrate these effects
into system-level models, especially when there may be interactions with other energy domains.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
9.4 Modeling Open Thermo-Fluid Systems 589

The choice of variables is a key step, and may be motivated by a particular application, the need to track processes for
a particular audience, user preference, as well as other requirements. The relation between molar and mass flow rate is
simply,
fm = fN ⋅ M𝑤 , (9.131)
where M𝑤 is the molecular weight in kg/mol. Another useful relation to keep in mind relates enthalpy, pressure, and
chemical potential:
h = u + P𝑣 = 𝜇 + Ts, (9.132)
which is used in forming the bond graph structure of the open thermodynamic C in Figure 9.13. This relation was used
in Example 9.5. The following examples further illustrate how these concepts can be used in modeling basic devices and
systems that involve open thermo-fluid processes.

Example 9.6 Ideal gas flowing in short pipe between tanks

An ideal gas flows from a line source with known conditions through a short pipe and into a rigid tank, as shown in
Figure 9.19. Develop a model that can be used to determine the states of the tank (ST , V T , NT ), the dynamic temperature
and pressure in the tank (TT , PT ), as well as the amount of gas that flows from the source, fN . Assume energy losses are
primarily due to mixing, which results in heating in the tank.

Figure 9.19 Flow of gas from a line source to a rigid tank through a short
pipe. Line Rigid tank

Mixing

Streamlined, isentropic flow

Solution
A bond graph is shown in Figure 9.20, indicating two states: entropy in the tank, ST , and moles of gas in the tank, NT .
The line (source) conditions are known: TL and 𝜇L = uL − PL 𝑣L − TL sL = hL + TL sL . The equations of state for the C, which
represents internal energy storage in the tank, are (from equations 9.53–9.55):
( 𝑣 )𝛾−1 ( )
s − so
TT = TTo o exp , (9.133)
𝑣 c𝑣
NR
P= ⋅ TT , (9.134)
𝑣T
𝜇 = cp TT − TT sT , (9.135)
where initial states of the tank are known and indicated by the “o” subscript.
From the bond graph, state equations are indicated as,
Ṡ T = fsm + fsT , (9.136)
Ṅ T = fN , (9.137)

Figure 9.20 Bond graph modeling flow and convection between connected line source and rigid tank through a short pipe.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
590 9 Thermodynamic Systems

so we need to find these entropy and molar flow rates. Since fsT = sT fN , sT = ST ∕STo , we see that it comes down to finding
the two flows on the two-port mixing element, Rm , that is:
fsm = fsm (𝜅, TT ), (9.138)
fN = fN (𝜅, TT ). (9.139)
In these relations, we assume a quasi-steady model for mixing so,
𝜅 = hL − ht ,
where hL is determined from known source conditions and ht = 𝜇T + TT sT , a function of states. Assume that the mixing
losses absorb all the (molar) kinetic energy of the flow entering the tank, which is Tp = p2 ∕2M𝑤 = M𝑤 𝑣2t ∕2, where M𝑤 is the
gas molecular mass and 𝑣t is the velocity of the flow (averaged across the pipe cross-section). Thus, 𝜅 = hL − ht = M𝑤 𝑣2t ∕2,
and thus,
√
𝑣t = 2(hL − ht )2 ∕M𝑤 .
We take,
( )𝛾−1∕𝛾 [ ]
hL = cp TL = cp To Po ∕PL exp (sL − so )∕cp ,
and assuming st = sL ,
( )𝛾−1∕𝛾 [ ]
ht = cp To Po ∕PL exp (sL − so )∕cp .
We can now write the molar flow rate as,
√
A
fN = (A∕𝑣t )𝑣t = 2(hL − ht )2 ∕M𝑤 ,
𝑣t
or,
√ ( )
√ h − ht
A A 2
fN = 2(hL − hT + hT − ht )2 ∕M𝑤 = (hL − hT ) 1 + T ,
𝑣t 𝑣t M𝑤 hL − hT
( )𝛾−1∕𝛾 [ ]
where hT = cp To Po ∕PL exp (sT − so )∕cp .
For the two-port Rm , 𝜅fN = TT fsm , thus,
𝜅fN
fsm = ,
TT
allowing us to determine the state equations as functions of state.
What is left then is to determined conditions under which assumptions would change and the flow could choke [14].

In the next example, we introduce the effect of mechanical expansion with a piston/cylinder element. The “open” effect
arises when the cylinder loses mass via an orifice. Heat transfer is also included via an electrical heater and conduction/
convection through cylinder and piston walls.

Example 9.7 Collapsing piston chamber with heating element and orifice
The initially closed cylinder in Figure 9.21 contains air with a movable piston. An electrical resistor can heat the contained
air. Develop a model that can account for the air being compressed by the piston under gravity, heated by the electrical
resistor, and escaping via an orifice that may open. The cylinder walls and piston are not insulated.
Solution
A bond graph for this system is shown in Figure 9.22. Here we have Tc = Tc (Sc , V c , Nc ) and Pc = Pc (Sc , V c , Nc ), and the states
[ ]T
are x = p, V c , Sc , Nc , where N is the moles of air in the cylinder. The model could also be changed to track the mass of
air, mc = Nc ⋅ M𝑤 (M𝑤 the molar mass of air).
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
9.4 Modeling Open Thermo-Fluid Systems 591

Air at STP

Leakage flow

Vs (t)
Figure 9.21 A pneumatic cylinder with an electric heater and piston, losing mass when the orifice is opened.

Ta
Se Rh Effective heat transfer Se R:b

fsh mg

b
F
Rs A
Vs ·· Tc −Pc ·· ṗm
Se
i
R 0 C T 1 vm I:m
fse Ṡc –˙ c
V
Electric heater fsc μc Ṅc

sc Tc s
T 1 T :A
fN c

hc = μ + T s −fN o Pa

0 Se

hc fN o Orifice flow
ht κo vt
–
Se 1 Ro
fN o

Figure 9.22 Collapsing chamber using open thermodynamic “C” model.

State equations from the bond graph are:

ṗ m = A(Pc − Pa ) − mg − Fb ,
V̇ = A(pm ∕m),
Ṡ = fse + fsc − fsh ,
Ṅ = −fNo = −M𝑤 ⋅ ṁ o ,
where we can assume throat conditions will be defined by atmospheric conditions, ṁ o = mass flow rate via the orifice (as
in equations 9.112 and 9.113), and M𝑤 = molar mass of air (kg/mol). Note that the choice to track moles or mass may be
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
592 9 Thermodynamic Systems

dependent on the problem and/or personal preference. The equations of state account for changes in N in the T and P from
equations 9.53 to 9.55,
( ) ( )
V o Nc 𝛾−1 Sc ∕Nc − Sco ∕Nco
Tc = To ⋅ exp ,
No V c c𝑣
( )𝛾 ( )
V co Nc Sc ∕Nc − So ∕Nco
−Pc = −Pco ⋅ exp ,
Nco V c c𝑣
( ) ( )
( ) V co Nc 𝛾−1 Sc ∕Nc − Sco ∕Nco
𝜇c = uco − c𝑣 Tco + c𝑣 + R − Sc ∕Nc Tco ⋅ exp ,
Nco V c c𝑣
where V co and Nco are the initial volume and moles of air in the cylinder, Tco is the initial temperature, and uco = Uco ∕Nco
is the initial specific internal energy.
The electrical heater provides entropy flux via the relation,
Vs (t) ⋅ i Vs2 (t)
fsh = = ,
Tc Rs (Tc ) ⋅ Tc
where R(Tc ) in case the resistance has some dependence on temperature. Further formulation and detailed development
are required to enable studying the effect of combined heating, heat transfer, piston collapse, and loss of mass via the orifice
flow. This is demonstrated in Solved Problem A-9-3.

9.5 Multicomponent Systems

Multicomponent thermodynamic systems are prevalent in such important engineering phenomenon as chemical reactions
and adsorption. In this regard, the chemical and adsorption properties of ideal gases are of particular importance since many
industrial processes are carried out in the gaseous state in near ideal gas conditions. Furthermore, even under nonideal
conditions, ideal gas theory provides a framework for more realistic gas models. The essence of ideal gas behavior is that
molecules just collide and do not attract or repel each other. This implies that internal energy can be modeled as additive
over the n components,
∑
n
USVNi = Ni u i ,
i

where ui is the molar-specific internal energy for a single component. Internal energy for an for an ideal gas component
can be expressed in terms of specific heat and temperature as (equation 9.47),
T
ui = ui (T) = uoi + c𝑣i dT.
∫To

Similarly, entropy of a multicomponent ideal gas is,

∑
n
S= N i si ,
i
T ( )
c𝑣i V
si = soi + dT + R ln ,
∫To T Ni 𝑣o
where we have used equation 9.46. A parametric form of the fundamental internal energy relation for a mixture of ideal
gases can then be expressed as,
n [ ]
∑ T
USVNi = uoi + c dT ⋅ Ni , (9.140)
i=1
∫To 𝑣i
[ ( )]
∑
n T
c𝑣i V
S= soi + dT + R ln ⋅ Ni , (9.141)
i=1
∫To T Ni 𝑣o

with temperature T playing the role of a parameter to be eliminated in order to explicitly obtain internal energy.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
9.5 Multicomponent Systems 593

For constant specific heats, temperature can be explicitly eliminated to yield,

USVNi (S, V, N1 , … , Nn ) = Nu(s, 𝑣, x1 , ..., xn )
⎧ ( )R∕c𝑣
⎡ 𝑣 ∏ n ( ) ⎤⎫
⎪ ⎢
s − so ⎪
− 1⎥⎬ ,
o xi
= N ⎨uo + c𝑣 To xi exp (9.142)
⎪ ⎢ 𝑣 c𝑣 ⎥
⎩ ⎣ i=1 ⎦⎪
⎭
where xi = Ni ∕N is the mole fraction and (⋅) represents the molar averaged property, that is,
∑
n
(⋅) = (⋅) xi . (9.143)
i=1
This leads directly to constitutive relations of temperature and pressure from,
𝜕USVNi 𝜕u
T= = ,
𝜕S 𝜕s
𝜕USVNi 𝜕u
−P = = ,
𝜕V 𝜕𝑣
Chemical potential can be determined by,
𝜕USVNi
𝜇i = ,
𝜕Ni
and is often expressed in terms of temperature and pressure via Gibbs free energy,
𝜕G
𝜇i = ,
𝜕Ni
where G(T, P, N1 , ..., Nn ) = U − TS + PV. Using 9.140, 9.141, and the mechanical equation of state P = NRT∕V yields,
[ ]
𝜇i = 𝜇iO + RT ln(P∕Po ) + ln(xi ) . (9.144)
The quantity 𝜇iO (T) is a reference potential function of temperature only (reference pressure Po is often taken as 1 atm).
For constant specific heats, it is,
𝜇iO (T) = 𝜇oi + (cpi − soi )(T − To ) − cpi T ln(T∕To ),
where cpi = c𝑣i + R and 𝜇oi = uoi − To soi + RTo . Equations of state for a multicomponent ideal gas with constant specific
heat can then be expressed as summarized in Table 9.6.
The chemical potential leads directly to the concept of a chemical equilibrium constant. Chemical reactions with n chem-
ical components can be expressed,
∑
n
0⇄ 𝜈i Mi ,
i=1
the 𝜈i are stoichiometric coefficients (positive for products; negative for reactants) and the Mi are component elements.
At chemical equilibrium, we have,21
∑
n
𝜈i 𝜇i = 0,
i=1

Table 9.6 Equations of state of multicomponent ideal gas with constant specific heats.

[ ]R∕c𝑣 ( )
𝜕USVNi 𝜕u 𝑣o ∏
n
xi s − so
T= = = To xi exp (9.145)
𝜕S 𝜕s 𝑣 i=1 c𝑣

𝜕USVNi 𝜕u RT
−P = = =− (9.146)
𝜕V 𝜕𝑣 𝑣
𝜕USVNi [ ]
𝜇i = = 𝜇iO (T) + RT ln(P∕Po ) + ln(xi ) (9.147)
𝜕Ni

21 We will discuss this in more detail when we take up chemical reactions in Section 9.7
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
594 9 Thermodynamic Systems

which results in,

[ ( )]
∑n
P
𝜈i 𝜇i (T) + RT ln
O
x = 0,
i=1
Po i
and thus,
( )𝜈i ( )
∏
n
P ∑ n
𝜈i 𝜇iO
x = exp − = KP (T), (9.148)
i=1
Po i i=1
RT
where KP is an equilibrium constant based on pressure and defined by the above expression.

Example 9.8 Chemical equilibrium of CO2 in a vessel

As a review of chemical equilibrium, consider one mole of CO2 is enclosed in a rigid vessel and heated to a temperature of
2200 K at a pressure of 1 MPa. If CO2 disassociates according to the reaction,
1
CO2 ⇄ CO + O2 ,
2
find the equilibrium mole fractions of the system if the equilibrium constant at 2000 K is given as Keq (2200 K) = 0.0594
with Po = 1 atm.
Solution
The stoichiometric coefficients are:
𝜈CO ∶ −1
2
𝜈CO ∶ 1
1
𝜈O ∶
2 2
and thus the equilibrium constant is,
x ( )1∕2
1 1∕2
xCO
O P
Keq = 1 2 = 0.0594. (9.149)
x Po
CO2
If 𝜀 is degree (amount) of disassociation, we also find,
NCO = 1 − 𝜀
2
NCO = 𝜀
1
NO = 𝜀.
2 2
The total number of moles is,
1
N = NCO + NCO + NO = 1 + 𝜀,
2 2 2
and mole fractions are,
1−𝜀
xCO = ,
2 1 + 12 𝜀
𝜀
xCO = ,
1 + 12 𝜀
1
2
𝜀
xO = .
2 1 + 12 𝜀
Using equation 9.149 results in,
(𝜀∕2)1∕2 ( )1∕2
1 MPa
1 1∕2 0.101 MPa
= 0.0594,
(1 − 𝜀)(1 + 𝜀) 2
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
9.5 Multicomponent Systems 595

which can be solved iteratively to yield,

𝜀 = 0.0691.

The equilibrium mole fractions are then,

xCO = 0.9,
2
xCO = 0.067,
xO = 0.033.
2

Another important multicomponent system is adsorption of gaseous components on solids. If we consider Na moles of
solid adsorbent and Ns moles of sorbate as the adsorbed phase, internal energy of the adsorbed phase can be expressed as,

U = USVNa Ns (S, V, Na , Ns ),

where V is the adsorbent-sorbate volume. This relation, by itself, has no special significance to adsorption pro-
cesses. However, following Gibbs, if we assume a thermodynamically inert adsorbent with adsorption occurring on a
two-dimensional surface, it is useful to make the basis of internal energy, entropy, and volume be the adsorbent in the
absence of sorbate [212]. Then, neglecting volume change due to adsorption, internal energy can be expressed in terms of
just three extensive parameters as,

U = USNa Ns (S, Na , Ns ).

Finally, since surface area, A, is directly proportional to the amount of adsorbent Na , we have a classical form of internal
energy as,

U = USANs (S, A, Ns ).

Extensive properties can then be defined as,

𝜕USANs
T= ,
𝜕S
𝜕USANs
−𝜋 = ,
𝜕A
𝜕USANs
𝜇s = ,
𝜕Ns
where 𝜋 is called the spreading pressure and its physical meaning corresponds to difference in surface energy (surface
tension) between a clean sorbate-free surface and one with sorbate.
In a completely analogous fashion to simple thermodynamic gases, we can obtain equations of state for spreading pres-
sure and specific internal energy as functions of temperature and molar-specific surface area (analogous to molar specific
volume). For low amounts of adsorption (high a and T) we have an ideal gas like relation as,
RT
𝜋= ,
a
du
= ca ,
dT
where a = A∕Ns , u = USANs ∕N, and ca is a molar-specific heat at constant area (this quantity is often taken as equal to c𝑣 ,
the gas-phase specific heat). Using a Gibbs relation

du = Tds − 𝜋da,

and assuming constant specific heat, we can then obtain a fundamental internal energy equation,
[ ( ( )R∕c ( ) )]
a a s − so
USANs = Ns uo + ca To o exp −1 , (9.150)
a c𝑣
with constitutive relations as summarized in Table 9.7.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
596 9 Thermodynamic Systems

Table 9.7 Equations of state of ideal gaseous adsorption on a solid.

( a )R∕ca ( )
𝜕USAN 𝜕u s − so
T= = = To o exp (9.151)
𝜕S 𝜕s a ca
( a )(ca +R)∕ca ( )
𝜕USAN 𝜕u s − so
−𝜋 = = = −𝜋o o exp (9.152)
𝜕a 𝜕a a ca
( a )R∕ca ( )
𝜕USAN s − so
𝜇= = u − Ts + 𝜋a = uo − ca To + (ca + R − s)To o exp (9.153)
𝜕Ns a ca

Example 9.9 Chamber of powdered adsorbent

A chamber of powdered adsorbent is filled with a gas at a set pressure P and temperature T. Obtain the number of moles
of gas adsorbed as a function of P and T if ideal gas type relations can be used for both the gas and adsorbed phases.

(a) Adsorbent-gas chamber (b) Bond graph description

Figure 9.23 Adsorbent in equilibrium with sorbate.

Solution
The bond graph of Figure 9.23(b) gives a clear picture of what relations need to be obtained. By applying pressure and
temperature on the gas, the chemical potential 𝜇g is known (two efforts determine the third for a three-port thermodynamic
C element). At equilibrium, sorbate chemical potential and temperature equals gas chemical potential and temperature.
Adsorbent efforts 𝜇s and T then determine spreading pressure 𝜋.
As discussed previously, chemical potential for an ideal gas can be expressed in terms of pressure and temperature as,
𝜇g = 𝜇gO (T) + RT ln(P∕Po ).
In an analogous manner sorbate, chemical potential can be expressed as,
𝜇s = 𝜇s∗ (T) − RT ln(a∕ao ),
where 𝜇s∗ is solely a function of temperature. Equating these two potentials yields,
𝜇gO (T) + RT ln(P∕Po ) = 𝜇s∗ (T) − RT ln(a∕ao )
where 𝜇s∗ (T) = 𝜇so + (ca − so + R)(T − To ) − Tca ln(T∕To ) and 𝜇so = uo − Tso + RT. This expression can then be solved for
Ns (a = A∕Ns ) as,
( O )
Ns 1 𝜇g − 𝜇g∗
= exp P,
A ao Po RT
or,
q = KP,
( )
with q = Ns ∕A and K = (1∕ao Po ) exp (𝜇gO − 𝜇g∗ )∕(RT) . Thus, a linear relationship between pressure and adsorbed phase
arises from an ideal gas type equation of state for spreading pressure. This relationship is called Henry’s law and K is termed
Henry’s constant.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
9.6 Open System Effects and Diffusion 597

9.6 Open System Effects and Diffusion

The systems studied up to this point that involve fluid flow have emphasized internal flows such as in conduits, pipes, and
valves. The relationships describing how rates of flow related to driving forces in those cases can also be used for modeling
leakage flows. Many engineering applications of interest involve fluids (liquids or gases) that can also flow (and interact)
with each other (mixing), or as they flow into and through other media. The latter include permeable membranes, tissue,
soils, etc. Within the scope of this book, it is of interest to adopt ways to model such effects using a lumped-parameter
form. This section describes ways for modeling storage and flow of liquids through membranes, soils, or similar forms of
porous solids. For example, a semi permeable membrane as in Figure 9.24(a) is meant to separate a pure solvent (water)
from a solution (such as salt in water). A bond graph in Figure 9.24(b) shows how we can adopt a thermodynamic open C to
represent the storage of energy in the solvent and solution. A word bond graph “Membrane” element is used to represent a
partition. We show how single or multiport R elements can be used to model the membrane in such a system. In this case,
only the flow of the solvent is shown flowing through a semi permeable membrane (designed, say, to block salt).
Some basic principles are reviewed as needed to model these types of system. The examples should lay groundwork for
extension into applications that are beyond the scope of this discussion. Emphasis here is on modeling systems where
there may be mass or molar transfer processes by diffusion of liquid through porous or permeable materials. Since analo-
gies to other physical processes can provide useful insights, these are also discussed. References are cited that provide
more in-depth introductions, as the intent here is to understand how these systems can be modeled within a bond graph
framework.

9.6.1 Preliminary Concepts

In this section, we are concerned with mixtures in liquid state that may flow from one region or compartment to another,
possibly as part of an engineering process (e.g., desalination), or in a biophysical system. These liquids can be completely
defined by specifying the temperature, pressure, and the composition of the liquid (constituent components). It is helpful to
define some concepts useful when modeling systems that involve these types of substances, especially when defining how
bond graph elements can be used for these purposes. In some cases, emphasis may be placed on isothermal problems, and
the pressure state may be more of interest than the volume state, especially when pressure is also held constant. For this
reason, the Gibbs potential, G(T, P, Ni ), can be useful in defining and deriving basic relations.

9.6.1.1 Measures of Content

The number of moles of the ith component, Ni , relative to total moles in a system, can be measured in terms of concentration,
Ci = Ni ∕V, as well as in mole fraction xi = Ni ∕N. Thus, Ci = xi C, where C = N∕V is the total concentration. Note that the a
“C” notation is used to designate concentration to distinguish from C used as capacitance in a modeling context. In binary
mixtures where x1 + x2 = 1, it is common to take x2 as x and x1 as 1 − x. A binary solution is dilute when one species
(1, solvent) has a much larger mole fraction than the other (2, solute).

9.6.1.2 Properties of Solutions

As for gases, categorizing a liquid as ideal enables us to consider the free energy of constituents independently. Thus, we
could write the Helmholtz free energy for a mixture of two gases or liquid constituents as [213],
F(T, V, N1 , N2 ) = F(T, V, N1 , 0) + F(T, V, 0, N2 ),

Membrane

Membrane
Water Solution

(a) (b)

Figure 9.24 (a) A semi permeable membrane separates the water solvent from a solution. (b) The word membrane element stands in
for more detailed bond graph representations for modeling flow between a water solvent and solution.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
598 9 Thermodynamic Systems

and the entropies,

S(T, V, N1 , N2 ) = S(T, V, N1 , 0) + S(T, V, 0, N2 ).
This partitioning is helpful in modeling systems with multiple components.
As an example, consider a binary solution of Ne moles of ethanol and N𝑤 moles of water so the mole fractions are
xe = Ne ∕N and x𝑤 = N𝑤 ∕N, respectively, where N = Ne + N𝑤 . We can determine basic properties of the solution such as
density and viscosity in terms of the constituent mole fractions; i.e., 𝜂 = 𝜂e xe + 𝜂𝑤 x𝑤 and 𝜌 = 𝜌e xe + 𝜌𝑤 x𝑤 . The symbol “𝜂”
has been used for viscosity rather than “𝜇,” which is used for chemical potential in this context.

9.6.1.3 Chemical Potential of Real and Ideal Solutions

The chemical potential of a real solution with i components is given by [214],
𝜇i = 𝜇io (T, P) + RT ln(𝛼i ) = 𝜇io (T, P) + RT ln(𝛾i xi ), (9.154)
where 𝛼i is the activity, which also defines the activity coefficient, 𝛾i = 𝛼i ∕xi , and xi is the mole fraction. The activity coef-
ficient provides a measure of how much a solution deviates from ideal behavior [214]. When 𝛾i = 1, Raoult’s law holds so
the solution is ideal, otherwise it is real.
Chemical potential, 𝜇 (units of J/kg or J/mole), is identical with partial molar Gibbs function. In any reversible process,
the decrease in Gibbs function (or chemical potential) is equal to the net work done. The term 𝜇io (T, P) is called the standard
chemical potential of i. It is assumed it is dependent on temperature and pressure only, so will be denoted simply as 𝜇io . For a
solvent in the treatment of solutions, it is always taken to be the chemical potential of pure solvent and denoted 𝜇i∗ . In such
cases, the activity of pure solvent is then unity [215].
Guggenheim [213] outlines certain properties that can be described using chemical potential. First, the process of trans-
ferring one mole of a substance reversibly and isothermally from a large quantity of one phase to a large quantity of another
phase is a decrease −Δ𝜇i in 𝜇i and the net work done by a system.
Second, when two phases are at the same temperature, the condition for equilibrium distribution of the substance i
between two phases is that the 𝜇i should have the same value in both phases. This condition is valid even when the two
phases are at different pressures, for example, when separated by a semipermeable membrane, provided always that they
are at the same temperature, T.
An ideal or dilute solution is one where one species (solvent) is present in an amount much larger than the other (solute).
Let the solvent comprised of N1 moles have chemical potential 𝜇1o , to which is added a solute of N2 moles. The Gibbs potential
for a dilute solution in this case can be written [21],
( ) ( )
N1 N2
G(T, P, N1 , N2 ) = N1 𝜇1o + N2 𝜇 ∗ + N1 RT ln + N2 RT ln ,
N N
where N = N1 + N2 and 𝜇 ∗ () is an unspecified function of P and T. It is valid to assume behavior of dilute liquid solutions
of arbitrary density follow that for ideal gases, and so entropy in mixing is accounted for by the latter two terms [21]. In
these cases, where N2 ≪ N1 , you can approximate the latter two terms to give,22
( )
N2
G(T, P, N1 , N2 ) = N1 𝜇1o + N2 𝜇 ∗ − N2 RT + N2 RT ln .
N1
Now, the two constitutive relations become,
𝜕G
𝜇1 (P, T, x) = = 𝜇1o + RT ln(1 − x) ≈ 𝜇1o − RTx (solvent), (9.155)
𝜕N1
𝜕G
𝜇2 (P, T, x) = = 𝜇 ∗ + RT ln x (solute), (9.156)
𝜕N2
where x = N2 ∕N1 is the mole fraction of the solute.

9.6.1.4 Membrane Processes

A membrane is a permselective barrier, meaning it is semipermeable between two phases, typically referred to as the feed and
permeate. A semipermeable membrane allows molecules of a solvent but not a solute to pass. Permselectivity is a measure

22 For example, using ln(x) ≃ 2(x − 1)∕(x + 1) for x > 0, you can show ln(N1 ∕N) ≃ −N2 ∕N1 .
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
9.6 Open System Effects and Diffusion 599

Figure 9.25 Membrane processes have flows driven by forces related to concentration (C), Membrane
pressure (P), temperature (T), and electric potential (V ). Permeate

Feed
Membrane driving forces

of a membrane’s ability to separate, or the quality of the separation. The selectivity is often measured by the retention or the
separation factor [216]. The retention factor, for example, is expressed in terms of concentrations of the feed and permeate;
that is, R = (Cf − Cp )∕Cf = 1 − Cp ∕Cf . Thus, modeling the influence of membranes requires constitutive relations between
the flow (or flux, permeation rate) to measures of fluid concentrations.
Transport across a membrane can only occur when a driving force (chemical potential or electrical potential differences)
acts on the individual components in the system. These potential differences arise due to pressure, concentration, temper-
ature, or electrical potential, as inferred by Figure 9.25.

9.6.1.5 Osmosis
Refers to the process in which flow of a solvent occurs through a semipermeable membrane that separates two solutions of
different concentrations. Osmosis tends toward equal concentrations on each side of the membrane, the flow of the solvent
(i.e., osmosis) being from less concentration (richer in solvent) toward that with higher concentration.

9.6.1.6 Osmotic Pressure

Consider a membrane permeable to a solvent (such as water, i = 1). A small amount of solute (such as sugar or salt, i = 2)
is introduced on the one side (b) of this semipermeable membrane, where pressure PB may vary, as in Figure 9.26, while
the other side (a) contains the solvent and is kept at constant pressure, Pa . Diffusion is at equilibrium when the chemical
potentials of the solvent, 𝜇1 (), on each side of the membrane are equal, that is, 𝜇1 (Pa , T, 0) = 𝜇1 (Pb , T, x). The effect of the
small amount of solute can be approximated using equation 9.155,
𝜇1 (Pa , T, 0) = 𝜇(Pb , T, x) = 𝜇1 (Pb , T, 0) − RTx, (9.157)
where x = N2 ∕(N1 + N2 ) ≃ N2 ∕N1 is the mole fraction of the solute. Use tangent linearization to approximate 𝜇1 (Pb , T, 0)
in equation 9.157 about Pa ,
𝜕𝜇1 (Pb , T, 0) ||
𝜇1 (Pb , T, 0) ≈ 𝜇1 (Pa , T, 0) + | ⋅ (Pb − Pa )
𝜕Pa | Pa
≈ 𝜇1 (Pa , T, 0) + 𝑣1 ⋅ (Pb − Pa )
≈ 𝜇1 (Pa , T, 0) + 𝑣1 ⋅ ΔΠ,
where 𝑣1 is the molar partial volume of the solvent, and ΔΠ is referred to as the osmotic pressure, the pressure required to
stop the flow of solvent to the solution side at equilibrium, as illustrated in Figure 9.26. Substituting into equation 9.157,
𝜇1 (Pa , T, 0) = 𝜇1 (Pa , T, 0) + 𝑣1 ⋅ ΔΠ − RTx,
which implies,
𝑣1 ⋅ ΔΠ = RTx. (9.158)

Figure 9.26 Osmotic pressure during equilibrium in diffusion. Osmotic

pressure
b

a direct
osmosis
= constant Reverse
osmosis
Solvent Solution
Semipermeable
membrane
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
600 9 Thermodynamic Systems

Multiplying this relation by N1 gives,

N1 𝑣1 ΔΠ = RTN1 (N2 ∕N1 ) ⇒ ΔΠ ⋅ V 1 = N2 RT, (9.159)
which is the classic van’t Hoff relation for dilute (ideal) solutions. From this relation, we can write,
ΔΠ = RT(N2 ∕V 1 ) = RTC2 , (9.160)
showing how osmotic pressure directly depends on the solute concentration, C2 . This concept will be useful when describ-
ing membrane models in Section 9.6.3.

9.6.1.7 Reverse Osmosis

If the pressure on the solution side is increased beyond the osmotic pressure (a property of the concentration), osmotic flow
will reverse and pure solvent will pass from the solution, through the membrane, and into the solvent phase. This is referred
to as reverse osmosis, a process essential in water desalination [215]. Indeed, the direct relation between osmotic pressure
and concentration of the solution (equation 9.160) explains why pressure needs to be increased throughout desalination.

9.6.2 Energy Storage Representations

As was previously demonstrated in modeling multicomponent systems in Section 9.5, the internal energy stored in a
solution composed of multiple species can be represented by a multiport C element. For example, for a binary solution
composed of constituents i = 1 and i = 2, the chemical potentials would take the form,
𝜇1 = 𝜇1 (S, V, N1 , N2 )
𝜇2 = 𝜇2 (S, V, N1 , N2 ).
In the bond graph form, this coupled energy storage case can be represented as in 9.27(a). For the special case of an
ideal solution, in which components do not interact, the multiport C may be decoupled (or reticulated) into two separate C
elements having the explicit constitutive relations,
𝜇i = 𝜇i∗ + RT ln(xi ), i = 1, 2, (9.161)
where 𝜇i∗ is solely a function of temperature and pressure (or alternatively entropy and volume), and using mole fraction
xi of each constituent. This model can be represented by the bond graph form in Figure 9.27(b), where two distinct C
elements can be used to emphasize this partition of energy. In this latter case, the temperature and pressure ports on the
C elements are not shown, implying constant temperature and pressure. Otherwise, they should be included as needed to
show dependence on those properties.

9.6.3 Diffusion Processes in System Models

Diffusion can be defined as the relative motion between difference stores of a constituent induced by forces that depend on
available stored energy or content. It can be helpful to draw comparison between fundamental phenomenological equations
used for various types of physical processes, as in Table 9.8. In these analogous relations, a flux (per unit area) is indicated
by J as,
dX
J = −A , (9.162)
dx

(a) (b)

Figure 9.27 Bond graphs for (a) nonideal and (b) ideal binary solutions, the latter with mechanical and thermal ports on C elements
suppressed.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
9.6 Open System Effects and Diffusion 601

Table 9.8 Analogous physical phenomenological relations (laws).

Flux type Physical relation Coefficient Attribution

Heat flux ̇
Q∕A = Jh = −𝛼 dT∕dx 𝛼 = thermal diffusivity Fourier
Electrical i∕A = Ji = −(1∕R) dV∕dx R = electrical resistance Ohm
Mass flux ̇
m∕A = Jm = −D dCi ∕dx D = diffusion coefficient Fick
Volume flux Q∕A = J𝑣 = −Lp dP∕dx Lp = permeability coefficient Darcy
Momentum flux F∕A = Jp = −𝜂 d𝑣∕dx 𝜂 = absolute viscosity. Newton

a) Note that in other contexts, absolute viscosity is indicated by “𝜇,” which in the present context designates
chemical potential.

where A is a phenomenological coefficient and dX∕dx is a driving force (or effort) gradient along distance x. For example,
recall Fourier’s law (see Chapter 2, Section 2.6),
̇ dT
Q∕A = Jh = q′′ = −k ,
dx
̇
relates a heat transfer flux, q′′ = Q∕A, where A is the area over which heat is transferred, to a temperature gradient.
Adolph Fick (in 1855) based his empirical law for diffusion on the work of Fourier and Ohm’s law for electrical conduction
[122, 217]. A one-dimensional form of Fick’s law for molar flux, Jd , along an x-axis takes the form [217],
dCi
Jd = −D , (9.163)
dx
where A is the area across which diffusion occurs, Ci is a component i’s concentration (Ci = Ni ∕V), and D is the diffusion
coefficient. This model implies no convection along the same direction and also indicates a dilute solution (although the
form of Fick’s law is not as restrictive as this may suggest [217]).
It should be emphasized that the phenomenological relations assume steady-state conditions, and typically that the mem-
brane is saturated with a solution. Different types of forces may contribute to causing particles within the membrane to move
at the same velocity (lumped approximation). In this case, the total flux for a given constituent i is an algebraic sum effect
of the forces, Fi scaled by a proportionality constant [218], Ji = Lii Fi . But truly capturing the coupled effect of all forces on
all constituents in a solution requires a from such as [219],
∑
Ji = Li j Fj , (9.164)
ij

where the Onsager reciprocal relations require the phenomenological coefficients follow, Li j = Lji [156, 214, 219]. Further
we note that for the steady-state (or equilibrium) diffusion case, the entropy production 𝜎d is,
∑
𝜎d = T ⋅ fs = Xi J i , (9.165)
i

where Xi refers to the driving forces. Note that,

∑∑ ∑
𝜎d = Li j Fj = Li j Fi Fj ≥ 0,
i ij ij

requiring that Li j + Lki are positive (or at least non-negative). This means that Lii > 0 and that off-diagonal elements are
constrained to satisfy conditions such as, Lii Ljj ≥ (Lik + Lki )2 ∕4 [219]. Given that these parameters are related and must
often be experimentally determined, such constraints are essential when building models for analysis. Further, and for
our modeling purposes, these principles suggest how n coupled membrane processes can be modeled using multiport R
elements.
In the case of two components, there are two equations from 9.164, and four coefficients: L11 , L12 , L21 , and L22 . According
to Onsager, the coupling coefficients L12 and L21 are equal, which means that only three coefficients are unique. Along
with the additional restrictions above, this means, L11 ≥ 0, L22 ≥ 0, and L11 L22 ≥ L212 . Note that the coupling coefficients
(L12 = L21 ) can be either positive or negative. Usually there is positive coupling, so the flux of one component will increase
the other [216], and this decreases membrane selectivity.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
602 9 Thermodynamic Systems

(a) (b)
∑
Figure 9.28 (a) A general n-port R for modeling coupled diffusion, representing total dissipated power, 𝜎d = Ji Xi = T ⋅ fs . (b) A
2-port R model for a membrane permeable on to water (w) and a solute (s).

Dissipative R elements thus provide a basis for integrating diffusion process models within an open thermodynamic
system modeling framework. A general R can be expressed as in Figure 9.28(a). First and foremost, it should be made clear
that this is a lumped-parameter approximation, requiring the same assumption made previously in thermodynamic systems
modeling. The systems of interest may also have sources and/or “stores” of energy and mass, which can be represented by
either ideal sources and potential (internal) energy storage (C) elements. It is then possible to represent processes that
involve mass (or molar) diffusion. Extensions can be made to include additional effects [217, 218].
For a true bond graph form, equation 9.165 will take the form,
∑
𝜎d = T ⋅ fs = Δ𝜇i ⋅ Ji , (9.166)
i

where equations 9.164 are now written,

∑
Ji = L∗i j Δ𝜇j . (9.167)
j

Consider then for a membrane permeable only to a solvent (e.g., water, w) and a solute (s),
J𝑤 = L∗11 Δ𝜇𝑤 + L∗12 Δ𝜇s (9.168)
Js = L∗21 Δ𝜇𝑤 + L∗22 Δ𝜇s . (9.169)
These equations correspond to a 2-port R as in Figure 9.28(b). This model can be used in a wide range of coupled diffusion
flows between two regions 1 and 2 (e.g., intra- and extra cellular environments), with a membrane characterized by three
parameters: L∗11 referring to water permeability, L∗22 to solute permeability, and L∗12 to a coupling coefficient.
A wide range of applications assume dilute solutions, often modeled by the classical Kedem and Katchalsky (K-K)
equations [220]. Kedem and Katchalsky formulated the coupled phenomenological flow equations in the form [216, 220],
J𝑣 = Lp (ΔP − 𝜎ΔΠ), (9.170)

Jd = Cs (1 − 𝜎)J𝑣 + 𝜔ΔΠ, (9.171)

where here J𝑣 is the total volume flux, Jd is the solute flux, ΔP the pressure differential, and ΔΠ the osmotic pressure. In
this form, the three (membrane) transport properties are solvent permeability, Lp , solute permeability, 𝜔, and reflection
coefficient, 𝜎. These parameters provide insight into membrane properties and can also be related to frictional coefficients
corresponding to dissipative effects due to relative flow of solvent and solute as well as flow relative to the membrane
[221, 222]. When the two-port R constitutive relations (equations 9.168 and 9.169) are expressed in linear resistive form (by
inverting causal form),
[ ] [ ][ ]
Δ𝜇s Rss Rs𝑤 Js
= , (9.172)
Δ𝜇𝑤 R𝑤s R𝑤𝑤 J𝑤
the resistance values can be related to membrane parameters (or friction coefficients) as shown by Kiritsis [223].
Equations 9.168 and 9.169 provide a basis for constructing true bond graph forms, although not as common as the
discipline-specific K–K equations. A relation between these two forms can be made by using the dissipation function in
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
9.6 Open System Effects and Diffusion 603

equation 9.166, which must also hold for dilute solutions implied by the K–K equations 9.170 and 9.171. Thus, we require,
𝜎d = T ⋅ fs = J𝑤 Δ𝜇𝑤 + Js Δ𝜇s = J𝑣 ΔP + Jd ΔΠ,
where [216, 224],
J𝑣 = J𝑤 𝑣𝑤 + Js 𝑣s = L11 ΔP + L12 ΔΠ,
J
Jd = s − J𝑤 𝑣𝑤 = L21 ΔP + L22 ΔΠ,
Cs
with Cs the mean permeable solute concentration, and 𝑣𝑤 and 𝑣s are the partial (molar) volumes of the water and solute,
respectively. The value of Cs can be found by alternate forms [216, 224],

C −C
1 2
[ ]
1 1 2
Cs = [s 1 s2 ] = Cs + Cs , (9.173)
ln Cs ∕Cs 2

where 1 and 2 represent the conditions on each side of a membrane.

A less restrictive form of the K–K equations which hold for dilute solutions will take the form of 9.168 and 9.169. A
complete derivation is given by Papanek [224] and resulting equations can be expressed as,
J𝑤 = g11 Δ𝜇𝑤 + g12 Δ𝜇s , (9.174)
Js = g21 Δ𝜇𝑤 + g22 Δ𝜇s , (9.175)
with,
[ ]
1 Lp
g11 = + 𝜃𝜎𝑣s , (9.176)
𝑣2𝑤 (1 + 𝜃(1 − 𝜎)) 1 + 𝜃(1 − 𝜎)
[ ]
Cs Lp (1 − 𝜎)
g12 = − 𝜔𝑣s = g21 , (9.177)
𝑣𝑤 (1 + 𝜃(1 − 𝜎)) 1 + 𝜃(1 − 𝜎)
[ ]
Cs Lp (1 − 𝜎)2
g22 = C −𝜔 , (9.178)
1 + 𝜃(1 − 𝜎) 1 + 𝜃(1 − 𝜎) s

where 𝜃 = 𝑣s Cs . Papanek [224] also derives explicit relations between each of the parameters Lp , 𝜎, and 𝜔 to the L∗i j coeffi-
cients (and Cs ). Similarly, and for dilute solutions, equations 9.170 and 9.171 guide experimental testing to determine K–K
parameters (see Mulder [216] for details). For example, when there is no osmotic pressure, ΔΠ = 0, and there is volume
(hydraulic) flow due to the pressure difference, ΔP. This permits experimental determination of Lp , the hydrodynamic
permeability by,
J |
Lp = 𝑣 || .
ΔP |ΔΠ=0
With water as the solvent, Lp takes on values less than 50 for reverse osmosis, from 50 to 100 for ultrafiltration, and greater
than 500 for microfiltration (all in units of L/(m2 ⋅hr⋅atm) [216]. Likewise, solute permeability, 𝜔, can be determined by,
J |
𝜔 = s || .
ΔΠ |J𝑣 =0
Kedem and Katchalsky [220] developed the foundational approach and understanding for modeling permeability of
membranes, as discussed in the previous paragraphs. Further, Katchalsky would later work with Oster et al. [225] to show
how such problems could be understood using network thermodynamics and a bond graph approach. They argued strongly
for use of bond graphs over linear graphs or circuit analogs.23
Now, the case of “simple” diffusion across a membrane can refer to those cases where pressure effects are negligible (with
implied thermal considerations). In this case, Fick’s law (Table 9.8) mass (or molar) flux for a given constituent i can be
written directly as,
Δ𝜇
ṁ i ∕A = Jdi = Di i = Pi ΔCi , (9.179)
Δx

23 See pp. 6–7 of Oster et al. [225].

 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
604 9 Thermodynamic Systems

Δμ Jdi
μi,a μi,b
Se 1 Se
fi,a fi,b
Membrane
(a) (b)

Figure 9.29 (a) Membrane with diffusion between region a and region b containers. (b) Representation of single constituent diffusion
between regions a and b.

where Pi is the permeability for constituent i for a given membrane. This relation suggests the use of a single-port resistive
element for lumped-parameter modeling of the diffusion for each constituent, as illustrated in Figure 9.29(b). In the case
of planar membrane layer of thickness l, area A, the permeability can be taken as, Pi = DA∕l, where D is the diffusion
coefficient. Thus, a linear resistance is defined as, Ri = 1∕Pi . In the case where diffusion is only driven by concentration
differences, it is common to use concentration as an “effort” variable in a pseudo-bond graph formulation, as will be shown
in Section 9.6.4.
This discussion has emphasized the example of diffusion in membranes and how they can be integrated into a bond
graph system modeling framework. Membrane science and technology is a highly specialized field of study including
knowledge and methods from many disciplines. Thermodynamics, fluid mechanics, and transport science are essential.
The interested reader can find books such as that by Mulder [216] useful in building some of the essential basics and
terminology. For example, Mulder classifies most membranes as porous, nonporous, and carrier-mediated according to
membrane structure and separation principle. Porous membranes are used in microfiltration and ultrafiltration, achieve
separation by discriminating particle size, and consist of a polymeric matrix having pore structures wide ranging geome-
tries and size ranging between 2 nm and 10 μm. The total membrane thickness determines the resistance to transport for
larger pore sizes (microfiltration), while ultrafiltration membranes are asymmetric and transport resistance is dominated
by a top-layer [216]. Nonporous membranes can separate molecules of approximately the same size from one another.
This separation is also dependent on differences in solubility (amount of solute that dissolves in a given amount of solvent
to form a saturated solution) and/or diffusivity. So in this case, the membrane material determines the extent of selectivity
and permeability. Nonporous membranes are used in process such as pervaporation, vapor permeation, gas separation, and
dialysis. These much denser membranes are, of course, porous as well at the molecular level.
A simple way to distinguish these membranes in the modeling approach discussed here is to introduce solubility, S, as
a thermodynamic parameter and as a measure of the amount of penetrant absorbed by the membrane under equilibrium
conditions [216]. In particular, the solubility of gases in polymers is very low and can be described by Henry’s law. That is,
the equilibrium concentration of a gas in a polymer can be given by [216, 226],

C = SP, (9.180)

where P is the pressure. This provides a way to relate external pressure at the surface to the concentration within the
membrane. This relation can be helpful in building lumped-parameter models, as shown by Paterson and Lutfullah [227]
and discussed in Section 9.6.4.

9.6.4 Models with Diffusion and Stored Internal Energy

We can now illustrate construction of some system models using the multiport C and R models. These models can be used to
simulate diffusion and transport processes in membranes, solutions, biological systems, and reaction diffusion schemes. As
in any physical system modeling problems, it is necessary to gather information about specific geometry and transport prop-
erties, as well as the interconnection of assumed model elements. Further, we need to establish environmental conditions
that can be used to set inputs. The utility of bond graph methods for modeling network thermodynamics was recognized
early on by Oster et al. [225, 228]. Additional works such as by Thoma and Atlan [229, 230] and Lachenbruch and Diller [231]
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
9.6 Open System Effects and Diffusion 605

have demonstrated application of the methods. More recent work by Gawthrop and Pan [232] shows extension of the meth-
ods to bioelectrical systems. The emphasis here is on lumped-parameter models, although in some cases approximation of
the distributed effects due to diffusion may need to be considered.

9.6.4.1 Coupled Diffusion Between Solvent and Solute Flow Across a Membrane
These applications refer to the common case described in Section 9.6.3 and in Figure 9.28. This model can be used to
describe situations with binary solutions such as water (solvent) and a solute. As shown in Section 9.4.3, we again use the
“matter” port of a thermodynamic C to track flow of moles or mass. Construction and use of bond graphs is analogous to
other energy domains, enabling formulation of state equations. A 0-junction is used represent a point of common chemical
potential (or concentration in some cases), with net flows modeling accumulation of one type of matter into a port of a C
element for that species. In these cases, R and C lumped-parameter elements are used in combination to better approximate
distributed diffusion effects.
Examples include the models described by Lachenbruch and Diller [231, 233] on perfusion of kidneys for cryopreser-
vation (with cryopreservation agent, CPA, as solute) and microdialysis systems by Kiritsis et al. [223, 234] (with ethanol
as solute). For example, Lachenbruch developed a lumped-parameter subcomponent model of a single kidney tissue
compartment surrounded by a membrane that is permeable to water and CPA, but impermeable to electrolyte in the
solution. Coupled osmotic flow of water and CPA across the membrane is modeled by a 2-port R, while storage of water
and solute (CPA) inside the tissue is modeled by single port C elements, as shown in Figure 9.30.
Along with the membrane resistive relations discussed previously, the constitutive relations for the two compliance
elements would take the form,
[ ] [ ]
Ns Ns + Nimp
𝜇s = RT ln and 𝜇𝑤 = −RT ,
NT NT
where Ns is the number of solute moles, N𝑤 for water, and Nimp impermeant [231]. State equations in this case would depend
on the source taken, say, as the respective (constant) chemical potentials for solute and water for a bathing solution, and
thus of the form,
Ṅ s = Js and Ṅ 𝑤 = J𝑤 .
with J𝑤 and Js from equations 9.168 and 9.169, respectively.
For cases where some storage of water and solute is to be considered within the membrane (a lumped approximation
of distributed effects), the form used by Kiritsis [223] can be used as in Figure 9.31. Here the solvent is again water but

Figure 9.30 A model for storage of water and solute (s = cryopreservation agent, CPA) in kidney
tissue as proposed by Lachenbruch and Diller [231].

Figure 9.31 Two binary solutions separated by a membrane.

 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
606 9 Thermodynamic Systems

the specific case here considered ethanol as solute. The bond graphs take similar forms and the only difference will be
in the particular parameters for the membrane elements.

9.6.4.2 Extensions to Include Effect of Pressure

In contrast to the above examples, Atlan and Thoma [230] described situations with combined hydraulic and osmotic type
flows. In such cases, they distinguish membranes having very small pores (“osmotic pores”) as those requiring an osmotic
pressure to stop flow even when hydraulic pressure difference is zero. In such a “chemical equilibrium,” it is the chemical
potential difference that drives the flow and an osmotic pressure must be introduced to stop it. Such cases are particularly
suited for a true bond graph from. They suggest that there is a need for equal pressures on each side of a membrane for
“larger pores” to have equilibrium.
The bond graph in Figure 9.32 shows only the diffusion of water through a semipermeable membrane (between two
storage regions), assuming salts have been blocked, as formulated by Thoma and Atlan [229]. Note in this case the use of a
transformers to introduce pressure and temperature influences into the chemical potential relations,
𝜇A = 𝜇𝑤A − TA s𝑤A + PA 𝑣𝑤A ,
𝜇B = 𝜇𝑤B + TB s𝑤B − PB 𝑣𝑤A .
The bond graph in Figure 9.32 shows only “osmotic” flow. Thoma and Atlan [229] also show how a parallel hydraulic
resistor can be used to model a “leaky” membrane. In that case, the two pressures PA and PB would “drop” across an
equivalent hydraulic resistor. Similar methods extending from purely diffusion cases were studied by Kiritsis [223] for a
microdialysis membrane probe, where the internal pressure is too high, causing “ballooning” of the elastic probe. This can
also be modeled as a “leaky membrane” since the higher internal pressure stretches the membrane and possibly the size of
the pores. It is thus not a purely diffusion condition.

9.6.4.3 Examples of Pseudo-Bond Graphs for Simple Diffusion Cases

The foregoing examples have examined system level problems with lumped-parameter approximations for membranes
with inherent diffusion characteristics. Some problems with simple diffusion can also be studied using basic pseudo-bond
graphs. Consider the example shown in Figure 9.33, where a permeable membrane is represented by a Π-model. This model

Figure 9.32 Bond graph model for diffusion of a water through a semipermeable membrane (as in [229]).

A B
Figure 9.33 Diffusion of a nonelectrolyte solute from A to B via a permeable membrane. The membrane is modeled by a
lumped-parameter Π-model to approximate distributed resistance and capacitance effects (models of this type are discussed in
Chapter 6).
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
9.6 Open System Effects and Diffusion 607

Figure 9.34 A three-lump approximation for a permeable membrane between an infinite permeant source and a solution in
pseudo-bond graph form. Note that concentration across each R element is denoted by ΔC k .

is formed by distributing the total membrane resistance, Rm , into two R elements (Rm ∕2) on each side of a single membrane
capacitive element Cm . Note that the capacitive element in Figure 9.33 generally requires a constitutive relation that follows
equation 9.161, here taking a linear capacitance, such as at an equilibrium, Cm = 𝜕N∕𝜕𝜇, for a given temperature T.
Models of this form follow analogy to circuits, as is common with pseudo-bond graphs for thermal systems, and take con-
centration as effort and in this way a true set of power variables is not used. Nevertheless, this approach can be convenient
and conceptually straightforward in some cases. Consider, for example, the case for Fickian diffusion alone in a membrane,
to be modeled by three lumped elements as shown in Figure 9.34. Following Patterson and Lutfullah [227], we replace chem-
ical potential with concentration as the effort variable, forming a strictly pseudo-bond graph model. As such, we consider
isothermal conditions and no pressure-induced effects.
As shown in Figure 9.34, the membrane is exposed on the one side to an infinite permeant modeled as an effort source
with known concentration, Cs , while on the other side is a solution bath modeled by a linear capacitance with state, Nl .
The model as formed would result in a set of linear equations to be derived as implied by the indicated causality: four
states, N1 , N2 , N3 , and Nl . The parameter values for the membrane resistances can be derived from the overall membrane
permeability, Pm = 1∕Rm .
The individual R and C values can be set in different ways when making lumped approximations of this type. The total
membrane resistance for Fickian diffusion can be found by Rm = l∕(DA). For a given R element, the (local) concentration
difference, ΔCk , relates to the molar flow rate, fk , by
ΔCk = Rk fk
where k = 1, 2, 3, 4 in Figure 9.34. It is common then to simply partition Rm (as before) based on the number of lumps.
For the three-lump system, however, the values would be distributed recognizing that R2 and R3 in the bond graph of
Figure 9.34 have twice the value of R1 and R4 , thus, taking R as the base value,
R1 + R2 + R3 + R4 = R + 2R + 2R + R = Rm ⇒ R = Rm ∕6.
Thus, R1 = R4 = Rm ∕6 and R2 = R3 = 2Rm ∕3. The linear capacitance relating concentration and moles would be split
evenly between the three lumped elements: C1 = C2 = C3 = Cm ∕3.
If there is diffusion of multiple phases in membranes, beads, or layered structures, a distribution coefficient, 𝛼, can be
defined by the ratio C∕Co , where Co is the concentration of the equilibrium solution. In these cases, permeability of a layer
would be defined by P = DA𝛼∕l, and capacitance is defined by, C = 𝛼V o , where V o is the total volume of the membrane or
of a lumped element (see [227]).

Example 9.10 Gas diffusion in ethylene–propylene membrane

A gas diffusion cell is shown in Figure 9.35(a), as used by Paul and DiBenedetto [226] to create “simple diffusion” conditions
across different amorphous polymers films. Various gases were fed into a top chamber via the gas inlet, inducing diffusion
across the film (membrane) and into a lower chamber volume. The test cell included means for measuring the total quantity
of gas that permeated through the film (at STP) over time, using a modified time-lag technique method first proposed by
Daynes [235] in 1920. The measurements were used to determine diffusion coefficients and other membrane properties.
a) Assume the planar membrane can be modeled by lumped-parameter approximation assuming diffusion follows Fick’s
law, as in the bond graph form shown in Figure 9.34. Determine expressions for the R and C elements in an n-lump
model as well as proper end conditions for the lumped membrane model.
b) Paul and DiBenedetto provide one example of the total volume of nitrogen gas diffusing over time through an
ethylene–propylene copolymer (EPR 3418, run no. 79) under the conditions given in Table 9.10. Selected points
(digitized from their Figure 9.3) are shown in Table 9.9. They estimated that the diffusion coefficient is D = 3.09 × 10−7 .
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
608 9 Thermodynamic Systems

Feed side conditions

Filter Feed
Gas inlet Film

Membrane
Thickness, Diffusion coeffcient, D

Area, A
To pressure measurement/
compensation Permeate side conditions
Permeate

(a) (b)

Figure 9.35 (a) Gas diffusion cell for measuring membrane properties of amorphous polymers (after Paul and DiBenedetto [226].
(b) Model of simple diffusion through test film (membranes) highlighting key parameters and physical variables.

Table 9.9 Volume collected versus time of nitrogen gas permeated through a ethylene–propylene copolymer (EPR 3418, run no. 79).

Time (s) 0 1.24 2.25 3.21 4.23 5.28 6.23 7.29

3
Volume (cm ) 0.0002 0.0100 0.0258 0.0425 0.0597 0.0772 0.0935 0.1104

Table 9.10 Data from Paul and DiBenedetto [226].

Parameter Symbol Value Units

Diffusion coefficient D 3.09 × 10−7 cm2 /sec

Solubility S 7.44 × 10−4 cm3 (STP)/(cm3 cmHg)
Area A 45.6 cm2
Thickness l 0.1013 cm
Collecting volume V1 1.498 cm3
Pressure (source) Ps 53.37 cmHg
Temperature T0 285.5 K

Develop a simulation that predicts the amount of nitrogen gas that diffuses through the membrane over time to steady
state. Compare the results over the selected values in Table 9.9 (from [226]). Use the additional data provided in
Table 9.10.

Solution (a)
The bond graph of Figure 9.34 can be used to formulate models with any number of lumps, N, as well as a load capacitance,
CL based on the “collecting” volume. As given in Table 9.10, CL will be determined by the collecting volume of 1.498 cm3 ,
while each membrane capacitance is based on the total membrane volume, Vm = A ⋅ l divided by the number of lumps, N.
However, we use the 𝛼 distribution factor based on the solubility, S, proposed by Paterson and Lutfullah [227], 𝛼 = Po ST∕T0 ,
where T0 is 273.15 K. Now each lump capacitance becomes, Ci = 𝛼Vm ∕N, i = 1, N. Similarly, the resistance of each lump is
Ri = Rm ∕N, where Rm = l∕(DA𝛼).
The concentration source input is determined by the pressure, Ps = 53.37 cmHg, or Cs = Ps ∕RT, where T is the operating
temperature, 285.5 K. With these given parameters, a bond graph of the form Figure 9.34 (shown for N = 3) can be developed
with different numbers of lumps, N, to approximate diffusion across the membrane.
Solution (b)
Three cases are shown in Figure 9.36 for N = 1, 2, 5 and compared with measured values of nitrogen gas volume provided
by Paul and DiBenedetto [226]. The simulations converged for N = 5. The use of a distributed-parameter H-line model for
diffusion in this membrane element is demonstrated in Solved Problem A-9-4.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
9.7 Systems with Chemical Reactions 609

0.12
N=5
N=2
0.10 N=1
Data (Paul and DiBenedetto*)

0.08
qL cm3

0.06

0.04

0.02

0
0 1 2 3 4 5 6 7 8
Time, hr

Figure 9.36 (a) Gas diffusion cell for measuring membrane properties of amorphous polymers (*after Paul and DiBenedetto [226].
(b) Model of simple diffusion through test film (membranes) highlighting key parameters and physical variables.

9.7 Systems with Chemical Reactions

Homogeneous chemical reactions can be depicted as interaction between reversible storage of energy in a C field and
irreversible dissipation of available energy in an R field. The matter can be best discussed by considering a specific example
with two reactants (A, B) and one product (C) as,

𝜈A A + 𝜈B B ⇄ 𝜈C C, (9.181)

where 𝜈i with i = A, B, C are stoichiometric coefficients. Extensions to more reactants and products are straightforward. Bond
graph models for this reaction are given in Figure 9.37.

Storage Equations
For nonideal solutions, the relationship between chemical potential and the number of moles of each chemical species
within the reaction mixture is represented by the multiport C element in Figure 9.37(a). In general, chemical potentials are
functions of all species (and other thermodynamic variables) or,

𝜇A = 𝜇A (NA , NB , NC , S, V)
⋮ (9.182)
𝜇C = 𝜇A (NA , NB , NC , S, V).

For the special case of ideal solutions,24 in which components do not interact, the multiport capacitor may be reticulated
into three single port C elements with constitutive relations,

𝜇i = 𝜇i∗ + RT ln(xi ) i = A, B, C, (9.183)

where 𝜇i∗ is solely a function of temperature and pressure (or alternatively entropy and volume) and xi is the mole fraction.
This is depicted in Figure 9.37(b).
For perfect gas mixtures, equation 9.147 applies or,

𝜇i = 𝜇iO + RT ln(P∕Po ) + RT ln(xi ),

24 Refer to Denbigh [214] as a reference for this and a general reference for the entire section.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
610 9 Thermodynamic Systems

(a) Nonideal solution with reaction

(b) Ideal solution with reaction - mechanical and thermal ports of C elements suppressed

(c) Reticulation of irreversible portion of reaction

Figure 9.37 Bond graph representation of 𝜈A A + 𝜈B B → 𝜈C C. The source of the modulation ẋ m is determined from the set
(Ṅ A , Ṅ B , Ṅ C , V̇ ). For simplicity, this modulation is not shown explicitly on the bond graph.

which implies a perfect gas mixture is ideal with,

𝜇 ∗ = 𝜇iO + RT ln(P∕Po ).
We will also find it useful to express chemical potentials for perfect gas mixtures in terms of concentrations, Ci , rather
than mole fraction xi as,
𝜇i = 𝜇io (T) + RT ln(Ci ∕Co ), (9.184)
with,
𝜇io (T) = 𝜇iO (T) + RT ln(T∕To ), (9.185)
and where Ci = Ni ∕V is the concentration of the ith component and Co = Po ∕RTo is a reference concentration.25

25 Usually Co = 0.1 MPa∕(8314 J/kmol K × 298 K) = 0.0404 kmol/m3 .

 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
9.7 Systems with Chemical Reactions 611

As in equation 9.154, the method usually adopted in dealing with real solutions of equation 9.182 is based on
equation 9.183 as,
𝜇i = 𝜇i∗ + RT ln(ai ) = 𝜇i∗ + RT ln(𝛾i xi ), (9.186)
where ai is the activity of component i and 𝛾i is called the activity coefficient. In general, 𝛾i may be a function of temperature,
pressure, and mole fractions of all substances in the reaction. A component i approaches ideal behavior as xi → 0 and as
xi → 1. In order to uniquely define both 𝜇i∗ and 𝛾i , 𝛾i is usually constrained either as,
𝛾i → 1 as xi → 1,
or,
𝛾i → 1 as xi → 0.
The product of activity coefficient 𝛾i and xi is identically the activity ai of the component i. An activity coefficient yi can
also be defined in terms of 9.184 as,
𝜇i = 𝜇io + RT ln(yi Ci ∕Co ), (9.187)
where,
yi → 1 as Ci ∕Co → 0.

Rate Equations
For a reaction such as described by 9.181, the law of definite proportions for chemical reactions suggests a common factor
can be used to quantify the total reaction rate as,
Ṅ Ṅ Ṅ
𝜉̇ = A = B = C = J, (9.188)
𝜈A 𝜈B 𝜈C
where 𝜉 is termed the degree of advancement of the reaction. This relation represents a constraint, which is imposed by the
transformers in Figure 9.37, on the flows of Rch as,
Jr = Jf = J = 𝜉.̇ (9.189)
It remains now to find the constitutive relations,
J = J(xm , At , T), (9.190)
Ṡ = S(x
̇ m , At , T), (9.191)
where xm are modulation variables determined from the set (Ni , V), i = A, B, C, and At is the total affinity [225],
At = Af − Ar , with forward and reverse affinities defined as,
Af = 𝜈A 𝜇A + 𝜈B 𝜇B , (9.192)
Ar = 𝜈C 𝜇C . (9.193)
Thermodynamics imposes constraints on equations 9.190 and 9.191. From Gibbs equation and the bond graph description
in Figure 9.37, we see that the entropy generated in the reaction is,
1
Ṡ = At J. (9.194)
T
The second law of thermodynamics requires that,
Ṡ ≥ 0,
during an admissible reaction. For this to be true, reaction rate J times affinity At must be positive. This occurs if J is solely
a function of At which, as a dissipative function, resides only in the first and third quadrants.

Perfect Gas Mixtures

It is common that reaction rate J can be expressed as the difference of two terms,
J = 𝔳f − 𝔳r , (9.195)
where 𝔳f and 𝔳r are forward and reverse reaction velocities or rates. Although not a thermodynamic requirement, this form
has meaning in terms of kinetic theory [236]. These velocities are often best expressed as proportional to reactant and prod-
uct concentrations raised to some power. In some cases, the power coefficients reflect the stoichiometry and fundamental
collisional dynamics of the reaction, but this is not usually the case. In general, these power coefficients are arrived at
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
612 9 Thermodynamic Systems

experimentally. In many cases, the forward velocity is simply proportional to powers of reactant concentrations and reverse
velocity is simply proportional to powers of product concentration as,
𝛼A 𝛼B 𝛽C
J = 𝜅f C A C B − 𝜅r C C , (9.196)
where 𝜅f , 𝜅r are forward and reverse rate “constants,” respectively, which are functions of temperature but not concen-
tration. For example, in the formation of hydrogen iodide from hydrogen and iodine gas described by the stoichiometric
equation,
1 1
H + I ⇄ HI,
2 2 2 2
experimental evidence shows that the reaction rate may be expressed as
J = 𝜅f [H2 ][I2 ] − 𝜅r [HI]2 , (9.197)
where [ ] denotes the constituent concentration.
Thermodynamic restrictions on the parameters 𝜅f , 𝜅r , 𝛼A , 𝛼B , 𝛽C in equation 9.196 can be determined by using equation
9.184 to express concentrations in terms of chemical potentials as
[ ]
𝜇i − 𝜇io
Ci ∕Co = exp , i = A, B, C. (9.198)
RT
Substituting this result into equation 9.196 yields,
[ ] [ ]
𝛼A +𝛼B 𝛼A (𝜇A − 𝜇Ao ) + 𝛼B (𝜇B − 𝜇Bo ) 𝛽C 𝛽C (𝜇C − 𝜇Co )
J = 𝜅f C o exp − 𝜅r Co exp .
RT RT
Since at thermodynamic equilibrium total reaction rate J is identically zero, we have
[ ]
𝜅f 𝛽C −𝛼A −𝛼C 𝛽C (𝜇Ceq − 𝜇Co ) − 𝛼A (𝜇Aeq − 𝜇Ao ) − 𝛼B (𝜇Beq − 𝜇Bo )
= Co exp , (9.199)
𝜅r RT
where the suffix “eq” indicates equilibrium values and since 𝜅f ∕𝜅r is solely a function of temperature and not a function of
concentration, this implies
[ ]
𝛽C 𝜇C − 𝛼A 𝜇A − 𝛼B 𝜇B eq = 0. (9.200)
On the other hand, the fundamental condition for chemical equilibrium (J = 0) of a single reaction is obtained by min-
imizing Gibbs free energy with respect to degree of advancement 𝜉 at constant temperature and pressure which results in
total affinity equal to zero at equilibrium or,
[ ]
Aeq = 𝜈A 𝜇A + 𝜈B 𝜇B − 𝜈C 𝜇C eq = 0. (9.201)
Comparing equations 9.196 and 9.201 then yields the following thermodynamic conditions for the power coefficients 𝛼A ,
𝛼B , 𝛽C ,
𝛼A = n𝜈A , (9.202)
𝛼B = n𝜈B , (9.203)
𝛽C = n𝜈C , (9.204)
or the power coefficients must be proportional to stoichiometric coefficients with the same proportionality constant in
each case. This should not be surprising since stoichiometric coefficients are arbitrary to a multiplicative constant, that
is, stoichiometrically 12 H2 + 12 I2 = HI is just as valid as H2 + I2 = 2HI. It remains for kinetic data to set these coefficients
absolutely.
As a consequence of relations 9.202–9.204, equation 9.199 reduces to,
[ ] [ o ]n
𝜅f n(𝜈C −𝜈A −𝜈B ) n(𝜈C 𝜇Co − 𝜈A 𝜇Ao − 𝜈B 𝜇Bo ) n(𝜈C −𝜈A −𝜈B ) At
= Co exp − = Co exp . (9.205)
𝜅r RT RT
This expression is in the form of an equilibrium constant Kc in terms of concentration as,
𝜅f
= (Kc )n , (9.206)
𝜅r
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
9.7 Systems with Chemical Reactions 613

where
[ ]
𝜈C −𝜈A −𝜈B Aot
Kc = Co exp − . (9.207)
RT
Besides the equilibrium requirement,
J = 0 @ At = 0,
we also have a dissipative requirement that reaction rate J be positive when At is positive and conversely negative when At
is negative. In order to ensure this, we note that,
[ ]
𝔳r
J = 𝔳f 1 − , 𝔳f > 0, (9.208)
𝔳f
[ ]
𝔳f
J = 𝔳r − 1 , 𝔳r > 0. (9.209)
𝔳r
Using 9.208, 9.209, 9.198, and 9.202 to 9.204, we then have two equivalent expressions for J,
[ ( )]
nA
J = 𝔳f 1 − exp − t , 𝔳f > 0, (9.210)
RT
[ ( ) ]
nA
J = 𝔳r exp − t − 1 , 𝔳r > 0. (9.211)
RT
From either of these expressions, we see that dissipation is ensured as long as,
n > 0. (9.212)
Also note that for 𝔳f , 𝔳r strictly positive, J is a function of At that lies in the first and third quadrants.
The terms 𝔳f and 𝔳r are concentration dependent as,
𝛼A 𝛼B
𝔳f = 𝜅f C A C B ,
𝛽C
𝔳r = 𝜅r C C .
In the bond graph, these concentration terms are provided by the modulation set,
xm = {NA , NB , NC , V},
indicated by modulation of the R element in Figure 9.37(c).
In a more general case, the forward and reverse reaction velocities are functions of both product and reactant concen-
trations. This would occur, for example, when a product concentration autocatalyzes the product formation rate or when
reactant concentration facilitates reverse reactions. For instance, the formation of phosgene from carbon monoxide and
chlorine has the following stoichiometric representation
CO + Cl2 ⇄ COCI2 ,
but reaction rate is given experimentally as,
J = 𝜅f [CO][Cl2 ]3∕2 − 𝜅r [COCI2 ][CI2 ]1∕2 ,
where we note that reverse reaction velocity is a function of both product (COCl2 ) and reactant (Cl2 ). A generalization of
9.196 is then,
J = 𝔳f − 𝔳r = 𝜅f . (9.213)
In an analogous manner to the derivation of 9.202–9.204 and 9.212, we have the following thermodynamic restrictions
on the parameters 𝛼 and 𝛽,
𝛼A − 𝛽A = n𝜈A , (9.214a)
𝛼B − 𝛽B = n𝜈B , n > 0, (9.214b)
𝛽C − 𝛼C = n𝜈C ,
and equations 9.205–9.212 still hold.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
614 9 Thermodynamic Systems

Energy

Reactants

Products

Degree of advancement of reaction

Figure 9.38 Energy barrier for a generalized system of reactants and products.

Imperfect Gases and Solutions

For imperfect gases and solutions, the product of activity coefficients and concentrations can be expressed in terms of
chemical potentials as,
[ ]
𝜇i − 𝜇io
yi Ci ∕Co = exp , i = A, B, C, (9.215)
RT
which results from 9.187 with yi an appropriate activity coefficient. Rates of reaction are not typically proportional to powers
of concentrations. Instead generally satisfactory results can be obtained on the basis of equations of the form,
J = 𝜅f (yA CA )𝛼A (yB CB )𝛼B Φ − 𝜅r (yC CC )𝛽C Φ, (9.216)
where Φ is a general function of all the concentrations. In order to satisfy thermodynamic constraints, equations 9.202–9.204
through 9.212 still apply. Note that at equilibrium, we have
[ ]n
𝜅f (yC CC )𝜈C
= = Kcn , (9.217)
𝜅r (y C )𝜈A (y C )𝜈B
A A B B
where Kc is the equilibrium constant for the reaction in terms of concentrations and activity coefficients.

Temperature Dependence of Rate Constants

The rates of most chemical reactions are very sensitive to temperature. A form that is commonly used to account for
temperature effects is the Arrhenius equation [236],
[ ]

𝜅 = Ko exp − a , (9.218)
RT
where Ko and a are constants. Ko is commonly called the pre-exponential or frequency factor and a is called the activation
energy. As depicted in Figure 9.38, there is a relation between the forward and reverse activation energies (af and ar ) and
internal energy change per mole of reaction Δu.
The activation energy af represents transition state energy which must be overcome by reactants to form products and
conversely ar represents activation energy for a reverse reaction. The difference in these two energies is proportional to
total reaction energy Δu as,
ar − af = nΔu. (9.219)
Note that n can be set to unity with appropriately scaled stoichiometric coefficients.

Example 9.11 Experimental apparatus revisited

Consider again the laboratory apparatus of Example 9.2, but this time the chamber has been filled with equal parts of
hydrogen H2 and iodine I2 gas which is in mechanical equilibrium with mass, m, at temperature T, with initial volume
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
9.7 Systems with Chemical Reactions 615

of V o . After the gases are introduced, they react according to the equation
H2 + I2 ⇄ 2HI.
Extend the bond graph from Example 9.2 to include the chemical reaction effects and develop state equations that can
be used to predict dynamic response or P = chamber pressure, T = chamber temperature, V = chamber volume, and
xHI = mole fraction of hydrogen iodide (HI). Discuss the constitutive relations needed in order to set up a numerical solution
of this problem.
Solution
The bond graph from Figure 9.4 is extended as shown in Figure 9.39. In this model, there is mechanical translation energy
storage (I element) and thermodynamic energy storage (C element). In addition, there is an entropy producing reaction (R
component).

Figure 9.39 Model for H2 + I2 ⇄ 2HI reaction in a adiabatic closed system of

Example 9.2.

The chemical transformer relates hydrogen iodide mole production rate to total reaction rate and reverse affinity to hydro-
gen iodide potential as: NHI = 2J and Ar = 2𝜇HI . Using causality and standard bond graph equation formulation, state
equations can be derived as,
ṗ = −mg + A(P − PA ), (9.220)

V̇ = A𝑣 = Ap∕m, (9.221)
Ṅ I = −J, (9.222)
2

Ṅ H = −J, (9.223)
2

Ṅ HI = 2J, (9.224)

Ṡ = J(𝜇I + 𝜇H − 2𝜇HI )∕T, (9.225)

2 2
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
616 9 Thermodynamic Systems

where
NI = moles of iodine
2
NH = moles of hydrogen
2
NHI = moles of hydrogen iodide
S = chamber entropy
𝜇I = chemical potential of iodine
2
𝜇H = chemical potential of hydrogen
2
𝜇HI = chemical potential of hydrogen iodide
J = total reaction rate

In order to complete state equation derivation, we need constitutive relations for chamber pressure, chamber tempera-
ture, chemical potentials, and total reaction rate. Assuming the chemical constituents can be approximated as ideal gases
with constant specific heats26 and using 9.145–9.147 results in the following constitutive relations for the C element:
[ ] [ ]
𝜈 xI xH xI R∕c𝑣 s − so
T = To o x 2 x 2 xHIHI exp , (9.226)
𝜈 I2 H 2 c𝑣
P = RT∕𝜈, (9.227)
𝜇I = 𝜇 O (T) + RT[ln(P∕Po ) + ln(xI2 )], (9.228)
2 I2
𝜇H = 𝜇 O (T) + RT[ln(P∕Po ) + ln(xH2 )], (9.229)
2 H2
𝜇HI = N O (T) + RT[ln(P∕Po ) + ln(xHI )], (9.230)
I2 HI
where
xI2 = NI2 ∕N
x H2 = NH2 ∕N
xHI = NHI ∕N
𝑣 = V∕N
s = S∕N
NI2 = NI2 + NH2 + NHI
c𝑣 = xI2 c𝑣I + xH2 c𝑣H + xHI c𝑣HI
2 2

so = xI2 soI + xH2 soH + xHI soHI

2 2

𝜇IO (T) = 𝜇oI + (cpI − soI )(T − To ) − cpI T ln(T∕To )

2 2 2 2 2

𝜇HO (T) = 𝜇oH + (cpH − soH )(T − To ) − cpH T ln(T∕To )

2 2 2 2 2

𝜇HI
O
(T) = 𝜇oHI + (cpHI − soHI )(T − To ) − cpHI T ln(T∕To )
soI = 0.260685 MJ∕(kmol ⋅ K) (from [237])
2

soH = 0.130680 MJ∕(kmol ⋅ K)

2

soHI = 0.206589 MJ∕(kmol ⋅ K)

𝜇oI = −15.302 MJ∕(kmol ⋅ K)
2

𝜇oH = −38.962 MJ∕(kmol ⋅ K)

2

𝜇oHI = −35.236 MJ∕(kmol ⋅ K)

cpI = c𝑣I + R = 37.311 kJ∕(kmol ⋅ K) (from [237], averaged from To to 700 K)
2 2

cpH = c𝑣H + R = 29.139 kJ∕(kmol ⋅ K) cpHI = c𝑣HI + R = 30.110 kJ∕(kmol ⋅ K)

2 2

To = 298.13 K
Po = 0.1 MN∕m2
𝑣o = RTo ∕Po = 24.79 m3 ∕kmol

26 In actuality, these specific heats vary with temperature.

 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
9.9 Problems 617

In order to obtain reaction rate J for the R element, we find that the data of Bodenstein (see [238]) between temperatures
of 556 and 781 K indicates that 9.197 holds and forward and reverse rate “constants” can be approximated as Arrhenius
functions of temperature 9.218 with,
Kof = 5.96 × 1010 m3 ∕(kmol ⋅ sec)
af = 1610 MJ∕kmol
Kor = 1.4 × 1011 m3 ∕(kmol ⋅ sec)
With ar determined from equation 9.219. Equations 9.197, 9.210, and 9.211 can then be used to obtain J as
[ ][ ]
NH NI [ (
nAt
)]
2 2
J = Kof exp(−af ∕RT) 1 − exp − . (9.231)
V V RT
A numerical simulation study of this problem is left as Problem B-9.13.

9.7.1 Summary
This section introduced ways for incorporating the effect of homogeneous chemical reactions with bond graph representa-
tions of thermodynamics in a system. The thermodynamic multiport C enables representation of associated internal energy
storage. A key step is adoption of ideal transformer elements to capture chemical reactions, with modulation by stochio-
metric coefficients. Modulation is also essential in representing irreversible effects that depend on state, notably matter
distribution and volume.

9.8 Chapter Summary

The system concept as introduced in thermodynamics was first introduced in Chapter 1 as a guide to how we model systems.
Thermodynamic principles also help us track the influence of heat and matter change, and this was demonstrated in this
chapter. Thermodynamic effects in system modeling were introduced early in Chapters 2 and 3, with emphasis on “closed
systems.” That constraint was relaxed in this chapter to enable showing how bond graph methods can continue to be used
to model a wide range of physical systems. This especially includes those that may have open system effects that involve
convected mass and energy between system elements.

9.9 Problems
Problem B-9.1 Linear elastic substance thermodynamics The equation of state of an ideal elastic substance relates the
tension force, FT , to temperature, T, and length, L,
( )
L L2
FT = KT − o2 ,
Lo L
where K is a constant and Lo (the value of L at zero tension) is a function of temperature only.
a) Show that the isothermal Young’s modulus is given by,
FT 3KTL2o
Y= +
A AL2
b) Show that the isothermal Young’s modulus at zero tension is given by,
3KT
Yo = ,
A
c) Show that the linear expansivity is given by,
3 3
FT 1 L ∕Lo − 1
𝛼 = 𝛼o − = 𝛼o − ,
AYT T L ∕L3o + 2
3

where 𝛼o is defined as the value of the linear expansivity at zero tension, or,
1 dLo
𝛼o =
Lo dT
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
618 9 Thermodynamic Systems

d) When a sample of rubber is stretched L = 2Lo , determine FT , Y , and 𝛼, given the following values for this sample:
T = 300 K, K = 1.33 × 10−2 N/K, A = 1 × 10−6 m2 , 𝛼o = 5 × 10−4 K−1 .
e) Develop a bond graph of a mass-rubber-band system in which the mass, M, hangs under gravity while restrained
by the rubber band modeled in the previous steps. Formulate state equations.
f) Use an equilibrium solution to determine the mass value that will stretch the rubber band to L = 2Lo (as in step d).
g) Linearize about the equilibrium condition and determine the system undamped natural frequency

Problem B-9.2 Gas in a cylinder with piston and a small driven paddle 27 Recall the system in Example 3.7 with a gas
contained in a cylinder and a movable piston, now fitted with a small paddle that is driven through a magnetic coupling
as shown in Figure B-9.2. An external motor provides drive torque 𝜏, resulting in an input rotational shaft speed 𝜔.

Walls either insulated

or conductive
Known input P, T

Magnetic coupling Gas

is external
temperature Movable piston

Figure B-9.2 Gas in cylinder with a small paddle

driven via magnetic coupling through a fixed wall.

There is no significant velocity induced in the gas within the cylinder, but the paddle does cause the pressure to increase
at a rate,
dP 2 𝜏𝜔
=
dt 3 V
where V is the contained gas volume.
a) Develop a bond graph that describes the mechanical and thermal interactions in this system. Assume all walls are
insulated.
b) Show that the energy difference of any two states can be determined by reference to this model. In particular, deter-
mine the difference in energy between states C and A and between D and B.
c) Explain why this process can only proceed vertically upward in a P − V diagram, and never vertically downward.

Problem B-9.3 Mechanical compression of argon gas – extended Consider the laboratory apparatus of Example 9.2.
Complete the following analysis and design evaluation:
a) Recreate the simulation results given in the solution of Example 9.2.
b) Confirm that for the conditions given, the mass will repeatedly fall and rise with equal amplitude for every subse-
quent cycle (there are no losses in this system).
c) Show how the mass, m, can be chosen to achieve the desired temperature and pressure states.
d) Compare the results from the assumed adiabatic case to an alternative design in which the walls of the chamber
are made from a stock aluminum alloy that can conduct heat. Select a wall thickness about 1/5 the piston diameter.
What is the difference in the peak pressure and temperature values that can be achieved for such an alternative
design?

Problem B-9.4 Compression of ethane Ethane is contained in an insulated cylinder as shown in Figure B-9.4, with a
weighted piston sealing the top end. A pin holding the piston is removed.
a) Determine the final pressure, volume, and temperature assuming the ethane is initially at 35 F, has a quality of 90%,
and a volume of 0.01 ft3 . Assume the piston area is 0.5 in2 and the total mass of the piston and weights is 500 lbm.
The ambient pressure is 1 atm.

27 Inspired by Example 1 in Callen [21].

 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
9.9 Problems 619

Piston Patm

Pin

C2H6

Figure B-9.4 Compression of

contained ethane in an insulated
cylinder by a weighted piston.

b) Develop a bond graph model of this system to allow dynamic response analysis. Identify state variables and write
state equations.
c) Simulate the response of the system and determine the variation of the pressure, volume, and temperature over
time. Compare to the final value results determined in part (a).

Problem B-9.5 Flow from a large tank through smooth nozzle Air at T1 flows from a large tank through a converging
nozzle of area A2 as shown in Figure B-9.5. The discharge is to temperature and pressure (T2 , P2 ). Assume isentropic
conditions and that the velocity inside the tank is essentially zero.

Figure B-9.5 Air flows from a

large tank to external conditions
via a converging nozzle.

a) Sketch a basic bond graph of the system.

̇ as in equations 9.112 and 9.113, and show that they can be expressed in the
b) Derive the mass flow rate relations, m,
form,
√
( )1∕𝛾 ( )(𝛾−1)∕𝛾
P1 P2 P2
̇
m∕A 2 = C1 √ 1− , (P2 ∕P1 ) < (P2 ∕P1 )c
T1 P1 P1
P1
̇
(m∕A2 )max = C2 √ , (P2 ∕P1 ) ≥ (P2 ∕P1 )c ,
T1
where
√
2𝛾
C1 =
R(𝛾 − 1)
√
𝛾 ( 2 )(𝛾+1)∕(𝛾−1)
C2 =
R k+1
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
620 9 Thermodynamic Systems

̇
c) Define fn (P2 ∕P1 ) = (m∕A ̇ max as a function of P2 ∕P1 varying from P2 ∕P1 from 0 to 1, normalizing m.
2 )∕(m) ̇ Use values
for air as an ideal gas for k and R.28

Problem B-9.6 Comparing true and pseudo bond graphs – 1 The problem of two bodies separated by a thermally
conductive barrier has been examined in various problems in this text (see Example 2.19 in Chapter 2). The system
shown in Figure B-9.6 shows two such bodies with a thermally conductive barrier insulated from the environment.

Insulated walls

T1 T2

Conductive barrier

Figure B-9.6 Two bodies with a

thermally conductive barrier to
allow heat transfer and insulated
from the environment.

a) Develop a bond graph model using “true” bond graphs; that is, with temperature, T, as effort and entropy flow rate,
fs , as the flow variable. Apply causality and identify the state variables.
b) Develop a bond graph using a pseudo-bond graph approach where heat transfer flow rate, fQ = Q, ̇ is taken as the
flow variable. Apply causality and identify state variables.
c) Compare the required constitutive laws required for deriving complete state equations for the models developed in
(a) and (b).
d) Simulation study: Set up a specific scenario, choosing material properties for each body and the thermal barrier so
that a simulation can be constructed using each model approach. Assume that T1 ≠ T2 and illustrate how entropy
increases.
e) Extend this case study as in Example 2.19 of Chapter 2 to include an electrical heat source into body 1. Discuss the
difference in how the source is modeled in each of the modeling approaches taken in (a) and (b).

Problem B-9.7 Cylinder leaking ethylene An uninsulated cylinder of known fixed volume V c contains ethylene at a
known initial pressure Po and temperature To , where To is the temperature of the surroundings. The cylinder valve is
opened slightly so the gas leaks to the surroundings.
a) Consider the valve is open only slightly, so the ethylene leaks slowly. After a very long time, the pressure has dropped
to P1 . Determine the mass that has escaped given this change in pressure and the heat transferred during the process.
Assume V c = 2 ft3 (0.0566 m3 ), Po = 1000 psi (6.9 MPa), To = 80 F (27 C), P1 = 500 psi (3.45 MPa).
b) Discuss how this problem will change when the valve is opened to allow a higher rate of flow. Sketch a bond graph
to account for open system effects. Identify state variables and write state equations for modeling this process as a
dynamic process.

Problem B-9.8 Measuring 𝜸 = cp ∕c𝒗 for a gas using forced resonance The approach for measuring the ratio of specific
heats due to Rüchardt [239] is described in Problem A-9-2. That method relies on estimating 𝛾 through a measure-
ment of the unforced period of oscillation of a ball interacting with the gas of interest. A more accurate method due to
Clark and Katz [240] uses forced response of a contained gas. As shown in Figure B-9.8, an electromagnetically driven
piston is driven at frequencies close to the system resonance frequency, which depends on 𝛾. By sweeping the drive
frequency of the input voltage, Vs (t) = Vo sin 𝜔t, around the resonance, more accurate and repeatable estimates of 𝛾
can be determined.

28 Note: this compact form of these relations is suggested by the relations in Blackburn et al, Chapter 8 [75].
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
9.9 Problems 621

x
Coil 1 Coil 2

Driven
piston Area, A

Ta is external CL
temperature

Figure B-9.8 Closed piston-cylinder in equilibrium.

In their apparatus, Clark and Katz included an additional electromagnet (not shown) to offset the weight of the piston
and to reduce the effect of friction.
a) Develop a bond graph model of the system shown in Figure B-9.8, assuming ideal gas, effective friction b between
the piston and chamber wall, the effective driving force from the coils can be modeled by an ideal force source,
Fc (t) = Fo sin(𝜔t − 𝜑). Also assume that the compression of the gas can be assumed adiabatic, and the total mass of
the gas is mg .
b) For the model formed in part (a), derive state equations.
c) Linearize the equations from part (b) and derive a transfer function between the position x and the input force Fc .
Assume the system is at equilibrium when the piston is centered and V o is the volume on either side of the piston in
this case, with Po the associated pressure. V 1 and V 2 are then the volumes during motion, along with the respective
pressures, P1 and P2 .
d) Use the transfer function in part (c) to determine the undamped natural frequency, and compare with the result
from Clark and Katz [240] (and the result from Rüchardt, Problem A-9- 2),
2𝛾A2 Po
𝜔2n = .
mV o

Extended modeling case study. Develop a more complete model of the system shown in Figure B-9.8 that can be
used to address design questions that would come up in prototyping. In particular, the model should explain how the
coils should be selected and sized, how to model the variable-reluctance effect for predicting the induced (driving)
force, etc.

Case studies

Problem B-9.9 Valve-controlled pneumatic hydraulic cylinder and load The pneumatic valve-controlled ram shown
in Figure B-9.9 drives a mass, M, over a surface with damping, b, via a very stiff piston rod. Air is supplied from condi-
tions shown (s) released into the exhaust conditions (e). Assume that the servovalve is controlled by x. The hydraulic
valve is modulated by x. Assume any discharge coefficients take the form Cd (x).
During operation, the position of the servovalve x would be changed (either in open or in closed loop control) to create
desired motion of the load. For this problem, consider initially that the valve is moved unidirectionally to cause motion
of the mass in one stroke.
a) Develop a bond graph of this system, making sure to treat both chambers “a” and “b” of the pneumatic ram. Use
true bond graph approach, and account for changes in the mass of air in each chamber.
b) Develop state equations for the model in part (a).
c) Study the companion articles by Shearer [241, 242] on modeling and analysis of these types of “pneumatic servo-
motors.” Explain how Shearer’s approach in accounting for the changing mass in the chambers aligns with the use
of a thermodynamic accumulator model rather than a true bond graph approach.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
622 9 Thermodynamic Systems

Supply
Exhaust conditions,
conditions,

Servovalve

Pneumatic
cylinder
M

Figure B-9.9 A pneumatic valve-controlled piston (ram).

Problem B-9.10 Wall-climbing robot A glass wall-climbing robot uses a single electric motor to actuate suction bellows
and walking linkages, as shown in Figure B-9.10. The bellows suction foot action is described in Problem B-3-48.
If such a robot is to be battery powered, it can be desirable to estimate the power required.29 In the wall climber shown
in Figure B-9.10 as single motor drive drives two pneumatic bellows that cycle the pressure inside suction cups that
serve as feet on different sides. For example, the figure shown the left-front foot and rear-right foot under suction while
the other two feet are “stepped” forward the walking linkages (not shown) while contained air is vented (released) to
atmosphere. It is evident that each bellows-dual-foot subsystem constitutes a cycled open thermodynamic system.
The question to be answered in this problem: What level of model is needed to predict that suction pressure formed
and the effect traction forces induced by the suction feet?

LF
Hold
Step

Suction L Release
d

Hold
Step

Figure B-9.10 Pneumatic climbing robot

showing alternating suction/release on left
and right feet to enable climbing.

a) Begin by completing the model study in Problem B-3-48 on a single suction foot, assuming the system is closed.
b) During a suction process, the bellows expand trapped air to create suction. This produces an effective normal force,
Fn , along the rim of the cup to create an effective traction force for climbing. Assumes a known motion to specify
𝑣d and the parameters’ given in Table B-9.11.
c) Predict the peak pressure generated when there is a full stroke of 5/8 inches at point “d” and the frequency is 2.5
cycle/sec.
d) Estimate the maximum force required to hold up the body (on a wall; note the weights specify values measured
horizontally).
e) What is the minimum 𝜇 that will allow the system to climb vertically?

29 Some commercial wall-cleaning robot products use continuous suction pumps that require AC power.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
9.9 Problems 623

Table B-9.11 Parameter data from wall-climbing robot toy.

Parameter Description Nominal value (inches)

Db Mean dia of bellows 3/4

Lb Full length of bellows 3/4
Ds Suction cup outer dia 1 5/8
hs Suction cup approx height 1/4
Dcf Inner dia of tube to front suction cup 3/32
Lcf Length of tube to front suction cups 2
Dcr Inner dia of tube to rear suction cup 3/32
Lcr Length of tube to rear suction cups 3/4
L Distance between from and rear suction cups 3/4
Wf Weight on front feet 3/4
Wr Weight on rear feet 3/4
hcg Approx height to CG 3/4

f) How much difference will there be between operation at room temperature (25 ∘ C) a hotter day, say 40 ∘ C.
g) Show how the model will change if there is some leakage in the suction feet. Is it possible to estimate how much
the traction force would be degraded?

Problem B-9.11 Tea leaf caffeine release Consider a case of pure Fickian diffusion to model caffeine release from a
fully swollen tea leaf as in Williams [243]. As illustrated in Figure B-9.11(a), assume there is release from both faces
of a planar tea leaf, assumed as infinite baths. The leaf is assumed to be very thin so edge effects are ignored. The
diffusion from the tea leaf can be modeled using a pseudo-bond graph with concentration as effort and molar flow
rate of caffeine as flow (as in Example 9.10). In the preliminary bond graph of Figure B-9.11(b), the infinites baths are
modeled as large capacitive elements and the tea leaf as a word bond graph element. It is assumed that the tea leaf is
initially “charged” with qo moles of caffeine.

Planar tea leaf

‘swollen’ thickness
C̄ 0 C̄ L
C 0 tea leaf 0 C
N˙ 0 f1 f2 N˙ L

Solutions (infinite bath)

(a) (b)

Figure B-9.11 (a) Caffeine release from a tea leaf, shown modeled as a planar diffusion membrane with
solution baths on each side. (b) Word bond graph.

a) For a membrane model of the tea leaf, determine the membrane resistance, Rm , and capacitance, Cm , assuming a
diffusion coefficient D = 5.07 × 10−7 cm2 /sec, area A = 0.159 cm2 , and l = 0.01008 cm.
b) Develop a lumped-parameter bond graph model and determine the number of lumped elements needed to reason-
ably match the results given for fractional release shown in Table B-9.12. For these results, Williams [243] assumed
an initial charge qo = 1.6 × 10−5 mmol in the swollen leaf. The bath capacitors were assumed to be free of caffeine
initially. Show the difference between modeling the bath capacitors with large volumes on order of 1000 cm3 as
pure concentration sources. Such a model would enable studying finite-size bath conditions.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
624 9 Thermodynamic Systems

Table B-9.12 Fractional release of caffeine from tea leaf, q∕qo , from Williams [243] (analytical solution results).

Time (sec) 1.141 4.775 8.366 12.000 15.592 19.310 22.859 26.451 30.127
Fractional release 0.172 0.345 0.459 0.547 0.620 0.682 0.732 0.776 0.812

Problem B-9.12 Microdialysis probe modeling A microdialysis probe is used in for monitoring the chemical
constituents of fluids in a sampled media, such as tissues [244, 245] or soil [246], as well as for drug delivery. For
example, a dialysis membrane probe (on the order of 200 μm outer diameter) is placed into a target environment to be
sampled as shown in Figure B-9.12. The inlet line carries a perfusate solution into the probe and the outlet line carries
the dialysate out of the probe, forming a unidirectional exit flow, Qout , which is collected for chemical quantitative
analysis.
Consider a microdialysis probe as shown in Figure B-9.12 meant to study delivery of an ethanol–water (binary) solu-
tion into a background solution (external environment). The ethanol–water solution of known concentration enters
the inlet line at flowrate Qi n, and the outlet flow carries a solution that is analyzed to assess delivery of the solute
(ethanol) [223].

Inlet line Outlet line

for analysis

Permeable ‘hydraulic’ flow

dialysis
diffustion
membrane
Pext, external
environment
pressure

Figure B-9.12 Schematic of a typical microdialysis probe used

for sampling of analyte concentrations and for drug delivery into
tissue as well as sampling of other permeable materials such as
soils.

a) Propose a bond graph model for how ethanol and water diffuse across the membrane between the inside of the
probe and the external environment (e.g., see Figure 9.31 for ideas).
b) As suggested in Figure B-9.12, a challenge to design of the probe is that there are pressure-induced effects given
the pressure built up inside the probe. Recommend how these effects can be integrated with the diffusion model,
including representing elastic effects (and energy stored) of the probe material.
c) Show how the models in parts (a) and (b) can be integrated to begin building a complete microdialysis probe model.
Apply causality and identify key state variables.
d) Formulate a set of state space equations for the model in part(c).

Problem B-9.13 Mechanical compression of a gas Complete a simulation of the model equations 9.220–9.225 describ-
ing hydrogen iodide production, as initiated in Example 9.11.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
625

References

1 Paynter, H.M. (1961). Analysis and Design of Engineering Systems. Cambridge, MA: MIT Press.
2 Hayt, W.H. Jr. (1989). Engineering Electromagnetics. New York, NY: McGraw-Hill, Inc.
3 Blackstock, D.T. (2000). Fundamentals of Physical Acoustics. New York, NY: Wiley.
4 Panton, R.L. (1984). Incompressible Flow. New York, NY: Wiley.
5 Malvern, L.E. (1969). Introduction to the Mechanics of a Continuous Medium. Englewood Cliffs, NJ: Prentice-Hall, Inc.
6 Newton, I. (1968). The Mathematical Principles of Natural Philosophy, Translation of A. Motte in 1729. London:
Dawsons of Pall Mall.
7 Brown, F.T. (2006). Engineering System Dynamics: A Unified Graph-Centered Approach, 2e. Boca Raton, FL: CRC Press.
8 Lanczos, C. (1970). Variational Principles of Mechanics, 4e. Toronto: University of Toronto Press (also a Dover book,
1986).
9 Holman, J. (1980). Thermodynamics. McGraw-Hill.
10 Adler, C.G. (1987). Does mass really depend on velocity, dad? American Journal of Physics 55 (8): 739–743.
11 Popov, E.P. (1968). Introduction to Mechanics of Solids, Prentice-Hall Civil Engineering and Engineering Mechanics
Series. Englewood Cliffs, NJ: Prentice-Hall.
12 Meriam, J.L. and Kraige, L.G. (1986). Engineering Mechanics. New York, NY: Wiley.
13 Feynman, R.P., Leighton, R.B., and Sands, M. (1963–1965). The Feynman Lectures on Physics. Reading, MA:
Addison-Wesley Publishing Company.
14 Fox, R.W. and McDonald, A.T. (1978). Introduction to Fluid Mechanics, 2e. New York, NY: Wiley.
15 Young, W.C., Budynas, R.G., and Sadegh, A.M. (2012). Roark’s Formulas for Stress and Strain, 8e. New York:
McGraw-Hill.
16 Crandall, S.H., Karnopp, D.C., Kurtz, E.F. Jr., and Pridmore-Brown, D.C. (1968). Dynamics of Mechanical and
Electromechanical Systems. New York: McGraw-Hill (republished R.E. Krieger Publishing Company, Malabar, FL,
1982).
17 Merritt, H.E. (1967). Hydraulic Control Systems. New York: Wiley.
18 Incropera, F.P. and DeWitt, D.P. (1985). Fundamentals of Heat and Mass Transfer, 2e. New York: Wiley.
19 Mills, A.F. (1999). Heat Transfer, 2e. Upper Saddle River, NJ: Prentice-Hall.
20 Van Wylen, G.J. and Sonntag, R.E. (1973). Fundamentals of Classical Thermodynamics, 2e. New York: Wiley.
21 Callen, H.B. (1985). Thermodynamics and An Introduction to Thermostatistics. New York, NY: Wiley.
22 Wiberg, D.M. (1971). Schaums Outline of Theory and Problems of State Space and Linear Systems. New York:
McGraw-Hill Book Company.
23 van der Schaft, A.J. and Schumacher, J.M. (2000). An Introduction to Hybrid Dynamical Systems. Springer-Verlag
London Ltd.
24 Liberzon, D. (2003). Switching in Systems and Control. Boston, MA: Birkhäuser.
25 Bronson, R. (1973). Schaum’s Outline of Modern Introductory Differential Equations. New York: McGraw-Hill Book
Company.
26 Hairer, E., Nørsett, S.P., and Wanner, G. (1993). Solving Ordinary Differential Equations I: Nonstiff Problems. Berlin:
Springer-Verlag.
27 Chapra, S.C. and Canale, R.P. (2002). Numerical Methods for Engineers: With Software and Programming Applications,
4e. Boston, MA: McGraw-Hill.

Modeling of Physical Systems: Simulation and Control, First Edition. Raul G. Longoria and Joseph J. Beaman.
© 2025 John Wiley & Sons Ltd. Published 2025 by John Wiley & Sons Ltd.
Companion Website: www.wiley.com/go/Modeling_of_Physical_Systems/1e
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
626 References

28 Gear, C.W. (1971). Numerical Initial Value Problems in Ordinary Differential Equations. Englewood Cliffs, NJ:
Prentice-Hall.
29 Shampine, L.F. and Allen, R.C. (1973). Numerical Computing: An Introduction. Philadelphia, PA: W. B. Saunders and
Company.
30 Shampine, L.F. and Gordon, M.K. (1975). Computer Solution of Ordinary Differential Equations: The Initial Value
Problem. San Francisco, CA: Freeman, Cooper and Company.
31 Woodson, H.H. and Melcher, J.R. (1968). Electromechanical Dynamics, Part I: Discrete Systems. New York: Wiley.
32 Fitzgerald, A.E. Jr., Kingsley, C., and Umans, S.D. (2003). Electric Machinery, McGraw-Hill Series in Electrical
Engineering. Power and Energy, 6e. Boston, MA: McGraw-Hill.
33 Shearer, J.L., Murphy, A.T., and Richardson, H.H. (1971). Introduction to System Dynamics. Reading, MA:
Addison-Wesley Publishing Company.
34 Wyatt, J.L. and Chua, L.O. (1977). A theory of nonenergic N-ports. International Journal of Circuit Theory and
Applications 5 (2): 181–208.
35 Paynter, H.M. and Longoria, R.G. (1998). Efficient computer models of automotive fluid couplings and torque convert-
ers. ASME Fluid Machinery Forum, Fluids Engineering Division Summer Meeting, Washington, DC (21–25 June 1998).
FEDSM98-5128.
36 Hambley, A.R. (2018). Electrical Engineering: Principles and Applications, 7e. Hoboken, NJ: Pearson.
37 Chua, L.O., Desoer, C.A., and Kuh, E.S. (1987). Linear and Nonlinear Circuits. New York: McGraw-Hill.
38 Karnopp, D.C., Margolis, D.L., and Rosenberg, R.C. (2012). System Dynamics: Modeling, Simulation, and Control of
Mechatronic Systems, 5e. Hoboken, NJ: Wiley.
39 Rosenberg, R.C. and Karnopp, D.C. (1983). Introduction to Physical System Dynamics. New York: McGraw-Hill, Inc.
40 Paynter, H.M. (1992). An epistemic prehistory of bond graphs. In: Bond Graphs for Engineers (ed. P.C. Breedveld and
G. Dauphin-Tanguy), 3–17. Amsterdam: Elsevier.
41 Howell, J.R., Mengüç, M.P., and Siegel, R. (2016). Thermal Radiation Heat Transfer, 6e. Boca Raton, FL: CRC Press.
42 Burr, A.H. (1981). Mechanical Analysis and Design, 1e. New York, NY: Elsevier Publishing Co., Inc.
43 Gieck, K. and Gieck, R. (1990). Engineering Formulas, 8e. New York: McGraw-Hill.
44 Zemansky, M.W. and Dittman, R.H. (1997). Heat and Thermodynamics, 7e. New York: McGraw-Hill Book Company.
45 Sissom, L.E. and Pitts, D.R. (1998). Schaum’s Outline of Theory and Problems of Heat Transfer, 2e. New York:
McGraw-Hill.
46 Chironis, N.P. (1991). Mechanisms and Mechanical Devices Sourcebook. New York: McGraw-Hill.
47 Thoma, J.U. (1979). Hydrostatic Power Transmission. Morden, Surrey, England: Trade and Technical Press.
48 Taylor, C.F. (1985). The Internal Combustion Engine in Theory and Practice, vol. 2. Cambridge, MA: The MIT Press.
49 Cellier, F.E. (1991). Continuous System Modeling. New York: Springer-Verlag.
50 Zhang, S., Wang, X., and Zeng, Z. (2020). A simple no-equilibrium chaotic system with only one signum function for
generating multidirectional variable hidden attractors and its hardware implementation. Chaos: An Interdisciplinary
Journal of Nonlinear Science 30 (5): 053129.
51 Gelb, A. and Vandervelde, W. (1974). Applied Optimal Estimation. MIT Press.
52 Gelb, A., Kasper, J.F. Jr., Nash, R.A. Jr. et al. (1974). Applied Optimal Estimation. Written by: Technical Staff, The
Analytic Sciences Corporation (ed. A. Gelb). Cambridge, MA: MIT Press.
53 Ogata, K. (1997). Modern Control Engineering, 3rd edition or later. Upper Saddle River, NJ: Prentice-Hall, Inc.
54 Cannon, R.H. Jr. (1967). Dynamics of Physical Systems. New York: McGraw-Hill.
55 Takahashi, Y., Rabins, M.J., and Auslander, D.M. (1972). Control and Dynamic Systems. Reading, MA: Addison-Wesley
Publishing Company.
56 Schultz, D.G. and Melsa, J.L. (1967). State Functions and Linear Control Systems. New York: McGraw-Hill.
57 Rabinowicz, E. (1995). Friction and Wear of Materials, 2e. New York: Wiley.
58 Wong, J.Y. (2001). Theory of Ground Vehicles. New York: Wiley.
59 Routh, E.J. (1892). Dynamics of a System of Rigid Bodies. New York: Macmillan.
60 Hurwitz, A. (1895). On the conditions under which an equation has only roots with negative real parts. Mathematische
Annalen 46 (5): 273–284.
61 Dorf, R.C. and Bishop, R.H. (2017). Modern control systems, 13e. Hoboken, NJ: Pearson.
62 Den Hartog, J.P. (1956). Mechanical Vibrations, 4e. New York: McGraw-Hill (also as Dover reprint, 1985).
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
References 627

63 Beaman, J.J. and Rosenberg, R.C. (1988). Constitutive and modulation structure in bond graph modeling. Journal of
Dynamic Systems, Measurement, and Control 110 (4): 395–402.
64 Wyatt, J., Chua, L., Gannett, J. et al. (1981). Energy concepts in the state-space theory of nonlinear n-ports: Part
I-Passivity. IEEE Transactions on Circuits and Systems 28 (1): 48–61.
65 Nayfeh, A.H. (2004). Perturbation Methods. Weinheim: Wiley-VCH.
66 Pun, D., Lau, S.L., and Cao, D.Q. (1999). Stability analysis for linear time-varying systems with uncertain parameters:
application to steady-state solutions of nonlinear systems. Journal of Vibration and Control 5 (6): 925–939.
67 Krysinski, T. and Malburet, F. (2011). Mechanical Instability, 1e. Somerset: Wiley.
68 Goldstein, D. (1980). Classical Mechanics, 2e. Reading, MA: Addison-Wesley.
69 Neimark, J.I. and Fufaev, N.A. (1972). Dynamics of Nonholonomic Systems (Translated from the Russian by J.R.
Barbour). Providence, RI: American Mathematical Society.
70 Karnopp, D.C. and Rosenberg, R.C. (1968). Analysis and Simulation of Multiport Systems: The Bond Graph Approach to
Physical System Dynamics. Cambridge, MA: MIT Press.
71 Lawrence, A. (2001). Modern Inertial Technology: Navigation, Guidance, and Control, 2e. New York, NY: Springer.
72 Haug, E.J. (1989). Computer Aided Kinematics and Dynamics of Mechanical Systems. Needham, MA: Allyn and Bacon.
73 Paynter, H.M. (1979). Multiport dissipators as ideal work-into-heat converters. Proceedings of the 10th Annual
Pittsburgh Conference (Edited W.G. Vogt and M.H. Mickle), 1843–1845.
74 Breedveld, P.C. (1984). Physical systems theory in terms of bond graphs. PhD thesis. Enschede, The Netherlands:
Twente University.
75 Blackburn, J.F., Reethof, G., and Shearer, J.L. (1960). Fluid Power Control. Cambridge, MA: The MIT Press.
76 Daugherty, R.L., Franzini, J.B., and Finnemore, E.J. (1985). Fluid Mechanics with Engineering Applications, 8e.
New York: McGraw-Hill.
77 Willson, B. and Traver, A.E. (1987). The use of control volume analysis and non-potential junction concepts to model
liquid piston engine dynamics. 1987 American Control Conference, 1436–1443.
78 Paynter, H.M. (1972). The dynamics and control of Eulerian turbomachines. Journal of Dynamic Systems, Measurement,
and Control 94 (3): 198–205.
79 Paynter, H.M. and Longoria, R.G. (1997). Two-port canonical bond graph models of lossy power transduction. 1997
IEEE International Conference on Systems, Man, and Cybernetics. Computational Cybernetics and Simulation, volume 2,
1533–1537.
80 Lambeck, R.P. (1983). Hydraulic Pumps and Motors: Selection and Application for Hydraulic Power Control Systems.
New York: Dekker.
81 Lynch, W.A. and Truxal, J.G. (1962). Signals and Systems in Electrical Engineering, Brooklyn Polytechnic Institute
Series. New York: McGraw-Hill.
82 Agarwal, A. and Lang, J. (2005). Foundations of Analog and Digital Electronic Circuits. Amsterdam: Morgan Kaufman
Publishers is an imprint of Elsevier.
83 Tiersten, M.S. (1969). Force, momentum change, and motion. American Journal of Physics 37: 82–87.
84 Siegel, S. (1972). More about variable mass systems. American Journal of Physics 40: 183–185.
85 Timoshenko, S. and Young, D.H. (1948). Advanced Dynamics. New York: McGraw-Hill Book Company.
86 Caughey, T.K. (1960). Sinusoidal excitation of a system with bilinear hysteresis. Journal of Applied Mechanics 27 (4):
640–643.
87 Paynter, H.M. (1966). Positive/negative feedback in amplification and control. The Lightning Empiricist, volume 14,
Jan-July 1966.
88 Paynter, H.M. (1970). Simulation of physical and functional systems. Proceedings of the 16th International ISA
Aerospace Instrumentation Symposium, 80–90.
89 Karnopp, D. (1983). Computer models of hysteresis in mechanical and magnetic components. Journal of the Franklin
Institute 316 (5): 405–415.
90 Ewins, D.J. (2000). Modal Testing: Theory, Practice, and Application, 2e. Baldock, Hertfordshire, England: Research
Studies Press.
91 Guillemin, E.A. (1963). Theory of Linear Physical Systems: Theory of Physical Systems from the Viewpoint of Classical
Dynamics, Including Fourier Methods. New York: Wiley.
92 Dorf, R.C. (1967). Modern Control Systems, 1e. Reading, MA: Addison-Wesley Publishing Company.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
628 References

93 Franklin, G.F., Powell, J.D., and Emami-Naeini, A. (2006). Feedback Control of Dynamic Systems, 5e. Upper Saddle
River, NJ: Pearson Prentice Hall.
94 Brogan, W.L. (1991). Modern Control Theory, 3e. Englewood Cliffs, NJ: Prentice Hall.
95 Mixl, D.F. (1969). Random Signal Analysis. Reading, MA: Addison-Wesley.
96 Hogan, N. (1985). Impedance control: an approach to manipulation: Part I–Theory. Journal of Dynamic Systems,
Measurement, and Control 107 (1): 1–7.
97 Brown, F.T. (1972). Direct application of the loop rule to bond graphs. Journal of Dynamic Systems, Measurement, and
Control 94 (3): 253–261.
98 Thomson, W.T. (2003). Theory of Vibration with Application. London: Taylor & Francis.
99 Meirovitch, L. (1967). Analytical Methods in Vibrations. New York: Macmillan.
100 Ljung, L. (1999). System Identification: Theory for the User, Prentice-Hall Information and System Sciences Series, 2e.
Upper Saddle River, NJ: Prentice Hall PTR.
101 Pintelon, R. and Schoukens, J. (2012). System Identification: A Frequency Domain Approach, 2e. Hoboken, NJ: Wiley.
102 Tangirala, A.K. (2015). Principles of System Identification: Theory and Practice, 1e. Boca Raton, FL: CRC Press.
103 Graham, D. and McRuer, D. (1961). Analysis of Nonlinear Control Systems. New York: Wiley.
104 Gelb, A. and Vander Velde, W.E. (1968). Multiple-Input Describing Functions and Nonlinear System Design. New York:
McGraw-Hill.
105 IEEE Std C57.149-2012 (2013). IEEE Guide for the Application and Interpretation of Frequency Response Analysis for
Oil-Immersed Transformers, 1–72. IEEE Committee.
106 ASTM D3580-95 (2015). ASTM Standard Test Methods for Vibration (Vertical Linear Motion) Test of Products, 1–4.
ASTM Committee.
107 Smith, G.A. and Triplett, W.C. (1954). Experimental flight methods for evaluating frequency-response characteristics of
aircraft. Transactions of the American Society of Mechanical Engineers 76: 1383–1390.
108 Lees, S. and Hougen, J.O. (1956). Pulse testing a model heat exchange process. Industrial and Engineering Chemistry 48
(6): 1064–1068.
109 Law, V.J. and Bailey, R.V. (1963). A method for the determination of approximate system transfer functions. Chemical
Engineering Science 18: 189–202.
110 Bendat, J.S. and Piersol, A.G. (1985). Random Data: Analysis and Measurement Procedures, 2e, rev. and expanded.
New York: Wiley.
111 Bendat, J.S. and Piersol, A.G. (1993). Engineering Applications of Correlation and Spectral Analysis. New York: Wiley.
112 Bendat, J.S. and Piersol, A.G. (1980). Engineering Applications of Correlation and Spectral Analysis. New York: Wiley.
113 Juang, J.-N. (1994). Applied System Identification. Upper Saddle River, NJ: Prentice-Hall PTR.
114 Doebelin, E.O. (1980). System Modeling and Response: Theoretical and Experimental Approaches. New York: Wiley.
115 Balabanian, N. and Lepage, W.R. (1956). What is a minimum-phase network? Transactions of the American Institute of
Electrical Engineers, Part I: Communication and Electronics 74 (6): 785–788.
116 Guillemin, E.A. (1953). Introductory Circuit Theory. New York: Wiley.
117 Huelsman, L.P. (1993). Active and Passive Analog Filter Design: An Introduction, McGraw-Hill Series in Electrical and
Computer Engineering. Electronics and VLSI Circuits. New York: McGraw-Hill.
118 Molloy, C.T. (1962). Application of four-pole parameters to torsional vibration problems. Journal of Engineering for
Industry 84 (1): 21–34.
119 Huelsman, L.P. (1963). Circuits, Matrices, and Linear Vector Spaces, McGraw-Hill Electronic Sciences Series. New York:
McGraw-Hill.
120 Harman, W.W. and Lytle, D.W. (1962). Electrical and Mechanical Networks; An Introduction to Their Analysis.
New York: McGraw-Hill.
121 Pipes, L.A. (1946). Applied Mathematics for Engineers and Physicists, 1e. New York: McGraw-Hill.
122 Moore, R.K. (1964). Wave and Diffusion Analogies. New York: McGraw-Hill, Inc.
123 Paynter, H.M. and Ezekiel, F.D. (1958). Water hammer in nonuniform pipes as an example of wave propagation in
gradually varying media. Transactions of the American Society of Mechanical Engineers 80 (7): 1585–1592.
124 Wylie, C.R. and Barrett, L.C. (1982). Advanced Engineering Mathematics, 5e. New York: McGraw-Hill Book Company.
125 Bedford, A. and Drumheller, D.S. (1994). Introduction to Elastic Wave Propagation, 1e. Chichester: Wiley.
126 Oldenburger, R. and Goodson, R.E. (1964). Simplification of hydraulic line dynamics by use of infinite products.
Journal of Basic Engineering 86 (1): 1–8.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
References 629

127 Rocke, R.D. (1966). Transmission matrices and lumped parameter models for continuous systems. PhD thesis.
California Institute of Technology.
128 Astleford, W.J., Holser, J.L., and Gerlach, C.R. (1972). Analysis of Propellant Feedline Dynamics (Project No. 02-2889,
NASA Contract NAS8-25919). Technical Report. San Antonio, TX: Southwest Research Institute.
129 Paynter, H.M. (1962). Lumped structures method and apparatus and approximation of uniform media by lumped struc-
tures. US Patent 3, 044, 703.
130 Goodson, R.E. (1970). Distributed system simulation using infinite product expansions. Simulation 15 (6): 255–263.
131 Chapman, C.P. (1966). Derivation of the Mathematical Transfer Function of a Electrodynamic Vibration Exciter. NASA
Report No. 32-934. Technical Report. Pasadena, CA: Jet Propulsion Laboratory, California Institute of Technology.
132 Kochenburger, R.J. (1972). Computer Simulation of Dynamic Systems, 1e. Englewood Cliffs, NJ: Prentice-Hall, Inc.
133 AIEE Feedback Control Systems Committee (1951). Proposed symbols and terms for feedback control systems.
Electrical Engineering 70 (10): 905–909.
134 Mayr, O. (1970). Origins of Feedback Control. Cambridge, MA: MIT Press.
135 Paynter, H.M. (1970). System graphing concepts. Instruments and Control Systems 43 (7): 77–78.
136 Ogata, K. (1995). Discrete-Time Control Systems. Englewood Cliffs, NJ: Prentice-Hall, Inc.
137 Thaler, G.J. and Brown, R.G. (1960). Analysis and Design of Feedback Control Systems, 2e. New York: McGraw-Hill.
138 Van de Vegte, J. (1994). Feedback Control Systems. Englewood Cliffs, NJ: Prentice-Hall, Inc.
139 Black, H.S. (1977). Inventing the negative feedback amplifier: six years of persistent search helped the author conceive
the idea “in a flash” aboard the old Lackawanna Ferry. IEEE Spectrum 14 (12): 55–60.
140 Beckwith, T.G. (2007). Mechanical Measurements / Thomas G. Beckwith, Roy D. Marangoni, John H. Lienhard V , 6e.
Upper Saddle River, NJ: Pearson Prentice Hall.
141 Figliola, R.S. (2011). Theory and Design for Mechanical Measurements / Richard S. Figliola, Donald E. Beasley, 5e.
Hoboken, NJ: Wiley.
142 D’Azzo, J.J. and Houpis, C.H. (1988). Linear Control System Analysis: Conventional and Modern. New York:
McGraw-Hill Book Company.
143 Kuo, B.C. (1975). Automatic Control Systems. Englewood Cliffs, NJ: Prentice-Hall.
144 Nise, N.S. (2015). Control Systems Engineering, 7e. Hoboken, NJ: Wiley.
145 Evans, W.R. (1948). Graphical analysis of control systems. Transactions of the American Institute of Electrical Engineers
67 (1): 547–551.
146 Ziegler, J.G. and Nichols, N.B. (1942). Optimum settings for automatic controllers. Transactions of the American Society
of Mechanical Engineers 64 (8): 759–765.
147 Friedland, B. (1986). Control System Design: An Introduction to State-Space Methods. New York: McGraw-Hill.
148 Astrom, K.J. and Murray, R.M. (2021). Feedback Systems: An Introduction for Scientists and Engineers, 2e. Princeton,
NJ: Princeton University Press.
149 Luenberger, D.G. (1964). Observing the state of a linear system. IEEE Transactions on Military Electronics 8 (2): 74–80.
150 Luenberger, D. (1971). An introduction to observers. IEEE Transactions on Automatic Control 16 (6): 596–602.
151 Joseph, D.P. and Tou, T.J. (1961). On linear control theory. Transactions of the American Institute of Electrical Engineers,
Part II: Applications and Industry 80 (4): 193–196.
152 Kalman, R.E. (1960). Contributions to the theory of optimal control. Proceedings of 1959 Mexico City Conference on
Differential Equations, Mexico City, 102–199.
153 Astrom, K.J. and Hagglund, T. (1995). PID Controllers, 2e. Research Triangle Park, NC: International Society for
Measurement and Control.
154 Members of the staff of the Department of Electrical Engineering (MIT) (1943). Magnetic Circuits and Transformers: A
First Course for Power and Communication Engineers. New York, NY: Wiley.
155 Woloszko, J., Stalder, K.R., and Brown, I.G. (2002). Plasma characteristics of repetitively-pulsed electrical discharges in
saline solutions used for surgical procedures. IEEE Transactions on Plasma Science 30 (3): 1376–1383.
156 Onsager, L. (1931). Reciprocal relations in irreversible processes. I. Physical Review 37: 405–426.
157 Onsager, L. (1931). Reciprocal relations in irreversible processes. II. Physical Review 38: 2265.
158 Busch-Vishniac, I.J. and Paynter, H.M. (1991). Bond graph models of acoustical transducers. Journal of the Franklin
Institute 328 (5): 663–673.
159 Haus, H.A. and Melcher, J.R. (1989). Electromagnetic Fields and Energy. Englewood Cliffs, NJ: Prentice Hall.
160 Jefimenko, O.D. and Walker, D.K. (1980). Electrets. The Physics Teacher 18 (9): 651–659.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
630 References

161 Busch-Vishniac, I.J. (1999). Electromechanical Sensors and Actuators. New York: Mechanical Engineering Series,
Springer.
162 Wangcharoenrung, C. and Longoria, R.G. (2003). A bond graph model of outer hair cell active force generation.
International Conference on Bond Graph Modeling and Simulation (ICBGM) 2003, 101–106.
163 Wangcharoenrung, C. (2005). Development of Adaptive Transducer Based on Biological Sensory Mechanism. The
University of Texas at Austin.
164 Shampine, L.F. and Reichelt, M.W. (1997). The matlab ODE suite. SIAM Journal on Scientific Computing 18 (1): 1–22.
165 Krause, P.C. and Wasynczuk, O. (1989). Electromechanical Motion Devices. New York: McGraw-Hill.
166 Krause, P.C., Wasynczuk, O., Sudhoff, S.D., and Pekarek, S. (2013). Analysis of Electric Machinery and Drive Systems,
3e. Hoboken, NJ: Wiley.
167 Watson, J.K. (1980). Applications of Magnetism. New York, NY: Wiley.
168 Chai, H.-D. (1998). Electromechanical Motion Devices. Upper Saddle River, NJ: Prentice Hall PTR.
169 Nasar, S.A. (1981). Schaum’s Outline of Theory and Problems of Electric Machines and Electromechanics, Schaum’s
Outline Series. New York: McGraw-Hill.
170 Hammond, P. and Sykulski, J.K. (1994). Engineering Electromagnetism: Physical Processes and Computation. New York,
NY: Oxford University Press.
171 Campbell, P. (1994). Permanent Magnet Materials and Their Application. New York, NY: Cambridge University Press.
172 Furlani, E.P. (2001). Permanent Magnet and Electromechanical Devices: Materials, Analysis, and Applications. San Diego,
CA: Academic Press.
173 Kamerbeek, E.M.H. (1973). Electric motors. Philips Technical Review 33 (8/9): 215–234.
174 Timmerman, J. (1973). Two electromagnetic vibrators. Philips Technical Review 33 (8/9): 249–259.
175 Campbell, P. (2000). Comments on “Energy stored in permanent magnet”. IEEE Transactions on Magnetics 36 (1):
401–403.
176 Furlani, E.P., Lee, J.K., and Dowe, D. (1993). Predicting the dynamic behavior of moving magnet actuators. Journal of
Applied Physics 73 (7): 3555–3559.
177 Wells, D.A. (1967). Lagrangian Dynamics, Schaum’s Outline Series in Engineering. New York: McGraw-Hill.
178 Sommerfeld, A. (1952). Mechanics: Lectures on Theoretical Physics, vol. I (translated from 4th German edition).
New York: Academic Press.
179 Tait, P.G. and Steele, W.J. (1900). A Treatise on Dynamics of a Particle. London: Macmillan.
180 Denny, M. (2020). A uniform explanation of all falling chain phenomena. American Journal of Physics 88 (2): 94–101.
181 Denny, M. (2021). Balloon and chain: an instructive variable mass system. European Journal of Physics 42 (5): 025013.
182 Wong, C.W., Youn, S.H., and Yasui, K. (2007). The falling chain of Hopkins, Tait, Steele and Cayley. European Journal
of Physics 28: 385–400.
183 Cayley, A. (1856-1857). On a class of dynamical problems. Proceedings of the Royal Society of London 8: 506–511.
184 Beaman, J.J. and Breedveld, P.C. (1985). Physical modeling with Eulerian frames and bond graphs. ASME Winter
Annual Meeting.
185 Beaman, J.J. and Breedveld, P.C. (1988). Physical modeling with Eulerian frames and bond graphs. Journal of Dynamic
Systems Measurement and Control (ASME) 110: 182–188.
186 Redfield, R.C. (2006). Bond graphs of open systems: a water rocket example. Proceedings of the Institution of
Mechanical Engineers, Part I: Journal of Systems and Control Engineering 220 (7): 607–615.
187 Cherry, C. (1951). Some general theorems for non-linear systems possessing reactance. Philosophical Magazine 42:
1161–1177.
188 Aleksandrov, A.D., Kolomogorov, A.N., and Lavrent’ev, M.A. (1963). Mathematics: Its Content, Methods and Meaning.
MIT Press.
189 Feynmann, R.P., Leighton, R.B., and Sands, M. (1964). The Feynmann Lectures on Physics. Addison-Wesley.
190 Reddy, J.N. (1967). An Introduction to the Finite Element Method. New York: McGraw-Hill.
191 Diels, H. (1924). Antike Technik, 3e. Leipzig, Berlin: B.G. Teubner.
192 Senturia, S.D. (2001). Microsystem Design. New York, NY: Springer.
193 Madni, A.M., Wan, L.A., and Hammons, S. (1996). A microelectromechanical quartz rotational rate sensor for inertial
applications. 1996 IEEE Aerospace Applications Conference. Proceedings, volume 2, 315–332.
194 Madni, A.M., Costlow, L.E., and Knowles, S.J. (2003). Common design techniques for BEI GyroChip quartz rate
sensors for both automotive and aerospace/defense markets. IEEE Sensors Journal 3 (5): 569–578.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
References 631

195 Atkins, P. and de Paula, J. (2006). Atkins’ Physical Chemistry, 8e. New York: W. H. Freeman.
196 Mandl, F. (1988). Statistical Physics. New York: Wiley.
197 Clausius, R. (1879). The Mechanical Theory of Heat. London: Macmillan.
198 Tribus, M. (1961). Thermostatics and Thermodynamics: An Introduction to Energy, Information and States of Matter,
with Engineering Applications. Princeton, NJ: Van Nostrand.
199 Gibbs, J.W. (1906). The Scientific Papers of J. Willard Gibbs. London: Longmans, Green, and Co. (Dover reprint, 1961).
200 Lee, B.I. and Kesler, M.G. (1975). A generalized thermodynamic correlation based on three-parameter corresponding
states. AIChE Journal 21 (3): 510–527.
201 Haase, R. (1990). Thermodynamics of Irreversible Processes. New York: Dover.
202 Guggenheim, E.A. (1967). Thermodynamics: An Advanced Treatment for Chemists and Physicists, 5th rev. ed.
Amsterdam: North-Holland Publishing Company.
203 Falk, G. and Ruppel, W. (1976). Energie and Entropie. Springer-Verlag (in German).
204 Wallace, D.C. (1972). Thermodynamics of Crystals. New York: Wiley.
205 Ingrim, M.E. (1988). A discrete network representation of thermomechanical processes in continuous media. PhD
thesis. The University of Texas at Austin, Department of Mechanical Engineering.
206 Zanj, A., He, F., and Breedveld, P.C. (2016). An energy-based viscoelastic model for multi-physical systems: a bond
graph approach. 2016 IEEE International Conference on Systems, Man, and Cybernetics (SMC), 002214–002219.
207 Zanj, A., He, F., and Breedveld, P.C. (2018). Domain-independent thermoviscoelastic modeling framework: a physical
approach on thermoelasticity by bond graph. Journal of Thermophysics and Heat Transfer 32 (1): 61–79.
208 Shapiro, A.H. (1953). The Dynamics and Thermodynamics of Compressible Fluid Flow, vol. I. New York: Ronald Press.
209 Bett, K.E., Rowlinson, J.S., and Saville, G. (1975). Thermodynamics for Chemical Engineers. The MIT Press.
210 Katz, S. (1967). Mechanical potential drops at a fluid branch. ASME Journal of Basic Engineering 89: 737–736.
211 Breedveld, P.C. (1984). Essential gyrators and equivalence rules for 3-port function structures. Journal of the Franklin
Institute 318 (2): 77–89.
212 Ruthven, D.M. (1984). Principles of Adsorption Processes. New York: Wiley.
213 Guggenheim, E.A. (1952). Mixtures: The Theory of the Equilibrium Properties of Some Simple Classes of Mixtures,
Solutions, and Alloys. London: Oxford University Press.
214 Denbigh, K. (1981). The Principles of Chemical Equilibrium, 4e. New York: Cambridge University Press.
215 Merten, U. (ed.) (1966). Delasination by Reverse Osmosis. Cambridge, MA: The MIT Press.
216 Mulder, M. (1997). Principles of Chemical Kinetics. Dordrecht, The Netherlands: Kluwer Academic Publishers.
217 Cussler, E.L. (1997). Diffusion Mass Transfer in Fluid Systems. Cambridge: Cambridge University Press.
218 Spiegler, K.S. (1958). Transport processes in ionic membranes. Transactions of the Faraday Society 54: 1408–1428.
219 de (Sybren Ruurds) Groot, S.R. and (Peter) Mazur, P. (1962-1984). Non-Equilibrium Thermodynamics / S.R. de Groot, P.
Mazur, Dover edition. New York: Dover Publications.
220 Kedem, O. and Katchalsky, A. (1958). Thermodynamic analysis of the permeability of biological membranes to
non-electrolytes. Biochimica et Biophysica Acta 27: 229–246.
221 Kedem, O. and Katchalsky, A. (1961). A physical interpretation of the phenomenological coefficients of membrane
permeability. The Journal of General Physiology 45: 143–179.
222 Ginzburg, B.Z. and Katchalsky, A. (1963). The frictional coefficients of the flows of non-electrolytes through artificial
membranes. The Journal of General Physiology 47: 403–418.
223 Kiritsis, N. (1999). Modeling of microdialysis processes and systems used for in vitro experiments. PhD thesis. The
University of Texas at Austin, Department of Mechanical Engineering.
224 Papanek, T.H. (1976). The water permeability of the human erthrocyte in the temperature range +25∘ C to −10∘ C. PhD
thesis. MIT, Department of Mechanical Engineering.
225 Oster, G.F., Perelson, A.S., and Katchalsky, A. (1973). Network thermodynamics: dynamic modelling of biophysical
systems. Quarterly Reviews of Biophysics 6 (1): 1–134.
226 Paul, D.R. and DiBenedetto, A.T. (1965). Diffusion in amorphous polymers. Journal of Polymer Science Part C: Polymer
Symposia 10 (1): 17–44.
227 Paterson, R. and Lutfullah (1985). Simulation of transport processes using bond graph methods: I. Gas diffusion
through planar membranes and systems obeying Fick’s laws. Journal of Membrane Science 23 (1): 59–70.
228 Oster, G.F., Perelson, A.S., and Katchalsky, A. (1971). Network thermodynamics. Nature 234: 393–399.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
632 References

229 Thoma, J. and Atlan, H. (1985). Osmosis and hydraulics by network thermodynamics and bond graphs. Journal of the
Franklin Institute 319 (1): 217–226.
230 Atlan, H. and Thoma, J. (1987). Solvent flow in osmosis and hydraulics: network thermodynamics and representation
by bond graphs. The American Journal of Physiology 252 (6 Pt 2): R1182–R1194.
231 Lachenbruch, C.A. and Diller, K.R. (1995). A network thermodynamic model of kidney perfusion. PhD thesis. The
University of Texas at Austin, Department of Mechanical Engineering.
232 Gawthrop, P.J. and Pan, M. (2021). Network thermodynamical modeling of bioelectrical systems: a bond graph
approach. Bioelectricity 3 (1): 3–13.
233 Lachenbruch, C.A. and Diller, K.R. (1999). A network thermodynamic model of kidney perfusion with a cryoprotective
agent. Journal of Biomechanical Engineering 121 (6): 574–583.
234 Kiritsis, N., Longoria, R.G., and Gonzales, R.A. (2001). Network thermodynamic bond graph models for microdialysis
probe processes. International Conference on Bond Graph Modeling and Simulation (ICBGM) 2001.
235 Daynes, H.A. (1920). The process of diffusion through a rubber membrane. Proceedings of the Royal Society of London.
Series A, Containing Papers of a Mathematical and Physical Character 97 (685): 286–307.
236 Hammes, G.G. (1978). Principles of Chemical Kinetics. New York: Academic Press.
237 Chase, M.W. (ed.) (1988). JANAF Thermochemical Tables. Journal of Physical and Chemical Reference Data, vol. 14
(Supplement no. 1), 3e. Washington, DC: American Chemical Society.
238 Bamford, C.H. and Tipper, C.F.H. (1972). Chemical Kinetics. Amsterdam: Elsevier.
239 Rüchardt, E. (1929). Eine einfache methode zur bestimmung von Cp/Cv. Physikalische Zeitschrift 30: 58–59.
240 Clark, A.L. and Katz, L. (1940). Resonance method for measuring the ratio of the specific heats of a gas, Cp/Cv .
Canadian Journal of Research 18A: 23–38.
241 Shearer, J.L. (1956). Study of pneumatic processes in the continuous control of motion with compressed air-I.
Transactions of the American Society of Mechanical Engineers 78 (2): 233–241.
242 Shearer, J.L. (1956). Study of pneumatic processes in the continuous control of motion with compressed air-II.
Transactions of the American Society of Mechanical Engineers 78 (2): 243–249.
243 Williams, D.A. (1988). The simulation of transport and diffusion processes using network thermodynamics and bond
graph techniques. Master’s thesis. University of Glasgow.
244 Bungay, P.M. and Gonzales, R.A. (1996). Pressure-enhanced delivery of solutes via microdialysis. 26th Annual Meeting
of the Society for Neuroscience, 2076.
245 Bungay, P.M., Sumbria, R.K., and Bickel, U. (2011). Unifying the mathematical modeling of in vivo and in vitro
microdialysis. Journal of Pharmaceutical and Biomedical Analysis 55 (1): 54–63.
246 Miró, M. and Frenzel, W. (2005). The potential of microdialysis as an automatic sample-processing technique for
environmental research. TrAC Trends in Analytical Chemistry 24 (4): 324–333.
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
633

Index

0-junction 100 Brachistochrone problem 534

1-junction 100 Bridged-T RC network 370
1-port bond graph element 94 Bulk compressibility modulus 55f
1/4-car suspension 295 Butterfly capacitor 538
2-port C-element 96
c
a Cantilever with barbell 482, 537
Absolute stability 185, 404 Capacitive (C) element 95
Ackermann’s formula for pole-placement 433, 443 Capacitive plates 43
Activation energy 614 Capacitive position detection 538
Active elements 347 Casimir forms 480
Active suspension for ride control 465 Causal bind (or conflict) 117
Activity 598 Causal stroke 14, 110
Activity coefficient 598, 611 Causal structure 191
Adiabatic flow in a duct 583 Causality 14, 109
Admittance 310 Characteristic equation 180, 404
Admittance matrix 332 Characteristic impedance 350
Adpedance matrix 332 Chatter 184
Adsorption of gaseous components 595 Chemical equilibrium constant 593
Air pump 138 Chemical potential 560, 567, 598
Air-cushion vehicle 225 Chirp signal 326
Algebraic loop 123 Circular tank 51
Ampere’s law 44, 504 Closed-loop transfer function 376, 382
Angular electromechanical actuator 504, 544 Clutch-flywheel system 164
Arrhenius equation 614 Coefficient of thermal expansion 127
Auto-spectral density function 327 Coercivity (magnetic) 507
Collocated sensor 418
b Common effort junction 100
B-H curves 498 Common flow junction 100
Backward transmission matrix 332 Comparator (control) 387
Ballast control 464 Compensator types 396
Balloon expelling mass 213 Compensators in frequency domain 421
Ballscrew actuator 138 Composite constitutive relation 191
Bandwidth 321, 412 Composite dissipative loop 195
Bandwidth and crossover correlation 413 Composite modulation 206
Basic constitutive relation 191 Compressibility factor 568
Basic dissipative element 195 Compressible flow in a uniform conduit 585
Basic energy-storing element 193 Compressible nozzle flow 581
Basic modulated coupling element 206 Compression of ethan 618
Bellows-driven suction-cup 154 Concentration 597, 601
Belt-drive control 467 Conduction heat transfer 62
Binary solution 597 Conical tank 82
Block diagrams 170 Constant pressure specific heat 566
Boat stability gyro 468 Constant strain specific heat 573
Bode plots 304 Constant volume specific heat 566, 570
Modeling of Physical Systems: Simulation and Control, First Edition. Raul G. Longoria and Joseph J. Beaman.
© 2025 John Wiley & Sons Ltd. Published 2025 by John Wiley & Sons Ltd.
Companion Website: www.wiley.com/go/Modeling_of_Physical_Systems/1e
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
634 Index

Constitutive relation 95 Electrical heater 66

Continuity of energy 8 Electrical thermal circuit analogs 133
Control effort 389 Electrically linear assumption 491, 492
Control system gain 391 Electrolyte 605
Control system output 388 Electromechanical relay 503, 543
Control system sensitivity 391 Electromechanical solenoid 493, 521
Controllability 426 Electromechanically balanced inverted pendulum 545
Controlled ablation probe 479 Electroquasistatic 488
Convection bond 579 Electrostatic voltmeter 490
Convective interface 523 Elevator lift model 285
Converging nozzles 581 Energetic structure 188
Convolution 282 Energy derivative relations 483
Corner frequency 321, 410 Energy domains 6
Coupled-gear system 80 Engine propeller system 37
Cramer’s rule 177 Engine torque-speed curve 39
Crane-belt system 551 Engine torque-speed Norton equivalent 106
Critically damped 2nd order response 249 Engine torque-speed Thevinin equivalent 106
Cruise control 404 Enthalpy 562, 577, 579
Cryopreservation agent (CPA) 605 Entropic dissipative structure 193
Cylinder leaking ethylene 620 Entropy 560
Entropy flow source 66
d Entropy flowrate 61
Damping ratio 246 Equation of state 126
Dc tachometer bridge 463 Equilibrium 15
Decibel (dB) 304 Equilibrium causality 159
Delay time 262 Equilibrium concentration of a polymer 604
Demagnetization curve 506, 507 Error constants 402
Dependent energy storage element 110, 122 Error function (control) 389
Dependent sources 210 Error rate damping 402
Derivative (dependent) causality 111 Error signal 376
Derivative and rate feedback 401 Euler equation (thermodynamics) 563
Derivative control with filter 397 Euler equations (rigid body) 200
Differential capacitive acclerometer 542 Euler junction structure 201
Diffusion dynamics 351, 597 Euler solver 72
Diffusion multiport R 602 Euler-Lagrange equations 534
Discrete PID 459 Eulerian turbomachine (ETM) 210
Displacement flux density 491 Explicit field 475
Displacement-modulated constraints 516 Explicit state equations 118
Dissipative (R) element 99 Extensive variables 558
Distributed-parameter model 349
Dominant closed-loop poles 379, 412 f
Drag cup 36 Faraday’s law 44f
Drebbel’s incubator 461 Fast Fourier transform (FFT) 326
Driving-point impedance 331 Feedback control 376, 388
Dual Lagrange form 530 Feedback control benefits 390
Dual-inertia rotational coupling 338, 345, 370 Feedforward control 394
Fick’s law 601, 603
e Field-controlled dc motor 288
Eignevalue-eigenvector problem 272 Field-excited dc motor 254, 339
Electret 491 Field-excited dc servomotor 86
Electret microphone 491 Field-excited generator 234
Electric analog meter 85, 148, 290, 366, 395 Final value theorem 398
Electric capacitor 42 First law of thermodynamics 560
Electric Coulombic force 40 First order frequency response 301
Electric flux density 41 First order step response 242
Electric inductor 45 First order system 239
Electric polarization 491 Fluid capacitance relations 56
Electric resistor 47 Fluid capacitors 56
Electric spark ignition system 86 Fluid clutch 292
Electric voltage 41 Fluid forces in a bend 209
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Index 635

Fluid forces on a wall 209 Ideal gas equations of state 566

Flux linkage 45 Ideal gas flow into a tank 589
Flyball governor 233 Ideal gas in a cylinder 130
Force feedback control 390 Ideal gas with resistance heater 131
Force transmissibility 318 Ideal gaseous adsorption on a solid 596
Forced equilibrium 156 Ideal sensor model 171
Four terminal parameters 331 Immittance matrix 332
Frequency response 299 Impact hammer testing 327
Frequency response basic factors 305 Impedance 296
Friction fire 88 Impedance bond graph 310
Full-state variable feedback 425, 432 Impedance matching 315
Impedance matrix 331
g Impedance scaling 314
Gain crossover frequency 410 Imperfect gases and solutions 614
Gain margin 410 Implicit equations 122
Gas diffusion in membranes 607 Implicit field 475
Gauss’s law 41, 492 Improper transfer function 380
Generalized conservative effort 516 Impulse response 253
Generalized controlled sources 211 Independent energy storage element 110
Generalized coordinates 509 Inertive (I) element 96
Generalized displacement 93 Infinite-product expansion 357
Generalized electromechanical machine (GEM) 495 Information state 171
Generalized momenta 516 Ingot quenching 87
Generalized momentum 93 Input types 403
Gibbs free energy 562, 597 Input variable 4
Gibbs internal energy 560, 565 Integral (independent) causality 111
Gimbaled rate gyro 202 Integrating amplifier model 212
Gyrator (G) element 98 Intensive variables 558
Gyroscopic coupling 58, 201 Internal model approach 438
Intrinsic induction 498
h Inverted pendulum on a cart 466
H-line 350, 355 Inverted spring pendulum 512
Half-car vehicle suspension 236, 297 Isentropic flow 580
Hamilton’s principle 534 Isentropic stagnation properties 578, 580
Hamiltonian equations 530 Isothermal assumption 124
Heaviside step function 244 Isothermal flow in a conduit 582
Helmholtz free energy 562, 597
High-pass filter 324 j
Holonomic constraint 511 Junction elements 100
Homogeneous first order function 561 Junction structure 112
Hydraulic capacitor 51
Hydraulic control valve 208 k
Hydraulic damper 224 Kedem-Katchalsky (K-K) equations 602
Hydraulic inertia 52 Kinetic coenergy 12, 509
Hydraulic piston 104, 172 Kinetic energy 12, 26, 34, 52
Hydraulic pump filter 322 Ktesibios water clock 377, 378, 538
Hydraulic resistance 53
Hydraulic servovalve 59 l
Hydroelectric power system 148 Lagrange subsystem bond graph 514
Hydrogen iodide reaction model 615 Lagrange’s equations 508, 517
Hydrostatic drive system 139, 235, 284 Lagrangian 508
Hydrostatic transmission 234 Lagrangian strain tensor 573
Hysteresis modeling 230, 498, 508 Lagrangian subsystems 509, 514
Laplace transform 171, 175
i Legendre transform 23f, 490, 562
IC multiport 493 Lennard-Jones potential function 192
Ideal (or dilute) solution 598, 601 Linear diffusion resistance 604
Ideal (thermodynamic) machines 129 Linear observer design 442
Ideal gas 565 Linear quadratic regulator 453
Ideal gas equation of state 127 Linear superposition 241
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
636 Index

Linear thermoelastic beam 575 Momentum flux 213, 523

Linear thermoelastic equations of state 574, 617 Motion sensor modeling 364
Linearization 165 Motor fan loading 222
Linearized hydraulic tank 385 Motor speed control 399
Link-cart system 197 Motor-belt-fan dynamics 287
Loading effect 381 Motor-driven torsional drive system 315
Local stagnation properties 578 Motor-driven web/film windup drive 153
Logarithmic decrement 256 Multicomponent gas modeling 592
Loop gain 413 Multicomponent ideal gas 593
Loop shaping 419 Multiport causality 481
Lorentz magnetic force 44 Multiport dissipative element 479
Loud speaker 366 Multiport element 475
LRC circuit 49 Multiport kinetic energy storage 478
Multiport potential energy storage 476
m
Mach number 578 n
Magnetic circuit elements 501 N-th order system response 272
Magnetic circuit models 498, 500 Negative resistance 184
Magnetic field intensity 498 Negative stiffness 184
Magnetic flux 45, 498 Neutral stability 182, 409
Magnetic flux density 498 Newton’s law of cooling 65
Magnetic levitation 228, 422, 469 Non-collocated sensor 418
Magnetic permeability 44 Non-minimum phase 329, 410, 422
Magnetic susceptibility 498 Non-unique constitutive relation 191
Magnetization density 498 Nonconservative effects 510, 514
Magnetomotive force 498 Nonholonomic constraint 511, 514
Magnetoquasistatic 488 Nonideal solution with reaction 609
Mass with heavy chain 214 Nonideal solutions 609
Mass-spring system eigenvalues 273 Nonlinear damping 28
Matched load 353 Norton equivalent 106, 220, 342, 507
Matrix convolution integral 279
Matrix exponential 275 o
Matrix Ricatti differential equation 453 Observability 427
Maximum energy product 507 Observer for spring deflection 444
Maximum overshoot 263, 411 Onsager forms 480
Maxwell reciprocity 189, 483, 563 Onsager reciprocal relations 601
Mayer’s relation 566 Open loop control 394
Mayr’s criteria for feedback control 377 Open thermodynamic cylinder with piston and heating 590
Measuring ratio of specific heats 621 Open thermodynamic multiport C 584
Mechanical compression of argon gas 567, 571, 618 Open-loop poles 405
Mechanical equation of state 573 Open-loop zeros 405
Membrane 597 Operational amplifier 211
Membrane capacitive element 605, 607 Optimal control 451
Membrane permeability 604 Order of a system 113
Membrane phenomenological coefficients 601 Osmosis 599
Microdialysis probe diffusion 624 Osmotic flow 606
Mismatched load 353 Osmotic pressure 599
Mixed causality on multiports 480 Output variable 4
Mixed-energy multiport 494 Overdamped 2nd order response 248
Modulated dissipative element 207 Overhead crane (gantry) 470, 471, 552
Modulated sources 210
Modulated transformer 196, 199 p
Modulation 195 Parallel compensation 397
Modulation state 196 Partial fraction expansions 265
Molar flow rate 577 Passivity 190
Molar flux 600 Passivity structure 190
Molar specific entropy 561 Path independence in energy-storing multiports 489
Molar specific internal energy 561 Peak time 262
Mole fraction 561, 593, 597 Pelton wheel 214
Moment of momentum 210 Pendulum stability 182
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Index 637

Perfect gas mixtures 614 Relative permeability 498

Perfusion of kidneys 605 Relative stability 185, 409
Permanent magnet 491, 500 Relativistic mass 27
Permanent magnet models 506, 507 Reluctance (magnetic) 499, 500
Permanent-magnet dc motor 85, 174, 381, 383 Resonant frequency 306, 411
Permeability of free space 44 Resonant peak magnitude 306, 411
Permeance (magnetic) 499 Response effect of zeros 269, 414
Permittivity of free space 40, 492 Retention factor 599
Permselective barrier 598 Reverse osmosis 600
Phase angle 300 Reynolds number 53, 158, 582
Phase crossover frequency 410 Rise time 262
Phase margin 410 Rocket variable-mass model 212
Phase velocity 354 Rod in tension 25
Physical feedback 378 Root locus 405, 418
Physical phenomenological relations 601 Rotational damper 36
Pi-section 335 Rotational inertia 34
PID controller 396 Rotational spring 32
Piezoelectric materials 491 Routh-Hurwitz stability criterion 185, 404, 408
Piezoelectric multiport model 492 Runge-Kutta solvers 72
Pipe flow resistive model 82
Pipeline characterization 362 s
Planar rigid beam 478 S-plane plot 268
Plant (control) 375 Salient-pole machine 495, 520
Pneumatic joint actuator 290, 362 Second order system 246
Polar plot 302 Semipermeable membrane 598, 606
Pole-zero cancellation 415 Sensitivity function 391
Pole-zero transfer function form 379 Separately-excited dc machine 205
Poles and zeros 268, 414 Sequential causality assignment procedure (SCAP) 112
Polytropic processes 128 Series compensation 397
Positive-displacement machine (PDM) 210 Series impedance 334
Positive-displacement pump system 123, 472 Settling time 263
Power bond 12 Shaker table dynamics 297, 303, 368
Power-conjugate variables 13, 92 Shear modulus of elasticity 34, 352
Pressure vessel 83, 141 Ship powertrain model 294
Pressure-based depth-rate detection 368 Shunt admittance 334
Pressure-compensated hydraulic pump 472 Single-input-single-output feedback control 376
Probability density function 168 Single-port bond graph element 94
Process (control) 376 Slider-crank 231
Propagation operator 350 Sliding friction 78
Proper transfer function 380 Solubility 604
Pseudo-bond graph 134, 577 Solute flux 603
Pseudo-bond graphs for simple diffusion 606 Solute permeability 603
Pure thermal storage element (PTSE) 67 Solution 597
Pushrod-lifter-valve system 147, 217 Solvent 597
Solvent permeability 603
q Source causality 110
Quartz yaw rate sensor modeling 552 Specific heat 63
Specific internal energy 127
r Speed of sound 578
Rack and pinion 142, 149 Sphere falling in fluid 73
Random signal 326 Spherical tank 82
Ratio of specific heats, gamma 128, 566 Splittng and mixing flows 579
Rayleigh dissipation function 510 Spreading pressure 595
RC impedance bridge 362 Spring 23f, 33
Reaction rate constants 614 Spring coenergy 23
Real solution 598 Stability and eigenvalues 183
Reciprocal two-port 335 Stability margins 409
Reciprocity theorem 335 Stable equilibrium 162
Redundant state variables 431 Stagnation enthalpy 580
Reflection coefficient (diffusion) 603 State variable 4
 Downloaded from https://onlinelibrary.wiley.com/doi/ by ibrahim ragab - Oregon Health & Science University , Wiley Online Library on [30/04/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
638 Index

State variable pairs 114 Translational damper 28

State-determined system 4 Translational mass 26
State-space forms 70 Translational spring 23
State-variable control 424 Transmissibility ratio 318
Static equilibrium 15, 159 Transmission line 349
Stationary equilibrium 15, 159 Transmission matrix 312, 332
Statistical entropy 559 Transport delay 385
Statistical linearization 167 Triangular pulse 326
Steady-state error 403 Tustin’s approximation 459
Steady-state pmdc motor model 223 Twin-T RC network 371
Stefan-Boltzmann law 65, 564 Two-car train 292, 417, 471
Stochiometric coefficients 594, 609, 612, 613, 614 Two-dimensional spring-mass system 484, 518
Storm sewer system 54, 430 Two-port cantilever beam 477
Superconducting coils 546 Two-port heat transfer R 125
Surge admittance 350 Two-port holonomic constraint 199
Swept-sine testing 324 Two-port interconnections 341
System 4 Two-port models 330
System identification 324 Two-port non-holonomic constraint 200
System stability 181 Two-port parameters 331
System variable 4 Two-port thermodynamic C element 126
Two-story system 296
t
T-section 335 u
Tait-Steele falling chain 524 Undamped 2nd order response 256
Tank filled by controlled source 215 Undamped natural frequency 246
Tank-float multiport system 485, 538 Underdamped 2nd order response 250
Taylor series linearization 165 Uniform transmission line (U-line) 351
Tea leaf caffeine diffusion 623 Unit impulse input 244
Tetrahedron of state 93 Universal gas constant, R 564
Thermal capacitor 61 Unstable equilibrium 162
Thermal conductivity 63, 64 Unstable poles 269
Thermal convection heat transfer 65 Unstable valve 185
Thermal diffusivity 356
Thermal inertance 562
v
Thermal internal energy 60, 561
Vacuum cleaner system 139
Thermal motor model 88
Valve-controlled pneumatic cylinder 622
Thermal pmdc motor model 148, 223
Van der Waals gas 568, 571
Thermal radiation 65
Van’t Hoff relation 600
Thermal two-port C causalities 563
Variable-mass systems 213, 523
Thermodynamic accumulator 586
Variable-reluctance devices 506
Thermodynamic multiport C 561
Vehicle heading angle control 466
Thermodynamic temperatue 559
Thermoelastic systems 573 Vibration absorber 319
Thermostatics 558 Viscous damping 28
Thevenin equivalent 106, 152, 220, 341, 507 Voltage controlled voltage source 211, 347
Time and frequency domain relationship 410
Time constant 241 w
Time to first peak 262 W-line 351
Time-difference equation 280, 359 Wall-climbing robot 622
Time-domain specifications 261 Water-buoy system 361
Toroid inductor 46 Watt’s flyball governor 462
Torque meter 285, 321 Wave dynamics 351
Torque-driven wheel 150 Wavelength 354
Torsional shaft 33 Wheatstone bridge 217
Total volume flux 603 Word bond graph 9, 13, 92, 386
Transfer function approximation 328
Transfer functions 175 z
Transfer impedance 331 Zeros 268, 330
Transformer (T) element 98 Ziegler-Nichols tuning 424