Mathematical Modelling and Simulation

of Physical Systems

Lecture Notes in Course

TMR4275 Modelling, Simulation and Analysis of

Dynamic Systems

by

Eilif Pedersen and

Hallvard Engja

Department of Marine Technology

Norwegian University of Science and Technology

Trondheim, August 2014

Contents

PREFACE i

1 INTRODUCTION 1
1.1 Systems 1
1.2 Models 4
1.3 Modelling 5
1.4 Simulation 8
1.5 Summary 10
REFERENCES 10
PROBLEMS 11

2 Bond Graphs and Basic Element Models 13

2.1 Bond Graph Basics 13
2.2 Definition of Generalised Variables 15
2.3 Basic Bond Graph Elements 19
2.3.1 Basic 1-port Elements 20
2.3.2 Basic 2-port Elements 25
2.3.3 Junction Elements 28
REFERENCE 32
PROBLEMS 33

3 System Modelling 35
3.1 Mechanical Systems 35
3.1.1 Mechanical Sources 36
3.1.2 Mechanical Capacitor Elements 36
3.1.3 Mechanical Inertia Elements 43
3.1.4 Translational Resistors 44
3.1.5 Mechanical Transformers 47
3.1.6 Modulated Mechanical Transformers 49
3.2 Mechanical System Modelling 50
3.2.1 Systematic Construction Method 51
3.3 Torsional System Modelling 56
3.4 Electrical Systems 60
3.4.1 Electrical Sources 61
3.4.2 Electrical Capacitor 62
3.4.3 Electrical Inertia-Inductor 62
3.4.4 Electrical Resistance 63
3.4.5 Electrical Transformer 64
3.5 Electrical System Models 65
3.6 Hydraulic Systems 70
3.6.1 Fluid Flow Sources 70
3.6.2 Fluid Capacitors 75
3.6.3 Hydraulic Actuators 79
 3.6.4 Hydraulic Resistance 82
3.6.5 Fluid Inertance 84
3.7 Hydraulic Systems Modelling 86
3.8 Thermal Systems 92
3.8.1 Conduction 93
3.8.2 Convection 95
3.8.3 Radiation 96
3.8.4 Thermal Storage Elements 97
3.9 Thermal System Modelling 98
PROBLEMS 103

4 Development of State Equations 109

4.1Causality Assignment for Multiports 115
4.2Assigning Causality to System Models 122
4.3Development of state equations 126
4.4Differential Causality. 134
4.5Incomplete Causality Assignment - Algebraic loops 140
4.6Non-linear models 148
4.7Output Equations 151
4.8Bond Graph to Block Diagram Conversion 156
4.8.1 Block diagrams 156
REFERENCES 166
PROBLEMS 166

5 System Analysis Methods. 169

5.1 Linearisation of Non-linear Systems. 170
5.2 Solution of Linear State Equations. 174
5.3 Natural Response of First-order Systems 175
5.4 Natural Response of Second-order Systems 180
5.5 Forced Response 192
5.5.1 Step Function and Sinusoidal Sources 193
5.6 Step Response of a First-order system. 195
5.7 Step Response of a Second-order System 197
5.8 Forced Response of Sinusoidal Inputs 203
5.8.1 General comments 208

6 Numerical Solutions 209

6.1 Method of solution 210
6.1.1 Computational formula and interpretation of the method 210
6.1.2 Errors and Stability 213
6.1.3 The improved Euler method 220
6.1.4 The modified Euler method 221
6.2 Explicit Runge-Kutta methods 222
6.2.1 Some explicit Runge-Kutta methods 224
6.2.2 Stability of ERK-methods 224
6.3 Stiff systems 226
6.3.1 Implicit Runge-Kutta methods 227
 6.4 Multistep methods 229
6.4.1 Error and order of linear multistep methods 230
6.4.2 Stability of linear multistep methods 232
6.5 High-order systems 234
6.6 Practical considerations 236
REFERENCES 238

7 Multiport fields 239

7.1 Energy storing fields 239
7.1.1 C-fields 239
7.1.2 Causal considerations for C-fields 245
7.2 I-fields 247
7.2.1 Causal considerations for I-fields 250
7.3 Mixed Energy-Storing Fields 251
7.4 Resistive fields 252
7.5 Multiport transformers 253
REFERENCES 256
PROBLEMS 257

8 Multibody dynamics 259

8.1Introduction 259
8.2Multidimensional dynamics 259
8.3A modelling procedure 270
8.4Lagrangian or Hamiltonian IC-field modelling 278
8.5General rigid body bond graphs 284
8.5.1 Application to Rotordynamics 291
REFERENCES 297
PROBLEMS 299

9 Thermodynamic systems modelling 307

9.1 Basic principles 308
9.1.1 Conservation equations 309
9.2 Bond graph representation 312
9.3 Basic compressible fluid elements 315
9.3.1 Thermofluid accumulator 315
9.3.2 Pseudo-bond graph restrictor 318
9.4 Thermal Incompressible Fluid Models 320
9.4.1 Basic element Set 320
9.5 Multicomponent two-phase thermodynamic systems using pseudo bond graphs 327
9.5.1 Fundamentals 330
9.5.2 Modeling multicomponent two-phase systems 332
9.5.3 Constitutive laws for multicomponent mixtures 342
9.5.4 Calculation of thermodynamic data for multicomponent mixtures from a funda-
mental relation. 345
9.6 Summary 357
REFERENCES 358
PROBLEMS 360
Index 363
 PREFACE i

PREFACE

For engineers and physical scientists working in system analysis, design or automatic con-
trol, the development of a computable mathematical model simulating physical processes is a
fundamental task. In the last few years, personal computers have become so powerful and
available that it is now possible for most people to use them to solve complex simulation
problems easily and efficiently. An important assumption for utilizing simulation as a prob-
lem solving tool is that a mathematical representation or model is available of the problem at
hand. Thus, the well-learned engineer should not only know how to use the computer, but
also know how to abstract a physical reality to a mathematical model.

This book is thus for those who want to learn how to construct mathematical models of phys-
ical systems and how a model is converted from a passive device into an active device
through analytical solution and computer simulation. It is written for people who are simply
curious what simulation can do to those professionally concerned, such as practising engi-
neers, undergraduate and graduate students. This means that it is written for those interested
in the subject called dynamic systems, which has a prominent place in technology, although it
seldom is emphasized in university curricula. The purpose of this book is therefore, to
present a unified approach to the subject of modelling of physical systems such that the
reader can develop competence in building mathematical models and in using them to evalu-
ate system performance under varying rules of operation, varying parameters or varying
structure. This work is devoted to the art of modelling dynamics found in design and opera-
tion of various technical processes and systems, in order to persuade the reader that mathe-
matical modelling can be the framework of an economical, quantitative, and scientific
approach to solve technical problems in industry. From this stems important characteristics
of the book

• a unified approach to the construction of dynamic models from different energy

domains using the concepts of bond graphs
• state-space representation of all the obtained mathematical models

The text has been written to demonstrate that there is a subject called modelling, which can
be learned and needs to be learned. It is an important subject because the skills of modelling
are so closely related to the more general skills of problem solving. Learning to model is
bound up with learning to solve problems and to think imaginatively and purposefully. A lot
of engineers are building models and using them to make important decisions and predic-
tions, without ever having learned to criticize a model or distinguish a good model from a
bad one.
ii PREFACE

It has been said that writing imposes a discipline or thinking, modelling takes the process one
step further by quantifying our thinking and establishing interaction of related thoughts.

System dynamics is concerned with the fundamental aspects common to the unsteady state
behaviour of systems and as such is fundamental to a wide variety of problems in engineer-
ing. Although, applications are diverse, there are basic similarities between response and
methods of analysis.

By adapting a generalized approach to system modelling we learn one technique which is

applicable to many different system types which traditionally are taught in different courses
as if they do not have any thing to do with each other.

Therefore, the subject of this text is designed to allow an engineer to recognize that the
underlying principles are the same for widely diverse applications and can therefore, be con-
sidered in a similar and unified fashion.

This text is therefore a unified engineering treatment of mechanical, electrical, fluid and ther-
mal dynamic systems, intended to stimulate the analytical approach to system engineering by
removing obstacles and presenting a unified procedure for developing mathematical models
of a great variety of physical systems. The process of dynamic investigations, its fundamental
concepts, its basic building blocks, and its applications are the subjects in this book. The uni-
fying approach used to construct dynamic models of physical systems in this book, is a
graphical notation called bond graphs.

A bond graph is a graphical notation for representing physical and engineering components
and the actions and interactions, static and dynamic, taking place within an energetic system.
An important aspect of bond graphs is its organized approach to modelling. They allow all
types of physical systems to be expressed with just a few basic elements, and provides rules
and procedures for studying the interaction within and between subsystems. A major virtue
of the bond graph representation is that a model may be developed in successive steps of
increasing detail from virtually a purely descriptive model to a completely quantitative one.
Even computers can read bond graph descriptions and can automatically derive equations
and simulate systems.

Our goal has been to present a useful state-of-the-art application-oriented tool for studying
physical dynamic systems. We have found, and we hope you will too, that systems dynamics
is fun. You will have to draw on your knowledge from physics, mathematics and all areas of
engineering. Quite a challenge! In this text we have tried to present a logical development of
abstracting and analysing dynamic systems which may be summarized as follows:

• to learn to construct realistic and reasonable models of physical systems

 PREFACE iii

• to learn techniques for solving the equations that make up the mathematical model
• to gain an appreciation for the dynamics and the time-dependent response of physical
systems
In an attempt to show the broad applicability of the bond graph method, numerous and
diverse examples is worked out in detail. The physical interpretation of system response is
emphasized.

The text attempts to put within one cover the techniques and principles of bond graphs and
their utilization in solving important engineering problems. A certain amount of original
material has been included, although, most of the content has been derived from the litera-
ture.

Finally it must be said that no book gets its final form through the efforts of its authors alone.
There are always many others deserving credit for stimulation, inspiration, support and help.
In the case of this book we wish to extend our most heartily gratitude to the bond graph
brotherhood and in particular, Professors Dean Karnopp and Don Margolis at University of
California, Davis. A special thanks to Berit Ulleberg for helping us with typing and retyping
the manuscript.

Trondheim, August 2003.

Eilif Pedersen Hallvard Engja

Preface to 2nd edition

In the new edition three new chapters has been added covering an introduction to multibond
or vector bond graphs, multibody dynamics and thermodynamic systems modeling. The
added features closely follows the changes in topics of the lecture series which has evolved
during the last years. In addition the figures has been given a brushup, numerous problems
has been added and finally numerous small and larger errors and spelling mistakes has been
corrected. Thanks to Ola Pedersen for updating the figures. We hope that this updated ver-
sion will introduce, extend and demonstrate the advantages of the bond graph method in even
more and challenging applications and engineering diciplines.

The 2010 version has been slightly updated and numerous errors has been removed.

The 2014 version has been revised somewhat and some errors has been removed.
iv PREFACE
 INTRODUCTION 1

Chapter 1

INTRODUCTION

As technology advances, the problems the engineering community must face are getting
more sophisticated day by day and it is safe to predict that this trend will continue. Because
of the increasing complexity of man-designed systems, and the need to cope with this com-
plexity has caused the climate of engineering activity and education to shift towards an
increased demand for quantitative methods of analysis and deeper understanding by the stu-
dents and practicing engineers.

As a result, current engineering problems require deeper insight into the problems and which
in turn demands more precise formulation and exact solution. All this points to a basic and
mathematically rigorous approach to problem solving. This analytical approach to engineer-
ing problems allows both considerably wider scopes in investigating alternative designs and
more efficient operation. In addition, as systems become more complex, incorporating ever-
increasing degrees of automation, there will be greater need for the analytical approach to
problems associated with their design and operation. Theoretical engineering analysis might
well be defined as an orderly method of attacking new situations by considering fundamental
principles with a well ordered analytical thought process.

The concept of systems has become increasingly important in engineering design and analy-
sis. Therefore, the object of this text is to develop familiarity with the process of investigating
the behaviour of physical systems due to changes in physical structure, input stimuli and
parameter values.

1.1 Systems
Webster’s Dictionary defines a system as: a collection of objects united by some form of
interaction or interdependence in an orderly working totality. From this then we can define a
physical system as an arrangement of interacting physical devices which functions as a whole
and which store and transform energy. It is the connection, or more important the inter-con-
nection between the parts which leads to the complex nature of systems.

The boundary of a system is chosen for convenient conceptual separation of the system from
its environment. Sometimes, what is considered a system on one scale may be only a compo-
nent of a much larger system. How to determine how small a system is and when a system
2 INTRODUCTION

becomes a component is one of those philosophical questions which really has no definite
answer.

Due to its importance, this book will place emphasis on system aspects. That means it is no
longer sufficient to pay detailed attention to the design of the various elements which go to
make up an overall system, but it is important to consider the elements in relation to each
other. This is what we call a system viewpoint. The essence of what may be called the system
viewpoint is to concern oneself with the operation of a complete system rather than with just
the operation of the component parts. That is to make sure that not only will the parts of the
system work, but also the system as a whole will perform its intended function. Even if each
element or subsystem is optimised from a design or operational viewpoint, the total system
performance may be sub-optimal owing to interactions between the parts. That is, perfectly
good components can be assembled into an unsatisfactory system.

A great deal of the design of engineering systems can be carried out based on steady state
analysis, but dynamic behaviour is increasingly important as systems become more complex
and interactive. In particular optimised automatic control can only be achieved when the
dynamics of the system to be controlled is sufficiently well understood. Prediction of the
dynamic behaviour of systems before they are built is of vital importance and is the key to
confidence that they will work properly in the environment they are intended for. Thus, there
is increasing interest in the field described as dynamic systems, which means the study of
how things change with time, and the forces that cause them to do so. More perhaps than any
other field, the study of dynamic behaviour links the engineering disciplines and has become
a keystone to modern technology.

Generally system studies can be considered to consist of three main phases, namely

• establishing system specifications

• system design
• system analysis

In the synthesis or system design mode, the object is to develop a system concept which is
believed to satisfy a set of specified or implied requirements. The design parameters are cal-
culated from the specified outputs. The system analysis mode aims to understand how a pro-
posed or existing system operates, that means predicting the behaviour at the design stage in
order to test a concept for its capacity to satisfy demands placed on it. The analysis mode is
characterized by the fact that all system inputs and design parameters are specified and it cal-
culates the output or response. In a design study, the analysis mode is used to perform design
evaluations iteratively for making case studies. The difference between the design mode and
the analysis mode is illustrated in Fig. 1.1.
 Systems 3

Stationary System Output

inputs unknown specified
(known)
Design mode

Input System Output

specified known unknown

Analysis mode

Fig. 1.1 Illustration of design and analysis modes

This shows that the analysis mode is the link between design and operation.

System analysis, can in general, be divided into two broad aspects:

• Modelling theory - the process of generating a model of the system from a defined
system structure.
• Behavioural theory - the process of analysing the solution characteristics of the sys-
tem model to simulate the physical behaviour of the system as a function of a change
in parameters, structural features and inputs and compare the predicted performance to
the specifications.

Except for simple systems, the best resource for effective system modelling, analysis and
evaluation exists when the human investigator effectively couples his or her unique capabili-
ties with a computer. Personal computers have become so powerful and available that it is
now possible for most people to use them to solve large complex system analysis problems
easily and efficiently. At the same time it must be realized that, in general, a computing
machine cannot formulate problems and choose the proper method of solution. In fact, the
computer must be given specific instructions on the procedure to be followed, after which it
carries out the specific operations very rapidly. Less we are overly impressed with the present
and future capabilities and applications that are offered by computers, it is important to rec-
ognize that even advanced computers cannot provide imagination and creativity in model
conception and abstraction. Furthermore, the ability to evaluate, compare, and make judge-
ments about real system and predicted model behaviour, and then to innovate and remodel in
order to produce improved and more accurate system models are human enterprises. Hence
the entrance of the computer into the engineering arena only reinforces the need for rigorous
and complete treatment of the process of abstracting a physical reality into a model that is
4 INTRODUCTION

adequate and suitable for computer evaluation. This means that a model is the central ele-
ment in any system analysis.

1.2 Models
There is no single accepted meaning of the word model. Its meaning varies from one branch
of learning to another. In this book our usage of the word model is given by the following
definition - A model is a representation of a system. From this we can generalize the defini-
tion of a model to be a representation that mimic another object under study and should con-
tain only those features of the system which the model builder feels are relevant and
significant for the goal in mind.

Models are therefore not reality, and a model no matter how complex, is only a representation
of reality and should never be confused with it. Models can predict nothing with certainty.
Their proper function is to develop judgement, not a substitute for it. In summary, always
remember that models are, in general, partial in the sense that they never represent the entire
system. One therefore, should keep in mind what has been omitted and what assumptions
have been made in constructing the model. So, what is being called a model is actually a
reflection of the modellers understanding of reality, its components and their interrelations.
One of the things that students and teachers of modelling finds difficult to accept is that there
is no unique model and no unique solution to a real world problem. However, it is fair to state
that a good model will represent all aspects of the simuland that bear on the purpose for
which it is to be used, not more. The concept of representing some system, or idea with a
model, is so general that it is difficult to classify all the functions models fulfil. At least, five
legitimate and common uses can be recognized:

• an aid to thought
• an aid to communication
• and aid to training and instruction
• a tool of prediction
• an aid to experimentation

Broadly speaking, models can be classified into two main types or categories:

• material or physical models

• non-material or formal models

As the name suggests, material models have a physical existence with a direct relation to the
real-world problem. For example, a scaled model of a system is well known to engineers
since they are used extensively in laboratories to generate information concerning specific
attributes of the physical process.
 Modelling 5

In contrast, non-material or formal models are abstract representations. A mathematical

approximation to the physical attributes of interest is an example of an abstract model. In this
text we restrict our attention to dynamic, lumped-parameter models which arise naturally in
many common engineering areas.

Creating a model or modelling can thus be defined as the process of translating the most
important aspects of a real or proposed engineering system from its real environment to an
expression by which it is more conveniently studied. It is the first and most important step,
but also one of the most difficult tasks in applying quantitative analysis to the study of sys-
tems.

1.3 Modelling
Modelling is often thought of simply as a process involving the application of physical laws
and principles to establish a set of equations which describe a given system. Looking through
the literature on the subject it appears that it is not simply, but reveals the enormous diversity
of approaches used and hence the difficulties associated with developing relevant computa-
ble models and in a form which engineers can understand. In spite of this fact, the means for
constructing a suitable model is often given very little attention in the literature or in engi-
neering education. It usually never states the teaching how to do. The teaching of how to do it
is not the same as teaching of models. It is not at all clear that the teaching of models by
exposing the inexperienced to the ad hoc ways of doing it, contributes much to the develop-
ment of creative model building. Someone have said - “If your students are hungry, don’t
give them fish, but teach them how to fish”. It can be argued that all intellectual activities
involve modelling of sorts since we always make conceptual simplifications in order to com-
prehend a real-world object. The construction of models with an acceptable validity requires
a profound understanding of the system being modelled and the area or discipline to which it
belongs. It is important to realize, that the modelling process is by no means to be divorced
from the physical process to which it is applied. Physical system modelling differs from its
forerunner applied mathematics by its emphasis on the physical system to be modelled and
not on the mathematical solution technique.

Without doubt the most important result of developing a model of a system is the understand-
ing that is gained of what really makes the system “tick”. The insight enables us to strip away
from the problem the many extraneous confusing factors and to get to the core of the system.
Modelling therefore, is an integral part of the practice of engineering and it is very important
to human thought processes and to our ability to solve problems. Properly done, model build-
ing forces us to organize, evaluate and examine the validity of thoughts. Model building is
the cornerstone of any scientific activity. The ability to model will grow with experience and
this text will hopefully form a foundation for such growth.
6 INTRODUCTION

Many people think the use of models is recent. Modelling is not new, the conceptualization
and development of models have played a vital part in mankind’s intellectual activity since
we began to try to understand and manipulate the environment. We have always used the idea
of models in attempting to represent and express ideas. As a matter of fact, the progress and
history of science and engineering are most accurately reflected in the progress of man’s abil-
ity to develop models of natural phenomena, ideas and objects. Thus, the art of modelling
consists of the ability to analyse a problem, abstract from it essential features, select and
modify basic assumption that characterize the system, and then enrich and elaborate the
model until a useful approximation without undue complexity results. Certain principles of
approach can be learned and it is the intention to illustrate this as we progress through the
book.

We will attempt to set the modelling activity within a methodological framework, gradually
developing the model-building process. However, modelling poses some formidable peda-
gogical problems. Therefore, many different approaches have been advocated for learning
the modelling process. In some cases, models are described and the descriptions are usually
justifications of the model and not the process by which one creates the model. In other cases,
those who do the teaching have a predisposition towards the mathematical manipulation
rather than the creation of the model. The selection is not easy. In what follows we are pre-
senting you some arguments why we have made the particular selection of a modelling
method for the entire text.

An important aspect of modern engineering systems is their great diversity and they often
include several different physical domains that interact. To adequately model such systems
we as engineers must be prepared to deal with a number of physical domains. The modelling
phase of such multi-disciplinary systems is best done by using an interdisciplinary approach
valid for all kinds of physical systems. It would be convenient if we could refer to compo-
nents of a system generally without having to identify the physical medium such as mechani-
cal, electrical, fluid, etc. This indicates a need for a transparent language to be able to think,
to reason and communicate in a multi-disciplinary manner.

A very powerful realization of such an approach is found in the bond graph language. It
forms a unified conceptual and operational framework for the study of a very wide variety of
system types and thus provides a general paradigm for modelling of dynamic systems based
upon energy and power flow.

The bond graph language is an extension of the electrical network formalism, describing
physical systems as ideal physical elements interconnected through a network of power
exchange bonds. Letters represents the components, and a line resembling the diagrams used
in chemistry depicts the bonds; hence the name bond graphs. They were invented by Profes-
sor H.M. Paynter at MIT in the late 1950s [1], and brought into applications [2] by his then
 Modelling 7

PhD-students, now Professors at University of California, Davis and Michigan State Univer-
sity, Dean C. Karnopp and Ronald C. Rosenberg, and have been under constant development
ever since. These three persons are inseparable as the originators of the bond graph language.
In [3] Paynter states the birthday of bond graphs writing:

“So it was that on April 24, 1959, as the writer was about to give a seminar lecture at
Case Institute (now Case-Western) on “Interconnected Engineering Systems”, he awak-
ened early that morning with the 0,1-junctions somehow finally planted in his head! Thus
on that date the Bond Graph system was complete and constituted a formal discipline”.

Today bond graphs are recognized as one of, if not the most promising method for modelling
of dynamic systems. Bond graphs are used by a significant fraction of mechanical and elec-
trical engineers, and even by some life scientists. The method has been extended to model
heat flow, compressible fluid flow, chemical reactions and into modeling for control, to name
a few. It has been promulgated by thousands of researchers and professional conference and
journal papers. Since 1991 the biennial International Conference on Bond Graph Modeling
(ICBGM) [4] has been one of the major contributors to the dissemination of the Bond Graph
Method.

There are many advantages in using the bond graph method in engineering activity and for
teaching dynamic system modelling. The most important ones are:

• One of the fundamental advantages of using bond graphs is that we can use the same
symbolism to represent the power interaction in a large selection of physical systems.
• On a graphical form bond graphs display the energetic structure of complex systems
with several energy domains in a way which is close to what we may call the physics
of the system.
• A physical based sign convention can be shown directly on the graph, which is impor-
tant when interpreting numerical results from simulations.
• The method is equally applicable for linear and non-linear systems.
• A unique feature of bond graphs is the display of causality on the graph. That is, it
indicates which variable for an element or module is the independent variable or input
and which variable is dependent or output variable.
• The bond graph method gives an algorithmic procedure for converting the graph into
mathematical equations.
• It can be directly entered and processed by a computer.

Due to its unified and systematic approach to model development, we will in this book use
the method of bond graphs as our methodological framework for all our modelling activity.
8 INTRODUCTION

1.4 Simulation
Once a model is constructed, it has to be converted from a passive device into an active
device which can be used to infer the behaviour of the actual object by analytical use of the
object model. That is to perform a quantitative analysis, which means finding the explicit
solution satisfying the given mathematical formulation. Quantitative analysis can be divided
into two types:

• Analytical methods
• Computational methods.

In the analytical method the solution to a given mathematical formulation is obtained in

terms of an analytical expression involving the parameters and the independent variables of
the formulation. Only for a small class of mathematical formulations is it possible to obtain
exact analytical solutions. In contrast, computational methods of analysis using a computer
yield solutions for specific numerical values of the parameters. These methods of analysis are
applicable to a wide class of mathematical formulations and system size. This is what we call
simulation.

Simulation is a very wide-open and somewhat ill-defined subject. According to Webster’s

Collegial Dictionary, to simulate is to feign, to attain the essence of something without the
reality. With some trepidation, we shall in this book define simulation as follows:

Definition: Simulation is the process of conducting experiments with a model of a real sys-
tem for the purpose to understand the behaviour of the system or of evaluating various strat-
egies for the operation of the system.

Once the performance requirements for a system are determined and a tentative design based
on these requirements is complete, there are always nagging questions such as:

• Will the actual system perform as required?

• What will happen if the operating conditions change?
• By how much does a critical factor have to change before a critical response results?
• How sensitive is one factor to changes in another?

By contrasting some of the major uses of computer simulation, to alternative methods of

answering the questions above, one can begin to get a feel for its usefulness. In addition to
providing a flexible and frequently less expensive vehicle for answering the questions high-
lighted above, simulation has other benefits. It can be used for design studies, and multiple
runs using various configurations can be used to home in on design parameters. “What if”
studies are easily and fruitfully done via simulations. The effects of constraints, whether due
 Simulation 9

to physical properties or safety policies can easily be determined. Simulation has a general
applicability and is the natural link between design and operation. Simulation must therefore,
be considered as a central theme in future engineering education and practice.

Simulation is therefore, one of the most important and useful tools available to those respon-
sible for the design and operation of complex processes or systems. It allows the user to
experiment with systems which would be impossible or impractical otherwise.

Also, and important use of simulation is learning. There is no better way to learn the behav-
iour of a system than to experiment with a model of a real-world system. Sun Yat-Sen
“Father of the Chinese Republic” is reported to have said:

“To understand is hard.

Once one understands,
action is easy”.

Simulation is a powerful tool for alleviating the hard problem by improving the understand-
ing. Another important advantage of simulation is that it permits sensitivity analysis by
manipulation of system parameters and variables. By experimenting on a model of a complex
system, we can often learn more about its internal interaction than we could through manipu-
lation of the real world system itself owing to our control of, and the measurability of the
model organisational structure, and the ease of parameter variations.

The rapid increase in computing power available to the average user of simulation and the
advent of user friendly simulation programs has made it possible for engineers to design and
optimise systems, but nagging questions about the accuracy and reality of simulation results
remain. The ideal case in which simulation results can be verified against at least some par-
tial experimental results is to our experience more the exception than the rule. Standard prod-
ucts which evolve slowly over time can certainly be epitomized using simulations verified
through comparison with experience on previous versions of the product, but we will argue
simulation has most to contribute in the preliminary design of conceptually new systems.
Thus simulation is supposed to expedite the design of a proof of concept prototype and only
at a much later stage would it be possible to do any detailed verification studies. How can it
then be, that unverified simulations, which in some mathematical sense may be very inaccu-
rate, can yet be used fruitfully in the design of new systems? The answer lies in the concept
of a model of a system. In any case the model will only attempt to represent partially and in
approximation the behaviour of the real system. There is a wide range of models which could
produce useful insight to the system behaviour, as well as a wider range of models which
would produce misleading results. Therefore, it is clear that an inaccurate simulation of the
equations of a useful model is better than an accurate simulation based on equations from a
poor model. The basic news is that a lot of people are building models and using them to
make important decisions and predictions, without ever having learned to criticize a model or
10 INTRODUCTION

distinguish a good model from a bad one. Model building and simulation is thus, in effect,
techniques for bringing order and predictability out of an apparent chaos of multiple varia-
bles.

1.5 Summary
Modelling, mathematical analysis and simulation as a problem solving methodology can be
characterized at three distinct physical levels such as:

• Overall system
• Sub-systems
• Physical processes

At each physical level the problem solving methodology consists of three important phases:

• Modelling
• Numerical analysis and algorithm selection
• Conducting experiments with a model

Thus, for modelling and simulation to be an effective problem solving methodology, it is

required that the analyst have an insight into all three of the above mentioned phases. Only in
this way can there be a degree of assurance that the analysis of outputs constitute meaningful
and useful solutions.

The process of dynamic investigation - its fundamental building block and its applications -
is the subject of this book. The emphasis in the succeeding chapters is on the formulation of
models and the use of those models in determining response using both analytical and com-
puter simulation techniques.

REFERENCES
1. Paynter, H. M., Analysis and Design of Engineering Systems, The M.I.T. Press, Cam-
bridge, Mass., 1961.
2. D. C. Karnopp, D. L. Margolis, R. C. Rosenberg, System Dynamics - Modeling and Sim-
ulation of Mechatronic Systems, 3rd ed., A Wiley Interscience publication, 2000.
3. Paynter, H. M., An Epistemic Prehistory of the Bond Graphs, in P. C. Breedveld and G.
Dauphin-Tanguy (Eds.), Bond Graphs for Engineers, Elsevier, Amsterdam, pp. 3-17,
1992.
4. SCS - The Society for Computer Simulation International, Proceedings of the Interna-
tional Conference on Bond Graph Modeling and Simulation, 1991 - continuing
 Summary 11

PROBLEMS

1-1 A ship is unloading its oil cargo to a tank on shore. The pumps used for this operation is
two centrifugal pumps operating in parallel. If you where to construct a dynamic system
model of this system to help plan the time it takes to unload the ship, how may and what
state variables do you think you need for your model? What input variables and parame-
ters do you want to use for your model

1-2 Suppose your tanker was to carry low quality crude oil on its next voyage, and in order to
save fuel your company would like to limit the use of cargo heating during the journey.
You was given the responsibility for this operation and decided to develop a dynamic
model to analyse the case. What subsystems would your model consists of and what math-
ematical equations would you select to describe the physics of each of the subsystems?
What might you consider for input, outputs and state variables for a simple dynamic mod-
el to be able to predict the average and lower temperature of your cargo during the jour-
ney?

1-3 One of the speed boat teams participating in the races for World Championship would like
to have the whole bridge including the captains chair mounted on a vibration free frame
to allow the captain a cup of tea while cruising in 80 knots. Discuss the development of a
mathematical model for analysis of the dynamics of the tea-cup. What physical laws can
you use to describe this case? What subsystems would you introduce to describe the total
system? What is the links between the subsystems? What inputs, outputs and state varia-
bles might prove useful in this case?

1-4 A power station consists of the water reservoirs, the water tunnels, the expansion channel
and the water turbine connected to the generator which delivers electricity to the electric
grid. Discuss the development of a mathematical model for the dynamics of the system
when the load on the generator changes quickly. Include the control system of the gener-
ator. What subsystems can you identify for this case? What is the physical and mathemat-
ical links between them? What are the physical laws that you would like to use to derive
the model equations? What is the inputs, outputs, state variables and parameters that you
will suggest to use? How would you solve the model equations that you have developed?
12 INTRODUCTION
 Bond Graphs and Basic Element Models 13

Chapter 2

Bond Graphs and Basic Element Models

In this chapter we will begin the formalisation of a general system of notation for bond
graphs as a general approach to modelling interacting energetic systems based on identifying
the energetic structure in a system. The fact that interacting physical systems transmit power
is used to establish a uniform classification of the variables associated with power and
energy. These variables are then used to define a small number of ideal elements that make it
possible to describe in principle all possible lumped parameter models of physical systems.

2.1 Bond Graph Basics

The most important feature of physical systems theory and the method of bond graphs are the
concept of what we call reticulation, that is, a net like formation or decomposition of a sys-
tem into elemental physical properties. In this approach, the assumption is that it is possible
to separate and concentrate certain properties of an object, and then describe this object as a
system of interrelated properties or interconnected basic ideal elements. With the adjective
basic we mean an element that can be used to display a basic physical phenomenon and that
the element cannot be further reduced or subdivided into other elements. Thus, the basic ele-
ments form the building blocks for constructing component and system models. We use the
adjective ideal to imply that a basic element does not exist as a realisable physical compo-
nent. However, ideal elements can be interconnected to represent actual systems.

Thus, the bond graph approach to modelling focuses on interconnecting ideal basic elements
in such a way that the interconnected elements will be able to predict the behaviour of actual
devices and systems within acceptable limits of accuracy. It is this ability to model actual
systems with ideal elements that makes bond graphs so useful to engineers.

B
A

Fig. 2.1 Schematic representation of a system

14 Bond Graphs and Basic Element Models

Thus, a system can be defined in the context of bond graphs as a collection of interacting
basic elements, united at a finite number of interfaces as represented schematically by the
linkage system, shown in Fig. 2.1. With the proper length of the linkages this system can be
used to convert rotational motion of link A to a rocking motion of link B. The fact that the
behaviour of real physical systems which engineers design, analyse and understand is con-
trolled by the flow, storage and interchange of various forms of energy. This means that the
unsteady behaviour of dynamic systems is governed by the way in which energy is handled
by the system. Thus a suitable unifying concept, which can be used to characterise systems in
various engineering disciplines, is energy. The components, which make up a system, may
thus be thought of as energy manipulators or operators. The components or energy operators
may be grouped into groups constituting a system. This then means that when elements inter-
act they exchange energy with each other through these interaction points which in bond
graphs are called power ports or simply ports. Thus, the concept of an energy port - a point of
energetic interaction between system parts and a system and its environment is central to the
bond graph approach. The power interaction between two ports is represented by a single
line, which is called a power bond or only bond as indicated in Fig. 2.2.

System B
System A
SB,2
Bond SB,1

Ports SB,3

Fig. 2.2 Basic bonding

This is the reason that the graphical representation of a system obtained by using bonds and
energy ports is called a bond graph.

A power bond is considered to transmit power instantaneously and without loss of power
from one port to another and hence generalise the notion of resistance less wires or infinity
stiff rods, which in other areas of physics are considered to conduct power instantaneously
and without loss from one location to another.

The question of sign convention is a nuisance in all system studies, and does not disappear
even when we use bond graphs. The direction in which the power flow is assigned a positive
value is indicated by a half-arrow on one end of the bond as shown in Fig. 2.3. This minimal
indication fixes all positive directions of the physical variables in a system. The power direc-
tion is interpreted as power flowing in the direction of the half-arrow whenever the power is
positive.
 Definition of Generalised Variables 15

A full arrow on a bond is called an active bond and it indicates a signal flow at very low
power. Thus, the notation for an active or signal bond shown in Fig. 2.3 is identical to that for
a signal in a block diagram.

Power bond Signal bond

Fig. 2.3 Sign convention on power and signal bonds

The essential features of the bond graph modelling approach can now be summarised in two
basic principles, namely the

• reticulation principle - meaning decomposition into elemental physical properties or

basic elements.
• power postulate - meaning representing the exchanges of energy between basic ele-
ments as a network of power flows along lines called power bonds.

2.2 Definition of Generalised Variables

It is a well-known fact that the power interaction between two elements at their joining inter-
face cannot be rigorously described with only a single coupling variable. Rather to deal with
the power transfer that describes the energy flow between elements, it is necessary to con-
sider two variables at each coupling port. These variables are selected so that their product is
the instantaneous power being transmitted through the port. In the bond graph language these
two variables are called power variables since their product is power, and have been given
the generalised names of effort denoted by the letter e , and flow denoted by the letter f . This
means that there are three quantities associated with a power bond, effort, flow and the power,
given by the relation

P  t  = e  t f  t  (2.1)

The power variables e and f can be indicated on each side of a bond, by the convention that
the effort variable e is written above or to the left of a bond and the variable f is thus written
below and to the right of the bond as shown in Fig. 2.4

In addition to the two power variables, we will find it useful to define two additional varia-
bles which are important in describing energetic relations. One is the generalised momentum
denoted by the letter p , and defined as the time integral of an effort. That is
16 Bond Graphs and Basic Element Models

e
e f
f

e - above bond e - left of bond

f - below bond f - right of bond

Fig. 2.4 Notation convention for effort and flow on bonds

p  t  =  e  t  dt + p  0  (2.2)
0

where p  0  represents the generalised momentum at t = 0 , i.e. an initial condition.

The other variable is called the generalised displacement q , and is defined as the time inte-
gral of a flow variable. Thus

q  t  =  f  t  dt + q  0  (2.3)
0

where q  0  represents the generalised displacement at t = 0 , i.e. the initial condition.

As the effort and flow variables are called power variables, the momentum and displacement
variables are called energy variables since stored energy can be expressed in terms of these
variables. This fact can be seen from the following development:

The energy E  t  , which has passed into or out of a port is the time integral of the power
flowing in or out of a port and can be written in the form

t t

E  t  =  P  t  dt =  e  t f  t  dt (2.4)
0 0

From Eq. 2.2 and Eq. 2.3 we can write

edt = dp and fdt = dq (2.5)

For cases where a flow is expressible as a function of momentum in the general form
 Definition of Generalised Variables 17

f = f  p  , the energy expression can be written as

E  p  =  f  p  dp (2.6)
0

When cases occur where an effort is a function of displacement in the form e = e  q  , the
energy can be expressed as

E  q  =  e  q  dq (2.7)
0

In mechanics, E  p  is called kinetic energy and E  q  is what we call potential energy. It

will be demonstrated in this text as an impressive fact that only these four variables e , f , p
and q will be needed to model physical systems.The choice of which variable to call effort
and which to call flow is basically arbitrary in bond graphs, but the choice determines which
type of analogy is built into the system. However, the founding fathers of bond graphs
adopted an analogy, which has become a standard.

The particular analogy employed seems natural from the point of view of basic physical laws
according to traditional modes of expression. That is, mechanical force, electromotive force,
and magneto motive force are considered analogous. In Table 2.1 is listed a number of energy
domains and the variables categorised as effort, flow, displacement and momentum in each
domain

In order to convey a preliminary feeling for energetic bonding, it might be useful at this stage
to introduce what is called the word bond graph of a representative system. Word bond
graphs are preliminary feeling for energetic bonding. Word bond graphs are preliminary a
heuristic means of exploring the possible energetic interconnections within a system prior to
a detailed analysis.

.We will illustrate the basic concepts of bond graph modelling through a word bond graph
18 Bond Graphs and Basic Element Models

Energy Effort Flow Momentum Displacement

domain (e ) (f ) (p ) (q )

Electrical Voltage Current Flux linkage Charge

[V] [A] [Vs] [ As] or [ C ]

Mechanical Force Velocity Linear Distance

translation [N] [m  s ] momentum [m]
[ kgm  s ]

Mechanical Torque Angular Angular Angle

rotation [Nm] velocity momentum [rad]
[ rad  s ] [Nms]

Hydraulic Pressure Volume flow Pressure Volume

[Pa] rate momentum 3
[m ]
[ m3  s ] [ N  m2s ]

Thermal Temperature Entropy flow (not defined) Entropy

[K] [J  s ] [J ]

Magnetic Magneto- Flux rate (not defined) Flux

motive force [ Wb  s ] [ Wb ]
[A ]

Chemical Chemical Rate of (not defined) Advancement

potential reaction of reaction
[ J  mol ] [ mol  s ] [ mol ]

Table 2.1 : Identification of variables

construction of the multi energy electric motor driven pump system shown schematically in
Fig. 2.5.

Fig. 2.5 Schematics for a electric motor driven pump

This system consists of three energy domains:

• electrical
• mechanical
• hydraulic
 Basic Bond Graph Elements 19

As a first modelling step we identify the main physical components or subsystems by names
which in this particular problem arebattery

• electric motor
• shaft with a flexible coupling
• hydraulic pump

For each of these components we then identify the ports through which the components
exchange power with the environment or other components. Then those ports which
exchange power with each other are connected with a bond. The resulting word bond graph
of the example system is shown in Fig. 2.6.

u Mm Flexible Mp P
Battery DC-motor Pump
i ωm couping ωp Q

Fig. 2.6 Word bond graph of electric motor driven pump system

From Fig. 2.6 it can be concluded that a bond graph pictorially display the power coupling
between subsystems and between the system and the environment. The above example
shows that bond graphs can handle different types of energy and power in a compatible fash-
ion. The key choice initially in constructing a model is to identify the major components or
subsystems, and bonds are then introduced to mark each distinct power interaction between
the components.

In order to develop our model in a quantitative direction we need to define a basic set of ele-
ments, which describe the energetic properties within a system. This will be done in the next
section.

2.3 Basic Bond Graph Elements

Having made an introduction to the basic idea behind the bond graph approach to system
modelling and defined the generalised variables which we will use to describe system behav-
iour, the next step in our development will be to introduce a collection of basic ideal elements
which represents all the ways a system can process energy. Roughly speaking these elements
account for energy supply, energy storage and energy dissipation in addition to energy trans-
duction or energy conversion from one mode to another. In this manner, it is possible to build
up a catalogue of basic elements with distinct energy handling properties. These basic ele-
ments will serve as the building blocks in modelling systems in detail. They will in certain
20 Bond Graphs and Basic Element Models

cases represent mathematical abstractions of real physical components, and in other cases are
they used to model physical effects. Friction is such an effect, which it is not possible to iden-
tify as a component.

In order to be able to model systems, we also need a means by which we can hook the ele-
ments together. For this purpose we will define a junction structure which may be viewed as
a switchboard for the routing of power into appropriate paths. The basic set of elements is
characterised by the number of energy ports they have, and by their distinct energy handling
properties. An element with a single port is called a 1-port, an element with two ports is
called a 2-port and so on. In general they are also referred to as multiports.

2.3.1 Basic 1-port Elements

We will begin with defining the simplest members of the multiport family, the 1-port ele-
ments, which represent

• energy supply
• energy storage
• energy dissipation

We will in this chapter just define the basic elements and then in a later chapter expose you to
a number of physical components in the different domains which can be represented by the
defined basic elements.

Sources
Let us start the presentation of the basic elements with the elements which we will use to
model energy supply to a system, namely energy sources. By an energy source is meant a
device which is capable of delivering power continuously to a system. The power, which a
source delivers, is provided by an energy reservoir external to the system under considera-
tion. Ideal sources are used to describe interactions between a system and its environment, or
to describe an effort or a flow which is constrained to have a prescribed value.

It is necessary to define two idealised sources:

• Source of effort

that is one where the effort is a specified function of time and is called an effort-source
with the mnemonic code S e
 Basic Bond Graph Elements 21

• Source of flow

that is one where the flow is a specified function of time and is called a flow-source with
the mnemonic code S f

In Fig. 2.7 is shown the bond graph symbol for the two sources.

e e
Se Sf
f f

Fig. 2.7 Bond graph of effort and flow sources

Since an energy port is a point at which other elements can be connected to the element, it is
indicated with a free bond.

As mentioned, the sources are considered to be ideal. An ideal effort source is assumed to be
capable of imposing a prescribed effort on the system regardless of the flow. Similarly, an
ideal flow source is assumed to be capable of imposing a prescribed flow regardless of the
effort.

Although most sources supply energy to systems, there is no restriction for any source to
absorb power which means power is flowing into the source. We will treat the sign in detail
when we in later chapters model systems.

1-port Capacitor Element

The next group of elements we are defining are those which store energy. Most systems can
store energy in two forms, as, for example potential and kinetic energy in mechanical sys-
tems. Let us then first consider the 1-port device in which a single valued static constitutive
relationship exists between an effort e , and a displacement q variable. Such an element is
called a 1-port capacitor and given the mnemonic code C and is therefore often called a C-
element The defining constitutive relation and the bond graph symbol for a capacitor is
shown in Fig. 2.8

e
e
f
C
q =  f dt

Fig. 2.8 Bond graph representation of a capacitor element

The capacitor is characterised by a constitutive relation of the form

22 Bond Graphs and Basic Element Models

q = c  e  (2.8)

in the general case, and

q = Ce (2.9)

in the linear case, where C is the constant capacitance parameter. The capacitor element is an
idealisation of physical components such as springs, capacitors, and liquid storage tanks.

As a sign convention for a capacitor we will establish the rule that the half arrow will always
point towards the C as indicated in Fig. 2.8. This then means that positive power associated
with the capacitor implies power flow into the element and negative power indicates power
flow out of the element. As discussed previously, the C-element is considered ideal, which
means that the element can store and give up energy without loss.

The energy stored in a capacitor element at any time t can be represented by the expression

E  t  =   e  f  dt + E  0  (2.10)
0

where E  0  is the stored energy at time zero.

Substituting the inverse of the constitutive relation for the effort e , into Eq. 2.10 and remem-
ber the definition of q , we can write the energy expression as

q
–1
E t =  c  q  dq + E  q  0   (2.11)
q0

This stored energy is represented by the crosshatched areas under the constitutive relations
for both linear and non-linear relations illustrated in Fig. 2.9. For a linear capacitor element
the stored energy becomes

q
2
q q
E q =  ---C- dq + E  q  0   = ------- + E  q  0  
2C
(2.12)
q0
 Basic Bond Graph Elements 23

Linear Non-linear
e e

E q Eq
q =  f dt q =  f dt

Fig. 2.9 Stored energy in linear and non-linear 1-port capacitors

1-port Inertia Element

The other energy storage element come into being when we find a single valued static consti-
tutive relationship between a momentum p , and a flow f . This element is in bond graph ter-
minology called a 1-port inertia indicated with the mnemonic code I and hence also called an
I-element. In Fig. 2.10 is shown the bond graph symbol and a graphical illustration of the
defining constitutive relation for the inertia element.

f
e
f
I

p =  e dt

Fig. 2.10 Bond graph representation of an inertia element

The inertia element is characterised by a constitutive relation of the form

p = I  f  (2.13)

in the general case, and

p = If (2.14)

in the linear case with I as the constant inertia parameter. The inertia element is an idealisa-
tion of physical objects such as a mass, inductance in electrical systems or inertia effects in
hydraulic systems. As for the capacitor element, the positive power direction is into the ele-
ment as shown in Fig. 2.10.

The inertia, as the capacitor, is considered to be ideal, which means that the element can store
and give up energy without loss. With a positive power flow into the element the stored
24 Bond Graphs and Basic Element Models

energy at any time t can be computed from the expression

p
–1
E p =  I  p  dp + E  p  0   (2.15)
p0

where E  p  0   is stored energy at time equal zero in the general case. This stored energy is
represented by the crosshatched area under the constitutive relations for both linear and non-
linear relations as illustrated in Fig. 2.11.

Linear Non-linear
f f

p =  e dt p =  e dt

Fig. 2.11 Stored energy for linear and non-linear 1-port inertias

For a linear inertia element the stored energy becomes

p
2
p p
E p =  --I- dp + E  p  0   = ----
2I
- + Ep 0 (2.16)
p0

In Eq. 2.11 and Eq. 2.16 we confirm what was indicated in Eq. 2.6 and Eq. 2.7 earlier that the
p and q variables specify the energy storage in a system.

1-port Resistor Element

The last 1-port element is an element defined by a single valued static relationship between
an effort e and a flow f , and is called a 1-port resistor, and given the mnemonic code R. This
and only this element handles power dissipation. The R-element is characterised by a static
relationship between e and f of the form

e = R  f  (2.17)

in the general case, and

e = Rf (2.18)
 Basic Bond Graph Elements 25

in the linear case. The constant R is called the resistance parameter.

The bond graph symbol and an example graph of a constitutive relation for a resistor is
shown in Fig. 2.12.

e
e
R
f
f

Fig. 2.12 Constitutive relation and bond graph representation of a resistor

From Fig. 2.12 we can see that positive power flows into the resistor as for the C and I ele-
ment. According to this sign convention we can conclude that power is dissipated if the con-
stitutive relation fall in the first and third quadrants of the  e f  -plane as shown in Fig. 2.12.
Thus, unlike the C and I elements, the R-elements do not store energy but directly opposite, it
dissipates energy. We of course interpret this to mean that the energy has been converted to a
form that we do not intend to account for subsequently. The R-element will be used to model
all kinds of energy dissipation such as mechanical and hydraulic friction and electrical resis-
tors.

As an aid to remember which variables relate the 1-port elements, we can use what Paynter
called the tetrahedron of state, shown in Fig. 2.13

 dt
C
p R q

I
 dt
f

Fig. 2.13 Tetrahedron of state for 1-port elements

2.3.2 Basic 2-port Elements

Up to this point, we have focused our attention on elements which exchange energy through
a single port. We will now turn our attention to elements where energy is conducted into or
away from them via two power bonds and therefore, called 2-port elements. In Fig. 2.14 is
26 Bond Graphs and Basic Element Models

shown an illustration of a 2-port with through power convention.

e e
TP
1 2

f 1 f 2

Fig. 2.14 Bond graph of a general 2-port element

We assume these elements to be ideal in that they neither store nor dissipate, but transmit
power instantaneously from one bond to the other. This means that the power is always equal
on the two bonds. Thus, in order to determine the number of possible basic two port ele-
ments, we will consider the case of the simplest linear 2-port and write the transmission
matrix in the form

e2 e
= a b 1 (2.19)
f2 c d f1

By using the power constraint that the power on each bond is equal, then two solutions are
possible

1) b = c = 0 and d = 1a (2.20)

and

e2 e1
= a 0
f2 0 1  a f1

2) a = d = 0 and c = 1b (2.21)

and

e2 0 b e1
=
f2 1  b 0 f1

This means that only two basic 2-port elements exist where for one the transmission matrix is
symmetric and in the other case we have a skew-symmetric transmission matrix.

The first solution leads to the constitutive laws

 Basic Bond Graph Elements 27

e 1 = me 2
(2.22)
mf 1 = f 2

having renamed 1  a to m . An elements with this property is called a transformer with a

mnemonic code TF as shown in Fig. 2.15.

e m e
TF
1 1

f1 f
1

Fig. 2.15 Transformer element

The parameter m is called the transformer modulus and displayed on the graph above the
mnemonic TF, as shown in Fig. 2.15. A transformer is hence an element, which transform an
effort on one port into an effort on the other port with a magnitude depending on the modu-
lus. The corresponding flows on the two bonds are also directly related as shown by Eq. 2.22.

The electrical transformer is probably the only transformer you have heard of previously. In
later chapters, we will introduce you to a number of physical components which can be mod-
elled as transformers.

The second solution corresponds to the relations

e 1 = rf 2
(2.23)
rf 1 = e 2

where 1  b is renamed r . An element with such a property is called a gyrator with a mne-
monic code GY and is symbolised as shown in Fig. 2.16.The gyrator takes its name from its
gyroscopic counterpart in mechanics. This element was introduced by B. H. H. Tellegen in
[3] .

e r e
GY
1 2

f
1 f
2

Fig. 2.16 Bond graph of a gyrator element

For this element the parameter r , is called the gyrator modulus and as for the transformer,
the modulus is displayed above the mnemonic GY, as shown in Fig. 2.16. As opposed to the
transformer, the gyrator is an element, which transform an effort on one bond into a flow on
the other bond with a magnitude which depend on the gyrator modulus.
28 Bond Graphs and Basic Element Models

Since a gyrator relates an effort variable to a flow variable, it can be viewed as a diagonalis-
ing element. For example, the properties of the following bond graph elements are distin-
guishable at their external bonds

GY GY = TF
GY C = I
GY I = C

Fig. 2.17 Gyrator characteristics

The physical domain on each side of a transformer or gyrator does not have to be the same,
and for most real devices modelled by 2-port elements they are different. This means that
these elements are used to couple or interface between different energy domains.

Physical devices which perform gyration, are far less common than those which perform
transformation. However, there are more devices that can be modelled as gyrators than you
would imagine, and we will expose you to these physical devices later when we treat each
physical domain in detail.

In some cases is it possible to vary the parameters m and r without a power requirement.
When this is the case, we speak of modulated transformers and gyrators with the mnemonic
codes MTF and MGY. This is an instance in which a signal bond is used to represent the var-
iable modulus and they are then represented as shown in Fig. 2.18.

m(t) r(t)

MTF MTY

Fig. 2.18 Modulated transformer and gyrator

2.3.3 Junction Elements

The structure of a system is dependent not only on the elements contained within the system
boundary but also their interconnection or topology. In other words, simply being given a set
of elements does not provide sufficient information from which to deduce the nature of the
system.

What we must know about the system is the physical arrangement of the elements, that is the
 Basic Bond Graph Elements 29

system topology. In the formalism of bond graphs an ideal junction is considered as non-
energy or more correctly asynergic, which means that it neither stores nor dissipates energy,
but merely transmits it. It means it transmits power instantaneously between its ports without
storing or dissipating energy, that is, the instantaneous power out equals instantaneous power
in. Because the power transmission is assumed to be instantaneous the output and input
power fluxes must be related through a static relation. Thus, bond graph junctions are defined
to be power-continuous and may be viewed as a switchboard for the energy, routing incident
energy into appropriate paths instant by instant.

The description of power flow in a bond as the algebraic product of effort and flow leads to
the existence of two completely symmetric asynergic junctions. One of the two junctions is
called a zero-junction with mnemonic code 0 and the other junction is called a one-junction
with mnemonic code 1. The two junction types are shown in Fig. 2.19.

e3 f3 e3 f3

e1 e2 e1 e2
0 1
f1 f2 f1 f2

Fig. 2.19 The two junction types

The key property of a 0-junction is that it defines a unique common effort on all the bonds
adjoining the junction. With the inward power sign convention shown in Fig. 2.19, the con-
stitutive relations can be written as

e1 = e2 = e3 (2.24)

and

f1 + f2 + f3 = 0 (2.25)

This means in words that a 0-junction has equal effort on all bonds adjoining, and the alge-
braic sum of all flows equal zero. The 0-junction thus represents Kirchoff's current law and
which in mechanics go by the name of geometric compatibility.

The other junction element, the 1-junction is a multiport element where effort and flow are
interchanged compared to the 0-junction. This is why they are given the mnemonic code 0
and 1. With inward power flow convention, the constitutive relations for the 1-junction can
be written as
30 Bond Graphs and Basic Element Models

f1 = f2 = f3 (2.26)

and

e1 + e2 + e3 = 0 (2.27)

Thus, the 1-junction has equal flow on all bonds adjoining, and the algebraic sum of all the
efforts are equal to zero. The 1-junction represents Kirchoff's voltage law and which is
named dynamic force balance in mechanics.

A major feature Paynter gave to the bond graph notation is the materialisation or realisation
of the constraint due to the interconnections between lumped elements by representing inter-
connection laws as power conserving multiport elements. He once wrote: "Now the bond
graph approach to multiport system dynamics replaces the Kirchoff's laws of electric cir-
cuits, and analogously the force equilibrium and geometric compatibility condition of
mechanics by the ideal energy junctions (0, 1) which are physically real and valid as the
component elements themselves.” The two junction elements are laying much of the founda-
tion for the bond graph method and which for a large part distinguishes bond graphs from the
conventional modelling methods.

As shown in Fig. 2.19 the two junction elements are shown as 3-port elements, but there is no
upper limit of bonds which can adjoin a junction element. In some cases, 2-port versions will
occur which will demand some attention on the power direction. If we have a through power
direction, the junctions can be reduced to a single bond as shown in Fig. 2.20

0 and 1
e1 = e2 e1 = e2
f1 = f f1 = f
2 2

Fig. 2.20 2-port junction with through power direction

In the case with an inward power direction the junctions can not be reduced since there is a
sign difference on the two bonds.

0 and 1
e1 = e2 e1 = -e 2
f 1 = -f 2 f1 = f2

Fig. 2.21 2-port junctions with inward power direction

 Basic Bond Graph Elements 31

The constitutive relations for the 1-port, 2-port and the two 0- and 1-junction elements are
summarised in Table 2.2.
32 Bond Graphs and Basic Element Models

Bond graph element Constitutive relations

Se
e e = e t

Sf f = f t
f

e General : q =  C  e 
f
C
Linear : q = Ce

e General : p =  I  f 
f
I
Linear : p = If

e General : e =  R  f 
f
R
Linear : e = Rf

e1 m e2 e 1 = me 2
TF
f f mf 1 = f 2
1 2

e1 e2 e1 = f2 r
GY
f
1
f
2 f1 r = e2

e1 = e2 = e3
e3 f3
f1 – f2 – f3 = 0
e1 e2
0
f1 f2

f1 = f2 = f3
e3 f3
e1 – e2 – e3 = 0
e1 e2
1
f1 f2

Table 2.2 : Basic bond graph elements

REFERENCE
3. B.H.H. Tellegen, The gyrator, A New Electric Network Element, Philips Res. Rept., Apr.,
1948.
 Basic Bond Graph Elements 33

PROBLEMS
2-5 Construct tetrahedra of state similar to that of Fig. 2.13 for the following physical do-
mains: mechanical translation, mechanical roatation, hydraulic systems, electrical sys-
tems. Replace e , f , p and q with their physical counterpart variables, and list the
specific names and dimensions of each variable.

2-6 An non-linear damper is found to have the following relation between the force F and
the relative velocity v between the two ends of the damper: F = Av n ,where A and n
are positive constants. Graph this relation for different value of A and n , and introduce
the required modifications in the case that –   v   . Find the relation for the velocity
as a function of the force.

2-7 A cone turned upside down is used as a liquid storage tank. The cone angle is 60  . At
normal operation the tank contains a fluid volume V 0 . What type of bond graph element
can represent this hydraulic system?

2-8 The main feature of a DC motor is such that the magnetic field is set up so that the motor
torque T e is proportional to the product of the magnetizing flux  and the armature cur-
rent i a , that is
T e = k T i a

What type of bond graph element may represent this device?

2-9 For the air suspension damper part given below, please develop the constitutive relation
and draw the bond graph symbol representing this component. Assume isentropic com-
pression and expansion of the air.
F
v

P,V

2-10 In the figure below a hydraulic device often found in automobile suspension is shown.
Here you can assume that the mass of the piston is equal to zero. What kind of bond
graph element may represent this physical component? Please write down the constitu-
tive laws for the element.
k x
h P,V

Q
34 Bond Graphs and Basic Element Models

If the device was to be used as an pneumatic damper, what type of element would you use
then? Write down the constitutive equations for this situation also, assuming that the tem-
perature of the air in the cylinder is constant.

2-11 Morrisons and others found that the force on a cylinder section with a diameter D and
a length l submerged into water wan be written as

1 d
F = C D --- DLv v + C M V v
2 dt

What type of bond graph elements would you use to represent the equation above and
how would you connect them ? The solution contains four different elements.

2-12 For a hydraulic pipe the pressure loss from the input to the output of the pipe can be writ-
ten as
L v2
P in – P out = f  ----    -----
D 2

where f is the frictions factor, L and D is the length and the diameter of the pipe,  is
the density of the fluid and v is the average velocity of the fluid in the pipe. What type
of bond graph element is this, and please write down the constitutive equations for this
element. If the friction factor was an arbitrary function of the fluid velocity v , what type
of element would this have been. Discuss the constitutive relations for this new element.

2-13 A electric diode is important component in most electronic systems and its mathematical
description can be given as

u  nu
id = Is  e d T – 1 
where i d and u d is the diode current and voltage drop, I s is the brake current and n and
u T is two diode parameters dependent of diode temperature. What kind of bond graph
element can be used to model this component? Derive the possible constitutive laws this
element. If the voltage drop in the expression above is modified to u d' = u d – Ri i , how
would this influence your model and the expressions for the constitutive laws?

2-14 A magnet is moved in and out of an electric coil having n turns and connected to a light-
bulb. From physics you know that this will cause the bulb to light up. What single bond
graph element can be identified to model the relation between the dynamics of the mag-
net, i.e. the force and velocity experienced, and the induced voltage and current? Derive
the constitutive law for this element.

2-15 Albert Einstein states in his theory of relativity that the relation between the momentum
p and the velocity in a relativistic frame is p = mv  1 – v 2  c 2  – 1 / 2 . Identify what bond
graph element this is and show the constitutive laws possible.
 System Modelling 35

Chapter 3

System Modelling

In this book we have defined modelling as the process by which a problem as it appears in
the real world is interpreted and represented in terms of a bond graph. For short this means
that we as modellers must use creativity, intuition, experience, proper assumptions and
approximations to decide what physical elements and effects are important to include in the
model. It is important in this phase to recognise which basic element is the counterpart to the
individual physical effects in the system at hand. Therefore, due to the importance of relating
the physical reality to the basic 1-port and 2-port elements defined previously, we will in this
chapter indicate which physical components or physical effects in the different energy
domain which corresponds to the basic elements. Once we have acquired this knowledge, we
are ready to use it to develop the ability to construct bond graph models of systems. As soon
as the fundamentals of model construction have been performed on simple systems, we can
then apply these principles to the creation of more complex systems in later chapters. We
will first start with the most restrictive class of systems, namely those systems, which are said
to be “single-energy-domain” systems. Four representative energy domains will be dis-
cussed such as mechanical translation and rotation, fluid, electrical and thermal. For each
particular energy domain we shall find it convenient to divide the manner of proceeding into
a checklist of steps that will lead us from a schematic diagram or sketch of the system to a
bond graph model in a systematic unmistakable manner.

We begin with a study of modelling dynamics of mechanical systems since such systems are
quite well suited to the introduction of the reader to bond graph system modelling. Thereaf-
ter we will use a similar procedure to systems of the other energy domains.

3.1 Mechanical Systems

Here the term mechanical is used in a restrictive sense in that fluid and thermal systems are
treated separately.

In general, mechanical systems consist of a combination of several different mechanical ele-

ments, properly designed and arranged to work together as a whole. The motion of mechan-
ical elements can be described in various dimensions as translational, rotational or a
combination of both. In the following, we will consider system examples containing all three
36 System Modelling

forms of motion.

The motion of translation is defined as a motion that takes place along a straight line. This
simple motion provides good examples for an introduction to a generalised approach to
mechanical system modelling with bond graphs. Despite the apparent simplicity of transla-
tional systems, quite a few engineering design and analysis decisions are based on such mod-
els.

The rotational motion of a body can be defined as its motion about a fixed axis. A rotational
mechanical system is analogous to a translational one with the replacement of mass, force,
velocity and distance by moment of inertia, torque, angular velocity and angle respectively.
It is usually more difficult to sketch and visualise the directions of torque applied in a tor-
sional system than is forces applied in a rectilinear system.

3.1.1 Mechanical Sources

Now then, before we start to model let us illustrate a number of mechanical example ele-
ments that are counterparts to the 1-port and 2-port basic elements. We will make the illus-
trations in the same order as we defined the basic elements in Chapter 2. That means that we
will start with sources.

Examples of effort sources Se, are

• Excitation forces - F  t 
• Excitation torques - T  t 
• Gravity forces - Force = mass (m) times gravitational acceleration (g)

Likewise, velocity sources S f , are models of given velocities from the system environment
such as from cams, electric motors and internal combustion engines.

3.1.2 Mechanical Capacitor Elements

Next in line are mechanical capacitors or components which can be modelled as C-elements.
Mechanical 1-port capacitors are elements of zero mass experiencing a deformation when
acted upon by a force or torque. The property of a material which enables it to regain its
original shape and dimensions when the load is removed are what we call elasticity. An ele-
ment exhibiting compliance thus defines what we call a pure spring, which can be said to be
any element which is used primarily because of its elastic deflection.

The most common spring or capacitor is the translational spring known as the coil spring,
which may be made of wire with round, square or rectangular section. In Fig. 3.1. is shown a
 Mechanical Systems 37

coil spring, the symbol used to represent it and the bond graph symbol.

F F
x 1

e=F
. . C
f = x 1 - x2
x 2

F F

Fig. 3.1 Axially loaded helical spring, schematic and the bond graph

Many real physical structures of various shapes and forms can often be modelled success-
fully as ideal translational springs. The constitutive relation which relates the applied force
and the resulting deformation , is given as

 = x1 – x2 = g  F  (3.1)

For a linear ideal spring the constitutive relation takes the form

 = x2 – x1 = C  F (3.2)

where the constant C is known as the compliance or flexibility, while the inverse of C is
known as the spring constant or stiffness, often denoted by k.

Another example of a translational spring element is a bar which is subjected to a tensile or

compressive force F as shown in Fig. 3.2

F F

Fig. 3.2 Tension or compression bar

When the force F , is applied to the bar with cross-sectional area A and length L , the bar
becomes longer or shorter depending on the direction of the force. The elongation, , can
then be determined from the following expression
38 System Modelling

L
 = ----------- F (3.3)
EA

where E is the modulus of elasticity. This means that the compliance is identified to be

L
C = ----------- (3.4)
EA

As translational capacitors are elements which exhibits a linear deflection when subjected to
a force, torsional capacitors are elements which exhibits an angular deflection when exposed
to a torque. One of the most typical examples of a torsional capacitor is a cylindrical shaft.
For such a shaft with length L and diameter d , the ends will rotate when a twisting moment
or torque T t , is applied as shown in Fig. 3.3.

d
Tt Tt

Fig. 3.3 Torsional capacitor

The total angle of twist  in radians is given by the expression

L
 = ----------- T t (3.5)
JG

where J is the polar moment of inertia and G is the modulus of elasticity in shear. For a
solid circular cross-section the polar moment of inertia is given by the relation

4 4
d r
J = -------- = -------- (3.6)
32 2

The compliance for a solid circular shaft can then be expressed as

32L 2L
- = -------------
C = ------------ (3.7)
d G
4 r 4 G

A powerful tool for determining the deflection of simple members as well as more complex
strucutres can be found in the theory of elastic energy through the theorem of Castigliano.
This theorem is based on strain energy relationships which in accordance with the principle
 Mechanical Systems 39

of conservation of energy is equal to the work done during the deformation of a structure.
Thus, the theorem of Castigliano states that the displacement corresponding to any force may
be found by obtaining the partial derivative of the total strain energy with respect to any of
the external forces. This partial derivative is the displacement of the point of application of
that force in the direction of the force. The terms force and displacement are to be readily
interpreted as e’s and q’s since they apply equally to moments and angles.

In the form of equations, the theorem of Castigliano is stated as

U U
i = and i = (3.8)
 Fi  Ti

where U is the total strain energy, and  i and  i are the “displacements” in the directions of
the “forces”, F i and T j .

Since the total strain energy of a system is the starting point for the application of the theorem
of Castigliano, let us develop the strain energy expressions for tensile, flexural and torsional
loadings.

In an axially loaded slender bar of constant cross-sectional area A , length L and modulus of
FL
elasticity E , the deformation  = ------- , and the strain energy denoted by U becomes
EA

2
F L
U =  F d = ----------- (3.9)
2EA
0

For a tensile bar with varying EA and or F , the strain energy U becomes

2
1 F
U = ---  ------- dx (3.10)
2 EA

For a uniform beam of length L loaded by a bending moment as shown in Fig. 3.4, the beam

θ
T

Fig. 3.4 Beam with bending moment loading

40 System Modelling

undergoes an angular rotation  proportial to the bending moment. Thus, from elementary
strength of material we obtain that

TL
 = ------- (3.11)
EI

where EI , the product of modulus of elasticity E in tension and section moment of inertia I ,
is called the bending stiffness. Since the rotation  builds up linearly with T , the strain
energy stored is

2
U = ------ = ---  ---------
T 1 T L
(3.12)
2 2  EI 

For a beam with varying EI and or varying T the energy becomes

2
1 T
U = ---  ------ dx (3.13)
2 EI

For the case with a uniform circular shaft carrying a torque T , the shaft undergoes a total
twist  , in length L , proportional to the torque. From elementary strength of materials we
have that

L
 = ------- T (3.14)
GJ

where GJ , the product of the modulus of elasticity in shear and the cross sectional polar
moement of inertia, is called the torsional stiffness. Since the twist  builds up linearly with
the torque T , the strain energy stored is

2
1 1T L
U = --- T = --- --------- (3.15)
2 2 GJ

For a shaft with nonuniform properties and varying loading the stored energy can be
expressed as

2
1 T
U = ---  ------- dx (3.16)
2 GJ
 Mechanical Systems 41

Example 3.1
Let us now use the theorem of Castigliano to determine the constitutive relations for a couple
of flexible structural examples. The first example is a simply supported beam with a force F
, at the midpoint between the supports and with a moment diagram as shown in Fig. 3.5

l F l
2 2

(a)

Mx

(b)

Fig. 3.5 Loaded beam and moment diagram

Equations for the moment is given as


 F
--- x ,
l
0  x  ---
 2 2
M x =  (3.17)
 Fl F l
---  x  l
 ----
- – --- x
4 2
,
2


Substituting the bending moment relations given in Eq. 3.17, into Eq. 3.13 results in

l2 l
1  
2 2 2
  ----4- – -----2-
F x Fl Fx
U = --------- 
2EI   ----------- dx +
4
dx 

(3.18)
0 l2

which after integration yields the total strain energy as

2 3
F l
U = ------------ (3.19)
96EI

Taking the partial derivative of U with respect to the force F , results in an expression for the
deflection  at the midpoint as

3
Fl
 max = ------------ (3.20)
48EI
42 System Modelling

This result can be found in numerous textbooks giving tables of deflections of simple struc-
tural elements due to different loadings.

Example 3.2
In our next example, let us look at a little more complicated problem as the one shown in Fig.
3.6.

y
l F l
2 2

I1

h I2 I2

Q x

Fig. 3.6 Frame structural example

In this problem we are asked to determine the displacement at x = l , y = 0 due to a force

F at x = l  2 , y = h .

We indicate an imaginary force Q , at the point where we are to determine the deflection as
shown in Fig. 3.6. At any point on the legs the moment is

M y = Qy (3.21)

and for the horizontal members the moment at any point in the left half of the member is

Fx
M x = ------ + Qh (3.22)
2

The total strain energy is the sum of the energy in each of the three members. This gives

h l2
2 2
M M
U = 2  ----------- dy + 2  --------- dx (3.23)
2EI 2 2EI
0 0

and after substituting the given moments and integrating, the energy becomes
 Mechanical Systems 43

2 3 2 3 2 2 2
U = ------------ + --------  ---------- + ---------------- + --------------
Q h 1 F l FQhl Q h l
(3.24)
3EI 2 EI 1  96 8 2 

The displacement asked for is then

2
U Fhl
x = = ----------- (3.25)
Q 8EI 1

when Q is equal to zero.

The displacement at the force F can be determined from

3
U Fl
y = = -------------- (3.26)
F 48EI 2

when Q equal zero. As can be seen,  y is equal to the results obtained for the simply sup-
ported beam example, as it should.

3.1.3 Mechanical Inertia Elements

Inertia is the property of matter causing a resistance to change in motion. The quantitative
measure of inertia is mass. A pure translational inertia is a mechanical element whose parti-
cles are all rigidly connected together so that they translate with identical motion at all times.
Such a representation is also called a lumped ideal inertia.

The symbol for a translational inertia is a block and a rotational inertia is represented symbol-
ically as a disk. The two inertia elements are shown symbolically with their bond graph sym-
bols in Fig. 3.7.

v
ω ω
F m

F Mt
v I ω
I

Fig. 3.7 Symbols of translational and rotational inertia

The constitutive relationship for the translational inertia element is given by Newton’s sec-
ond law of motion as
44 System Modelling

d  mv 
F = --------------- (3.27)
dt

where F is force, m is mass and v is velocity. Rewriting Eq. 3.27 to the form

Fdt = mdv (3.28)

and integrating on both sides, we obtain

 F dt = mv = p (3.29)

From Eq. 3.29 the constitutive relationship is given by momentum p , as a sinlge-valued

function only of the absolute velocity of the element. That is

p = mv = mx· = Ix· (3.30)

For the translational inertia, it is important to note that the velocity of the element is meas-
ured with respect to a fixed reference frame, usually the earth.

The constitutive relationship for the rotational inertia can be developed in the same manner
as was done for the translational inertia and is written mathematically in the form

pr = Jm  (3.31)

where p r is the rotational momentum, J m is the mass moment of inertia and  is the angular
velocity.

The mass moment of inertia depends on the geometric composition about the axis of rotation
and its density. For instance, the inertia of a circular disk about its geometric axis is given by

4
1 2   d  t
J m = --- mr = ---------------------- (3.32)
2 32

where m is the mass of the disk, r and d is the radius and diameter of the disk,  is the den-
sity of the material and t is the thickness of the disk.

3.1.4 Translational Resistors

Mechanical systems are all more or less subject to energy dissipation, which is generally
 Mechanical Systems 45

referred to as damping. In reality, damping is a complicated phenomena. The force of damp-

ing can come from a variety of phenomena, singly or in combination. Some examples are
components such as automobile shock absorbers, friction between imperfectly lubricated
parts or energy absorption within a solid body. The schematic diagram for energy dissipation
is often represented by a dashpot such as shown to the left in Fig. 3.8.

v2 v1
e=F
F F R
f = v1 - v2

Fig. 3.8 Translational resistance: schematic and bond graph symbol

In mechanical systems, energy dissipation by frictional forces occur whenever there is

motion between two elements. The characteristics of the frictional forces between two con-
tacting surfaces often depend on such factors as the composition of the surfaces, the pressure
between the surfaces, their relative velocity, and others so that the exact mathematical
description of the frictional force is difficult. For practical purposes, frictional forces can be
divided into three basic categories:

1. Viscous friction represents a retarding force, which is a function of the velocity. The
mathematical expression of viscous friction is

n
F f = sign  v v (3.33)

in general, and

Ff = v FN v (3.34)

in the linear case where  v is the viscous frictin coefficient and F N the normal force.

2. Static friction represents a retarding force that tends to prevent motion to start. It is
defined as the static frictional force that exists only when the object is stationary but has a
tendency to move. It is also denoted as “stiction”. The constitutive relation for this case
46 System Modelling

is as shown in Fig. 3.9. where the static friction force is equal to the tensile forces until a

Fig. 3.9 Characteristic curve for static friction

maximum or minimum is reached, hence

 F max = F N  s
Ff =  (3.35)
 F min = – F N  s

with  s being the static friction coefficient.

3. Colomb sliding friction is a retarding force with constant amplitude with respect to veloc-
ity as shown in Fig. 3.10.

Fig. 3.10 Characteristic curve for Colomb friction

The amplitude of the frictions force can be described as

F f =  C F N sign  v  (3.36)

where  C is a coloumb friction factor often name the dynamic friction coefficient and F N is
the normal force.

Some of these friction model have a discontinuity at zero velocity. The friction force at zero
velocity is just a force source or a Se-element and not a unique function of velocity and actu-
ally no energy is dissipated. For handling such cases it is often approximated with a conti-
nous function of which we will focus on

F f = F N tanh  slope  v 
 Mechanical Systems 47

where slope is a very large constant. The use of continous functions results in straightfor-
ward models that are easy to simulate for large forces and velocities. For small forces and
velocities however the slope constant must be considered carefully. An initial guess for
slope is 103.

A model often found useful in many applications is the combination

of the static plus the coulomb plus the viscous friction stated as

F f = F N   s tanh  sv  – ---   s –  c  tanh  s  v – -------   +

1 6.5
(3.37)
 2   s  
tanh  s  v + -------  +  v v
6.5
  s  

where  s  c  v is the static, coulomb and viscous friction coefficients, s is the slope and
F N is the normal force.

The types of friction described for translational motion can be carried over to the motion of
rotation, but in that case we have a relationship between a torque and angular velocity. Gen-
erally modelling of friction is quite complicated and poses challanges both computationally
and and descriptionally. Care must be taken when these models are used.

3.1.5 Mechanical Transformers

Mechanical transformers are devices that transmit energy from one part of a system to
another in such a way that force, torque, velocity and displacement can be altered. There are
a number of devices, which can be modelled as a transformer and we will in the following
indicate some examples.

Probably the simplest mechanical transformer is the pivoted bar shown in Fig. 3.11

lL lR
vR FL m FR
vL
TF
vR
vL

FL FR

Fig. 3.11 Pivoted bar and corresponding bond graph

Assuming the bar to be infinitely stiff we can from taking moments around the pivot write the
relationship
48 System Modelling

l
F L = F R ---R- (3.38)
lL

and from velocity relations we obtain that

lR
---- v L = v R (3.39)
lL

Eq. 3.38 and Eq. 3.39 show that the constitutive relations for the pivoted bar is in fact those
for the transformer. The transformer modulus is shown to be the ratio of the distance from
the pivot to the two ends, namely

l
m = ---R- (3.40)
lL

A gear train is a mechanical device that transmits motion by means of successively connected
cogwheels and transmits energy in such a way that torque and speed are altered. In Fig. 3.12
is shown two coupled gear wheels.

T1 ω1
TL m TR
ω1 TF ω2
ω2 T2

Fig. 3.12 Gear train

The relationship between the torques T 1 and T 2 , angular velocities of the two gears  1 and
 2 , and the radius r 1 and r 2 or the teeth numbers Z 1 and Z 2 can be expressed by the rela-
tions

r Z
T 1 = ---1- T 2 T 1 = ----1- T 2 (3.41)
r2 Z2

and

r1 Z1
----  1 =  2 -----  1 =  2 (3.42)
r2 Z2

From Eq. 3.41 and Eq. 3.42 we can deduce that a gear train can be modelled as a transformer
 Mechanical Systems 49

and the ratio of radiuses or the number of teeth on each wheel becomes the transformer mod-
ulus m , as

r Z
m = ---1- m = ----1- (3.43)
r2 Z2

The same principles can be used to derive expressions for the transformer modulus for belt
driven shafts.

3.1.6 Modulated Mechanical Transformers

As we discussed previously physical 2-port components occur where the modulus is not con-
stant. An example of such a component is the well-known piston-crank mechanism used in
all piston engines and shown in Fig. 3.13.

b
a


F
c
x
Tt
v p = x·

Fig. 3.13 Piston-crank mechanism

In most textbooks the relationship between the force F on the piston and the twisting torque
T t on the crank is obtained by decomposing the force along the piston rod which again is
decomposed into a force along and normal to the crank. The force normal to the crank times
the length of the crank then becomes the twisting torque. In this text, we will use a different
approach by developing the relationship between the piston velocity v p and the rotational
velocity c of the crank. To do that we will start with the relationship between the distance x
and the angle . From the geometry we find that

2 2 2
x = a  cos  + b – a sin  (3.44)

Differentiating both sides of Eq. 3.44 with respect to time and recognising that

dx d
------ = v p ------ =  c (3.45)
dt dt
50 System Modelling

We obtain the following result

 1 a sin  
2 2
v p =  – a  sin  – --- ----------------------------------  c (3.46)
 2 2
b – a sin  
2 2

The term within the parentheses is the transformer modulus which in this case is a function of
the angle  , and hence not a constant, but varies with time.

The relation between the force and torque due to conservation of power becomes

 a sin  - 
2 2
1 ---------------------------------
 – a  sin  – --
-
2 2 F = T t (3.47)
 b – a sin  
2 2

If we denote the varying transformer modulus as m    , the modulated transformer can be

represented by the bond graph shown in Fig. 3.14.

m(θ)

F Mt
MTF
vp ω

Fig. 3.14 Bond graph modulated transformer modelling a piston-crank mechanism

3.2 Mechanical System Modelling

With the bond graph foundation and building blocks in place, we are ready to start develop-
ing bond graph models of physical systems. Due to its simplicity we will start with an exam-
ple of a system in rectilinear motion and then continue on with a system in torsional motion.
In each case, we are going from a schematic diagram model to a bond graph model. We are
not at this stage covering the step from a real world problem to a schematic diagram.

The lines of action that are in common use in proceeding from a physical system representa-
tion such as a sketch or schematic diagram to a bond graph model are:

• direct inspection method

• systematic modelling procedure
• combination of the inspection and the systematic construction method which is proba-
bly most used
 Mechanical System Modelling 51

The inspection method is the most demanding since it requires a physical understanding from
the modeller, but enhance our ability to construct schematic diagrams of systems in other
energy domains which have the same bond graph model. In addition, the direct inspection
method is of vital importance when checking the validity of resulting models obtained from a
systematic procedure. For many systems, the direct inspection approach is the easiest and
shortest way to obtain the bond graph model. However, there will always be systems that are
too complicated for the direct method alone and the systematic procedure is helpful or even
necessary. We will now outline a systematic construction method in six steps that always
work.

3.2.1 Systematic Construction Method

1. Since the velocity is the same at any connection point between elements, then estab-
lish a 1-junction for each distinct velocity.

2. Add existent inertia elements to their respective 1-junctions.

3. For each 1-port element establish the relative velocity over the element by bonding a
0-junction between appropriate 1-junctions. Adjoin the pertinent 1-port element to
the 0-junction.

4. Remove all 1-junctions with zero velocity and all the bonds connected to these 1-junc-
tions.

5. Assign power directions on all bonds.

6. Simplify the bond graph by replacing 2-port 0- and 1-junctions with through power
directions by single bonds.

We will start our modelling effort with the aid of the systematic construction method and
thereafter use the direct inspection method.

Let us then begin with a simple example by considering the spring-mass-damper system with
force excitation on the mass and base velocity excitation as shown in Fig. 3.15.The compo-
nents of the system are identified by a spring with a spring constant k , a viscous damper with
damping coefficient b , and an inertia with mass m . The inertia is acted upon by a force
F  t  and the base is excited by a varying velocity v b  t  . The behaviour is defined by the
displacement x m with the positive direction upward as indicated by the arrow. The displace-
52 System Modelling

ment x m is measured from the unstreached position of the spring.

F(t)

xm ,vm
g m

k b
v b (t)

Fig. 3.15 A spring-mass damper system

We start modelling the given system by observing that there are two points where the ele-
ments join and hence have a common velocity, which we denote by a 1-junction. These
points are at the inertia m and at the base. So we start by setting off two 1-junctions as
shown in Fig. 3.16a. The next step is then to bond to these two 1-junctions the elements,
which have their common velocity. The elements which shear the velocity v m is the mass m ,
the gravity force mg and the excitation force F  t  . The mass m is modelled by an I -ele-
ment, the excitation force F  t  is modelled as a source of effort, Se . The velocity condition
v b  t  is imposed as a source of velocity, and this bonding is shown in Fig. 3.16b,

So to step three and the force-generating elements, the spring C and the damper R . These
elements both connects to the mass on one end and the base on the other. That means that we
shall insert a 0-junction between the two 1-junctions and bond these three junction elements
together. Then we are to connect the 1-port C-element to one of the 0-junctions and the 1-
port R-element to the other 0-junction as shown in Fig. 3.16c. Really what this last step por-
tray is that the relative velocit v m – v b  t  is acting on the spring and the damper, and this rel-
ative velocity is computed by the 0-junction, according to its constitutive relation, and
bonded to the C and R elements as shown in Fig. 3.16c.

With respect to power directions, remember that we have already defined positive power
direction into the C, I and R elements. Assigning power direction to the remaining bonds can
be done by assigning a positive direction on the real system and then transfer them to the
bond graph, or assigning a positive direction arbitrarily on the bonds and when necessary
find out what it corresponds to on the real system. We will in this example use the assumed
directions in Fig. 3.15 and assign the power directions according to these assumptions. To
aid us in assigning proper power directions we will use free-body diagrams with force and
velocity directions added. The positive velocity directions are the same as on the schematic
drawing in Fig. 3.17a. To assign positive direction of the forces, we will assume that
v m  v b , meaning tension in the spring is positive. The free-body diagrams with assigned
positive velocities and forces according to our assumptions are shown in Fig. 3.17b. Our
 Mechanical System Modelling 53

next step is then to transform this information into the assignment of positive power direc-
tions on the bond graph.

Se :F(t) Se :F(t)

vm
1v mg:Se 1v I:m mg:Se 1 I:m
m m

1/k:C 0 0 R:b
vm - vb

1v 1v 1v
b b b

Sf :vb (t) Sf :vb (t)

(a) (b) (c)

Fig. 3.16 Example system bond graph

To do so, let us start with the mass and observe the arrows indicating the force and velocity
directions on this element. The velocity v m is shown positive upwards and all the forces are
shown positive downward. Remembering that positive power means that force and velocity
must have the same direction. This means that the half arrows are all pointing out from the 1-
junction representing v m , as shown in Fig. 3.17c.

Turning to the base, we find that the velocity v b and the forces from the spring and damper
are in the same direction, and therefore positive power is flowing into the base. The resulting
force in the velocity source and the velocity v b are in opposite directions which means that
positive power flow is into the velocity source for our assumption of positive directions. If
the initial assumption is wrong, then the resulting calculated values will be negative, indicat-
54 System Modelling

ing an incorrect choice of reference.

F(t)
F(t)
xm ,vm
g m vm
m
k b
v b (t)
mg
vm
(a)
Tension k b p
positive vb
Se :F(t)

vm vb (t)
mg:Se 1 I:m

(b)
1/k:C 0 0 R:b

1v
b

Sf :vb (t) (c)

Fig. 3.17 Free-body diagam and complete bond graph

 Mechanical System Modelling 55

The complete bond graph with power directions on all the bonds is shown in Fig. 3.18.

Se :F(t)

vm
mg:Se 1 I:m

1/k:C 0 0 R:b

1v
b

S f :vb (t)

Fig. 3.18 Complete bond graph

The power directions shown on the bond graph indicates that with the chosen positive refer-
ence directions, power is flowing into all the sources, which we know is not possible if the
system is to have a sustained motion.

We will now make an attempt to construct the bond graph model directly using the direct
inspection method and our physical insight and bond graph knowledge. Looking at the sche-
matic drawing in Fig. 3.17a, we recognise that the force F  t  , mass m , the gravitational
force mg and the upper part of the spring and damper are all experiencing the common
velocity v m . This statement can be pictured in bond graph language as shown in Fig. 3.19a.
Next we observe that the spring and damper are experiencing the same relative velocity
v m – v b . Equal velocity means that the C - and R -elements are connected to a 1 -junction.
The same relative velocity is obtained by inserting a 0 -junction between the 1 -junction rep-
resenting v m and a 1 -junction representing the base velocity v b as shown in Fig. 3.19b. The
final bond graph ends up as shown in Fig. 3.19c, where the 1 -junction representing v b is
removed since we have a through power direction. As you can see the bond graph shown in
Fig. 3.19c looks different from the one shown in Fig. 3.18, but they are equal. The two 0 -
junctions shown in Fig. 3.18 are replaced by one since they both compute the same velocity
56 System Modelling

difference which again is experienced by the C - and R -elements.

Se:F(t) Se:F(t)
vm 1/ k
C:
mg:Se vm mg:Se vm
1 I:m 0 1 1 I:m
vm – vb
1/ k
R: C:
b
1v 0 1
b vm – vb
R:
b
Sf:v b  t  1v
b

(a) (b) (c)

Sf:v b  t 

Fig. 3.19 Alternative bond graph model of spring-mass-damper system developed using
the direct inspection method.

3.3 Torsional System Modelling

As we have introduced you to system modelling using a translational mechanical system it is
now of interest to model a torsional system. In Fig. 3.20 is shown a schematic diagram of a
torsional system, which could be the shafting system for a ship.

J1
T1 (t) T2 (t)

k1
Tp J2
b2
k2 J4 J5
Tp J3
k3
b1 b3
ωp

Fig. 3.20 Torsional example system

The system has a propeller with inertia J 1 and a loading torque T p , which is a non-linear
2
function of the propeller speed. Typically this function is of the form T p = K p , where K
is a constant (or close to) and  p is propeller angular velocity. The system contains a gearbox
 Torsional System Modelling 57

in order to reduce the engine rotational velocity to one that is proper for the propeller. The

Tp

T1 (t) b2 T2 (t)
k1

k2 k3

b1 b3

Fig. 3.21 Free-body diagram of torsional system example

engine model consists of two cylinders with rotational inertia’s J 4 and J 5 . The two cylin-
ders each have a driving torque T i  t  and a damper to ground representing bearing friction.
In addition there is a damper indicated between the two cylinders representing material
damping.

It is often very instructive and helpful to draw a free-body diagram for mechanical systems,
and we select to do so here too and is shown in Fig. 3.21. Observe the sign convention on the
rotation and the torques.
58 System Modelling

1 1
ω1 ω2

1 1 1
ω3 ω4 ω5

(a)

R 1 0 1 I
ω1 ω2 C R

I C TF C Se 1 Se

I 1 0 1 0 1
ω3 ω4 ω5

I R I R
(b)

Fig. 3.22 Initial bond graph model

The first point in the modelling process is to set off a 1-junction for each inertia in the system
as shown in Fig. 3.22a. To each of the 1-junctions we connect the respective inertia elements,
and insert the torque-generating C-elements through 0-junctions between appropriate 1-junc-
tions. Between the 1-junctions representing the velocity of the two gear wheels is inserted a
transformer according to section 3.1.5. The loading on the propeller is given as a relation
between the torque and velocity or e and f which from our definition of the basic elements
can be modelled by a resistor connected to the 1-junction representing the propeller velocity
p .

The excitation torques from the engine are modelled as effort sources and connected to their
appropriate 1-junctions. The friction to ground is modelled as resistors connected to the 1-
junctions representing the velocity of each crank mass.
 Torsional System Modelling 59

The shaft complience and damping between the two crank masses are experiencing equal rel-
ative angular velocity   5 –  4  . Thus, a 0-junction bonded to the two 1-junctions represent-
ing  4 and  5 , yields the relative velocity on the third bond. Since the C and R elements are
experiencing this common relative velocity, they are connected through a 1-junction. The
resulting bond graph is shown in Fig. 3.22b.

The power directions can be assigned arbitrarily or as a result of selections done drawing the
free-body diagram. First as an example, assigning positive power flow from the engine
towards the propeller and the final bond graph with assigned power directions will be as
shown in Fig. 3.22b. We will now try to find out the positive rotational directions corre-
sponding to the chosen positive directions. As an aid, let us look at a three-inertia schematic
shown in Fig. 3.23 where positive displacements and moment directions are shown to be
clockwise.

T3 , ω3 T 4 , ω4 T5 , ω5

(a)

I C I C I

(b) 1 0 1 0 1
3 4 5

Fig. 3.23 Tree inertia schematic and bond graph.

With the assumed positive directions, let us assume that the angles  5   4   3 , and imag-
ine that we grab disk 5 with the right hand and disk 4 with the left hand and turn disk 5 an
amount  5 –  4 in the positive  -direction. To find out the effect of this twist on disk 5, we
lift our right hand. It is hoped that you agree that disk 5 will rotate in the negative  -direc-
tion. This indicates that there is a twisting torque acting in the negative  -direction on disk
5. Since the angle  5 and twisting torque T 5 are in opposite direction the power is negative
on disk 5.

In the next step let us grab disk 5 again and twist it  5 –  4 again in positive  -direction.
This time we lift our left hand away from disk 4 and it will rotate in the positive -direction.
This means that there must be a positive twisting torque acting on disk 4. Thus, on disk 4 the
rotation and twisting torque are in the same direction indicating positive power flow into disk
4.
60 System Modelling

Continuing with disk 4 and disk 3 and so on we will obtain power flowing from right to left
as shown in Fig. 3.23b. It is left as an exercise for the reader that if we assume that
 3   4   5 the power will flow from left to right. Following these principles the result is
the bond graph model shown in Fig. 3.24..

R 1 0 1 I
ω1 ω2 C R

I C TF C Se 1 Se

I 1 0 1 0 1
ω3 ω4 ω5

I R I R

Fig. 3.24 Complete bond graph

Assigning power bond direction using the free-body diagram may be of great help initially,
especially for the unexperienced student. The convention used drawing the free-body dia-
gram is the right-hand rule for directions of rotation and torques. Selecting the initial rotation
direction for the engine shaft we draw the hollow cone arrow from the right inertia up til the
gear. At the the connecting shaft the direction is opposite as indicated with the hollow arrows
drawn along the propeller shaft. Selecting the direction of the external torque T 2 beeing pos-
itive we simply follow through the assignment of torques along the inertia’s and shaft ele-
ments as shown at the figure. Assigning power bond directions now is stratight forward
recognizing energy flow beeing positive and hence must be assigned into an element when
both arrows are in the same direction and negative and out of an element if opposite.

3.4 Electrical Systems

Electrical systems range from those that are concerned with the generation, transmission and
control of minute amounts of energy to those that focus on the utilisation and control of tre-
mendous amounts of energy. This means that the domain of electrical systems is an
extremely diverse field of study involving a wide range of systems. However, the behaviour
of many electrical systems may be understood by means of the concepts of electric circuits
and idealised electric elements. Many useful electrical systems are less complicated if one
understands the electrical phenomena that take place at the terminals of the electrical compo-
 Electrical Systems 61

nents and devices that make up the system. Such systems are adequately analysed by the
proper application of circuit theory.

We begin by defining basic circuit elements and thereafter we shall model circuits and simple
electrical systems. Advantages of circuit theory are that it provides a relatively simple
description of practical components.

3.4.1 Electrical Sources

There are a number of ways in which electrical power can be supplied to components and
systems. The most familiar to us is the electrical wall outlet which enforces a sinusoidal volt-
age (AC) across the power cord wires of most connected equipment. The voltage is inde-
pendent of fluctuations in the current. This means that the wall outlet can be modelled as a
voltage source.

Another familiar power source is a battery. In contrast to the wall outlet, a battery enforces a
constant voltage (DC) across the power cord of connected equipment over a reasonable range
of currents. A battery can therefore also be modelled as a voltage source.

In some cases, we may encounter that the voltage output from a power supply may decrease
as a function of the load current as shown in Fig. 3.25a.

es e L
u Se 1 [Load]
iL i
L
es
e
R
e i
R L

i L
R

Fig. 3.25 Load dependent voltage source

In this case we need to combine the source with a resistor as shown in Fig. 3.25b. With a con-
stitutive relation for the resistor as

e R = bi L (3.48)

the load voltage can be computed to be

e L = e S – bi L (3.49)

Current supplies are also used in electrical circuits but are not as usual as voltage supplies.
62 System Modelling

3.4.2 Electrical Capacitor

When two conducting plates are placed close to, but insulated from each other, they form
what is known as a capacitor and is symbolised by two short parallel conductive plates, as
shown in Fig. 3.26.

C
+ - e= u1 - u2
C
u1 u2 f=i

Fig. 3.26 Circuit symbol for a capacitor

The circuit parameter of capacitance is represented by the letter C, and is measured in farads
(F). The farad is such a large unit of capacitance that it is almost never used in practice. The
most frequently encountered values lie in the picofarad ( pF ) to microfarad ( F ) range.

In the literature the mathematical relation for the capacitor is usually written as a relationship
between the power variables. Mathematically this is written as

du
i = C (3.50)
dt

where i is the current, u is the voltage. Integrating both sides of Eq. 3.50 gives us

q = Cu (3.51)

where q is the charge of the capacitor. Eq. 3.51can be recognised as the constitutive law for
a 1-port capacitor.

3.4.3 Electrical Inertia-Inductor

In electrical systems, the inductor is the equivalent of the inertia element. Inductors are cir-
cuit elements based on phenomena associated with magnetic fields. The source of the mag-
netic field is charge in motion, or current. If the current is varying with time, the magnetic
field is varying with time. A time-varying magnetic field induces a voltage in any conductor
that is linked by the field. The inductor is symbolised graphically as a coiled wire shown in
 Electrical Systems 63

Fig. 3.27.

+ - e= u1 - u2
I
u1 u2 f=i

Fig. 3.27 Symbol for the inductor

The circuit parameter used to describe an inductor is called inductance and symbolised by the
letter L and measured in Henries (H).

In the electrical engineering literature the voltage drop across the terminals of the inductor is
written in the following form

di
u = L (3.52)
dt

It is not easy to recognise what basic bond graph element is represented here. To do that we
can rewrite Eq. 3.52 in the form

udt = Ldi (3.53)

and simply integrating both sides to obtain

 u dt =  = Li (3.54)

where  is the generalised momentum p and called flux linkage and is measured in Weber
turns (Wb).

Now we can recognise that Eq. 3.54 represents a generalised inertia, equivalent to the mass in
mechanical systems.

3.4.4 Electrical Resistance

The characteristic behaviour of metals such as copper and aluminium is that thermal energy
arises whenever current is caused to flow through a metal. This characteristic behaviour is
referred to as the resistance of the material to the flow of current. The circuit element used to
64 System Modelling

model this behaviour is the resistor with a circuit symbol as shown in Fig. 3.28.

+ - e= u1 - u2
R
u1 u2 f=i

Fig. 3.28 Circuit symbol for an electrical resistor

A resistive element is described by an algebraic relationship between the terminal voltage

and current written as

u 1 – u 2 = Ri (3.55)

Eq. 3.55 is known as “Ohm’s law”, after George Simon Ohm, a German physicist who estab-
lished its validity early in the nineteenth century.

3.4.5 Electrical Transformer

The electrical transformer makes a contribution in that the voltage-current relations in a cir-
cuit may be changed. Transformers consists of primary and secondary coils of wire wound on
a common magnetic core and is a device that uses magnetic coupling as a basis for its opera-
tion. A changing voltage applied to the input or primary coil causes a changing current to
flow thus creating a changing magnetic flux in the core. Due to the changing flux, voltage is
induced in the output or secondary coil. Since only changing currents are transformed, an
output circuit can be isolated from a direct-current (DC) component of the input. In Fig. 3.29
is shown the graphical symbol for the ideal transformer.

i2
N1 / N 2
u1 u2
u1 N1 N2 u2 TF
i1 i2

Fig. 3.29 Graphical symbol for a electrical transformer

The vertical bars are used to imply that an ideal transformer can be approximated by coils
wound on ferromagnetic cores.

The terminal behaviour of the ideal transformer can be described by two characteristics.
First the magnitude of the volts per turn will be the same for each coil, that is
 Electrical System Models 65

u u
-----1- = -----2- (3.56)
N1 N2

where u1 and u2 are the input and output voltages and N1 and N2 are the number of turns on
input and output coils. The second characteristic is that the ampere-turns will be the same on
each coil.

i1 N1 = i2 N2 (3.57)

where, i1 and i2 is the current on each side of the transformer.

From these two characteristics, we can write the constitutive laws for the ideal electrical
transformer as

N N1
u 1 = -----1- u 2 ------ i 1 = i 2 (3.58)
N2 N2

and the ratio of the number of turns in the two windings is then the transformer modulus

N
m = -----1- (3.59)
N2

By adjusting the ratio of turns on the two coils, we can obtain a voltage “step up” or “step
down” or a current step up or step down.

3.5 Electrical System Models

We are now ready to start modelling electrical systems and the first problem will be to
develop a bond graph of an electric circuit. In a later chapter are we going to model some
alternating current (AC) drive systems. As we did for mechanical systems, a systematic mod-
elling procedure will be presented. This systematic construction method is as follows:

1. At each node or junction in a circuit establish a 0-junction.

2. For each 1-port element establish the voltage difference over the element by bonding
a 1-junction between appropriate pair of 0-junctions. Adjoin the pertinent 1-port ele-
ment to the 1-junction.

3. Assign power directions to all bonds.

66 System Modelling

4. Delete the 0-junction representing ground potential and delete the bonds connected to
this 0-junction

5. Simplify the bond graph.

We will start the bond graph modelling with a simple circuit as the one shown in Fig. 3.30.

i R1 L1
a + + d
- -
+ b
- - -
i(t) C1 C2 L2
- + + +
g

Fig. 3.30 Electric circuit

Before we present the systematic method approach, let us see if we can obtain the bond graph
model by direct inspection of the circuit schematic before we use the systematic construction
method.

By inspecting the circuit we can identify three connection patterns:

1. The current source S f and capacitor C 1 , are in parallel and therefore have the same
voltage which means that they are both bonded to a 0-junction.

Sf 0

2. The resistor R 1 , and inductor L 1 , have the same current and hence are both bonded to
a 1-junction.

R
 Electrical System Models 67

3. The capacitor C 2 , and inductor L 2 , have the same voltage and therefore both are
bonded to a 0-junction.

0 I

By bonding together the three connection patterns we obtain the complete bond graph model
shown in Fig. 3.31.

Sf 0 1 0 I

C I C

Fig. 3.31 System bond graph

The remaining part is now to assign power direction to the remaining bonds. This task is eas-
ier in electric circuits than in mechanical systems. The polarity reference for the voltage is
indicated by plus and minus signs on the schematic drawing and the reference direction for
the current is shown by the arrow placed alongside the current. The assignment of the refer-
ence polarity for voltage and the reference direction for current is entirely arbitrary.

Sf 0 1 0 I

C I C

Fig. 3.32 Complete system bond graph using the inspection method

Thus, if the current polarity reference is in the direction of a voltage drop power is positive.
68 System Modelling

This means that assuming current flow from left to right and power assumed to be delivered
by the sources the power directions will be as shown in the completed bond graph shown in
Fig. 3.32.As the next step let us construct the bond graph of the circuit shown in Fig. 3.30
using the systematic construction method. Hopefully, the final result is the same in the two
cases. The starting point is to write a 0-junction for each node in the circuit as shown in Fig.
3.33a. The 1-port elements are then inserted between appropriate 0-junctions via a 1-junction
as shown in Fig. 3.33b. The ground potential e g is considered as the reference potential, i.e.
a voltage, and is therefore equal to zero. This means that all bonds connected to the 0-junc-
tion representing the ground potential have zero power and can therefore be removed.

Assuming positive power direction from left to right and from the two 0-junctions and into
the two C-elements and the I-element. The resulting bond graph is then as shown in Fig.
3.33c.

We have four 2-port 1-junctions with through power direction. This means that they can be
reduced to a single bond. The final bond graph is then as shown in Fig. 3.33d which is the
 Electrical System Models 69

same graph as shown in Fig. 3.32 as we expected.

0 0 0
ea eb ed

0 0
eg eg

(a)

R I

0 1 0 1 0

Sf 1 1 C C 1 1 C

0 0
(b)

R I

0 1 0 1 0

Sf 1 1 C C 1 1 C
(c)

Sf 0 1 0 I

C I C

(d)

Fig. 3.33 Bond graph for circuit - Example 2

70 System Modelling

3.6 Hydraulic Systems

Many engineering systems contain working fluids in motion under pressure. In practice, they
are connected with pipes or tubes to form networks in a wide variety of applications to
numerous to mention. The dynamics of fluids, liquids and gases is fundamentally an
extremely complicated thing to study. However, progress is made in understanding fluid
dynamics because some people with marvellous physical insight have been able to contrive
physical models which where simple enough to analyse, and yet which have represented
many of the important problems in fluid flow well enough to predict behaviour with high
accuracy. In hydraulic systems, it is in the stage of physical modelling that, to a large degree
the possibility and usefulness of system analysis are determined.

In spite of the multitude of hydraulic system applications, they can be divided into two main
categories:

• Low-pressure large flow systems often called fluid transfer systems.

• High-pressure small flow systems which most often are denoted as fluid power sys-
tems.

The so-called liquid transfer systems may be cooling systems or pure liquid transport sys-
tems. In many cases thermal effects are also involved in these systems. When that is the case
we call it thermofluid systems where fluid mechanics and thermodynamics are both needed in
our treatment. Modelling such systems can not be performed using the basic 1-port elements
defined previously and will therefore be treated in later chapters when we have learned about
multiport generalisation of the 1-port elements.

In the majority of liquid transfer systems the fluid flow is an end in itself. However, in fluid
power systems, the fluid flow is only an adjunct to a more fundamental purpose, transmit
power from one point to another. The fluid can be a liquid, usually oil or a gas such as air. At
this time in our development, only systems using liquids as the working fluid will be consid-
ered. Compressible fluid systems will be treated in a later chapter.

As was done for the mechanical and electrical domain, we will first introduce a number of
physical hydraulic components which can be modelled by the basic bond graph elements
before performing system modelling.

3.6.1 Fluid Flow Sources

In any hydraulic system, a flow of fluid is created by pumps. The pump is probably the oldest
fluid-energy-transfer device known for which there are two basic types:
 Hydraulic Systems 71

• positive displacement pumps

• rotor dynamic pumps

Positive displacement pumps can again be sub-divided into two main categories

• reciprocating pumps
• rotary pumps

each of which gives rise to a number of specific sub-types.

The basic reciprocating pump comprises a piston or plunger which is reciprocated in a cylin-
der as shown schematically in Fig. 3.34

ωp

Fig. 3.34 Reciprocating pump

With a pump stroke of L , a piston diameter D and pump speed p, the theoretical pump
·
delivery V p can be computed from the following expression

2
· D
V p = ------ L p = V R  p (3.60)
8

where V R is the pump volume per radian. Ideally, the volumetric flow delivered by such a
pump per revolution of the pump shaft, is independent of the pressure against which the
pump works.

From Eq. 3.60 we can model the reciprocating pump as a fluid flow source when  p is con-
stant as shown in Fig. 3.35

P
Vp :Sf
V

Fig. 3.35 Ideal constant RPM displacement pump

72 System Modelling

With variable  p the reciprocating pump can be modelled as a transformer

MP VR
P
TF
ωP V

Fig. 3.36 Ideal variable RPM displacement pump

with the pump displacement V R as transformer modulus.

In fluid power applications multi-cylinder reciprocating pumps are often used and in Fig.
3.37 is shown two examples of such pumps.

Valve plate
(does not rotate)

Swash plate (a) Valve plate

(does not rotate)

Cylinder block

 Valve plate
(does not rotate)

(b)

Fig. 3.37 Multi-cylinder axial piston pumps, a) swash plate pump, b) bent-axis pump

In Fig. 3.37a, the housing containing the pistons and the cylinder barrel is turned by the pump
drive shaft. The pistons are caused to reciprocate axially by a shoe plate to which the pistons
are connected by ball joints. The shoe plate bears against a swash plate situated at an angle to
the axis of the cylinder barrel. Thus, during a half-revolution of the cylinder barrel, the pis-
tons are permitted to extend by the angle of the swash plate while drawing in oil through the
intake ports. During the subsequent half-revolution of the cylinder barrel, where the pistons
are in contact with the acute angle of the swash plate, they move in the opposite direction,
discharging oil through the exit ports.
 Hydraulic Systems 73

The swash-plate pump can be made into a variable-displacement pump by having the angle
of the swash plate controllable. When the plane of the swash plate becomes perpendicular to
the axis of the cylinder barrel, the pistons will cease to reciprocate and their delivery will go
to zero. Maximum delivery will occur when the plane of the swash plate is situated at its
maximum possible angle with respect to the cylinder barrel axis.

The capacity of the swash plate pump can be determined from the formula

V· = A p n ------  p tan  = V p  p tan 

D
(3.61)
2

and the torque needed to drive the pump then becomes

T = V p tan   P (3.62)

where V· is the pump capacity (m3/s), T is the driving torque (Nm), A p is the piston cross-
sectional area (m2), n is the number of pistons, D is the pitch diameter (m),  is the inclina-
tion angle,  p is the pump rotational velocity (rad/s) and V p = A p nD   2  is the pump
displacement per radian (m3/rad).

The swash-plate reciprocating pump can be represented with a modulated transformer as

shown in Fig. 3.38.

tanα

M P
ωp
MTF
V

Fig. 3.38 Swash-plate reciprocating pump

The pump shown in Fig. 3.37b is called a bent-axis pump. The bent-axis pump works on the
same general principle as the swash-plate pump, but its pistons are caused to reciprocate axi-
ally by driving the cylinder barrel through a universal joint and connecting shaft situated at
an angle to the rotating driving flange into which the pistons are seated by ball-jointed con-
necting roads. By controlling the angle between the drive shaft and the cylinder barrel, the
bent-axis pump can be made to be a variable-displacement device. When the axes are paral-
lel, the pistons cannot reciprocate and the discharge is zero.

The capacity of the bent-axis pump can be determined by Eq. 3.61 when tan    is replaced
with sin    . That is
74 System Modelling

V· = V p  p sin  (3.63)

and required torque can be determined by

T = V p  p sin   P (3.64)

As for the swash-plate pump, the bent-axis pump can be represented with a modulated trans-
former as shown below

sin(α)

M P
ωp
MTF
V

Fig. 3.39 Bent-axis pump bond graph

Turbo machines
A turbo machine is a device which adds energy to a fluid or extracts energy from it. If the
machine adds energy to the fluid, it is called a pump, if it extracts energy, it is called a tur-
bine.

A typical rotary pump is the centrifugal pump where the delivery from the pump reduces as
the pressure against which it has to operate increases. This means that the delivery or output
flow is depending on the system to which it is connected and therefore, is not a true source as
we have defined it. However, since the centrifugal pump is a major unit of flow creation we
will present an elementary centrifugal pump performance model.

The description centrifugal pump is generally given to machines employing an impeller with
fixed blades housed in a suitably shaped casing. Fluid enters axially through the eye of the
impeller blades, and is whirled tangentially and radially outward until it leaves through all
circumferential parts of the impeller into the diffuser part of the casing. The pumping action,
or liquid transport from the inlet to the outlet side is thus achieved by virtue of an increase in
momentum applied to the liquid.

In spite of the long history of pump development, pump theory is still rather tentative and
schetchy, and a large portion of design still come from testing and experience. After all, tur-
bomachines have compex threedimansional geometries with strong viscous effects which
preclude any simple theory.
 Hydraulic Systems 75

At this stage we will therefore present the centrifugal pump in terms of performance charts.
Such charts are almost always plotted for constant shaft-rotation speed n in revolutions per
minute. The basic independent variable is taken to be discharge V· in m 3  min . The depend-
ent variables or output is taken to be pressure rise or P , power and efficiency  . In Fig.

P P


V·

Fig. 3.40 Centrifugal performance curve

3.40 typical performance curves for a centrifugal pump is given.

The pressure rise is appriximately constant at low discharge and then drops to zero. If we
approximate the pressure rise versus discharge performance curve to the following equation

P = P out – P in = a 0 – a 1 V· – a 2 V· 2 (3.65)

then we can construct a bond graph model of the centrifugal pump as shown in Fig. 3.41.

Se :a0

Pin :Se 1 Se :Pout

R:(linear) R:(non-linear)

Fig. 3.41 Bond graph model of a centrifugal pump

3.6.2 Fluid Capacitors

In fluid systems there are often elements that store flow part of the time for delivery at
76 System Modelling

another time. This fluid storage is made possible through a fluid capacitor.

Pa
PT - pressure at tonk bottom

AT AT - crossectional area

h - liquid level

h p - liquid density

g - gravitational acceleration
PT
Pa - atmospheric pressure

- inflow volume rate

V

Fig. 3.42 Fluid open tank capacitor

One of the simplest examples of a fluid capacitor is an open tank located in a gravity field
and supplied with a liquid through a port or pipe at the bottom as shown in Fig. 3.42.

With no fluid acceleration of importance and no fluid resistance effects, the pressure differ-
ence between the top and bottom, the tank must support the weight of liquid. That is

g
P T – P a = gh = ------  Ah  (3.66)
A

The term (Ah) is the volume V , of liquid in the tank and Eq. 3.66 may be written as

g g
P T – P a = ------  V· dt = ------ V (3.67)
A A

The atmospheric pressure P a will be considered as the zero reference pressure and we will
therefore only consider gage pressure. If we then eliminate P a and rewrite Eq. 3.67 in the
form

A
V = ------ P T (3.68)
g

it is easily recognised from the definition of a capacitor that the capacitance for the tank is

A
C T = ------ (3.69)
g

A second example of a phenomenon which may behave approximately like a pure fluid
 Hydraulic Systems 77

capacitance is a tank or volume as shown in Fig. 3.43.

Vi PT ,V Ve PT PT
0
Vi Ve

(a) (b)

Fig. 3.43 Volume of compressible liquid

All liquids can be compressed slightly if sufficient pressure is applied to the liquid. In high-
pressure systems, compressibility may be particularly important in spite of the fact that in
many practical situations, the fluid is considered to be incompressible.

Since fluids do not process rigidity of form, the modulus of elasticity must be defined on the
basis of volume. Thus, compressibility or stiffness k of the fluid is defined as the fractional
reduction in volume of fluid per unit increase of applied pressure. In equational form this
means that

1 dV
k = – --- (3.70)
VdP

where V , is the volume and P is the pressure.

It is customary in engineering to express this elastic property as the reciprocal of the com-
pressibility, termed the bulk modulus ,

P
 = –V (3.71)
V

A finite difference form of the expression for  is usually written as

P
 = – V ------- (3.72)
V

The bulk modulus is perhaps the one fluid parameter that causes concern in its numerical
evaluation due to effects which modify it, such as gas inclusive.
78 System Modelling

To increase the pressure of the content of a rigid vessel of volume V from pressure P 1 to P 2
requires an extra volume of liquid to be pumped into the vessel equal to

V
V = ---  P 1 – P 2  (3.73)


The relationship given in Eq. 3.73 is recognised as the constitutive relation for a capacitor
with the compliance given as

V
C = --- (3.74)


An other example of a hydraulic capacitor is a unit called an accumulator. It is a chamber

that can store a measurable volume of oil under pressure. The majority of fluid power accu-
mulators is what is known as the gas-charged accumulator, and is simply a gas filled bladder
above the oil as shown in Fig. 3.44.

Gas
P, T
V

Liquid

Fig. 3.44 Bladder-type accumulator

The gas chamber and the oil chamber are separated by a bladder of synthetic rubber. The gas
chamber is brought to a predetermined pressure with air or nitrogen. Oil entering the accu-
mulator displaces an equal volume of gas. When the pressure in the external circuit falls
below that in the accumulator, the gas expands and forces fluid into the circuit.

For gaseous charged accumulators, the state conditions can be calculated by assuming that
gas compression and expansion occur either

• isothermally
• isentropically
• polytropically

If compression or expansion occur slowly, the isothermal assumption is appropriate. If these

processes occur rapidly, the isentropic assumption applies. For an isothermal process we can
write
 Hydraulic Systems 79

PV = P 0 V 0 (3.75)

where P is the instantaneous pressure V is the instantaneous volume, P 0 is the reference

system pressure and V 0 is the reference volume.

The instantaneous volume of gas is given by

V = V 0 –  V· dt (3.76)

Substituting Eq. 3.75 for V into Eq. 3.76 and solving for the instantaneous pressure P we
obtain

P0 V0
P = ----------------------
- (3.77)
V 0 –  V· dt

which is the constitutive relation for the accumulator as shown in Fig. 3.45.

P0

V0 V

Fig. 3.45 Constitutive relation for isothermal gas process in gas-charged accumulator

Returning to the cases of isentropic or polytropic process, then the constitutive relation takes
the form

k
V0
P = ----------------------
- P0 (3.78)
V 0 –  V· dt

where k is the specific heat ratio for an isentropic process  ,or the polytropic constant n .

3.6.3 Hydraulic Actuators

An actuator is a component used to convert the fluid energy back into mechanical power.
There are two basic types of hydraulic actuators:
80 System Modelling

• Linear - hydraulic cylinder

• Rotary - hydraulic motors

Linear actuators, as their name implies, provide motion in a straight line and they are usually
referred to as cylinders.

A hydraulic cylinder consists of a piston and rod located within a ported cylinder as shown in
Fig. 3.46.

Pb Pr a
FL
A

Vb Vr

Fig. 3.46 Hydraulic cylinder

Thus, the hydraulic cylinder is a fluid power actuator that converts pressure and volumetric
discharge into force and linear velocity. A bond graph for the hydraulic cylinder is shown in
Fig. 3.47.

to load

FL vr
A (A-a)
Pb Fb Fr Pr
TF 1 TF
Vb vr Vr

Fig. 3.47 Bond graph for hydraulic cylinder

In Fig. 3.48 is shown an extended bond graph model of the hydraulic cylinder shown in Fig.
 Hydraulic Systems 81

3.46.

C to load C

FL vr
A (A-a)
Pb Fb Fr Pr
0 TF 1 TF 0
vr
Vb Vr

inflow/outflow I outflow/inflow

Fig. 3.48 Extended bond graph model of hydraulic cylinder

In this extended model the energy storage mechanisms are accounted for by adding associ-
ated inertial and capacitance effects. The piston and rod inertia is modelled by an I-element
connected to the 1-junction and the capacitive effects of the two volumes is modelled by C-
elements connected to the two 0-junctions.

The constitutive relation for the two C-elements can be written as


P l = -------------------- V l (3.79)
V l0 + Ax

and


P r = -------------------- V r (3.80)
V r0 – Ax

where  is the bulk modulus of fluid, V l0 is the start volume of piston side, V r0 is the start
volume of rod side, x is the piston displacement and V r and V l is the change in fluid vol-
ume in V r and V l respectively.

Rotary actuators, generally known as hydraulic motors are similar to hydraulic pumps, but
whereas a pump shaft is rotated to generate a flow, a motor shaft is caused to rotate by fluid
being forced into the driving chambers. Hydraulic motors are available as gear, vane and pis-
ton versions. Piston driven hydraulic motors are the ones which is mostly used. They may
have either a fixed or variable displacement and Eq. 3.62 and Eq. 3.64 developed for the pos-
itive displacement pumps also apply to positive-displacement hydraulic motors.
82 System Modelling

In Fig. 3.49 is shown the symbol for a hydraulic motor and a bond graph of the same unit.

Pi Pi
V α

Pm
1 MTF ωm

Pe Pe
V

Fig. 3.49 Hydraulic motor symbol and bond graph

3.6.4 Hydraulic Resistance

In normal flow of fluid in a pipe or through a passage of changing area, a drop of pressure
occurs in the direction of the flow. There are two basic types of fluid resistance commonly
encountered which we will look at in the following. They are

• resistance in laminar pipe flow

• resistance in turbulent pipe flow

These different types of flow have different value of the flow resistance. Many empirical for-
mulas are used for pipe friction calculations in engineering practice, but the mostly used
empirical relationship is of the form

1 L V· 2
P = f f  --- ----  -----2- (3.81)
2D A

where P is the pressure drop, f f is the friction factor, L is the length of the pipe, D is the
pipe diameter,  is the fluid density, V· is the volume flow rate and A is the pipe cross-sec-
tional area.

To determine which of the flow types, laminar or turbulent which occur in a pipe with diam-
eter D , velocity U of the flow and the kinematic viscosity v of the fluid, we use the dimen-
sionless quantity

UD
Re = --------- (3.82)
v

called the Reynolds number. There exists a value of this number called the critical Reynolds
 Hydraulic Systems 83

number which has a value of about 2100, and then

Re  2100 , the flow will be laminar

Re  2100 , the flow will be turbulent

The laminar flow regime is characterized by the fact that the friction factor f f , can be repre-
sented by a single equation of the form

64
f f = ------ (3.83)
Re

which is independent of surface roughness. Substituting the laminar friction factor into Eq.
3.81, then the relation for pressure drop in laminar flow becomes

128L ·
P = -------------------
4
V (3.84)
D

This shows that in laminar flow, the pressure drop varies linearly with the flow rate.

In the turbulent flow regime, a relationship exists between the friction factor and the Rey-
nolds number for every relative surface roughness   D . In Fig. 3.50 is shown a logaritmic
plot of friction factors against Reynolds number with relative surface roughness as parame-
ter.

-1
10
0.1
0.08
20
Friction factor

0.05 50

100

0.03 200

500
0.02 1000
2000 D
5000
10000
-2
0.01
10 50000
100000
1000000
1000000
3 4 5 6 7
103
10 10
104 10
105 10
106 10
107
Reynolds number

Fig. 3.50 Friction factor f for laminar and turbulent pipe flow
84 System Modelling

Pressure-Flow Relations of Hydraulic Valves

In Werbster’s New Collegiate Dictionary, valves are defined as: “Any device by which the
flow of liquid, air or other gas, etc. may be started, stopped or regulated by a movable part
which opens or obstructs passages”. An often-quoted definition is “an engineered obstruc-
tion in a pipe”. Although this definition is proposed in humour it is technically correct. The
earliest “engineered obstruction” is accredited to Leonardo da Vinci in the sixteenth century.

Fluid flow through this restriction causes a drop of pressure in the direction of flow. The
relationship between the pressure across and flow rate through most hydraulic resistances
follow at least approximately the orifice equation

V· = C d A --- P v
2
(3.85)


where V· is the rate of flow, C d is the discharge coefficient, A is the area of valve orifice,
P v is the pressure drop across the orifice and  is the density of the fluid.

The coefficient of discharge is a function of Reynolds number and the geometric configura-
tion of the valve port. It is a constant unless the valve opening is very small or the pressure
drop is very low. If the orifice edges are sharp C d = 0.6 to 0.65, and if they are rounded C d
will be 0.8 to 0.9 or even more. Orifice shape seems to have very little effect.

By shaping the ports of a valve, or by manipulation of the opening and closing movement,
almost any flow-characteristics can be obtained. Thus, by changing the area A in Eq. 3.85
by valve motion x , we obtain the following flow-pressure-valve-motion relation for a control
valve as

V· = C d A  x sign  P v  --- P v
2
(3.86)


and which can be represented by the modulated resistance element as shown in Fig. 3.51.

e = ΔP
f=V
R X

Fig. 3.51 Bond graph valve model

3.6.5 Fluid Inertance

Fluid inertance arises from inertial forces required to accelerate a fluid in a pipe. To visualise
 Hydraulic Systems 85

the concept of fluid inertance, let us consider slug flow as illustrated in Fig. 3.52. Slug flow
is a term used to convey the idea of a fluid in motion as a rigid mass.

V1 V2
P1 P2
Ineterance, I

V1 V2
l
P1

Fig. 3.52 Fluid inertance

Assuming slug flow, we can consider that the forces on the fluid slug and its velocity obey
Newton’s second law written as

d· dU
F = m x = m (3.87)
dt dt

The force acting on the slug can be expressed as

F = A  P1 – P2  (3.88)

where A is the cross-sectional area of the pipe and P 1 – P 2 is the pressure difference across
the slug.

With the mass of the fluid slug given as

m = LA (3.89)

where  is fluid density and L is pipe length and assuming that the velocity U can be
expressed as

V·
U = x· = --- (3.90)
A

then Eq. 3.87 can be written as

86 System Modelling

1d
A  P 1 – P 2  =  LA  ---  V·  (3.91)
Adt

Integrating both sides results in

L
  P1 – P2  dt = p = ------ V· (3.92)
A

and the fluid inertance is given as

l
I = ----- (3.93)
A

It is interesting to observe that the fluid inertance decreases when the diameter of the pipe
increases. Effects of fluid inertia should be considered in pipes with large rates-of-change in
flow.

3.7 Hydraulic Systems Modelling

As for the case of modelling a mechanical or electrical system, we can also use a systematic
construction procedure for hydraulic systems. This procedure is as follows

1. Establish a 0-junction for each distinctive pressure

2. Insert the component models via a 1-junction between the two appropriate 0-junc-
tions.

3. Add pressure and flow sources to appropriate 0-junctions.

4. Assign power directions.

5. Define all pressures relative to atmospheric reference pressure and remove the 0-junc-
tions representing reference pressure and their connected bonds.

6. Simplify the bond graph.

Let us now demonstrate the procedure by constructing models of hydraulic systems. We will
treat one example of what we call fluid transfer systems and another second example of a
fluid power system. In Fig. 3.54a is shown a simple fluid transfer system. The system con-
sists of a centrifugal pump which on the suction side is connected to an open tank placed a
distance h below the pump inlet.On the outlet of the pump is placed a control valve (Valve 1)
and a second valve (Valve 2) is placed in the bypass circuit of the pump. This valve is pres-
sure controlled in that Valve 2 opens when the pump output pressure exceed a set value. To
 Hydraulic Systems Modelling 87

the output valve, Valve 1, is connected a pipeline which ends in a tank with a circular cross-
sectional area.Before we start constructing the bond graph we will discuss some assumptions
which are important for the resulting model.

Pa

PA PB PD
PT
Valve 1
Hs Pa
Valve 2

(a) 0 Pa

PA PB PD
0 0 0 0 PT

0 Pa
(b)

Fig. 3.53 Hydraulic fluid transfer system - example

The discharge tank has a cross-sectional area which is such that the liquid level increases
considerable and hence do the stored energy in the tank. This means that the tank must be
modelled as a capacitor or C-element. For a tank with a large cross-sectional area where the
liquid level do not change much, it would be sufficient to model the tank as an constant effort
source. In the pipeline there are effects of fluid inertia, compressibility and resistance to flow,
all distributed throughout the line. In this example however we will construct a discrete
model but will return to models of distributed systems at a later stage. The line between the
pump and the tank will for illustrative purposes be represented as one lump and thus, one ele-
ment for each of the inertia, compressibility and resistance effects.

Let us start the modelling by establishing a 0-junction for each distinct pressure point which
in the pipe line means between each of the effects I and R as shown in Fig. 3.53b. Next we
insert a 1-junction between each 0-junction. To each of the mentioned 1-junctions we connect
the appropriate 1-port elements. Now, starting to explain the result from the left side of the
bond graph shown in Fig. 3.53, we find: Between P a and P A the effort source represents the
static head of the suction line. Between P A and P B is inserted the centrifugal pump model
presented in Fig. 3.41 in parallel with the bypass valve Valve 2 which is modulated by the
pump output pressure P B . The R-element between P B and P D represents the output valve
Valve 1. The compressibility effect of the liquid in the pipe assuming rigid pipe walls is mod-
elled by the C-element between P D and P a . It is important to recognise that a C-element in a
88 System Modelling

hydraulic system must always be referenced to the reference pressure which is normally the
atmospheric pressure. The inertia effects is modelled as an I-element between P D and P 1
and the flow resistance is modelled by the R-element between the points P 1 and P T . From
the assumptions discussed earlier the tank is modelled as a C-element between P T and P a .

Assuming positive flow directions from the inlet to the outlet tank, the bond graph model
shown in Fig. 3.54 can be simplified removing all 0-junctions representing the reference
pressure and eliminating junction elements with through power direction we end up with the
bond graph shown in Fig. 3.54.

Se R C I

Se 0 1 0 1 0 1 C

R R
R

Fig. 3.54 Hydraulic fluid transfer system - example cont.

From this graph it can be seen that the order of the I- and R-elements representing the physi-
cal effects in the pipeline is arbitrary since they ended up as bonded to the same 1-junction.

Hydraulic accumulator x·
Hydraulic cylinder
Primary Drive

Hydraulic Pump
m
A B D

E Inertia load
Filter
Hydraulic
Pressure Relief Valve motor
4-way Manual Control Valve
G
Reservoir Drain Tank D
E J
(a) (b)
Rotary Inertia Load

Fig. 3.55 Fluid power system example

 Hydraulic Systems Modelling 89

As a second example we will model a fluid power system. In Fig. 3.55 is shown a schematic
drawing of a hydraulic system driving an inertia load. The system consists of a variable dis-
placement pump which have a filter, pressure relief valve and accumulator located in front of
a 4-way manual control valve. The valve is made so that the load can be moved in both direc-
tions depending on the valve position. The drive system could be either through a hydraulic
cylinder or through a hydraulic motor. We will model both systems in this example.

One advantage of the bond graph approach is that subsystems can be modelled independently
and then assembled into an overall system model in much the same way as assembling the
actual hardware. Let us therefore, utilise this approach to construct a model of the example
system shown in Fig. 3.55a.The usual modelling procedure is to set off 0-junctions for each
distinct pressure and move on to step two and insert 1-junctions between a pair of 0-junc-
tions and bond C, I, or R-elements to appropriate 1-junctions. In this example we will intro-
duce a slightly different approach to system modelling.

The major components in the system is easily identified to be:

• Pump and drive unit

• Filter
• Accumulator
• Pressure relief valve
• 4-way control valve
• Cylinder with load mass
• Hydraulic motor with load inertia.

The pump and drive unit can simply be modelled as shown in Fig. 3.56.


T P
Sf MTF
 Q

Fig. 3.56 Pump unit bond graph

The primary drive in the pump unit is assumed to be a source delivering an angular velocity.
The pump itself is modelled as a modulated transformer since it is given that the pump has a
variable displacement. For simplicity we omit the modulus signal in the next figures. For the
filter, we will only consider the pressure drop through it and therefore, we can model it as a
single R -element. The accumulator model is discussed earlier in this chapter, where it was
presented as a C -element. Pressure relief valves are used in hydraulic circuits to limit the
90 System Modelling

pressure in the system to a predetermined value P set . That is, the valve is closed when the
pressure input to the valve is less than P set and fully open when the system pressure exceeds
P set . The bond graph model of the pressure relief valve is thus a modulated R -element as
shown in Fig. 3.54 with the system pressure as the modulating signal.

The 4-way control valve is an element which is often used in fluid power systems to manu-
ally or electronically control the speed and direction of actuators as in this case, a hydraulic
cylinder or hydraulic motor. In Fig. 3.57 is shown a circuit symbol, schematic drawing and
the corresponding bond graph model of the 4-way valve. The four R -elements shown in the

R x x 
1 v

xv=[-1,0,1] B 0 1 0 D

D x x  R 1 1 R x x 
3 v 4 v
B D B
G E
G
G 0 1 0 E

a) b)
R x2  xv 

Fig. 3.57 4-way valve, a) hydraulic circuit symbol, b) schematic drawing and
c) corresponding bond graph

bond graph represents the pressure drop flow relationships through each of the four parts in
the valve.The constitutive laws for all the R -elements must be written as a function of the
spool position x v .

Let us now go through the valve operation and explain how it functions. Starting with mov-
ing the valve spool down or in the positive x v -direction, fluid will now flow from the supply,
port B, through port D to the actuator, and from the exhaust side of the actuator through port
E to the drain port B. In the bond graph this means flow through R:x1 and R:x2. Moving the
spool up or in the negative x v -direction, the fluid flows from the supply port B through port
E to the actuator and from the exhaust side of the actuator through port D to the drain port
G. In the bond graph we now hav flow through the R -elements R:x3 and R:x4. This means
that only two ports are open simultaniously. That is, two ports are open when the displace-
ment x v of the spool is positive, and the other two ports are open when the spool displace-
ment is negative. We will deal with this fact later in the text when we are writing system
equations.
 Hydraulic Systems Modelling 91

The bond graph models for the linear and rotational actuators have been develped previously
as shown in Fig. 3.48 and Fig. 3.49 respectively. Now, when all the submodels are developed,
the total system model can be constructed by assembling them together as would be done by
the physical components. The resulting system model shown in Fig. 3.58.

Filter Accumulator 4-way valve Cylinder

R C R:x1 C
Pump Rel. valve
P
A
1 0 1 0 0 1 0 0 TF
P P P P
B B D D
MTF x3:R 1 1 R:x4
1 R I 1

Sf 0 0 1 0 0 TF
P P P P
G G E E

Se:PG Se:PG R:x2 C

Reservoir Drain tank

Fig. 3.58 Fluid power system model

As the system bond graph model in this example show, care in drawing the bond graph struc-
ture is important in order for it to resemble as much as possible the physical system structure
and thus enhance the information value of the graph.

If we exchange the linear motion hydraulic cylinder with a hydraulic motor with a rotary
inertia load as shown in Fig. 3.55b and at the same time simplifying the graph, i.e. removing
0 -junctions with athmospheric pressure, combining nodes and removing through-nodes, the
92 System Modelling

bond graph resulting is shown in Fig. 3.59.

R:filter C:Acc R:x 1

0 1 0 1 0 I:J
MTF
R:x3 1
R:Rel.valve R:x4 1 TF 1

Sf
R:x2 0

Fig. 3.59 Hydraulic motor with inertial load

3.8 Thermal Systems

Thermal systems is an interdisciplinary field employing the traditional disciplines of thermo-
dynamics, heat transfer, fluid mechanics, mass transfer and chemical kinetics. The field of
thermal systems is a very broad one involving many separate engineering disciplines.
Accordingly, developments in thermal systems have a major impact not simply on a large
number of different applications, but on a large fraction of industrial activity as well as our
daily lives. Taken as a whole, the field of thermal systems provides an integrating framework
within engineering sciences mentioned above. Even though thermal systems are common-
place in the engineering world, very limited effort has to date been invested in development
of modelling and simulation methods for such systems.

The field of thermal systems is wide-ranging and in part require sophisticated modelling con-
cepts. This will be treated in depth in later chapters. In this chapter, we will introduce bond
graph modelling of simple thermal systems through some heat transfer problems.

In modelling thermal systems temperature T is often proposed as the effort variable and
entropy flow s· as the flow variable. This is a natural choice in bond graph methodology
since their product gives power. However, in heat transmission work done by engineers, tem-
perature T has been used as the effort variable and heat flow rate Q· as the flow variable.
Since this choice of variables is well known by engineers, we will use them here and we will
find that they are very useful also when we in later chapters will model large complicated
thermal systems. The product of temperature and heat flow is not power. Indeed, heat flow is
itself a power-like variable. Bond graphs in which effort and flow variables are not power
 Thermal Systems 93

variables are called pseudo bond graphs. In what follows, we will define basic bond graph
elements for heat transfer problems and apply them in examples of system modelling.

It is conventional in the subject of heat transfer to denote all energy flows that arise because
of a temperature difference as heat transfer. For this reason, the subject of heat transfer might
be more properly called thermal-energy-transfer.

Heat transport is unconsciously viewed as a more or less static process. Here we will view
heat transport as a dynamic process which behaves very much as though heat is actually flow-
ing through equipment. Therefore, we think of heat transport as a process involving the flow
of heat. We will take heat flow dynamics to mean the dynamic behaviour of heat flow proc-
esses in general and the functional relationships between the thermal and system parameters
in particular.

Heat transfer or heat is energy in transit due to a temperature difference. There are three dif-
ferent types of heat transfer processes often referred to as modes. These modes are called

• conduction
• convection
• radiation

All three modes have in common that temperature differences must exist and that heat is
always transferred in the direction of decreasing temperature. On the other hand, they differ
entirely in the physical mechanisms and laws by which they are governed.

3.8.1 Conduction
Conduction is used for the heat transfer which occur when a temperature gradient exist in a
stationary medium, solid or fluid. As a rule, solids conduct heat better than liquids and liq-
uids better than gases. Heat conduction is passage of heat energy through a body resulting
from molecular motion, and not due to rays or relative movement of parts of the body. The
amount of energy being transferred by conduction can be determined by Fourier’s law. For
94 System Modelling

the one-dimensional plane wall shown in Fig. 3.60,

T1 T(x)
Qx T2

L x

Fig. 3.60 Heat conduction in a wall

having a temperature distribution T  x  the rate Fourier’s law as a equation is expressed as

dT
q· = – k (3.94)
x dx

The heat flux q· is the heat transfer rate in the x direction per unit area perpendicular to the
x
direction of transfer. The constant k is a transport property, known as the thermal conductiv-
ity. Thus thermal conductivity is a specific item which is a property of the material of which
the wall is composed. The sign convention used in writing Eq. 3.94 assumes the heat flux to
be positive if it is in the direction of the coordinate axis.

The term one-dimensional used above refers to the fact that only one coordinate is needed to
describe the spatial variation of the dependent variable. With the assumption that the temper-
ature distribution is linear through the wall the steady flow of heat can be expressed by the
relation

Q· = ------  T 1 – T 2 
kA
(3.95)
L

where Q· is heat flow, k is thermal conductivity, A is the area normal to the heat flow, L is
the length of the conductor and T 1 and T 2 are the temperature of left and right side respec-
tively.

Taking a closer look at Eq. 3.95, we can see that a thermal resistance may be associated with
the conduction of heat just as electrical resistance is associated with the conduction of elec-
tricity.

Defining resistance as the driving potential to the corresponding transfer rate then thermal
resistance for conduction takes the value
 Thermal Systems 95

T L
R t cond = ------
- = ------ (3.96)
Q· kA

and with a bond graph symbol as shown in Fig. 3.61.

e = ΔT
f=Q
R: kAL

Fig. 3.61 Thermal resistance in conduction.

3.8.2 Convection
Convection refers to heat transfer that will occur between a solid surface and a fluid in
motion due to a temperature gradient between the surface and the bulk of the fluid as illus-
trated in Fig. 3.62.

Heat flow rate Q·

Fluid flow rate V·

T2

T 1 surface

Fig. 3.62 Heat transfer by convection

If the fluid is caused to circulate by some external mechanism such as a fan or pump, the heat
transfer is referred to as forced convection. If the motion only takes place due to the buoy-
ancy caused by temperature gradients, the mechanism is referred to as free convection.

Thus, a fluid, gas or liquid is involved in convection heat transfer, and convection is the proc-
ess by which heat transfers across a surface into or from the fluid. Heat transfer by convec-
tion occurs on walls of rooms and pipes. For this type of heat transmission the following
equation known as Newton’s law of cooling, is in general use and is expressed by

Q· = hAT (3.97)

where h is called the film or heat transfer coefficient , A is the surface area and T is the
temperature difference. It is expected that h will depend on the fluid-dynamic problem and
upon the state of the fluid, but in a given situation, h is often nearly constant. As for conduc-
tion heat transfer, we define a thermal resistance for convection as
96 System Modelling

1
R t conv = ------- (3.98)
hA

with a bond graph symbol as given in Fig. 3.63.

e = ΔT
f=Q
R: hA1

Fig. 3.63 Thermal resistance in convection.

3.8.3 Radiation
The transfer of energy by electromagnetic waves is called radiation heat transfer. All matter
at temperatures greater than absolute zero will radiate energy. Energy can be transferred by
thermal radiation between a gas and a solid surface or between two or more surfaces. The
transfer between a wood stove to a person standing close is an example of surface-to-surface
radiation heat transfer. Inside the stove we hag gas-to-surface radiation exchange.

The two bodies illustrated in Fig. 3.64 at respective temperatures T 1 and T 2 can exchange
energy by electromagnetic radiation without the need of a transmission medium.

Q·
T1 T2

Fig. 3.64 Transmission of energy by thermal radiation

Radiation energy transmission is governed by the Stefan-Bolzman law, which is expressed by

Q· =   T 1 – T 2 
4 4
(3.99)

where the temperatures T 1 and T 2 are absolute temperatures. The coefficient  called the
Stefan-Bolzman constant depends on the geometries, colors, textures and areas of both sur-
faces. Generally, it is difficult to calculate the Stefan-Bolzman constant .

Since radiant heat transfer is not proportional to the temperature difference between the two
bodies, but the difference of each temperature to the fourth power, it can not be modelled by
a 1-port R-element as was done for conduction and convection. We will return to radiation
heat transfer modelling in a later chapter when we have introduced multiport generalization
of the basic 1-port elements.
 Thermal Systems 97

3.8.4 Thermal Storage Elements

All matter, whether solid, liquid, or gas, stores energy in the thermal agitation of its mole-
cules. This stored energy due to energy flow into a body of matter is commonly referred to as
internal energy and manifest itself with an accompanying increase in the temperature of the
body. The ability of a material to store thermal energy is measured by the specific heat at con-
stant volume c v , of the material which is formally defined as

c v  u (3.100)
T v

The term u is internal energy per unit mass and T is the temperature. Thus, the quantity of
thermal energy gained or lost by a body is proportional to the mass of the body, m b , specific
heat of the material c v and the change in the body temperature T = T b – T b  0  , with
T b  0  being the initial or reference temperature.

For the simple case of a lumped model as shown in Fig. 3.65, the rate at which net heat

Q· 1 T2

Fig. 3.65 Thermal storage element

energy is flowing into the body and being stored is proportional to the change in the body
temperature. In equational form this can be written as

t
·
Q2 =  Q 1 dt = mb cv  T2 – T2  0   (3.101)
t0

where m b is the body mass and T 2  0  is the body temperature at t = 0 .

Since Eq. 3.101 gives a relation between an integral of a flow Q· 2 which is a generalized dis-
placement, and an effort T 2 , this can therefore be interpreted as a capacitor with a capaci-
tance parameter equal to

C = mc v (3.102)
98 System Modelling

and the bond graph for a thermal storage element is as shown in Fig. 3.66.

ΔT
C:mbc v
Q

Fig. 3.66 Bond graph element for thermal energy storage

3.9 Thermal System Modelling

Thus, with the traditional thermal-electric analogy with temperature T analogous to voltage
and heat flow analogous to current we have shown there exists thermal sources, capacitors
and resistors. The heat conduction and convection resistors do not dissipate energy, but illus-
trate a relationship between effort and flow variables. In heat transfer thermal inertia, how-
ever, do not exist.

Since we are using the thermal-electric analogy, then the method for constructing bond graph
models of thermal systems is the same as for electrical circuits except that now we start writ-
ing a 0-junction for each node with a distinct temperature.

In the following we will construct bond graph models of two different simple thermal system
examples. We will start with the problem of heat transfer from hot water through a solid wall
to air as shown in Fig. 3.67.

Boundary layers

Tw
Temperature profile
T1
T2

Ta

Tw >T 1 >T 2 >T a

Fig. 3.67 Heat transfer through a wall

We wish to approximate the actual system of Fig. 3.67 by a lumped bond graph model. That
is, even though the temperature will actually vary in a continuous way through the different
layers as indicated in Fig. 3.67, we wish in our model to lump the different physical effects.
 Thermal System Modelling 99

As a first step in the modelling process we establish a 0-junction for each of the distinct tem-
peratures T w , T 1 , T 2 and T a , as shown in Fig. 3.68a. In the next step the temperature differ-

0 0 0 0
Tw T1 T2 Ta

(a)

0 1 0 1 0 1 0
(b)

R R R

Sf 0 1 0 1 0 1 0 Se
(c)

R R

Te :Sa 1 Se:Ta

(d)

Fig. 3.68 Bond graph model of heat transfer through a wall

ence for each 1-port element is established by a 1-junction bonded to appropriate 0-junctions
as shown in Fig. 3.68b. To each 1-junction is then bonded a 1-port R-element and to the 0-
junctions representing T w and T a are connected effort sources as shown in Fig. 3.68c. From
the assumption that T w  T 1  T 2  T a we conclude that the power direction will be from left
to right. This power direction leads to 2-port 0-junctions with through power flow. These
junctions can then be replaced by a single bond, and the resulting bond graph is as shown in
Fig. 3.68d, where all the elements are connected to a common 1-junction.
100 System Modelling

If we look at Fig. 3.67 it is obvious that without energy storage the heat flow is the same
through the water side surface, as through the wall and through the air side surface. This
means that the resulting model shown in Fig. 3.68 is correct.

Example 3.1
As the next example we will construct a bond graph model of the system shown in Fig. 3.69.

Ts

m· p TT m· p
V

T in  T out
Liquid inlet Liquid outlet
temperature temperature

Fig. 3.69 Heating of tank liquid

The system consists of a tank containing a volume V of a liquid with density  . To heat the
liquid in the tank, a pipe with flowing hot liquid runs through the tank. The surface area of
the pipe which is in contact with the tank liquid is A p and the inlet and outlet temperatures
are T in and T out respectively. The surface area of the tank submerged in liquid is A s and the
temperature of the surroundings is T s .

We are asked to construct a dynamic model which is to be used for investigating the heating
time of the tank liquid due to changes in system constants and variables.

Before we start constructing the bond graph model let us identify some obvious thermal
energy processes.

• Heat transfer by convection between the flowing liquid in the pipe and the tank con-
tent
• Energy storage in the tank liquid
• Heat transfer by convection between the tank liquid and the tank surrounding

The heat transfer between the liquid flowing in the pipe and the tank liquid occur continu-
ously along the pipe and as a result the temperature of the pipe water decreases continuously
from T in to T out . Since we are going to construct a lumped model we will use a mean tem-
perature T m for the liquid flowing in the pipe when we calculate the convective heat flow
from the pipe to the tank liquid.
 Thermal System Modelling 101

Let us now then start constructing the bond graph for the system. We start with setting off the
0-junctions representing the temperatures T in , T m , T ref and T s as shown in Fig. 3.70a.

0 0 0
Tref TT Ts

0 0
Tin Tm
(a)

C R

Tref :Se 0 1 0 1 0 Se:Ts

Tref TT Ts

R 1 R

Tin :S e 0 1 0
Tin Tm
(b)

C R

0 1

Tin :S e 1 R

R
(c)

Fig. 3.70 Heated tank bond graph model

We then insert convective R-elements connected to a 1-junction between T m and T T , and

between T T and T s as shown in Fig. 3.70b. As also shown in Fig. 3.70b, we have inserted a
102 System Modelling

R-element between T in and T m . This is done on the notion that there is a resistor which is
causing a drop in effort or temperature from | T in to T m .

Between the temperature T T and the reference temperature T ref is inserted a C-element rep-
resenting the internal energy storage in the tank liquid. The complete bond graph with
assigned power direction is shown in Fig. 3.70c. The remaining task is now to determine the
value for the C and R-elements.

Let us first determine the value of the R-element between T in and T m . The mean tempera-
ture T m we will define as the mean value of T in and T out as

T in + T out
T m = ----------------------
- (3.103)
2

There is a difficulty computing T m since the outlet temperature T out is unknown and
depends on system operation. In order to get around this problem we will use the following
equation.

Q· PT = m· p c  T in – T out  (3.104)

where Q· PT is the heat flow from the liquid flowing in the pipe to the tank liquid, m· p is the
mass flow rate in the pipe and c is the specific heat of the flowing liquid.

Solving Eq. 3.103 and Eq. 3.104 for T out we obtain

Q· PT
2T m – T in = T in – --------
- (3.105)
m· p c

Rearranging Eq. 3.105 we can write that

1 ·
T in – T m = ------------
· - Q PT (3.106)
2m p c

We have in Eq. 3.106 a relation between an effort  T in – T m  and a flow  Q· PT  , and then the
resistance of the R-element can be expressed as

1
R = --------------
- (3.107)
2m· p c p

The convective resistances between T m and T T and between T T and T s are of the form
 Thermal System Modelling 103

1
R = -------- (3.108)
UA

where U is the heat transfer coefficient and A is the heat transfer area.

The capacitance for the C-element is according to Eq. 3.100 given as

C = Vc T = m T c T (3.109)

where  is the liquid density, V is the tank volume and c T is the specific heat of the tank liq-
uid.

This concludes the model development for this example.

PROBLEMS
3-1 A “smart guy” has found the following bond graph in a book, but he is not sure what
kind of physical system it represents. Could you help him find an electrical and mechan-
ical system which have the bond graph as shown below

I R I

m
R 0 TF 0 1 R

3-2 In the figure below is shown an electric circuit diagram. Construct a bond graph model
for the circuit and draw a mechanical system which have the same bond graph.
L1 L2

C2
V(t)
C1 R1 C2 R2

3-3 A number of electrical circuits are given in Chapter 4. Construct the bond graph of each
of these circuits using both the procedure and the inspection method.
104 System Modelling

3-4 In the figure below is shown a Wheatstone-bridge circuit. The voltage e is zero when
the bridge is balanced, and we measure the value of e with an instrument which draws
zero current. Construct the bond graph for this device.

+
e
- V

3-5 Make a bond graph of each of these systems. Include inertia and energy dissipation
k0
F0(t)
k1 k2 v(t)
m1 m2 m3
Friction Friction Friction

k
m2
l2 J=0

l1
m1

3-6 Construct a bond graph for this two-degree-of-freedom system which represents a vehi-
cle. The system consists of an rigid beam with mass m and center of gravity in point C .
The beam has a rotational moment of inertia J C . Assume small motions only and make
a b

m C yC
JC

ka ba
kf bf
va(t)
vf(t)

a bond graph for this system. Reject the small motions assumption and make necessary
changes to the bond graph model to handle this case also.

3-7 In the figure below is shown a new type of pump consisting of an cylinder with inlet
and outlet pressure difference control valves and a special crank mechanism which
 Thermal System Modelling 105

transfers the translation in rod b given as a velocity v b  t  through two links connected
with a rotational spring to the cylinder rod a.

 
m m A
a st st V
0

J  L
 k 2 2 2


v t J  L
b 1 1 1
F m 0
b b

 

For the purpose of this model you can assume that the compression and expansion of the
gas in the cylinder is polytropic, i.e. follows the equation

V0  n
p = p 0  ---------
-
 V  t 

a) Make a bond graph of the system. State clearly all the directions of efforts and flows de-
fined
b) (After Chapter 4) Write the state equations for the system.
c) (After Chapter 4) Find an expression for the force F b in the figure.

3-8 In the figure below is shown a sketch of a hydraulic safetyvalve. Here the chamber A and
B has a fluid volume of V A and V B respectively, and the oil in those chambers is ass-
mumed to have a bulk modulus  . The variable system pressure is P S . Make a bond
graph for this hydraulic device.
k2
m2 k1
m1

B
A
VA
VB

3-9 In the figure below is shown a cylindrical rod which at the left end is inserted into a well
insulated vapor container. The right end is insulated such that you can assume zero heat
loss. The temperature of the surroundings is 20  C. The heat loss coefficient h to the

l=230 mm

T=20 C d=19 mm
106 System Modelling

surroundings can be set to 13.0 W  m 2 K . Make a bond graph model of this system and
simulate to find the temperature profile along the rod when it is made of

a) Cobber :  = 8 9 kg  m 3 , K = 380 W  mK and C p = 390 J  kgK

b) Glass :  = 3 0 kg  m 3 , K = 1 1 W  mK and C p = 880 J  kgK

b) Cast-iron :  = 7 2 kg  m 3 , K = 58 W  mK and C p = 540 J  kgK

Divide the rod into 4 lumps initially.

3-10 Given a pendulum as shown at the figure below, with a point mass m , hinged at point
o and with a spring which is supposed to always stay vertical. The rod and the vehicle
where the spring is fastened is supposed massless. Construct a bond graph of this system.

3-11 Assume small motions and construct a bond graph for the dynamics of a multistory
building as a multidegree of freedom system being shaken by an earthquake.

x3
x3 m3
k3
x2
x2 m2
k2
x1
x1 m1
x0 k1 x0

x· 0 = A sin t

3-12 Given an electric circuit with linear elements and a DC-motor below. The motor coeffi-
cient is K and T the torque developed,  the angular velocity of the rotor, u the voltage
 Thermal System Modelling 107

difference between the connectors and i the current drawn. The load of the motor is giv-
en by an inertia J and a linear friction Rfric. Draw a bond graph for this circuit.
R1 R2 La

i
+

Lt T
V Ct N1 N2 Ce u DC Rfric
Rt 
-

3-13 Given a hydraulic system shown below. Draw a bond graph for this system. Include in
your answer the assumptions you make modelling each component and write down all
constitutive equations including necessary logic. Draw a free-body diagram for the dou-
ble-action mechanical-hydraulic accumulator (E) and assign power bond directions ac-
cording to it..

C
A - displacement pump
D E B - safety valve
 = const A k
A m
a F G H C - air-filled accumulator
k
c m
a d J D - directional valve
b
B k
b E - double-action accumulator
F - control and safety valve
g G - hydraulic motor
m H - winch

3-14
x1
m1
l1a l1b Centre of gravity of beam 1 is
x 1a positioned in the bearing.
x 1b
Observe that the polar moment of
J1  0 inertias of the beams aren not
ks, bs k12 equal to zero.
kL=0
l2a l2b
x 2a x 2b
J2  0
xL
k2 m0
mL g

Construct the bond graph for this mechanism. Include a free-body diagram.
108 System Modelling
 Development of State Equations 109

Chapter 4

Development of State Equations

The system models we have developed up to this point have been on a graphical bond graph
form. These models can be entered directly into a computer which automatically transform
the bond graph model into a mathematical model that we can use to perform numerical anal-
ysis and time simulation. In Fig. 4.1 is shown how this can be done using the commercial

Fig. 4.1 Using computer tools to draw bond graph models, enter input paramters and to per-
form simulations.
110 Development of State Equations

software 20-SIM [1]. In this software tool the bond graph is entered graphically using a pow-
erful graphical editor, the computer prompts you for required model parameters for the model
and finally it allows for simulation setup, simulation and finally data exploration and presen-
tation. However, even if the computer takes care of all hard labor work, to use it wisely you
need to know how to do the right thing. If not you will find that the simulations will just
crash.

In this chapter we are going to illustrate how a bond graph model can be transformed in a
systematic manner from a bond graph into mathematical differential equations. The differen-
tial equations which govern the behaviour of a dynamic system, can be written in a number
of ways. That is, we may use a single n-th order equation, n first-order equations or various
combinations of orders. The type of mathematical model that we will use in our development
here is described as a state-determined system model and is described by ordinary differen-
tial equations of first-order form. Such a set of differential equations has become known as
the standard form.

There are several reasons for the importance of using state-determined system models in
engineering system analysis and control.

• A basic characteristic of the state determined system model formulation is that linear
and non-linear, time-invariant and time-varying, single-variable and multivariable
systems can all be presented in a unified manner.
• State-determined system models have proved useful over centuries of scientific and
technical work, and are nearly universal.
• The universal notation for all systems makes possible a uniform set of solution tech-
niques.
• State-determined system models are a convenient form to use in engineering system
analysis and control.
• Almost all known numerical methods for solving simultaneous ordinary differential
equations require the standard form.
• Models of this form are ideally suited for solution by computers and constitute the
basis of much of system theory.

The concept of state-determined systems and state-space representation have been associated
with bond graphs since their inception. Henry Paynter described state-determined systems as
idealized systems whose behaviour could be completely described by the state of parameters
 Development of State Equations 111

and represent such systems by a generalised system model as:

State model
Input Output
variables x, x variables

Fig. 4.2 Generalized system model

The notion of state is extremely useful in the discussion of dynamics and is central to virtu-
ally all quantitative modelling of dynamic systems. Concisely put, the dynamics of a system
comprise the changes of its state as time progress.

Since the concept of the state of a system form the basis for the differential equations to be
developed, let us therefore elude on the concept of the state of a system.

To characterise a system at an instant of time, we need the values of system properties at that
instant of time. With a property we mean any characteristic or attribute which in principle
can be quantitatively evaluated at any instant of time. There are a large number of properties
which can be attributed to a system, but some of these properties are not independent. That is,
a change in value of one such property affects the values of some other properties. Neverthe-
less, for each well-defined system it is always possible to identify many different sets of
independent properties, each of which satisfies the following two conditions:

• The values of properties in the set can be varied independently, that is, the value of
each property can be changed without affecting the value of any other property.
• The values of the properties in the set are sufficient to determine the values of all the
other properties of the system.

The properties in one such set are what we call the state. A state of a physical object is
defined as a quantitative measure of a physical condition of the object which remains
unchanged with laps of time if the object is suitably isolated. Thus, the notion of state repre-
sents a complete description of a systems status at a given time. Knowledge of the state at a
given time t 0 , and knowledge of the input after time t 0 , render possible determination of the
state at a later time t . This means that the state at a given time t , contain a complete history
of the system response after the time t 0 , as far as the history affect future response. Thus, the
state is a set that specify everything about a system at one instant of time.

The minimum independent set of variables ( x 1 x 2  x n ) describing the state of a system

are called state variables. These variables also have the following property:

If the current state is known and the future input is known, then the future state and the output
can be uniquely determined. That is, the state variables of a system completely characterise
112 Development of State Equations

the past of the system since the past input is not required to determine the future state and
output of the system. Modern approach to dynamic systems is fundamentally based upon this
concept of state.

Although the number of independent state variables required to describe a system state is
fixed, the choice of a set of state variables is not unique. This means that for one and the
same system many different state variables can be used. The individual state variables, how-
ever, must be linearly independent as mentioned previously. This means that there must not
be any set of constants c i , other than zero, that satisfies the equation

c1 x1 + c2 x2 +  + cn xn = 0 (4.1)

If n state variables are needed to define the state of a system, they can be considered as the n
components of a state vector x . State space is then defined as n -dimensional space with
x 1 x 2  x n as the coordinates, and the state of the system at any instant of time will be a
point in the n -dimensional state space. A series of such point forms a state trajectory.

As mentioned previously the state-determined system model is described by ordinary differ-

ential equations of first order, and for general non linear models we shall write the equations
in the form

dx 1
= g 1  x 1 x 2  x n u 1 u 2  u n 
dt
dx 2
= g 2  x 1 x 2  x n u 1 u 2  u n 
dt
(4.2)
.
.
dx n
= g n  x 1 x 2  x n u 1 u 2  u n 
dt

where

x i - state variables
u i - inputs to the system
g i - algebraic functions of the state variables, input variables and time

For the ease of expression, it is convenient to represent Eq. 4.2 in the general form
 Development of State Equations 113

dx
= g  x u t  (4.3)
dt

where

x1 u1
x2 u2
x = . and u = . (4.4)
. .
xn ur

The n  1 column vector x is the state vector, and the r  1 vector u is the input vector. If
the functions g i    has a linear mathematical structure, then the state equations can be writ-
ten in the matrix form

dx 1
dt
a 11 a 12  a 1n x 1 b 11 b 12  b 1r u 1
dx 2 a 21 a 22  a 2n x 2 b 21 b 22  b 2r u 2
dt =
. . + . . (4.5)
.
. . . .
.
a n1 a n2  a nn x n b n1 b n2  b rr u r
dx n
dt

or using vector and matrix notation we obtain the following linear form for a vector-matrix
state equation

dx
= Ax + Bu (4.6)
dt

where x and u are the state and input vectors respectively and A is n  n and B is n  r
coefficient matrices respectively.

We now have defined the format in which to express the model equations, and we are ready
to show the systematic procedure to convert a bond graph model into a set of state equations
using pencil and paper.

Before we can start deriving the mathematical equations from a bond graph we must prepare
114 Development of State Equations

the graph with additional information that will make the writing of equations follow an
orderly systematic pattern. These steps are:

1. number all the bonds in the graph

2. assign a power direction to each bond
3. assign causality to each bond

The numbering of each bond is done for practical purposes in order to be able to refer to e , f ,
p or q on a specific bond. Power directions are necessary when writing equations in order to
know if we are to add or subtract efforts and flows at the junctions. It is also necessary when
intepreting calculations in order to determine direction of movement of objects, direction of
flow, direction of forces etc. We are going to treat assigning of power directions in more
detail when we come to the point when we are going to write equations for specific systems.

Now we come to the point of assigning input-output causality to a bond graph. The concept
of causality or cause and effect is usually not discussed in explicit terms during equation for-
mulation when using traditional methods of mathematical modelling, but it is however, very
useful in bond graph modelling. Bond graph causality mainly represents the computational
direction of the bilateral signals of a bond.

The question of what are proper input to a system is often not unambiguous. In fact even
those who design or study systems, but normally do not write differential equations for them
or simulate them do not always perceive the importance of causal considerations in model
building. Application of causality is extremely valuable in understanding the dynamic struc-
ture of models. That is, it can be used both as an aid in generating the system equations and
as a guide to modellers seeking insight directly without having to obtain the system equations
explicitly.

Causality as a concept and as a tool is uniquely associated with the bond graph method and
we will in what follows exploit causality applied to bond graph multiports and system mod-
els. We shall look at causality in several settings such as

• procedure for assigning causality to a bond graph.

• use of causality to check the correctness of a proposed design.
• use of causal information as a guide in assembling sub-models into a system model.
• use of causal information to resolve difficulties in converting from a bond graph
model to mathematical equations.

For multiports each bond has both an effort and a flow, and it is possible for only one of these
two variables to be an input, the other must be an output. To know which one is the input or
independent variable and which one is the output or dependent variable, is the key to the
 Causality Assignment for Multiports 115

proper assembling of system equations or establishing the correct computational order of all
the variables. The decision of which variable is the input and which is the output, most often
called the causal assignment, is in bond graphs done in a unique fashion, in that each bond is
marked with a causal stroke. The causal stroke is a vertical line at one end of the bond and
perpendicular to it, indicating the direction of the effort-signal. Consequently, the open end
of a bond indicates the causal direction of the flow signal. Thus, the causal stroke is a 1 bit of
information which shows the causal sense of both signals on the bond simultaneously as indi-
cated in Fig. 4.3.

e
e
A B A B
f
f

e
e
A B A B
f
f

Fig. 4.3 Meaning of causal strokes

It should be noted that the causal direction has nothing in common with the orientation of
power indicated by the half-arrow. The causality concept is a unique useful feature of bond
graphs and we shall utilise it in several settings on the multiport and system level. Let us
therefore, start with causality assignment for the basic multiports defined previously. As we
shall see, the basic multiports are constrained with respect to possible causality and some
exhibit their constitutive laws in different forms for different causalities.

4.1 Causality Assignment for Multiports

Before we utilise the causality concept on the complete system model, we have to explore the
cause and effect for each of the basic bond graph elements that we have defined.

The basic elements which are simplest regarding causality assignment is the two source ele-
ments. From their definition, they impose either an effort or a flow on the system to which
they are attached. Thus, from the definition of the causal stroke with respect to the direction
of the bilateral signal-flow, the causal assignment for the two sources must be as shown in
116 Development of State Equations

Fig. 4.4.

e e
Se and Sf
f f

Fig. 4.4 Causality assignment for the sources

The causality indicated on the effort source S e , implies that it imposes an effort into the sys-
tem and the causality indicated on the flow source implies that it imposes a flow into the sys-
tem. This shows that the causality assigned is correct according to the definition of the
sources.

There is more freedom associated with the causality assignment to the storage elements, in
that there are two possibilities. The two possible causal assignments and the corresponding
relations between the effort and flow at the port for a capacitor is shown in Fig. 4.5.

General case: Linear case:

–1 1 q
C e =  C   f dt  e = ----  f dt = ----
C C
de
C f = d  C  e   f = C
dt dt

Fig. 4.5 Causal assignment on C-element

When the flow is input to the capacitor, then the effort is given as a static function of the time
integral of the flow. Due to the time integration of the flow variable, this type of causality is
called integral causality. If the effort is the input, then the flow is the derivative of a static
function of the effort and this type of causality is therefore called derivative causality.

The inertia element has a constitutive equation which is identical to a compliance element,
but with effort and flow interchanged. We therefore, obtain correspondingly two possible
causal assignments, and the corresponding relations between the effort and flow at the port of
 Causality Assignment for Multiports 117

an inertia element is as shown in Fig. 4.6.

General case: Linear case:

–1 1 p
I f = I   e dt  f = ---  e dt = ---
I I
d df
e =  I f   e = I
I dt dt

Fig. 4.6 Causal assignment on I-elements

In this case, when an effort is imposed on an inertia element then the flow is given as a static
function of the time integral of the effort. Due to the time integration this type of causality is
called integral causality. When the flow is imposed on an inertia element, the effort is the
derivative of a static function of the flow and this causality is therefore called derivative cau-
sality.

Since we have two choices of causal assignment for the two storage elements, it might be of
interest to investigate if there are any physical implications when using one or the other of the
two causal choices. The answer is yes, one choice of causal assignment restricts the nature of
the port signals. In order to find out what it is, let us perform experiments with a C- and I-ele-
ment. First we must impose the physical restriction to our experiment that the input power
available is finite. For the first experiment, we impose a step in effort on a C-element as
shown in Fig. 4.7

e
C
t

Fig. 4.7 Step input signal on C-element

Refering to Fig. 4.5, this is the derivative causality choice and means that the flow output is
proportional to the derivative of the input effort. From calculus we know that the derivative
of the step function at the start is infinite and also the flow which results is infinite and hence
violate our power restriction. This means physically that if we mount a capacitor and a switch
into the electrical outlet of our house and suddenly close the switch, we will blow the fuse
since the current is very high. If we, however, impose a step in flow on the C-element, the
effort will be proportional to the integral of the flow. For a linear capacitor the input-output
118 Development of State Equations

relations will be as shown in Fig. 4.8.

f e
C
t t

Fig. 4.8 Input-output of linear capacitor

Likewise, we are not able to impose a step input in flow on an inertial element. You can get a
feel for this fact when you try to push your car and want to bring it up to speed immediately,
the required force is larger than what you are able to supply. However, imposing a step in
effort will be possible without any difficulties. This means that the C and I elements on phys-
ical grounds prefer integral causality, and integral causality is therefore often called preferred
causality. From this we can conclude that nature integrates, only mathematicians differenti-
ate. We will show you in a later chapter what it means physically and mathematically when
we are forced to assign derivative causality to a system model.

As for the C and I elements, the 1-port resistor also have two choices of causal assignment.
The two possible causal assignments for an R-element and their implication on equation for-
mulation are shown in Fig. 4.9.

General case: Linear case:

R e = R  f  e = Rf
–1 1
f = R  e  f = --- e
R R

Fig. 4.9 Causal assignment on resistors

The 1-port resistor is usually indifferent to what causality is forced on it, since the relation
between effort and flow variables for this element is purely algebraic and does not involve
differentiation or integration. However, no rule without exception because there are cases
where only one possible causality is possible. One such case is dry friction shown in Fig.
 Causality Assignment for Multiports 119

4.10.

Fig. 4.10 Dry friction

In this case it is only possible to specify a velocity and then determine the force. The velocity
can not be uniquely determined from a known force.

In order to determine what causal assignments which are possible for the transformer, we
turn to the constitutive laws of the transformer given by Eq. 2.22. Using the convention of
writing the independent variable to the right of the equal sign, Eq. 2.22 indicates that e 1 is
the input or independent variable and e 2 is the output or dependent variable. This then corre-
sponds to the following causal assignment

e1 1/m e2
TF
f1 f2

Interchanging the input and output efforts in Eq. 2.22 yields the other causal assignment for
the transformer as

e1 m e2
f1
TF f2

Following the same reasoning for the gyrator as for the transformer and using the constitutive
laws given by Eq. 2.23 the following two possible causal assignments are possible

1/r
e1 e2
GY
f1 f2
120 Development of State Equations

and

e1 r e2
GY
f1 f2

The constraints, which are imposed in the definitions of the 0-junction elements, imply
directly the causal relationship between the e and f variables. For a 0-junction the efforts on
all the bonds are equal, so, as viewed from the junction, only one effort can be an independ-
ent variable and all the rest must be dependent. Thus, only one stroke is placed on the junc-
tion side of the bond as shown in Fig. 4.11.

For a 1-junction, the rules are of course reversed and the causal stroke is placed on the junc-
tion side of the bond on all but one of the bonds as shown in Fig. 4.11.

e2 f 2 e2 f 2

e1 e3 e1 e3
0 1
f1 f3 f1 f3

Fig. 4.11 Junction element causality

In spite of the fact that we have considered the possibilities which exist with respect to causal
assignment for each of the basic elements, it is not easy to see why the question of causality
is so important. It can only be confirmed by experience that it is very important , and that
bond graphs are unique with respect to the study of causality. Now that we have developed
system bond graph models and are going to convert them into equations, we hope you will be
convinced that causality is important..

In Table 4.1 is shown a summary of the possible causal assignments for the nine basic ele-
ments that we have stated is sufficient to model most engineering systems you will meet.
 Causality Assignment for Multiports 121

Name Bond-graph Symbol Causal relations

Effort source e  t  = given

Se

Flow source f  t  = given

Sf

Resistance e = R  f 
R

–1
R f =  Re

Capacitance –1
C e =  C   f dt 

d
C f =  e
dt C

Inertia –1
I f =  I   e dt 

I e = d  I  f  
dt

Transformer 1 2 e 1 = me 2
TF
f 2 = mf 1

1
TF 2 e 2 =  1  m e 1
f 1 =  1  m f 2

Gyrator 1 2
GY e 1 = rf 2
e 2 = rf 1

1
GY
2
f 1 =  1  r e 2
f 2 =  1  r e 1

0-junction 1 2 e2 = e1
0
e3 = e1
3
f1 = – f2 + f3 

1-junction 1 2 f2 = f1
1
f3 = f1
3
e1 = – e2 + e3 

Table 4.1: Causal assignment ot basic elements

122 Development of State Equations

4.2 Assigning Causality to System Models

Since we now have explored the possible causal assignment for each of the basic elements, it
is appropriate to apply this information to entire system models in an orderly fashion. The
traditional way to assign causality to a bond graph model is by using what is called the
Sequential Causality Assignment Procedure (SCAP). This procedure is derived from a phys-
ical based motivation. That is, we seek a causal orientation of the bond graph that is consist-
ent with a set of physical constraints imposed in a hierarchical fashion. The constraints are
imposed in the following order:

• each source element S e and S f , must have its appropriate causal orientation
• each junction element 0, 1, TF and GY must have consistent orientation
• storage elements C and I should be oriented with integral causality to the maximum
extent possible

The Sequential Causality Assignment Procedure is straight forward and orderly as given in
the steps below.

1. Assign the required causality to all the sources S e and S f in the model

2. Extend the causal implications through the graph as far as possible using the con-
straint elements 0, 1, TF and GY .

3. Choose any storage element C or I and assign to it integral or preferred causality and
repeat step 2.

4. Repeat step 3 until all storage elements have been assigned a causality. Certain stor-
age elements may have been forced to take differential causality and certain bonds
may not have any assigned causality.

5. Choose any unassigned R element, assign causality to it arbitrarily, and repeat step 2.

6. Repeat step 5 until all bonds have been assigned a causality.

In order that the causality assignment to a bond graph can be performed with ease and rapid-
ity, it is required that the causality assignment to the constraint elements 0, 1, TF and GY is
memorised. Practice is the key word. Therefore, let us consider two examples using the pro-
cedure.
 Assigning Causality to System Models 123

Example I
The first example is an acausal bond graph as shown in Fig. 4.12a.

I R C I R C

2 4 6 2 4 6
3 5 3 5
0 1 0 0 1 0

1 7 1 7

Se R Se R
(a) (b)

I R C I R C

2 4 6 2 4 6
3 5 3 5
0 1 0 0 1 0

1 7 1 7

Se R Se R
(c) (d)

Fig. 4.12 Causality assignment and complete augmentation of a bond graph.

In Fig. 4.12b the effort source Se on bond 1 is assigned the required causality which means an
effort input to the 0-junction. Since only one bond can set the effort at a 0-junction we turn to
step 2 and extend the causal implications to bond 2 and bond 3. The 1-junction has only one
effort variable determined, and the other two bonds cannot yet be assigned a causality. We
now turn to step 3 and assign integral causality to the C-element as shown in Fig. 4.12c. Inte-
gral causality on the C-element sets the effort on the 0-junction and we can extend the causal
implication to bond 5 and bond 7 which must have a flow input to the 0-junction. At the 1-
junction two effort variables are determined and bond 4 must then determine the flow at this
junction. In part d is shown the graph with all bonds numbered, power directions are added to
all the bonds and the causality is assigned to all the bonds.

Example II
The next example shown in Fig. 4.13 is a more extensive model which contain eight of the
124 Development of State Equations

nine 1-port elements we have defined so far. As we assign causality to this model we will dis-
cover the eventualities that we mentioned in the procedure for assignment of causality.

10 9 7 5 4 1
I 1 0 1 GY 1 Se
3 2
11 8 6

I C TF R R I R

17 15 12
18 16 14 13
R 1 0 1 I

(a)

10 9 7 5 4 1
I 1 0 1 GY 1 Se
3 2
11 8 6

I C TF R R I R

17 15 12
18 16 14 13
R 1 0 1 I

(b)

10 9 7 5 4 1
I 1 0 1 GY 1 Se
3 2
11 8 6

I C TF R R I R

17 15 12
18 16 14 13
R 1 0 1 I

(c)

Fig. 4.13 Extended causality assignment example - Part 1

In Fig. 4.13a causality is assigned to the effort source Se according to the definition of this
element. Since the 1-junction has only one effort variable determined, the other bonds cannot
yet be assigned a causality. We now turn to step 3 in the SCAP and in Fig. 4.13b the I-ele-
 Assigning Causality to System Models 125

ment on bond 3 is selected with integral causality. This sets the flow on the 1-junction and
forces bond 2 and bond 4 to have an effort input to the 1-junction. Since bond 4 is one port of
the gyrator, the other port or bond 5 can be directed. The causal implication can not be
extended any further and we return again to step 3. In Fig. 4.13c the I-element on bond 17 is

10 9 7 5 4 1
I 1 0 1 GY 1 Se
3 2
11 8 6

I C TF R R I R

17 15 12
18 16 14 13
R 1 0 1 I
(d)

10 9 7 5 4 1
I 1 0 1 GY 1 Se
3 2
11 8 6

I C TF R R I R

17 15 12
18 16 14 13
R 1 0 1 I
(e)

Fig. 4.14 Extended causality assignment example - Part 2

assigned integral causality. Immediately, bond 16 and bond 18 may be directed with an effort
input to the 1-junction since bond 17 is assigned integral causality.

You may wonder why we selected a storage element in the opposite end of the graph where
we started. The reason is to show that the result is independent of the order of the selection of
causality assignment of the storage elements.

To move on, we now choose integral causality on the I-element on bond 10 in part d. Imme-
diately bond 9 and bond 11 can be assigned the causality of effort input to the 1-junction
since bond 10 sets the flow input. Since bond 11 is one port of the transformer, the other port,
bond 12 can be assigned causality. Since bond 12 then sets the flow at the 1-junction, bond 13
and bond 14 must be directed to have effort input to the 1-junction. Now, since bond 14 and
126 Development of State Equations

bond 16 are directed with flow input to a 0-junction, bond 15 is forced to have an effort input
to the junction. Thus the C element at bond 15 is given integral causality which is desired.
Notice that the I-element on bond 13 was forced to take differential causality. The meaning of
this occurrence of differential causality will be discussed later in the chapter when we write
equations from the bond graph.

A look at part d again indicates that there are still some bonds that do not have causality
assigned to them. The reason is that we have not been able to extend the causality any further.
In order to progress we choose arbitrarily the causality on one of the R-elements. In this case
we choose bond 6 to set the flow at the 1-junction. Since bond 7 and bond 9 both cause a flow
at the 0-junction, the R-elements at bond 8 must cause an effort at the 0-junction. This com-
pletes the causality assignment for the bond graph. The implication that we had to choose
causality on one of the R-elements will be discussed later in the chapter.

4.3 Development of state equations

A bond graph, augmented by sign convention and causal strokes contain exactly the same
mathematical information as a differential equation. It therefore follows that converting an
augmented bond graph into differential equations is a straightforward, in fact completely
algorithmic process. We shall therefore, in this section concentrate our effort on using a basic
systematic procedure for generating the system differential equations.

The first step is to select the input variables which we obtain from the sources. These input
variables will then be the elements of the input vector denoted u . The input variables will
appear in the final state-space equations.

The next step will then be to select the state variables. Since the set of state variables is not
unique a number of variables are possible as long as they are independent. However, from the
fact that the state of a system at a given instant is related to the energy stored in the system at
that instant, a physically meaningful set of state variables can be selected. This set of state
variables consists of the generalised displacement q of a C-element with integral causality
and generalised momentum p in I-elements with integral causality since they both can be
used to compute the instantaneous stored energy.

Before we start to write the state-equations, it might be useful to show how we utilise causal-
ity to write the equations at the junctions. For a 0-junction, it is only one bond which can
indicate or is the cause of effort in the other bonds adjoining the junction. On the other hand
for a 1-junction, it is only one bond which can indicate or be the cause of the flow in the other
bonds adjoining the junction. The way we will use to write the constitutive laws for the junc-
 Development of state equations 127

tion elements is as shown in Fig. 4.15.

e2 = e1  e 1 causing e 2
2
1 3 e3 = e1  e 1 causing e 3
0

f2 = f1  f 1 causing f 2
2 f3 = f1  f 1 causing f 3
1 3
1

Fig. 4.15 Formulation based on causality

Example I
As the first example on converting a bond graph model into mathematical equations, we will
start with the simple spring-mass-damper system or series RLC circuit which both have the
same bond graph as shown in Fig. 4.16.

F(t)
R L I

m 1
x(t) +
3 2
C R 1 C
-
k b
4

Se

Fig. 4.16 Equation formulation, example 1.

The system has two storage elements with integral causality and one input source. This then
means that we have to write two differential equations to describe the system dynamic behav-
iour. The state vector x , and input vector u , then becomes

p1 Se4
x = and u = (4.7)
q2 0

Since the system we are looking at is considered linear, the formulation result we are seeking
is as follows
128 Development of State Equations

p· 1 = a 11 p 1 + a 12 q 2 + b 1 S e 4
(4.8)
q· 2 = a 21 p 1 + a 22 q 2 + b 2 S e 4

The first step in deriving the state-space equations is to write the derivative of a state variable
on the left-hand side of each equation and the defining effort or flow variable on the right-
hand side. Then following the causal-stroke information, we proceed through the graph until
all expressions are evaluated in terms of state variables, system constants and input quanti-
ties. This process is repeated for each state variable and the final result is a set of first order
state equations.

Starting with the expression for p· 1 , which according to the definition of the momentum var-
iable is written as

p· 1  e 1 (4.9)

from the definition of p .

Since we have a 1-junction, we know from the constitutive laws of this junction that the alge-
braic sum of all the efforts is equal to zero. Thus

e1 = e4 – e2 – e3 (4.10)

The next step is now to express the efforts on the right hand side of the equal sign in Eq. 4.10
in term of state variables, sources and system constants. Effort e 4 is given as an input from
the source as

e 4 = Se 4 (4.11)

For the other two efforts e 2 and e 3 the causal marks indicates that they are given by consti-
tutive laws of the R and C elements as

q
e 2 = -----2- (4.12)
C2

and

e3 = R3 f3 (4.13)
 Development of state equations 129

We now ask: who is causing f 3 ? From the bond graph shown in Fig. 4.16, f 1 is causing the
flow at the junction, and according to the constitutive law of the I element, f 3 can be written
as

p1
f 3 = f 1 = ----- (4.14)
I1

which when substituted into Eq. 4.13, makes e 3 become

p1
e 3 = R 3 ----- (4.15)
I1

Substituting for e 4 , e 2 and e 3 into Eq. 4.10, then the equation for p· 1 becomes

q p
p· 1 = Se 4 – -----2- – R 3 ----1- (4.16)
C2 I1

This is the first state equation since it contains only a source, state variables and system con-
stants on the right-hand side.

Now, for the second state equation we start writing that

q· 2  f 2 (4.17)

from the definition of q .

From the causal marks in Fig. 4.16 we know that f 1 is causing f 2 at the 1-junction and can
then substitute Eq. 4.14 for f 2 in Eq. 4.17. The second state equation then becomes

p
q· 2 = ----1- (4.18)
I1

The two state equations, Eq. 4.16 and Eq. 4.18 can since they are linear, be written in vector-
matrix form as
130 Development of State Equations

R 1
· – -----3 – ------
p1 I1 C2 p1
= + 1 Se 4 (4.19)
· 1 q 0
q2 ---- 0 2
I1

This vector-matrix equation can be written in short hand notation as

x· = Ax + Bu (4.20)

where x and u are given by Eq. 4.7 and

R 1
– -----3 – -----
-
A = I 1 C 2 and B = 1 (4.21)
1- 0
--- 0
I1

As have been shown, the conversion of a bond graph model into state equations does not
require any physical knowledge. It is only a systematic procedure utilising the causal strokes
and constitutive laws of the basic elements. The physical understanding of the system is all
embedded in the bond graph model.

One additional and little more extensive example could be useful to show the strength of the
formulation and systematic reduction pattern, which is expressed by using energy variables.

Example II
Let us therefore write the state equations for the bond graph model shown in Fig. 4.17.

I C I I

11 9 6 2
10 8 m 7 5 r 4 1
1 0 TF 1 GY 1 Se

12 3

R R

Fig. 4.17 Equation formulation

In this model four storage elements are found with integral causality and we must write four
 Development of state equations 131

state equations. There is only one source and the state vector x and input vector u can then
be written as

p2
p6
x = and u = S e 1 (4.22)
q9
p 11

Starting with the derivative of the first state variable p 2 we set out

p· 2 = e 2 (4.23)

Following the causal strokes at the 1-junction we find that bond 2 is causing the flow and all
the other bonds at the 1-junction is causing efforts. This means that we can write that

e2 = e1 – e3 – e4 (4.24)

The effort on bond 1 is set by the effort source S e , and the effort on bond 3 is given by the
constitutive law for the R-element as

e3 = R3 f3 (4.25)

Now we ask: who is the cause of f 3 ? Following the causal strokes we find that f 2 is causing
f 3 and we obtain

p
e 3 = R 3 f 2 = R 3 ----2- (4.26)
I2

To find an expression for e 4 we have to turn to the constitutive law for the gyrator which we
write as

e 5 = rf 4
(4.27)
rf 5 = e 4

From the second line in Eq. 4.27 we find that e 4 is proportional to f 5 . We then ask immedi-
ately: who is causing f 5 ? The causal strokes at the second 1-junction from the right, tells us
that f 6 is causing f 5 and we can write
132 Development of State Equations

p
e 4 = rf 5 = rf 6 = r ----6- (4.28)
I6

Substituting Eq. 4.26 and Eq. 4.28 into Eq. 4.24 the first state equation can be written as

p p
p· 2 = S e – R 3 ----2- – r ----6- (4.29)
I2 I6

For the second state equation we write

p· 6 = e 6 = e 5 – e 7 (4.30)

To find a relationship for e 5 we return to the first line of Eq. 4.27, and find that e 5 is propor-
tional to f 4 which is caused by f 2 and we obtain

p
e 5 = rf 4 = r ----2- (4.31)
I2

To find an expression for e 7 we have to use the constitutive law for the transformer, for
which we can write

e 7 = me 8
(4.32)
mf7 = f 8

From the first line of Eq. 4.32 we find that e 7 is proportional to e 8 which again is caused by
e 9 . This means that

e 7 = me 8 = me 9 (4.33)

From the constitutive law for the C-element we have that

q
e 9 = -----9- (4.34)
C9

and substituting Eq. 4.34 into Eq. 4.33 yields

 Development of state equations 133

q
e 7 = m -----9- (4.35)
C9

Finally substituting Eq. 4.31 and Eq. 4.35 into Eq. 4.30 results in the second state equation

p q
p· 6 = r ----2- – m -----9- (4.36)
I2 C9

For the third state equation we write

q· 9  f 9 = f 8 – f 10 (4.37)

For f 8 we turn to the second line of Eq. 4.32 and find that f 8 is proportional to f 7 which
again is caused bye f 6 , which through the constitutive law for the I-element can be written as

p
f 6 = ----6- (4.38)
I6

This results in

p
f 8 = mf 7 = mf 6 = m ----6- (4.39)
I6

From the causal strokes we find that f 10 is caused by f 11 since it sets the flow at the 1-junc-
tion. Thus, we obtain that

p 11
f 10 = f 11 = ------
- (4.40)
I 11

Now we substitute Eq. 4.39 and Eq. 4.40 into Eq. 4.37 and we obtain for the third state equa-
tion

p p 11
q· 9 = m ----6- – ------
- (4.41)
I 6 I 11

For the last and final state equation we can write

134 Development of State Equations

q p 11
p· 11 = e 11 = e 10 – e 12 = -----9- – R 12 ------
- (4.42)
C9 I 11

This completes the task of converting the bond graph model into a model of differential equa-
tions. We hope that you have discovered that causality is a key in converting a bond graph
into a set of first-order state equations. Are you impressed ? We are !

4.4 Differential Causality.

In some of the models that we construct, we will not be able to assign integral causality to all
the C and I elements in the model as was the case for the graph shown in Fig. 4.14. That
means that we are forced to assign differential causality to some of the storage elements.
Since differential causality is not what the C and I elements prefer, we would suspect that
something special would occur when we write the state equations. It sure does, and we will
show what happens through a couple of very simple example systems, shown in Fig. 4.18.

L R

m
+

k1 u(t) C1 C2
-
b
k2 Se
6

5 4
I 1 R
3

2 1
C 0 C

Fig. 4.18 Differential Causality Examples

From the augmented bond graph, we identify that C 1 and I 5 have integral causality, but C 2
is forced to have differential causality. If we alternatively had chosen integral causality on
C 2 , then C 1 would have been forced to have differential causality since only one bond can
set the effort at a 0-junction.

When C 2 as in the problem at hand, has differential causality it means that q 2 is not a state
variable because it can be expressed in terms of the other state variables of the model. To
obtain this interdependence let us start with the definition of q 2 as
 Differential Causality. 135

q2 = C2 e2 (4.43)

when we assume a linear relationship.

From the causal marks on the bond graph we find that e 2 is caused by e 1 which can be
expressed as

q
e 2 = e 1 = -----1- (4.44)
C1

from the constitutive relation. Substituting e 2 from Eq. 4.44 into Eq. 4.43 we obtain that

C
q 2 = -----2- q 1 (4.45)
C1

which shows that q 2 is related to q 1 and hence is not an independent variable and can there-
fore not be a state variable.

For some of you this interdependency of the two capacitive spring elements might not be a
surprise since you have learned that two springs i series or two electrical capacitors in paral-
lel can be replaced by one equivalent spring or capacitor.

Thus, returning to the bond graph shown in Fig. 4.18 we obtain the following state and input
vectors

q1
x = and u = S e 5 (4.46)
p5

Starting with writing the state equation for q 1 as

q· 1  f 1 = f 3 – f 2 (4.47)

The flow variable f 3 is according to the causal strokes caused by f 5 which can be expressed
as

p
f 3 = f 5 = ----5- (4.48)
I5
136 Development of State Equations

The flow variable f 2 is by definition equal to

·
f2 = q2 (4.49)

Since C 1 and C 2 are both assumed to be constant, the time derivative of q 2 from Eq. 4.45
can be substituted into Eq. 4.49 which yields

C ·
f 2 = -----2- q 1 (4.50)
C1

By substituting Eq. 4.48 and Eq. 4.50 into Eq. 4.47 we now obtain the differential equation

· p C ·
q 1 = ----5- – -----2- q 1 (4.51)
I5 C1

·
which is an implicit equation in that q 1 occur on each side of the equal sign. Implicit equa-
tions will always occur when we have models with dependent elements or energy storage ele-
ments with differential causality.

Rearranging Eq. 4.51 we obtain

C1 p5
q· 1 = ------------------
- ----- (4.52)
C1 + C2 I5

The second state equation can be written without any difficulties to yield

p q
p· 5  e 5 = S e 6 – R 4 ----5- – -----1- (4.53)
I5 C1

Let us try to confirm this interdependency of the two capacitive elements from a physical
point of view. Looking at the two schematic drawings in Fig. 4.18 we see that the spring-
mass system contain two springs in series and the electric circuit contains two capacitors in
parallel. Two springs in series means that the force across each one of them is equal, and two
capacitors in parallel experience equal voltage. Two elements with equal efforts, only one
can prescribe the common effort and the other is dependent. Thus, our physical reasoning
confirm the interdependency given in Fig. 4.45. It is of interest to note that using bond graph
models and causality assignment we will even before writing system equations know that
implicit equations will occur with different causality on an element. In some cases with non-
linear equations we might not be able to solve the implicit equations and hence utilise causal-
 Differential Causality. 137

ity to modify the model and thus eliminate differential causality.

Example II
From the previous example we know that when a storage element is forced to take differen-
tial causality it does not contribute with a state variable and when writing the system state
equations implicit equations will occur.

Let us model one more system with dependent elements or differential causality. This time
we will write the state equations for the systems shown in Fig. 4.19a and b. The mechanical
system and the electrical circuit have the same augmented bond graph shown in Fig. 4.19c.

L1 L2 R
l1 l2
F(t)
+
m1 m2
u(t) C
-
k b

(a) (b)

Se C
7
1
5 4 3
1 TF 1 R
6 2

I I
(c)

Fig. 4.19 Differential Causality Examples

The particular systems chosen is so that you as a reader easily can verify that the velocities of
the two masses are directly related since they are both connected to the rigid bar.

Inspecting the bond graph in Fig. 4.19c elements C 1 and I 2 have integral causality, but I 6
has been forced to take differential causality. Which one of the two inertia elements I 2 and
I 6 we force to take differential causality is indifferent.

Since I 6 has differential causality we know that p 6 is not an independent variable and there-
fore, not a state variable. We then start to find the relationship between p 6 and other state
variables. With assumed linear inertia we write
138 Development of State Equations

p6 = I6 f6 = I6 f5 (4.54)

since f 5 is causing the flow at the left hand 1-junction. From the constitutive relation for the
transformer we can write

e 5 = me 4
(4.55)
mf5 = f 4

were m = l 2  l 1 .

Substituting for f 5 from Eq. 4.55 and recognising that

p
f 2 = ----2- (4.56)
I2

we can write that

1 I6 1 I6
p 6 = ---- ---- p 2 and p· 6 = ---- ---- p· 2 (4.57)
m I2 m I2

This result is obvious when we look at Fig. 4.19a. With a stiff bar the velocity of the two iner-
tias must be related. We can now write the state equation for p 2 as

p· 2 = e 2 = e 4 – e 1 – e 3
q1 p2 (4.58)
= e 4 – ------ – R 3 -----
C1 I2

Substituting for e 4 from Eq. 4.55 into Eq. 4.58 we obtain

1 q p
p· 2 = ---- e 5 – -----1- – R 3 ----2- (4.59)
m C1 I2

The relation for e 5 can be found by summing efforts at the left 1-junction which becomes

e 5 = e 7 – e 6 = S e 7 – p· 6 (4.60)

With a constant transformer modulus m , the time derivative of p 6 in Eq. 4.57 when substi-
tuted into Eq. 4.60 then gives
 Differential Causality. 139

1I
e 5 = S e 7 – ---- ---6- p· 2 (4.61)
m I2

When we substitute e 5 from Eq. 4.61 into Eq. 4.59 we obtain that

· 1 1 I q p
p 2 = ---- S e 7 – -----2- ---6- p· 2 – -----1- – R 3 ----2- (4.62)
m m I2 C1 I2

As shown in Eq. 4.62 we ended up with an implicit equation as we expected. When we solve
for p· 2 we obtain the following state equation

· m 2 I2  1 q p
- ---- S e 7 – -----1- – R 3 ----2-
p 2 = --------------------- (4.63)
m I2 + I6
2  m C 1 I2 

The other state equation can easily be written as

p
q· 1 = ----1- (4.64)
I1

As we progress through the book we will encounter many systems with dependent elements.

General procedure - differential causality.

When differential causality is experienced the approach for solving the implicit equations
and write down the state equations can be formulated in a general procedure. The procedure
can be given as

1. Write down the constitutive law for each element with differential causality in the
form:

–1 –1
p = I  f  or q = C  e 

2. Use the bond graph to express the flow or effort variables f or e as functions of the
state variables of the model, i.e. the storage elements with integral causality, the
sources and the model parameters. Insert these expressions into the constitutive laws
above.
140 Development of State Equations

3. Differentiate the constitutive laws above with respect to time

–1 –1
dp d  I  f   dq d  C  e  
e = ------ = -----------------------
- and f = ------ = ------------------------
-
dt dt dt dt

especially paying attention to possible sources and other relations which is a function
of time.

4. Next write down the state equations and insert the expressions for the efforts or flows
from 3, when these variables shows up in the expressions.

5. Solve the set of implicit equations for each of the state variables. Observe that often
one and one state equations can be treated individually, but in general any number of
state equations may be involved.

It is an advise of an expert to closely follow the procedure outlined here when trying to write
equations for models having elements with differential causality. Trust us!

4.5 Incomplete Causality Assignment - Algebraic loops

In some cases models appear where it is not possible to assign complete causality on all the
bonds in a model by using sources and energy storage elements. That is, we end up with
resistance elements, or in some seldom experienced cases transformer or gyrators, that have
no causality assigned to them, and in order to complete the causal assignment we have to
choose the causality on one of the R-elements and then extend the causality as far as possible
as we did in Fig. 4.14e. This process has to be repeated until all elements have been assigned
a causal stroke. Writing the state equations in such cases may turn out to be a infinite prob-
lem, where you are running in circles. However, following a practice reading the causality
information already present in the bond graph, writing state equations even in these cases
becomes a triviality.

Let us show an example on uncompleted causal assignment by starting with a very simple
system as shown in Fig. 4.20. As shown in Fig. 4.20b, the two R-elements do not have cau-
sality assigned to them and it is not possible to extend the causality based on the effort source
and the C-element. In order to complete the causality assignment we must choose causality
on one of the R-elements and then extend the causal assignment as far as possible. It is arbi-
trary on which R-element we choose causality. As shown in Fig. 4.20c, bond 3 have been
chosen with flow input to the 1-junction and bond 4 is then forced to have an effort input
since only one element can set the flow at a 1-junction. In this example, we completed the
causal assignment with choosing causality once, but in other cases, we might have to choose
causality several times in order to complete our task. Let us now write the state equations for
the system to see the consequences of the fact that we had to choose causality on one of the
 Incomplete Causality Assignment - Algebraic loops 141

R-elements.

C C
C R1

+ 1 1
u(t)
-
R2
2 3 2 3
Se 1 R Se 1 R

4 4
(c)
R R
(a) (b)

Fig. 4.20 Incomplete causality example

Since the bond graph shown in Fig. 4.20c contain only one storage element with integral cau-
sality, we only need to write one differential equation. Then

e 1
q· 1  f 3 = -----3 = -----  e 2 – e 1 – e 4  (4.65)
R3 R3

Substituting known relations for the efforts in Eq. 4.65, we can write

q
q· 1 = f 3 = -----  Se 2 – -----1- – R 4 f 3
1
(4.66)
R3  C3 

Taking a closer look at Eq. 4.66, we find that f 3 is a function of itself. We have an implicit
equation or what is known as an algebraic loop. In fact, every time we meet the problem
where we have to choose causality on an element in order to complete the causal assignment,
an algebraic loop occurs. The causal bond graph tells us this important fact even before we
have started to write equations.

Let us now see how to brake the loop before we start writing the system differential equa-
tions. The first thing we do is to identify the input variables e and f to the system from each
of the R-elements involved in a particular loop. In the particular problem at hand the causal
strokes indicate that f 3 and e 4 are the inputs to the rest of the system. As the next step, we
express each of these variables in terms of sources, state variables and the other variables
involved in the loop. With n R-elements involved in a loop, the result is n , algebraic equa-
tions in n unknowns which we must solve.

Starting with e 4 we write the constitutive relation for the element as

142 Development of State Equations

e4 = R4 f4 = R4 f3 (4.67)

since f 4 is caused by f 3 . The variable e 4 is now expressed in terms of f 3 , the other input var-
iable, so we stop and start again writing the constitutive equation for the other R-element as

e3 q1
f 3 = ----- = -----  Se 2 – ------ – e 4
1
(4.68)
R3 R3  C1 

We now have two equations Eq. 4.67 and Eq. 4.68 in two unknown f 3 and e 4 , which when
solved yields

q
f 3 = ------------------  Se 2 – -----1-
1
(4.69)
R3 + R4  C 1

and

R4  q
- Se – -----1-
e 4 = ----------------- (4.70)
R 3 + R 4  2 C 1

The resulting state equation can now be written as

q1
q· 1 = f 1 = f 3 = ------------------  Se 2 – ------
1
(4.71)
R3 + R4  C 1

Looking at Eq. 4.71 the two resistors in the circuit combine into an equivalent resistor equal
to the sum of the two. This is a well-known fact from basic electric circuit theory.

The fact that bond graph causal assignment indicates when we have constructed a model con-
taining an algebraic loop is very important since we then know that we have an algebraic
problem on our hands before we start writing equations. In fact, for some non-linear systems,
we are not able to break the loop by solving for the variables of interest. When this is the
case, then we have to make some changes to the model, and causality is indispensable in
making the necessary changes to the model to avoid the algebraic loop. This will be shown in
the next example shown schematically in Fig. 4.21.

In Fig. 4.22a is shown the system bond graph with causality assigned as far as possible. The
two R-elements in the model do not have causality assigned to them, which means that there
is an algebraic loop in the model. By choosing causality on one of the R-elements, the causal
assignment can be completed as shown in Fig. 4.22b. Here f 5 is selected as the input to the
 Incomplete Causality Assignment - Algebraic loops 143

system model which again forces the causal assignment on the rest of the bonds and with
e 10 , as the input to the system model.

v(t) k1 b1
A m

a
b2 k2
B

Fig. 4.21 Incomplete Causality Example 2.

To start, we will assume that all the elements have linear constitutive relations. Thereafter,
we will assume that one of the dampers have a non-linear constitutive relation. This assump-
tion increases the difficulty in breaking the loop, and we will show how we can use causality
to make changes to the model and avoid the algebraic loop.

C:1/k1 R:b1 C R

2 vA 5 2 5
1 3 4 6 1 3 4 6
Sf 0 1 0 I Sf 0 1 0 I

7 7

TF:m TF:m
8 8

10 9 10 9
b2 :R 1 C:1/k2 R 1 C
vB
(a) (b)

Fig. 4.22 Bond Graphs for Incomplete causality Example 2

Since we know that there is an algebraic loop in the system, we start with breaking the loop.
We note that f 5 and e 10 , are the input variables to the system model from the two R-elements
involved in the loop. Starting with f 5 and writing the constitutive relation for R 5 as

1
f 5 = ----- e 5 (4.72)
R5
144 Development of State Equations

and following the causal strokes, the resulting expression becomes

1 q q
f 5 = -----  -----2- – m  -----9- + e 10  (4.73)
R5  C2  C9 

and we stop since f 5 is now related to e 10 .

Continuing with expressing e 10 , through the constitutive relation for the other R-element as

e 10 = R 10 f 10 = R 10 f 8 (4.74)

since f 10 is caused by f 8 .

Extending through the constitutive equations for the transformer with modulus m

1
e 8 = ---- e 7 and mf 7 = f 8 (4.75)
m

the following relation for e 10 results

p
e 10 = mR 10  ----6- + f 5 (4.76)
 I6 

The transformer modulus can be determined from the equations for the two velocities v A and
v B in terms of the geometric lengths a b and the angular velocity  of the vertical bar as

v A =  a + b  = f 7 and v B = b = f 8 (4.77)

Equating  in the two expressions above yields that

a+b
v A = ------------ v B (4.78)
b

and the transformer modulus m is then equal to

b
m = ------------ (4.79)
a+b

We now have two equations Eq. 4.73 and Eq. 4.76 with two unknown f 5 and e 10 which can
 Incomplete Causality Assignment - Algebraic loops 145

be solved to yield

q p
-  -----2- – ------ q 9 – m 2 R 10 ----6-
1 m
f 5 = ------------------------------- (4.80)
 R 5 + m R 10   C 2 C 9 I6 
2

and

mR 10 q q p
-  -----2- – m -----9- – R 5 ----6-
e 10 = ------------------------------- (4.81)
2
 R 5 + m R 10  C 2  C9 I6 

From the bond graph in Fig. 4.22 we find that there are three storage elements with integral
causality. The state vector x and input vector u , is then given as

p6
x = q2 and u = Sf 1 (4.82)
q9

Starting with writing the state equations in the order of the state vector, then

q
p· 6 = e 6 = e 4 = e 3 – e 7 = -----2- – me 8
C2
q
= -----2- – me 9 – me 10 (4.83)
C2
q q
= -----2- – m -----9- – me 10
C2 C9

Substituting in the value for e 10 solved for in Eq. 4.81 and the equation for p· 6 becomes

R5 q q p
-  -----2- – m -----9- – m 2 R 10 ----6-
p· 6 = -------------------------- (4.84)
R 5 + m R 10 C 2
2  C9 I6 

Continuing with the next two state equations yields

q q p
-  -----2- – m -----9- – R 5 ----6-
1
q· 2 = Sf 1 – -------------------------- (4.85)
R 5 + m R 10 C 2
2  C9 I6 

and
146 Development of State Equations

q q p
-  -----2- – m -----9- + R 5 ----6-
m
q· 9 = -------------------------- (4.86)
R 5 + m R 10  C 2 I6 
2 C9

which ends the derivation of state equations when algebraic loops are present. This is a stand-
ard case which arise in many situations, however more complex situations may occur when
more than 2 equations have to be solved simultaneously. It is therefore important to stick to
the algorithm outlined here to be able to solve the problem with minimal amount of work.

In summary, an extension to the sequential causality assignment procedure to apply when

algebraic loops are present can be stated as:

1. If you are forced to select causality on an R-element as a consequence of carrying out

the sequential causality assignment procedure: then set aside the power variable
 e 1 f 1  , meaning the effort or the flow of that bond which is “forcing” your system
as seen from the outside. ( e 1 f 1 being renamed variables of the global naming con-
vention used in the system bond graph)

2. Extend the causal information by using 0, 1, TF and GY elements as normal. For each
R-element which is causally fixed by this: set the aside the power variables
 e 2 f 2  e n f n  which is “forcing” your system as seen from the outside.

3. Write equations for each of these variables as a function of each other, the model
parameters, expressions for the sources and the state variables of the system. Then
solve the possibly nonlinear algebraic system for the variables e i f i i = 1n

e 1 = f 1  e 1 f 2 e 3 
f 2 = f 2  e 1 f 2  e 3 
e 3 = f 2  e 1 f 2 e 3 
4. Write the state equations as normal, and insert the expressions for the e i f i i = 1n
variables when they show up, and voilà! You are done!

Algebraic loops with nonlinear elements.

Now, let us look at a more complex case which also better demonstrate the problem we are
facing. Let us assume a non-linear constitutive equation of the form

2
e 10 = B 10 sgn  f 10  f 10 (4.87)
 Incomplete Causality Assignment - Algebraic loops 147

for the damper R 10 . The expression sgn  f 10  means the sign of the velocity f 10 which is
needed in order for the damping force to oppose the motion when the velocity is positive and
negative. Following the causal strokes as before the following relation for e 10 results

p6 2
e 10 = B 10 sgn  f 10  ----  f 5 + -----
1
(4.88)
m I6 

or

p6 2
e 10 = B 10 sgn  f 10  ----  f 5 + 2f 5 +  ----- 
1 2
(4.89)
m  I6  

The result is now a second order equation, and with the relation for f 5 , given in Eq. 4.73, it
becomes quite an algebraic problem to solve for f 5 and e 10 , and it is therefore left to the
reader to complete the solution. Check out the physical meaning of your solution!

It is regrettable that someone could construct a model, which contain such difficult algebraic
problems. Let us therefore investigate if there is something we can do to the model in order to
eliminate the algebraic loop and the problems that they cause. Some people say that we cheat
when we make some changes to a model. However, a model is the modeller's comprehension
of the real world, and he decides which elements to include and how they are connected.
Therefore, the modeller can also make changes to his model without cheating. As mentioned
previously, causality will be used to help us make a change, which will eliminate our prob-
lem. As a help, we will return to the bond graph shown in Fig. 4.22a and investigate what
additional elements could be added to the model in order to resolve our causal problem.

If there were a storage element at either of the two 1-junctions or at the 0-junction joining
bond 4, 5 and 6 which would resolve the causality at the junctions the problem would be
solved. That is a C-element at 0-junctions and an I-element at 1-junctions forces the causality
on the other bonds joining the junction. There are no obvious elements, which we have
ignored that we could add for the right 0-junction or the two 1-junctions. However, if we
include the mass moment of inertia of the vertical bar, the system bond graph will be as
shown in Fig. 4.23 and we can complete the causal assignment and avoid the algebraic loop.
In exchange, we have obtained an additional state equation. As we will discover later, addi-
tion of extra elements is not without a price, but the problem and consequences are in the
hands of the modeller.
148 Development of State Equations

C R

2 5
1 3 4 6
Sf 0 1 0 I

TF:
8

10 9
R 1 C
11

TF:m
12

13
1 I

Fig. 4.23 Extended Bond Graph for Incomplete Causality Example 2

4.6 Non-linear models

For the bond graph models which we up to now have converted to mathematical equations, it
has been assumed that each element in the model had a linear constitutive law. However, for
most real world problems, the elements all have more or less non-linear constitutive laws, or
even advanced algorithms associated. It is therefore proper that we illustrate with an example
the difference between writing linear and non-linear equations from a bond graph model.

We will illustrate the procedure by studying the simple example shown in Fig. 4.24a. The
system consists of a Newtonian mass suspended by an air-bellow spring and a deliberately
 Non-linear models 149

constructed non-linear damper. The augmented bond graph is shown in Fig. 4.24b.

F(t) R

3
m 2 1
C 0 I:m
x

b 4

S e :F(t)

(a) (b)

Fig. 4.24 Non-linear system example model

Before we start to derive the system state equations, we will define the constitutive laws for
the air spring and damper.

For the damper we have a square law between the damping force F d and velocity x· , written
as

F d = bx· x· (4.90)

We must be careful not to forget the absolute sign because the damping force must be oppo-
site to the velocity. Otherwise the damper will supply power with negative velocity. That is
not too good.

The bellow is filled with air which have the equation of state

PV = m air RT (4.91)

where P is the pressure, V is the bellow air volume, m air is the mass of the air in the bellow,
R is the gas constant and T is the air temperature.

The air-spring force F sp can now be determined assuming constant temperature T as

m air RT
F sp = PA = ----------------- A (4.92)
V

where A is the cross-sectional are of the bellow. By measuring the displacement of the New-
tonian mass from equilibrium then the volume can be expressed as
150 Development of State Equations

V = V 0 – Ax (4.93)

where V 0 is the volume at equilibrium. To determine the equilibrium volume we start with
the fact that

P 0 A = mg (4.94)

From the equation of state, Eq. 4.91, we can write

m air RT 0
V 0 = -------------------- (4.95)
P0

and by substituting for P 0 from Eq. 4.94 into Eq. 4.95 we obtain

m air RT 0
V 0 = -------------------
-A (4.96)
mg

The constitutive law for the C-element can then be written as

RT 0 m air
F sp = ----------------------------
- (4.97)
RT 0 m air
-------------------- – x
mg

Now we are ready to write the system state equations from the bond graph in Fig. 4.24b. As
we can see, we have two elements with integral causality and the state and input vectors
becomes

p1
x = and u = S e 4 (4.98)
q2

Starting with expressing the derivative of the state variables as we did for linear systems, we
obtain for the derivative of p 1

p· 1  e 1 = S e 4 – e 2 – e 3 (4.99)

Since we have non-linear relationships for some elements we can not use the linear relation-
 Output Equations 151

ships for e 2 and e 3 . Instead e 2 is given by Eq. 4.97 and e 3 is given by Eq. 4.90. Substituting
these equations into Eq. 4.99 we obtain

RT 0 m air p
- – b ----1- p----1-
p· 1 = S e 4 – ------------------------------ (4.100)
RT 0 m air I1 I1
-------------------- – q 2
mg

For the second state equation, we can write

p
q· 2  f 2 = f 1 = ----1- (4.101)
I1

Now we are confident that you have obtained the necessary skills to take a bond graph and
deduce the state equations no matter how complex the system may look. Don’t disappoint
us!

4.7 Output Equations

It has been repeated many times that the state variables in a dynamic system represent a vital
key to the analysis problem. However, the actual output that we want from a system need not
be and usually is not, identical to the chosen state of the system. When this is the case, then
we need in addition to the state equations, output equations that relate desired outputs to the
state and input variables.

Let the variables y 1  t  y 2  t   y m  t  be the m desired output variables of the system. The
output variables represent the link between the system and the outside world. In general, the
output variables are functions of the state-variables and the input variables. This then means
that we can always write a generalised vector equation relating the m-dimensional output
vector y , to the n-dimensional state vector x , and the r-dimensional input vector u , as:

y  t  = h  x u t  (4.102)

In expanded form Eq. 4.102 becomes

152 Development of State Equations

y 1 = h 1  x 1 x 2  x n u 1 u 2  u r t 
y 2 = h 2  x 1 x 2  x n u 1 u 2  u r t 
. (4.103)
.
y m = h m  x 1 x 2  x n u 1 u 2  u r t 

In the case when the output equations are linear, they can be written as a vector matrix equa-
tion in normal form as:

y1 c 11 c 12 . . c 1n x 1 d 11 d 12 . . d 1r u 1
y2 c 21 c 22 . . c 2n x 2 d 21 d 22 . . d 2r u 2
. = . . . . . . + . . . . . . (4.104)
. . . . . . . . . . . . .
ym c m1 c m2 . . c mn xn d m1 d m2 . . d mr u r

or in compact form as

y = Cx + Du (4.105)

As an illustrative example of the use of output equations, let us consider the spring-mass-
damper system shown in Fig. 4.25.

k2 b2

m1
x2 ,x 2

k1 b1

m2
x1 ,x 1

Fig. 4.25 Spring-mass system

A force F  t  excites the system and it is our task in this example to compute the instantane-
ous velocity and displacement for each inertia. This means the output vector y becomes
 Output Equations 153

x1
x2
y = (4.106)
v1
v2

Neither velocity nor displacement are state variables but we are going to demonstrate how
they can be expressed in terms of the state variables.

Now since we have gained some experience in constructing bond graphs let us use our phys-
ical understanding of the system and construct the bond graph by direct inspection. We start
with establishing two 1-junctions to represent the absolute velocities of the two inertias and
adjoin an I-element to each junction as shown in Fig. 4.26. Representing the excitation force
F  t  and the two gravity forces as effort sources, we adjoin them to appropriate 1-junctions.
The spring and damper connected to the ground on one end are experiencing the velocity dif-
ference  x· 2 – 0  and can therefore be adjoined directly to the 1-junction representing the
velocity x· 2 . For the spring and damper connected between the two inertia's, a 3-port 0-junc-
tion bonded to the two 1-junctions is used to compute the relative velocity between the two
inertias, and this relative velocity represented by a 1-junction is the same for the spring and
damper and therefore adjoined to this 1-junction. With the positive directions as indicated in
Fig. 4.26 and assuming tension in springs, the companion power directions are shown in Fig.
4.26 of the augmented bond graph of the system in Fig. 4.25.

C R
4 12
7 2
Se 1 I

11 3 C
9
0 1
10

8 R
6 1
Se 1 I
13

5 R
Se

Fig. 4.26 Augmented bond graph

From looking at the augmented bond graph shown in Fig. 4.26, we obtain the state vector x
and input vector u to be
154 Development of State Equations

p1
Se 5
p2
x = and u = Se 6 (4.107)
q3
Se 7
q4

Writing the state equations as

·
p 1  e 1 = Se 5 – Se 6 – e 8 – e 13
e 8 =  caused by  e 9 = e 3 + e 10
q (4.108)
e 8 = -----3- + R 10 f 10
C3
p p
f 10 =  caused by  f 9 = f 8 – f 11 = ----1- – ----2-
I1 I2

· p p q
p 1 = Se 5 + Se 6 –  R 10 + R 13  ----1- + R 10 ----2- – -----3- (4.109)
I1 I2 C3

· q q p p
p 2 = e 2 = Se 7 + -----3- – -----4- + R 10 ----1- –  R 10 + R 12  ----2- (4.110)
C3 C4 I1 I2

· p p
q 3  f 3 = f 9 = ----1- – ----2- (4.111)
I1 I2

· p
q 4  f 4 = ----2- (4.112)
I2

The state equations can now be written in matrix form as

R 10 + R 13 R 10
–  ----------------------
1
– -------- – ------ 0
I1 I2 C3
p· 1 R 10 R 10 + R 12 1 p1 1 1 0 S
1
-------- – ---------------------- ------ – ------ p
p·
e5
2
= I1 I2 C3 C4 2 + 0 0 1 (4.113)
S e6
q· 3 1- 1 q3 0 0 0
--- – ---- 0 0 S
q·4
I1 I2 q4 0 0 0 e7
1
0 ---- 0 0
I2
 Output Equations 155

The output equations for the position and velocities of the masses can be expressed as

x2 = q4 (4.114)

x1 – x2 = q3 (4.115)

x1 = q3 + x2 = q3 + q4 (4.116)

p1
v 1 = ----- (4.117)
I1

p2
v 2 = ----- (4.118)
I2

and the output equations can then be written in matrix form as

x1 0 0 1 1 p1
x2 0 0 0 1 p2
= (4.119)
v1 1  I1 0 0 0 q3
v2 0 1  I2 0 0 q4

We may use the matrix equation

y = Cx (4.120)

as a simplified representation for Eq. 4.119.

156 Development of State Equations

4.8 Bond Graph to Block Diagram Conversion

Up to now we have developed bond graph models of numerous physical systems and you
may have been convinced that bond graphs are particularly well suited for the modelling
process. In many areas however, block diagrams are used to represent the mathematical
structure of the models. Control engineers preferably use block diagrams to represent ampli-
fiers, instruments, the actuator dynamics and the control scheme itself. Using bond graphs as
the preferred method for describing physical models it is easy to see that being able to com-
bine both block diagram and bond graph models at the graphical level would be advanta-
geous.

1
--------
Ts
i
Control System

K +
p RPM
-
K s
d K
------------------ f
1+T s -----------------
K
r d 1+T s
f

I
Propulsion System


x
d

Se 1 TF 1 R

Fig. 4.27 Bond graph and block diagram combinations

Simply combining bond graph and block diagram models can easily be done by for example
representing the physical model by bond graphs and the control scheme using block dia-
grams. The coupling in this situation is done through modulated bond graph sources with sig-
nal inputs from the block diagram representing the control scheme. A simple example is
shown in Fig. 4.27, where the bond graph model is a simplified representationof a torque
driven system containing an inertia and resistive load driven through a gearbox represented
with a transformer. The driving torque is represented as a modulated effort source the block
diagram represents the control system for the rotational velocity system.

4.8.1 Block diagrams

We have in our development so far shown two methods of representing dynamic systems,
namely:

• bond graphs
• first-order differential equations
 Bond Graph to Block Diagram Conversion 157

In control engineering, dynamic systems are frequently represented in the diagrammatic form
of block diagrams. Since block diagrams are an important representation of system models,
we will in this section show how a bond graph model can be converted to a block diagram
model in a systematic manner.

Before we start developing rules for converting a bond graph model into a block diagram
model, it may be appropriate to give a short basic description of block diagrams for those of
you that nave not used them previously.

A block diagram represents in a diagrammatic form the flow of information and the functions
performed by each component in the system. Block diagrams represents linear mathematical
expressions, and in this case dynamic block diagrams are built up from the three components
illustrated in Fig. 4.28.

x t yt  x t + zt x t x t
G
-
y t xt

a) b) c)

Fig. 4.28 Basic components of a block diagram, a) transmission block, b) summer or com-
parator and c) splitting or pick off point

The different blocks which go to make up a block diagram are interconnected by lines which
are interpreted as if the scalar value runs along the lines as power. This means that in the
block diagram representation, signals appear as lines and functional relations as blocks.

The square block shown in Fig. 4.28a is the graphical symbol for multiplication. That is, in
this block the input quantity x  t  is multiplied by the function G in the block to obtain the
output y  t  . The relationship between the input signal x  t  which enters the block and the
output signal y  t  is express by the following equation

yt = G  xt (4.121)

where G is an element or system characteristic which relates a change in an element or sys-

tem output to the change in input.

The circle shown in Fig. 4.28b is used to indicate a summing operation. The arrowheads
pointing toward the circle indicate input quantities, while the arrowheads leading away, sig-
nifies the output. The sign at each input arrowhead indicates whether the quantity is to be
158 Development of State Equations

added or substracted. The mathematical relationship for the summing operation is

z t = x t – yt (4.122)

In Fig. 4.28c is shown what is known as a pickoff-point which indicates that the value of any
signal may be forwarded to two or more locations.

A suitable mathematical form of the function G is met through the use of a relationship
called the system function or better known as the transfer function, and which contains basic
information concerning the essential characteristics of an element or system without regard
to initial conditions or excitations. Thus, the concept of transfer functions is very important in
block diagram and system analysis work.

In this section we will therefore, introduce the concept of the transfer function and show how
it can be used to convert a bond graph model to a block diagram model. Later, we will dis-
cuss the use of the transfer function partial-fraction expansion to obtain system time response
and how the transfer function can be used to find the sinusoidal steady state response for a
system as the frequency of the excitation sources are varied.

The transfer function is based on the so-called operator notation. This operational form is
usually expressed in terms of the Laplace transform operator s , indicated symbolically by

L g t   G s  (4.123)

and spoken Laplace transform of g  t  . Thus, the transfer function is defined as

L  output  Ys Laplace transform of y  t 

G  s  = ------------------------- = ----------- = -------------------------------------------------------------- (4.124)
L  input  Xs Laplace transform of x  t 

where a single input is applied, all other inputs being set to zero.

The basic purpose of using the Laplace transform is to replace operations such as differentia-
tion and integration by simpler algebraic operations in the s domain. Later, we will show that
Laplace transform gives us a very systematic way for relating the time-domain behaviour of
a system to its frequency-domain behaviour. Here we will only cover what is required to con-
vert a linear bond graph model to a block diagram model. That means that we need to list
three transform operations:
 Bond Graph to Block Diagram Conversion 159

1. Multiplication by a constant
x t yt
K

L  y  t   = Y  s  = L  Kx  t   = KX  s  (4.125)

1. Differentiation
x t d yt
-----
dt

Ys  = s  X s (4.126)

1. Integration
x t yt
 dt

X s
Y  s  = ----------- (4.127)
s

This tells us that differentiation and integration in the time domain reduces to an algebraic
operation in the s domain

To convert a bond graph model to a block diagram model, it is necessary to obtain the block
representation for each of the nine basic elements in therm of the transfer function between
the input and output signals. In Table 4.2 the set of block equivalencies for the basic bond
graph elements are given. Also the different causality options are included for the nine basic
elements, including the derivative causality options for the storage elements. For the 0 - and
1 - junctions however, there exists several options based on the power flow directions and
causality assignments. We hope that you without too much problem will be able to write
down these options.

The 0- and 1-junctions each denotes a special pair of operations. One is to sum the flow or
effort signal respectively and the other is to distribute the common variable effort or flow
respectively. The 0-junction shown in Table 4.2 has e 1 as input and therefore, both e 2 and
e 3 is output, simply representing a splitter. The sum of the flow at the 0-junction should
however be equal to zero, indicating that positive or negative signs must be assigned to each
input according to this definition. Here this gives that the f 3 input must be negative and the
input f 1 must be set positive. The same procedure must be performed at the 1-junction shown
in Table 4.2 , resulting in input e 1 being assigned a positive and input e 3 being assigned neg-
ative sign.
160 Development of State Equations

Utilizing the signal equivalent of the bond graph any augmented bond graph, that is where
power direction and causality has been assigned, can be put directly into an equivalent block
diagram without the need to consider any equations. To see this, consider the bond graph in
Fig. 4.29. From the augmented bond graph we simply draw element boxes for each bond

I 1I 1
----
Is
e f
5 5 5 e f5
5
e e e e
1 3 1 3 +
1 3 1Sf 10 11
Sf 0 1 ft
+ - -
f f
1 3 f f
1 3
2 4 e f e f
2 2 4 4 e f2 e f
2 4 4
1C 1R 1
------ R
C R Cs
(a) (b) (c)

Fig. 4.29 Bond graph to block diagram transfer: (a) augmented bond graph, (b) block dia-
gram paths, (c) complete block diagram

graph element, and 0- and 1-junction, then replacing each bond with their effort e and flow f
signal equivalents. The causal marks on the bond graph indicates how the signal information
flows, and thus how the errors on the signal lines in the block diagram should be oriented.
For example: bond 3 indicates that e 3 is input to the 1-junction and therefor must have a flow
( f 3 ) output from the same 1-junction or equivalently, effort output and flow input to the 0-
junction. Carrying out the same analysis for each bond yields the signal graph in Fig. 4.29b.

Now, let us continue and replace each bond graph element box by its block diagram equiva-
lent set of operations. The R-element for a start, has flow input and effort output meaning that
the block simply returns e 4 multiplying the input f 4 by the constant R . The Sf element sets
the flow f 1 neglecting any effort, meaning that the signal e 1 can be removed and the block
diagram source inserted for f 1  t  as shown in Fig. 4.29c.

Representing the storage elements I and C block diagram equivalents requires some discus-
sion. Starting with the defining equations for an I-element with constant coefficients, namely

1 dp
f  t  = --- p  t  and e  t  = ------ (4.128)
I dt

Performing the Laplace transformation of these equations yields

 Bond Graph to Block Diagram Conversion 161

1
F  s  = --- P  s  and E  s  = sP  s  – p  0  (4.129)
I

which solved for E  s  in terms of F  s  with p  0  = 0 results in the transfer function

Fs 1
----------- = ---- (4.130)
Es Is

The same development performed for the C-element shows that this element has the transfer
function

Es 1
----------- = ------ (4.131)
Fs Cs

These transfer functions can be inserted into the element blocks for the I- and C-element
respectively, resulting in the complete block diagram for the bond graph as shown in Fig.
4.29c.

If there are more than two elements connected to a junction, it might be a little difficult to
draw a clear diagram following the procedure outlined above and shown in Fig. 4.29. A
slightly different method of drawing the diagram might be used, although it is not strictly
procedural. In this method the sign and causality on each bond is used to form a block dia-
gram from a bond graph. It is also helpful in understanding the physical meaning of the
causal stroke. Let us review this procedure by converting the bond graph of a series circuit of
162 Development of State Equations

a spring-mass-damper system as shown in Fig. 4.30.

Se
a) b)
1
e-signal line
4 2
C 1 I
3 f-signal line

R
Se
e1
c) d)
e4 + e3 +

- - - -
e2
1 1- 1-
----
Is
-----
Cs
---
Is
R

f3

Fig. 4.30 From bond graph to block diagram

We start the conversion process with laying out two parallel lines representing the e - and f -
signal between which the relations are to be inserted as shown in Fig. 4.30b. Since we in this
example have a 1 - junction element, which means summation of efforts, we draw a circle
indicating a summation on the e - signal line. Then, using the causal strokes and the sign on
each bond to find that e 1 is a positive input and the efforts e 3 and e 4 are negative inputs to
the inertia elements, which yields the output velocity f 2 as shown in Fig. 4.30c. According to
the causal strokes the velocity f 2 is the input to the C and R elements, e 3 and e 4 are the out-
put. The resulting block diagram is shown in Fig. 4.30d.

In control engineering block diagrams are usually drawn with the input source to the left and
the signal flow towards the desired output to the right. For the block diagram in Fig. 4.30d
with Se 1 as input and f 2 as the output, the block diagram take the form shown in Fig. 4.31.

1-
-----
Cs
-
1
Se ---- f(t)
+ - Is

Fig. 4.31 Block diagram - control engineering layout

 Bond Graph to Block Diagram Conversion 163

Example:

After we now have converted a couple of simple bond graph models to block diagram models
let us consider a couple of more complicated system models. The first system we will con-
sider is given by the bond graph model shown in Fig. 4.32a.

This is a model of a permanent magnet DC-motor with a voltage input driving a resistive and
inertia load.

The first step in the conversion process is then to draw the two parallel lines representing the
e and f signals, and since we in this example have two 1 -junctions, we draw two summing
junctions on the e -signal line. The next step is then to connect the element blocks between
the two signal lines. Causal strokes at the left-hand 1 -junction indicates that the sum of the
efforts e 1 , e 3 and e 4 is the input to the inertia element I 2 which yields the flow f 2 as output.
This output flow is again input to R 4 which gives e 4 as the output. We will wait a bit with
the effort e 3 and move over to the right hand 1 -junction and do as we just did at the left
hand-hand 1 -junction. The block diagram will then look as shown in Fig. 4.32b.

As shown in Fig. 4.32a, a 2-port gyrator is the element which connects bond 3 and bond 5.
Thus, inserting the block diagram for a gyrator from Table 4.2 into Fig. 4.32b, we obtain the
block diagram which is shown in Fig. 4.32c. Now, redrawing the block diagram with input to
164 Development of State Equations

the left and output to the right, we end up with a final block diagram as shown in Fig. 4.32d.

I I
2 6
r
1 3 5
Se 1 GY 1
4 7

R (a) R
Se
e1
e4 + e3 e5 e7
- -
e-signal + -
e6
R 1
---- (b) 1
---- R
Is Is

f2 f6
f-signal

Se

e4 + e3 e5 e7
- - + -

r r
1 1
R ----
Is
----
Is
R

f2 f6
(c)

r
-
1 1
Se +
---- r +
----
- Is - Is

R R
(d)

Fig. 4.32 Bond graph to block diagram d example

This completes our development of the process of converting a bond graph to a block dia-
gram. As you have seen it is only a procedure and require no writing of equations. It is possi-
ble to computerize, but the final layout of the diagram may need some consideration.
 Bond Graph to Block Diagram Conversion 165

Table 4.2: Block diagrams for some bond graph elements

Sources: Se et e

e
Sf
ft f

Storage elements: C
e
1
------
Cs f

C e
Cs f

I e
1
----
Is f

e
I
Is f

Dissipation R
e
elements: R f

e
R –1
R f

Transformer and 1 2
e
1
m –1
e
2

gyrators: TF f f
m –1
1 2

e e
1 2 1 m 2
TF f f
1 m 2

e e
1 2 1 r –1 2
GY f f
1 r –1 2

e e
1 2 1 r 2
GY f f
1 r 2

Junctions: 1 3
e
1 e
3

(not all options) 0 f

1 + f
3

2
e
+
2 f
2

e - e
1 3 1 + 3
1 f
1
f
3

2
e f
2 2
166 Development of State Equations

REFERENCES
1. Controllab Products Inc., 20-Sim Reference Manual, P.O.Box 217, 7500 AE Enshede ,
The Netherlands, 2000.

PROBLEMS
4-1 For the system below, construct the bond graph and write the state equations

g M

v(t)

4-2 Make a bond graph and apply causality. Identify the state variable vector x and input
variables vector u .
+
-

4-3 Make a bond graph of the following circuit diagrams, apply causality and identify state
variables and input variables.
+

I
-

(a) (b)
+ V -
+

I
-

(d=
(c)
 Bond Graph to Block Diagram Conversion 167

4-4 Make a bond graph of the circuit diagrams below, apply causality and identify state and
input variables

L1 R2

R1 C1
E

R2 C
+

R3
-

(a) (b)

4-5 Make a bond graph of the circuit diagram below, apply causality and identify state and
input variables
R1

R1 R2
+
+

R2 C V
-

C1 C2
-

(a) (b)
4-6 Make a bond graph of the circuit diagrams below, apply causality and identify state and
input variables

4-7 Make a bond graph of the circuit diagram, apply causality and identify state and input
+

V
-

variables.
168 Development of State Equations

4-8 Make a bond graph, write the state equations and simulate the system using the follow-
R1 L1

I(t) C1 C2 L2

ing data: R 1 equal to 10, 100, 1000 and 5000  respectively, C 1 = C 2 equal to
2  10 – 4 As  V , L 1 = 166 Vs  A , L 2 = 200 Vs  A and I  t  equal to a step function
with 1 A as amplitude. Simulate for 5s and evaluate especially the current in L 1 and L 2 ,
and also the power in R 1 .

4-9 Assign causality to the bond graph, and show by indicating the causal path which vari-
ables are algebraically related to each other. (several solutions).

I:Ia R:Rb C:Ca

r
Ra:R 0 GY 0 1 R:Rc

Write down the state equations for this bond graph, and suggest a physical system that
can be modelled using this bond graph.

4-10 Draw a block diagram of the complete model given in Fig. 4.27.

4-11 Draw block diagrams of each of the bond graphs developed in Problem 4-1 to 4-9 above.

4-12 Write down the state equations of the problem set in Chapter 3.
 System Analysis Methods. 169

Chapter 5

System Analysis Methods.

In the previous chapters we have developed models in terms of bond graphs and as first order
state equations. It then seems natural at this time to make use of the state model of the physi-
cal process as a tool for determining and enhancing our understanding of the dynamic
response of the system. That is, we will look at methods of analysis which will deduce by the
use of mathematics, numbers that in some way indicate to us essential information as to the
operation or state of the system. These numbers will indicate a physical condition such as
position, velocity, flow rate, pressure, temperature, current and voltage or any one of a multi-
tude of other physical quantities.

Inspecting the models we have developed up to now it is not surprising to find that some are
linear but a large portion of real world problems are nonlinear. Strictly speaking, linear sys-
tems do not exist in practice, since all physical systems are nonlinear to som extent. Linear
systems are idealized models that are often fabricated by the analyst for simplicity of analy-
sis.

It is not difficult in principle to find solutions to general state equations using computer sim-
ulation. Although computers are effective in realizing answers from our mathematical state-
space models, they give us very little insight as to the general properties of the solutions. In
case of nonlinear equations there is no known approach for solutions in general and we must
use numerical methods. However, the mathematical theory of linear systems is highly devel-
oped and allows us to make great generalizations. The results of linear analysis allow for
extended understanding of nonlinear phenomena also, and thus linear theory forms the basis
for system analysis. Some system characteristics even have real meaning only when working
with linear systems. We mentioned previously that a large portion of the problems we are
asked to analyse are nonlinear. How is it then possible to utilize linear mathematical methods
to analyse nonlinear systems? One way is to utilize the concept of linearisation. That means
that we approximate the system equations as linear within a certain range. Linearising non-
linear equations is important since it is then possible to apply linear analysis methods that
will produce information on the behaviour of nonlinear systems. It is, therefore, the specific
objective of this chapter to explore general linear system properties and relate them to the
performance characteristics of systems represented by the state equations.
170 System Analysis Methods.

5.1 Linearisation of Non-linear Systems.

Since there is no general theory for the solution of nonlinear equations one must usually rely
on computer simulation to solve the governing equations. However, a computer simulation is
for one specific case only, and does not allow us to generalise with any certainty about the
nature of solutions for other cases. Moreover, if we wish to apply one of the most powerful
analytical tools available, superposition, our system must be linear. That is, in order to utilise
analytical techniques which in the majority of cases are based on linear theory, we need to
approximate the nonlinear system by linear versions using linearisation techniques.

The fundamental goal of linearisation is quite simple. The linear differential equation that is
chosen should represent the system more accurately, over the range of operation under con-
sideration, than any other linear differential equation. In most cases it is not possible to write
a linearised equation that will adequately represent the system over all possible conditions of
operation. As a compromise, we therefore determine the normal operating condition for the
situation under study and write the linearised equations to be accurate for small deviations
around that condition.

To consider possible methods for linearly approximating a non-linear equation, several pos-
sibilities exist. The type of linearisation method which is widely used, is the so-called small-
signal method. The small-signal linearization method presents a general method of lineariza-
tion which is applicable to all types of nonlinearities except those with infinite derivatives. In
this method the local tangent to the non-linear curve is used as a linear approximation about a
chosen operation point. Such approximation can be applied at several different points over an
operating range to explore the effect on system characteristics.

A mathematical formulation of this concept is readily available from the Taylor series expan-
sion of a function Y = g  X  about a chosen operating point X 0

2
2
 X – X0 
Y = g  X 0  + dg  X – X0  + d g ---------------------- + (5.1)
dX X0 dX x0 2!

when the existence of the partial derivatives is assumed. The term d means that the deriv-
dX i
ative is evaluated at the operating point i = X 0 . In practice, the operating point is usually
defined in terms of nominal or steady state values of the process variables.

For a linear approximation, we retain only the first two terms of the series, and which are eas-
ily seen to give the equation of the tangent line at X 0 . Thus, the linearised relation for the
non-linear function takes the form
 Linearisation of Non-linear Systems. 171

Y = g  X 0  + dg  X – X0  (5.2)
dX X0

which is an approximation to Eq. 5.1 and is linear in X . The above linearisation technique is
depicted graphically in Fig. 5.1.

gX
y

g  X0  gX
x

X0 X

Fig. 5.1 Linearization via Taylor series approximation

The linearization is computed around an operating point  X 0 g  X 0   and the choice of differ-
ent operating points will lead to different approximation results.

The linearised version of the nonlinear function, the tangent here, is only valid for limited
variations about the operating point X 0 . The larger the excursion from the point X 0 , the
larger the error introduced by the linearisation.

If we use  in front of the letters to indicate the variation of the capital-letter parameters
from the point of interest or the reference, we can write from Eq. 5.2 that

y = bx (5.3)

where

y = Y – g  X i 
x = X – X i
g  X  -
b = d----------------
dX i

It is then advantageous to write the linearised equations in terms of deviation of the depend-
ent variable rather than in terms of the dependent variable itself.

When dealing with a multi-variable non-linear function, the linearisation procedure is con-
ducted in a similar manner. For example the function
172 System Analysis Methods.

Y = g  X 1 X 2   X n  (5.4)

has a linear approximation at the point  X 1 i X 2 i  X n i  given as follows

Y = g  X 1 i X 2 i  X n i  + g  X 1 – X 1 i  + g  X 2 – X 2 i  +
 X1 i  X2 i
(5.5)
 + g  X n – X n i 
 Xn i

where all the partial derivatives are evaluated at the operating point i =  X 1 i X 2 i  X n i 
so they are just numbers, and Y becomes a linear function of X 1 i X 2 i  X n i in Eq. 5.4.

If we as for the single variable function introduce  to indicate the variation of the parame-
ters from the operating point, we can then write from Eq. 5.5

y = Y – g  X 1 i X 2 i  X n i 
x 1 = X 1 – X 1 i
x 2 = X 2 – X 2 i (5.6)

x n = X n – X n i

and the linearised version of the function can then be written as

y = a 1 x 1 + a 2 x 2 +  + a n x n (5.7)

where

a 1 = g a 2 = g ... a n = g
 X1 i
 X2 i
 Xn i

We are now able to sum up by presenting the liearised version of a nonlinear system of dif-
ferential equations.

Let us start with a first order model of the form

dX
= g  X  U t  (5.8)
dt
 Linearisation of Non-linear Systems. 173

Suppose that X 0 is an equilibrium state corresponding to the constant input U 0 satisfying the
condition

0 = g  X 0 U 0 t  (5.9)

then new variables X and U can be defined by

X = X – X 0
(5.10)
U = U – U 0

Then from our previous development of linear approximation, we can write our linear state
model in the following form

g 1 g 1 g g 1 g 1 g
 1  1
 x1  x2  xn  u1  u2  un
x· 1 x 1 u 1
g 2 g 2 g 2 g 2 g 2 g 2
x· x 2 u 2
1 = x x  x + u  u2

 un (5.11)
1 2 n 1
: : :
: :  : : :  :
x· n x n u n
g n g n g n g n g n g n
 
 x1  x2  xn  u1  u2  un

or in the form

x· = Ax + Bu (5.12)

where the matrices A and B are referred to as the Jacobian matrices.

Example

Let us demonstrate the technique proposed by finding a linear approximation for the pressure
P in the thermodynamic ideal gas equation of state

PV = mRT (5.13)

where P is the pressure, V is the volume, R is the gas constant and T is the absolute tem-
perature. The reference conditions are P 0 , V 0 and T 0 , and with the temperature and volume
as independent variables we can write
174 System Analysis Methods.

P = P  T V  (5.14)

Application of Eq. 5.7 to obtain the variation of the pressure P from its reference value
yields

 
P = P  T V  T + P  T V  V (5.15)
T V V T

The partial derivatives are evaluated from the equation of state, Eq. 5.13 as follows

   mRT
------------ mR
P  T V  = = mR
-------- = -------- = a (5.16)
T V0 T V  V0 V V0 V0

and

 P  T V    mRT
------------
mRT mRT 0
= = – -----------
- = – -------------
- = b (5.17)
V T0 V V  T0 V2 T0 V 02

The linearised approximation for P is then

P = a  T + b  V (5.18)

and where the actual value of P will be dependent on the reference state P 0 and T 0 .

5.2 Solution of Linear State Equations.

The solution of ordinary differential equations is a well-established practice in mathematics.
In spite of this fact, we will repeat it here and we will attempt to interpret solutions in physi-
cal terms. The ideas of state variable analysis will be presented first in the context of a single
first-order and second-order system for simplicity. Thereafter, general n-th order systems will
be considered. The reason for treating first and second-order systems separately is that their
analysis form a basis for the understanding of the analysis.

The state equation for a first-order linear system with a forcing input may be written in stand-
ard form as

x· = ax + bu with x = x0 at t = 0 (5.19)
 Natural Response of First-order Systems 175

where x and u are the state and input variables, and a and b are constants. Eq. 5.19 arises in
many applications and serves as a building block for more complex equations. To obtain the
general solution of Eq. 5.19, we make use of the properties of linear systems, the superposi-
tion principle. That is, the solution is considered to consist of the sum of two components,
namely

x  t  = xc  t  + xp  t  (5.20)

The component x c  t  is called the complementary function, which is the solution of the
homogeneous equation, that is the solution to Eq. 5.19 when the external disturbance u  t  is
equal to zero. The resulting intrinsic or unforced behaviour of the system is also called the
free or natural response. The term free or natural response emphasises the fact that the
response is determined by the nature of the system itself and not by external sources of exita-
tion. The other component part of the solution x p  t  is called the particular integral, which is
the solution introduced by the independent input variable u  t  . The behaviour, which the
particular integral represents, is also called the forced response.

For a particular system, either the natural or the forced response may be most important. For
some systems both are important. This is determined entirely by the objective that the system
must acomplish.

That the solution consists of two parts indicates that in predicting the behaviour of physical
systems, we must take into account two different sources of energy. The behaviour due to
internal energy storage is referred to as the natural response of the system; the behaviour
determined by an external energy source we call the forced response. The term natural
response emphasises the fact that the efforts and flows are determined by the nature of the
system itself and not by external sources of excitation.

5.3 Natural Response of First-order Systems

First, we will determine the natural response by setting the forcing u  t  equal to zero. The
differential equation for x  t  is then said to be homogeneous and written as

x· = ax with x = x0 at t = 0 (5.21)

From an inspection of the homogeneous equation we are lead to search for a general solution
wherein x  t  and its derivative must simply be a constant times each other. This implies a
function whose derivative has the same form as the function itself. One function which has
176 System Analysis Methods.

this property is the exponential function, and it is therefore natural to assume a solution of the
form

t
x = Ae (5.22)

where A and  are arbitrary constants. In order to determine the value of the constants A
and  the assumed solution is substituted into Eq. 5.21 to obtain

t
Ae   – a  = 0 (5.23)

t
There are two ways this last equation can be satisfied. One way is for A to be zero since e
cannot be zero. If this is the case, then we do not have any dynamics at all and therefore an
uninterested trivial solution. Then, assuming A not being equal to zero, the algebraic expres-
sion

–a = 0 (5.24)

must be true, and is known as the characteristic equation of the system. As will be shown
later the characteristic equation plays an important part in the study of linear system
response.

Thus, with  = a from Eq. 5.24, the free natural or complementary solution to Eq. 5.21
takes the form

at
x = Ae (5.25)

In order to specify a unique solution, the value of the constant A must be pinned down. The
constant A can be determined from our knowledge that the solution must satisfy, that is the
initial condition that x = x 0 at t = 0 . Hence this yields A = x 0 . Thus substituting this
value for A into Eq. 5.25 gives a complete free or natural solution of our original differential
equations as

at
x = x0 e (5.26)

The natural solution given by Eq. 5.26 represents an exponential growth with a being posi-
tive, and an exponential decay with a being negative. The response x  t  is shown graphi-
 Natural Response of First-order Systems 177

cally in Fig. 5.2 for a  0 and a  0 respectively.

x t

x0 a0

a0

Fig. 5.2 First-order system response

This shows that a linear first order system must have a negative value for a , if it is to be what
we call stable. With the term stable we mean that when disturbed slightly and released, the
system will return to its initial equilibrium state. If a , being positive, the system is said to be
unstable.

If we look at the exponential part of Eq. 5.26, it indicates that the constant a must have the
unit of 1/time in order for the exponent of e to be non-dimensional. It is common practice to
refer to the reciprocal of a , as the time constant and is most often denoted by the Greek letter
 (tau); thus

1
 = time constant = --- (5.27)
a

If we use the concept of the time constant, we can write Eq. 5.26 as

t
– --

x  t  = x0 e t0 (5.28)

Since the time constant is an important parameter, it is worth mentioning several of its char-
acteristics. First of all, it gives a measure of how fast a system responds to a disturbance. The
smaller the time constant the faster the response, and it is also the time at which the response
has undergone 63,2% of its total change. It is of some importance to know that the value of
five time constants is often used as the time at which the variation is essentially over since it
is less than 1% of its final value.

The natural response goes to zero as time increases and it is therefore, also referred to as the
transient response of the system.
178 System Analysis Methods.

Example

Let us now look at the two physical systems shown in Fig. 5.3.

R
2
i(t) C R
3 1
Sf 0 C

(a) (b) (c)

Fig. 5.3 First-order system examples

In part a is shown a tank with a valve in the output pipe. We assume that the valve has the
following characteristic

V· out = K P T (5.29)

where V· out is the volume flow through the valve, K is a valve constant and P T is the pres-
sure difference over the valve.

In part b is shown an electric circuit with a current source, a capacitor and a resistance in par-
allel. The two systems have the same bond graph which is shown in part c.

The state equation for the electric circuit becomes

q1
q· 1 = S f 3 – -----------
- (5.30)
C1 R2

The homogenous equation, after S f 3 is set to zero, takes the form

q1
q· 1 = – -----------
- (5.31)
C1 R2

This equation has the same form as Eq. 5.21 and direct comparison shows that

1
a = ------------ (5.32)
C1 R2
 Natural Response of First-order Systems 179

and this means that the time constant for the circuit becomes

1
 = --- = C 1 R 2 (5.33)
a

which states that the time constant is proportional to the capacitance and the resistance.

Returning to the hydraulic tank and the state equation in this case becomes

q
q· 1  f 1 = S f 3 – K -----1- (5.34)
C1

We now have a nonlinear state equation and we need to linarise the homogenous equation

q
q· 1 = f 3 = – K -----1- (5.35)
C1

Linearizing yields

K
f 3 = – ------------------- q 1 (5.36)
2 C1 q1 i

and the linearised state equation becomes

K
q· 1 = – ------------------- q 1 (5.37)
2 C1 q1 i

In this case then the time constant is

2 C 1 q 1 i
 = ---------------------- (5.38)
K

One important point to discover in Eq. 5.38 is that the time constant now is dependent on the
operating point q 1 i . This is a characteristic feature of nonlinear systems. Think about this!
180 System Analysis Methods.

5.4 Natural Response of Second-order Systems

After we now have investigated the response of systems which can be modelled by a single
state equation, it is of interest to investigate the response of systems containing two inde-
pendent storage elements, and hence, described by two state equations. The reason for treat-
ing this case before we go on to consider the solution of n-dimensional linear systems is that
the solution of second-order systems is a basis for a general understanding of high order sys-
tem behaviour.

As an example, let us use the systems shown in Fig. 5.4, which have the same bond graph.

Se:F(t)
F t
R L

m
x· x V
C 1 I:m
k
b

:C

R:
1 /k

b
Fig. 5.4 Second-order system example

The two state equations describing the system behaviour take the form

q p
p· = Se – ---- – R ---
C I
(5.39)
p
q· = ---
I

or in matrix form

p· = – R  I – 1  C p + 1 (5.40)
Se
q· 1I 0 q 0

For the homogeneous solution, that is with Se = 0 we do as for the case with a first-order
equation, assume that the solution is of exponential form. That is, to assume that the displace-
ment or charge and momentum or flux linkage will be of the form

t t
p = Pe and q = Qe (5.41)

where P , Q and  are unknown constants.

 Natural Response of Second-order Systems 181

If the assumed solution really is a solution, it must satisfy Eq. 5.39 for all values of t. When
we substitute the assumed solution into Eq. 5.39 we generate the following expressions

t R t Q t
Pe = – --- Pe – ---- e
I C
(5.42)
t P t
Qe = --- e
I

t
Eliminating e from Eq. 5.42 we can write the following matrix expression

R 1
 + --- ----
I C P = 0 (5.43)
1 Q
– --- 
I

The existence of non zero solutions to a set of linear homogeneous equations is one of the
most fundamental results of linear algebra. This result can be stated in the formal lemma.

Lemma:

Let A be a n  n matrix. Then the homogeneous equation

Ax = 0 (5.44)

has a non-zero solution vector whose components are all non-zero

if and only if the matrix A is singular, which means that the determinant of the coeffi-
cients is equatl to zero.

This then means that in our case the determinant of the A -matrix must be equal to zero,
which is

R 1-
 + --- ---
I C = 0 (5.45)
1
– --- 
I

or in equation form this means that

2 R 1
 + ---  + ------ = 0 (5.46)
I IC
182 System Analysis Methods.

The determinant in Eq. 5.45 can be written as

I – A = 0 (5.47)

where I is the unit or identity matrix. This means that Eq. 5.46 can be determined directly by
substracting the A -matrix in Eq. 5.40 from the I -matrix.

Eq. 5.46 is the system characteristic equation and as the name implies, it determines the char-
acteristics of the system behaviour. The roots of this quadratic equation will then determine
the mathematical character of the response. Therefore, the first step in finding the natural
response is to determine the roots of the characteristic equation.

The two roots of Eq. 5.46 are

R 2 1
 1  2 = – -----   ----- – ------
R
(5.48)
2I  2I IC

and we call each root a characteristic value or eigenvalue from the German word eigenwert
meaning characteristic value.

If either root is substituted into Eq. 5.41, the assumed solutions will satisfy the given differ-
ential equations, Eq. 5.39, and it can be shown that their sum is also a solution of the form

1 t 2 t
p = P1 e + P2 e
(5.49)
1 t 2 t
q = Q1 e + Q2 e

The roots of the characteristic equation are determined by the system parameters, and the ini-
tial conditions determines the value of the constants P 1 , P 2 , Q 1 and Q 2 .

The roots  1 and  2 of the characteristic equation Eq. 5.46 are determined by the system
parameters R, I and C, and the nature of the roots depends on the radical, i.e. the expression
inside the square root sign. Clearly there are two principal cases: when the radical is real and
when it is imaginary, separated by the borderline situation when the radical is zero or
 R  2I  2 just equals 1  IC .

It will be instructive to interpret the physical significance of the solution corresponding to the
three distinct cases. If we say that the borderline value of resistance R is a critical value R c ,
the amount of damping in any system can then be specified in terms of a non-dimensional
ratio
 Natural Response of Second-order Systems 183

R actual resistance
 = ----- = ---------------------------------------- (5.50)
Rc critical resitance

referred to as the damping ratio.

By defining the undamped circular natural frequency  n , as

1
 n = ---------- (5.51)
IC

the critical resistance takes the form

R c = 2 ---I- = 2I n (5.52)

C

and the damping ratio becomes

R
 = ------------ (5.53)
2I n

The characteristic equation can now be written in the following form

2 2
 + 2 n  +  n = 0 (5.54)

The roots of the characteristic equation in the form of Eq. 5.54 can then be written as

2
 1  2 = –  n   n  – 1 (5.55)

The nature of the roots  1 and  2 will depend on the values of the damping ratio. There are
four possible outcomes:

 = 0  1  2 =  j n ,j = –1 (5.56)

2
01  1  2 = –  n  j n 1 –  (5.57)

 = 1  1  2 = –  n (5.58)
184 System Analysis Methods.

2
1  1  2 = –  n   n  – 1 (5.59)

We will now discuss each of these cases in some detail.

No damping (  = 0 ):

In the case of no damping the natural solution becomes

j n t – j n t
p = P1 e + P2 e
(5.60)
j n t – j n t
q = Q1 e + Q2 e

It is, however, more useful to interpret the solution if it is expressed in real form. To do so we
start with expressing p in real form. Using the Euler identity

 j n t
e = cos   n t   j sin   n t  (5.61)

then the expression for p , can be written as

p =  P 1 + P 2  cos   n t  + j  P 1 – P 2  sin   n t  (5.62)

Remember that this is a physical system, which means that each term of Eq. 5.62 must be
real. This imposes limitations on the values of P 1 and P 2 , which require that

P 1 + P 2 = real number
(5.63)
j  P 1 – P 2  = real number

To satisfy simultaneously Eq. 5.63 the necessary condition is that P 1 and P 2 must be com-
plex conjugate numbers.

If we let

A 1 + jB 1 A 1 – jB 1
P 1 = -------------------
- and P 2 = -------------------
- (5.64)
2 2

then Eq. 5.62 reduces to

p = A 1 cos   n t  + B 1 sin   n t  (5.65)

 Natural Response of Second-order Systems 185

which is a real function, and likewise

q = A 2 cos   n t  + B 2 sin   n t  (5.66)

The solution as given in Eq. 5.65 and Eq. 5.66 can also be written in the following form

p  t  = A p cos   n t –  p  (5.67)

and

q  t  = A q cos   n t –  q  (5.68)

where

– 1B
Ap = A1 + B1 and  p = tan -----1 (5.69)
A1

and

–1B
Aq = A2 + B2 and  q = tan -----2 (5.70)
A2

The constants A p , A q and  p ,  q are referred to as the amplitude and phase angle respec-
tively.

Thus, the natural response in the case of no damping is a sine wave of constant amplitude and
a frequency equal to the undamped natural frequency  n , as shown in Fig. 5.5.

-1

-2
0 2 4 6 8

Fig. 5.5 Natural response with no damping (  = 0 )

The oscillatory behaviour is possible because of the presence of two types of energy storage
186 System Analysis Methods.

elements, I and C, in the system.

The solution given by Eq. 5.65 and Eq. 5.66 contains four arbitrary constants A 1 A 2 B 1 and
B 2 which have to be determined. To specify a unique solution, two auxiliary conditions must
be given for each state variable. That is, we need to know the initial states usually at the time
equal zero. Given the initial states to be

p  t = 0  = p0 and q  t = 0  = q0 (5.71)

we obtain by setting Eq. 5.65 and Eq. 5.66 equal to p 0 and q 0 respectively, that

A1 = p0 and A2 = q0 (5.72)

In order to determine the two remaining constants B 1 and B 2 we differentiate the solution
equations Eq. 5.65 and Eq. 5.66, and equate them to the two state equations given by Eq. 5.39
at the time equal to zero. This yields

q
p· =  n B 1 = – ----0- (5.73)
C

and

p
q· =  n B 2 = ----0- (5.74)
I

Now solving for B 1 and B 2 we obtain

q0 p0
B 1 = – ---------
- and B 2 = -------
- (5.75)
n C n I

When the expressions for A 1 A 2 B 1 and B 2 are inserted into Eq. 5.65 and Eq. 5.66 respec-
tively, yields the following complete solution for p and q becomes

q0
p = p 0 cos   n t  – ---------
- sin   n t  (5.76)
n C

and
 Natural Response of Second-order Systems 187

p0
q = q 0 cos   n t  + -------
- sin   n t  (5.77)
n I

In Fig. 5.6 is shown how the time-series for q varies depending on the values for q 0 and p 0
respectively.

4
p0=0
Data:
p0=5
p0=-2.5 C=0.1
2 I =1
R=0
0
q0 = 1
q [-]

p 0 = – 2.5,0, 5

-2

-4
0 1 2 3 4 5 6 7

Fig. 5.6 Undamped response with varying initial conditions.

Underdamped ( 0    1 ):

For the case with the damping ratio greater than zero and less than one, the two roots  1 and
 2 of the characteristic equation are found to be complex conjugates of each other, that is

2
 1  2 = –  n  j n 1 –  (5.78)

In this situation, the response is said to be underdamped. When we are discussing the under-
damped response, it is convenient to express the roots  1 and  2 as

2
 1  2 = –   j d where d = n 1 –  (5.79)

is known as the damped natural frequency. The reason for this terminology is that whenever
there is a dissipative R-element in the system, the frequency of oscillation  d , is less than
 n . Thus, when the damping is non zero, the frequency of oscillation is said to be damped.

The underdamped response can then be written as

188 System Analysis Methods.

 – t + j d t   – t – j d t 
p  t  = P1 e + P2 e
(5.80)
 – t + j d t   – t – j d t 
q  t  = Q1 e + Q2 e

The exponential time function may be expressed as the product of two exponentials as

 – t   d t  – t  j d t
e = e e (5.81)

and Eq. 5.80 can then can be written as

– t j d t – j d t
pt = e  P1 e + P2 e 
(5.82)
– t j d t – j d t
qt = e  Q1 e + Q2 e 

From our development for the case of no damping the solution written in Eq. 5.82 can be
expressed alternatively as

– n t
pt = e A p cos   d t –  p  (5.83)

and

– n t
qt = e A q cos   d t –  q  (5.84)

The response given by Eq. 5.83 and Eq. 5.84 are called exponentially damped sinusoid. They
consists of a sine wave of frequency  d , and with a magnitude that is decreasing exponen-
tially with time. It is not strictly periodic, since the factor multiplying the trigonometric terms
is continuously decreasing. However, there are regularly spaced passages through the equi-
librium for p and q as given by Eq. 5.83 and Eq. 5.84. Hence we can speak of the pseudo
period

2
 p = ------ (5.85)
d
 Natural Response of Second-order Systems 189

as shown in Fig. 5.7. In the case of the damped sinusoid the time constant is defined in terms

2
–  n t
2X e
1
Maximum point

1 Tangent point

0
q [-]

-1

  
------- ------- -------
  
d d d
-2
0 1 2 3 4 5 6 7 8

Fig. 5.7 Exponentially damped sinusoid

–  n t
of the envelope Ae . Thus the time constant now becomes

1
 = --------- (5.86)
 n

In order to determine the constants A 1 A 2 B 1 and B 2 , we use the same procedure as we did
for the no damping case (  = 0 ), and obtain

A1 = p0
A2 = q0
 
B 1 =   -----n- – 2 -----n- p 0 – ---------- q 0
1 (5.87)
 d  d d C
1 
B 2 = -------- p 0 +  -----n- q 0
I d d

Critical damping (  = 1 ):

For the case of  equal to unity we have critical damping, which is the value of damping at
which the response loses its oscillatory character and the roots are real and equal, that is

 1  2 = –  n (5.88)

Since there are multiple roots, the solution then takes the form
190 System Analysis Methods.

–  n t
p = e  A1 + B1 t  (5.89)

and

–  n t
q = e  A2 + B2 t  (5.90)

which again represents an exponentially decaying response.

For this case with  = 1 , the four constants A 1 A 2 B 1 and B 2 in the solution becomes

A1 = p0
A2 = q0
1 (5.91)
B 1 = –  n p 0 – ---- q 0
C
1
B 2 = --- p 0 +  n q 0
I

Response curves for different initial conditions are shown in Fig. 5.8

1.4 p0= 0
p0= 5
p0=-5

0.6

0.2

-0.2
0 1 2 3

Fig. 5.8 Natural response with critical damping

A system with critical damping has the smallest damping possible for aperiodic motion, and
hence returns to rest in the shortest time without oscillation.

Overdamped (   1 ):

For the case when   1 , there is a relatively large amount of damping and naturally enough,
the system is said to be overdamped. The two roots of the characteristic equation in this case
 Natural Response of Second-order Systems 191

are real and unequal

2
 1  2 = –  n   n  – 1 (5.92)

and the general solution becomes

1 t 2 t
p = P1 e + P2 e
(5.93)
1 t 2 t
q = Q1 e + Q2 e

Often, however, it is written differently by making use of the identity that relates the expo-
nential with the hyperbolic forms. That is

x
e = cosh  x   sinh  x  (5.94)

Using this form Eq. 5.93 may be expanded to form

– n t
p = e   A 1 + B 1  cosh   d t  +  A 1 – B 1  sinh   d t   (5.95)

and

– n t
q = e   A 2 + B 2  cosh   d t  +  A 2 – B 2  sinh   d t   (5.96)

Since A 1 A 2 B 1 and B 2 are arbitrary constants, their sum and difference are also arbitrary.
Hence we can write this in the form

– n t
p = e  A p cosh   d t  + B p sinh   d t   (5.97)

and

– n t
q = e  A q cosh   d t  + B q sinh   d t   (5.98)

The response in this case becomes the sum of two decaying exponentials and the resulting
nonoscillary response is called aperiodic. After following the procedure to determine these
new constants as done for the former cases we obtain
192 System Analysis Methods.

–  n t 
p 0 cosh   d t  +  – -----n- p 0 – ---- q 0 sinh   d t 
1
p = e (5.99)
 d C 

and

–  n t 
q 0 cosh   d t  +  -------- p 0 +  -----n- q 0 sinh   d t 
1
q = e (5.100)
 I d d 

Typical response curves for different initial values are given in Fig. 5.9.

1.4 p0= 0
p0= 1
p0=-1

0.6

0.2

-0.2
0 1 2 3

Fig. 5.9 Overdamped response

Clearly, there is no essential difference in the character of the response in the overdamped
and critically damped case.

5.5 Forced Response

We have previously discussed the homogenious solution or natural response of first and sec-
ond order system models. In those system models we assumed that energy had been stored
initially in C and I elements, and the response was caused bye the release of the energy to
other elements. We now discuss the problem of finding the response generated by input sig-
nals from sources. That is, we will determine what is known as the particular solution or
forced response. We will not treat general forcing functions but will treat two input functions
which are frequently used to investigate system characteristics. The first case is to find the
response generated when a source is suddenly applied to the system. The second case is
determination of the steady-state behaviour of the system models when they are energised by
sinusoidal sources.
 Forced Response 193

5.5.1 Step Function and Sinusoidal Sources

An instantaneous change in a source or input variable is referred to as a step function, and is
expressed as

S = As us  t  (5.101)

where A s is the value of the function and u s  t  is the symbol for a unit step function where
the height is equal to one and starting from t = t 0 . In Fig. 5.10 is shown the step function as
a function of time.

As

0 t0 t

Fig. 5.10 Step function

A sinusoidal source produces an effort or flow that varies sinusoidally with time. We can
express a sinusoidally varying function using either the sine function or the cosine function.
There is no clear-cut choice for either function, so we will use the sine function throughout
our discussion. Using the sine function, we can write a sinusoidally varying source as fol-
lows:

S = A m sin  t –   (5.102)

To facilitate our discussion of the parameters in Eq. 5.102, we show the plot in Fig. 5.11 of
194 System Analysis Methods.

amplitude versus time

Am

Am

 t
----


Am
T

Fig. 5.11 A sinusoidal signal

The sinusoidal function is repetitive and is therefore called periodic, and the length of time it
takes the function to pass through all of its possible values is referred to as the period of the
function and is denoted by T . The reciprocal of T gives the number of cycles per second, or
frequency and is denoted by f , thus

1
f = --- (5.103)
T

A cycle per second is referred to as Hertz and is abbreviated Hz . Omega, denoted  , is used
to represent the angular frequency of the function, thus

2
 = 2f = ------ (5.104)
T

The phase angle in Eq. 5.102 is the angle  which determines the value of the function at
t = 0 . It should be obvious that t and  must carry the same units since they are added
together in Eq. 5.102. Since t is expressed in radians,  must also be, but is normally
given in degrees so it must be converted before added to t .

It should be apparent at this point that the sinusoidal signal is completely specified once its
maximum amplitude, frequency and phase angle are given.
 Step Response of a First-order system. 195

5.6 Step Response of a First-order system.

The starting point is the first-order state equation in the standard form as given in Eq. 5.19 as

x· = ax + bu with x = x0 at t = 0 (5.105)

with a step input bu = bu s = b and remembering that the time constant  = 1  a , Eq.
5.105 can be written in the form

dx 1
------ = – --- x + b (5.106)
dt 

since a  0 for a stable system.

Since the excitation is a constant, the final value of the unknown x , will be constant and the
final value must satisfy Eq. 5.106. When x has reached its final value, the derivative dx  dt
must be zero and we can conclude that

x f = b (5.107)

where x f represents the final value of the variable x .

We can solve Eq. 5.106 by separating the variables and start with multiplying both sides of
Eq. 5.106 by dt which allows us to write

------------ = – 1--- dt
dx
(5.108)
x – xf 

Integrating both sides of Eq. 5.108 gives us

–t  
x  t  = x f +  x  0  – x f e (5.109)

In the case when the initial state x  0  is zero, Eq. 5.109 reduces to

–t  
x  t  = xf  1 – e  (5.110)

From Eq. 5.110 we find that after  seconds the response is approximately 63% of the final
value x f and after 4 the response is approximately 98% of the final value. This shows that
196 System Analysis Methods.

the time constant is a measure of the systems’s speed of response.

The step response of the first-order system is shown graphically in Fig. 5.12

xt
x
f

0.63 x f

 2 3 4 5 6 7 t

Fig. 5.12 Step response of first-order system.

The importance of Eq. 5.109 is more apparent if we write it out in verbal form

The unknown The final  the initial the final  t

  – -------------------------------
time constant
-
variable as a = value of the  value of the – value of the
+   e
 
function of time variable  variable variable 

Fig. 5.12 shows us that after the application of the excitation the state increases exponentially
from zero to a final value x f . The rate of increase is determined by the time constant  .

Returning to Eq. 5.109, it shows that the problem of finding the step response of a first-order
system involve the computation of three quantities:

1. The initial value of the variable

2. The final value of the variable
3. The time constant
 Step Response of a Second-order System 197

5.7 Step Response of a Second-order System

In finding the step response of second-order systems, we will return to Chap. 5.4 and the sys-
tems shown in Fig. 5.4. The state equations describing the behaviour of these systems are
given by Eq. 5.39 and repeated here for convenience as

q p
p· = Se – ---- – R ---
C I
(5.111)
p
q· = ---
I

The excitation source Se in Eq. 5.111 is a step function and can then be expressed as

Se = A s u s  t  (5.112)

where A s is the height or the amplitude of the function and u  t  is the symbol for a unit step
function whose height is 1.

To find the solution for the two state variables, let us start with differentiating the first equa-
tion in Eq. 5.111 with respect to time to obtain

q· R
p·· = 0 – ---- – --- p· (5.113)
C I

Then by substituting for q· in Eq. 5.113 from Eq. 5.111 and rearranging, we obtain the fol-
lowing second-order homogeneous differential equation.

R 1
p·· + --- p· + ------ p = 0 (5.114)
I IC

In searching for a solution of this second-order equation, it is natural to assume the same
solution as given in Eq. 5.41

t
p = Pe (5.115)

Substituting this assumed solution back into Eq. 5.114 we obtain the following characteristic
equation

2 R 1
 + ---  + ------ = 0 (5.116)
I IC
198 System Analysis Methods.

which is, as expected, the same equation as given by Eq. 5.46.

As we discussed in section 5.4, the solution for p will depend on the roots of the characteris-
tic equation and given by Eq. 5.47. The response will be underdamped, critically damped or
overdamped according to whether   1 ,  = 1 or   1 respectively. Thus, the three possi-
ble solutions are:

Underdamped: (   1 )

–  n t
pt = e  A 1 sin   d t  + A 2 cos   d t   (5.117)

or

– n t
p  t  = Ae sin   d t –   (5.118)

Critically damped: (  = 1 )

– n t
pt = e  B1 t + B2  (5.119)

Overdamped: (   1 )

1 t 2 t
p  t  = D1 e + D2 e (5.120)

The two constants in each of the equations above must be determined from the initial condi-
tions p  0  and q  0  .

In Fig. 5.13 is shown the response p  t  to a step change in the input for various values of the
 Step Response of a Second-order System 199

damping ratio  .

1
=0.1
0.8 =0.25
=0.5
0.6 =0.707
=0.8
=0.9
0.4
=1

0.2

-0.2

-0.4

-0.6

-0.8
0 2 4 6 8 10 12

Fig. 5.13 Response of p  t 

By setting the derivative of p  t  with respect to time equal to zero, it can be shown that the
maximum overshoot occurs at

–1
cos 
t m = ------------------------- (5.121)
2
n 1 – 

The corresponding maximum value of p  t  is

–1
cos 
– ------------------
2
1–
p  tm  = n e (5.122)

Solving for q we start with differentiating Eq. 5.111 to obtain

p·
q·· = --- (5.123)
I

Substituting the first equation of Eq. 5.111 into the right hand side of Eq. 5.123 and rearrang-
ing yields the following non homogenous second-order differential equation

q
Iq·· + Rq· + ---- = A s u s  t  (5.124)
C

This differential equation is the same one we obtain when we use Newton’s second law
directly
200 System Analysis Methods.

Force = mass  acceleration

or Kirchoff’s voltage law:

the algebraic sum of all the voltages around any closed path in a cir-
cuit equals to zero

to set up the differential equation directly.

As discussed in section 5.2, the complete solution of Eq. 5.124 consists of two parts namely

q = complementary function + particular integral

Solving for the complementary function requires first writing of the homogeneous equation

q
Iq·· + Rq· + ---- = 0 (5.125)
C

which is formed from Eq. 5.124 with the right hand side equal to zero.

The characteristic equation associated with Eq. 5.125 is as should be expected equal to Eq.
5.116. Thus from the solution for p in Eq. 5.117, Eq. 5.119 and Eq. 5.120 we can write the
complementary function for q as

Underdamped: (   1 )

–  n t
qt = e  A 1 sin   d t  + A 2 cos   d t   (5.126)

or

–  n t
q  t  = Ae sin   d t –   (5.127)

Critically damped: (  = 1 )

– n t
qt = e  B1 t + B2  (5.128)

Overdamped: (   1 )
 Step Response of a Second-order System 201

1 t 2 t
q  t  = D1 e + D2 e (5.129)

As for the complementary function for p  t  the determination of the two constants must also
here be determined from the initial conditions p  0  and q  0  .

Having found the complementary function we must now find the particular integral of Eq.
5.124. Various procedures are available for doing this, some applicable no matter what exci-
tation may be, others useful only when excitation belongs to some suitably specialized
classes of functions. Using the method of undetermined coefficients to Eq. 5.124, it is natural
to guess the following solution

q = K (5.130)

where K is a constant since the right hand side of Eq. 5.124 is also a constant. Substituting
the guessed solution into the given equation Eq. 5.124, we obtain

K
---- = A s u s  t  (5.131)
C

and solving for K the required particular integral is

q = CA s u s  t  (5.132)

The solution of Eq. 5.124 is the sum of the complementary function and the particular inte-
gral. For the case of underdamped system the solution can be written as

–  n t
qt = e  A 1 sin   d t  + A 2 cos   d t   + CA s u s  t  (5.133)

The constants A 1 and A 2 are chosen to satisfy the initial conditions corresponding to a sys-
tem at rest initially, i.e. q  0  = 0 and q·  0  = 0 , obtaining

q  0  = 0  A 1 + 1  A 2 + CA s u s  t  = 0
(5.134)
q·  0  =  d  A 1 –  n  A 2 = 0

which may be solved using Cramer’s rule to give

202 System Analysis Methods.

–  –
A 1 = ------------n CA s u s  t  = ------------------ CA s u s  t 
d 1 – 2
(5.135)
–
A 2 = --------d- CA s u s  t  = – C A s u s  t 
d

The complete solution for this case can then be written as

–  n t  
q  t  = CA s u s  t   1 – e ------------------ sin   d t  + cos   d t   (5.136)
  1 – 2 

or as

–  n t
q  t  = CA s u s  t   1 – e cos   d t –    (5.137)

–1 
where the phase angle is  = tan  ------------------ .
 1 –  2

Similar expressions for the response can be derived following the same procedure as given
above for the no-damping, critically damping and over-damped cases.

In Fig. 5.14 is shown the response q  t  to a step change in the input for various values of the
damping ratio  , where the abscissa is the dimensionless time  n t . With dimensionless time
 n t , as abscissa, the curves are then a function only of the damping ratio  .

1.8
 = 0.1
1.6  = 0.25
 = 0.5
1.4  = 0.707
 = 0.9
 = 1.0
1.2
 = 2
 = 5
1  = 10

0.8

0.6

0.4

0.2

0
0 5 10 15

Fig. 5.14 Step response of second-order system

These curves shows that the amount of overshoot depends on the damping ratio  . For the
underdamped case   1 , the system oscillates around the final value. The oscillations
decrease with time, and the system response approaches the final value. The peak overshoot
 Forced Response of Sinusoidal Inputs 203

for the underdamped system is the first overshoot. For the underdamped and critically
damped case,   1 , there is no overshoot.

5.8 Forced Response of Sinusoidal Inputs

As we have previously stated any function can be applied as an input or forcing function to
our models, but in practice, sinusoidal functions plays a dominating role for system response
studies. This is partly because most functions can be broken down into sinusoidal compo-
nents using a Fourier series as representation, then each component can be studied separately
and the solutions added using the principle of superposition valid for any liner system.

For a first order linear system as shown in Fig. 5.3, the equation for the charge in the capaci-
tor is given as

q
q· = – -------- + S f (5.138)
CR

where S f now is assumed to be a sinusoidal function with adjustable forcing frequency  f .

Writing the forcing function S f as

– j f t
S f = I 0 cos   f t  = Re  I 0 e  (5.139)

where Re means the real value of ( ).

It can be shown that after the initial transients due to the free response and the initial condi-
tions has died away, we are left with a steady-state response which for linear systems will
have a frequency equal to the forcing frequency. Thus, assuming that the solution can be
expressed as

– j f t
q  t  = Re  Ae  (5.140)

and inserting this into Eq. 5.138 together with the expression for S f and at the same time sup-
pressing the Re -operator, this gives

j f t A –j t – j t
j f Ae = – -------- e f + I 0 e f (5.141)
RC
204 System Analysis Methods.

The only unknown here is A and solving for the constant A yields

I0 I 0 RC
A = ---------------------
- = ------------------------
- (5.142)
1 1 + jRC f
j f + --------
RC

This solution for A which is imaginary, means that the real amplitude and its phase angle is a
function of the forcing frequency  f . The meaning of having an imaginary value of the
amplitude is that it has a magnitude which is the length of the imaginary number, and a phase
angle which is the angle between the rotating forcing imaginary vector and the rotating
amplitude. Analysing this dependency of A shows that

A  CRI 0 when f  0

and

A0 when f  

By direct manipulation we can write the expressions for the phase-angle and magnitude of
the amplitude as

–1
 = – tan  RC f  (5.143)

and

RCI 0
A = -----------------------------------------
12
- (5.144)
 1 +  RC f  1 

Plotting A and  or the corresponding normalised variables as a function of  f on a log-

log scale results in the well known Bode plots. In Fig. 5.15 is shown the Bode plots express-
 Forced Response of Sinusoidal Inputs 205

ing this dependency, using Eq. 5.143 and Eq. 5.144.

Linear scale
45

0

-------------
RCI -45
0
-90

-135 Log scale

-3 -2 -1 0 1 2 3
10
-------- 10-
------- 10
-------- 10
-------- 10-
------- 10-
-------  10
--------
RC RC RC RC RC RC f RC

Log scale 0
10

A
-------------
RCI
0
-2
10

Log scale
-3 -2 -1 0 1 2 3
10 10 10 10 10 10  10
-------- -------- -------- -------- -------- -------- f --------
RC RC RC RC RC RC RC

Fig. 5.15 Bode plot for a first order system

Now turning to our second order example as shown in Fig. 5.4 the state equations for the
mechanical system are given as

q p
p· = Se – ---- – R ---
C I
(5.145)
p
q· = ---
I

Let us assume that the effort source represented by the Se-element of the bond graph can be
expressed as

j j t
Se = F 0 cos   f t  = Re  F 0 e  (5.146)

To find a forced solution we can again assume that the state variables p and q will also be
sinusoidal with the same frequency as the input and therefor can be expressed as

j j t
p  t  = Re  P 0 e 
(5.147)
j j t
q  t  = Re  Q 0 e 

Then, substituting Eq. 5.146 and Eq. 5.147 into Eq. 5.145 and neglecting the Re( ) operator
yields
206 System Analysis Methods.

Q R
j f P 0 + -----0- – --- P 0 = F 0
C I
(5.148)
P
j f Q 0 – -----0 = 0
I

This is now two equations for two variables making it possible to solve for the two variables
P 0 and Q 0 . Using Cramer’s rule this leads to

1
F 0 ----
C
0 j f j f F 0
P 0 = ------------------------------------- = --------------------------------------------
- (5.149)
1
-----
- 2
– f – j  2  f  R
---
 j – R --- ----
1
IC I
 f I C
1
– --- j f
I

and

R
j f – --- F 0
I
1--- F0
– 0 -----
I I
Q 0 = ------------------------------------- = --------------------------------------------- (5.150)
------ –  f – j  2 ---  f
1 2 R
 j – R --
-  --- 1
- IC  I
 f I C
1
– --- j f
I

Analysing the dependency of the displacement of the mass block Q 0 of the forced frequency
yields

Q0  F0 C when f  0

and

Q0  0 when 0  

and perhaps even more interesting

 Forced Response of Sinusoidal Inputs 207

F0 C 1
Q 0 = ------------------------
- when  f  ------ =  e
2 1 –  2 IC

where  e is the natural frequency of the system.

In Fig. 5.16 is shown the Bode plots expressing the forced response, i.e the amplitude and the
phase angle of this system.

 = 0
10
2  = 0.1
 = 0.25
Q  = 0.5
0-
----------
F C
0  = 0.707
0
10

-2
10
10
-1
10
0  10
1
f
-------

e

0
Q  = 0
0
 = 0.1
-50
 = 0.25
 = 0.5
-100  = 0.707

-150

-200
-1 0 1
10 10  10
f
-------

e

Fig. 5.16 Bode plots for a second order system

As you may observe, the amplitude is increasing when  f is close to the natural frequency of
the system. For system with very little damping, i.e.  is small, the amplitude goes to unlim-
ited when  f =  e . This phenomena is called resonance and can cause great damage to any
dynamic system.

For systems of higher order the procedure outlined here can be followed. This means that for
any state-space equations the procedure to apply goes as follows:

1. For linear systems having multiple inputs each input can be solved for symboli-
cally or computed individually and the responses simply summed. Take one input
and if it is sinusoidal represent it in complex exponential form as
j j t
u 1  t  = Re  U 1 e  (5.151)

2. Let each state variable be expressed as a complex exponential too, as

j j t
x 1  t  = Re  X 1 e  (5.152)
208 System Analysis Methods.

3. Insert the expressions for the input u 1  t  and all the state variables
x =  x 1 x 2  x n  to form the algebraic problem

 j j – a 11 X 1 +  – a 12 X 2 +  +  – a 1n X n = b 11 U 1
 – a 11 X 1 +  j f – a 22 X 2 +  +  – a 1n X n = b 21 U 2
(5.153)
:
 – a n1 X 1 +  – a 12 X 2 +  +  j f – a nn X n = b n1 U n

4. Use Cramer’s rule or computer software to solve for the complex expressions of
j f t
number for X 1 X 2  X n in Eq. 5.153. When multiplying each X i with e and
applying the Re() operator, the forced response of the system is found as a function
of the forcing frequency. Voila!

5.8.2 General comments

Up to now we have been able to develop closed expressions for the solution of our state equa-
tions. We have stressed that general not too complex methods exists for linear systems, but
the algebra for even second order systems are becoming hard to carry out by pen and paper,
and for higher order systems this is better left to the computer. Having formulated the state
equations either by giving the system as a bond graph or using the equation form, modern
simulation tools now gives this possibility. Some of them such as 20-sim even as symbolic
expressions.

When it comes to non-linear systems none of the methods discussed so far exists for solving
or analysing your model. This is why simulation is so important to analyse practical and
physical systems. Using simulation you can dig into the system and explore its properties just
as you do for linear systems.
 Numerical Solutions 209

Chapter 6

Numerical Solutions

We have in the preceding chapters been introduced to mathematical modelling of physical

systems by use of the bond graph method. We have seen that we normally end up with a sys-
tem of equations of the initial-value form. The task is now to use these mathematical equa-
tions to compute a solution to the equations. This solution process is often referred to as
digital simulation.

Relatively few practical differential equations can however, be solved in analytical or closed
form. More often, solutions must be obtained by numerical means. Simulation of dynamic
systems therefore, involves the numerical solution of differential equations, most often for-
mulated as an initial-value problem. The numerical algorithms or schemes that are used for
this purpose are called numerical integrators. There exists a large number of references on
numerical integrators such as [10], [11] and [12], and a wide range of methods and algo-
rithms are available as source-code, in numerical libraries, or included into software pack-
ages such as Matlab and others. These methods have different properties related to accuracy,
computational cost and type of equations they are suitable for.

Several widely used methods are described briefly in this chapter, since a comprehensive
treatment of these methods is beyond the scope and objectives of this text, we merely stat
some of the algorithms and comment briefly on their important characteristics.

An engineer faced with the necessity of solving differential equations numerically, may have
difficulty in deciding which of the various methods to use, since there are indeed many from
which to choose and the selection of which method to use depends largely on the properties
of the system to be simulated. In this chapter we will introduce the basic terminology which
is needed to understand what determines the properties of the different methods in order to
enable the user to select the best method for the problem at hand.

The solution to problems involving differential equations require that the dependent variable
and/or its derivatives be given at prescribed value of the independent variable time t. When-
ever these constraints are imposed at the zero value of the independent variable, the solution
task is said to be an initial-value problem. Constraints on initial-value problems are termed
210 Numerical Solutions

initial conditions.

The initial value problem can be stated in simple terms. Given the general first-order differ-
ential equation

y· = f  y u t  (6.1)

where y is the dependent variable or state variable, t is the independent variable and u is the
input variable, with the initial condition y  t 0  = y 0 . Find the unknown function y  t  that
satisfies both the differential equation and the initial condition.

6.1 Method of solution

Now, consider the solution of the first-order ordinary differential equation

dy
------ = f  t y  (6.2)
dt

with the initial condition

y  t0  = y 0 (6.3)

where y 0 is the value of y  t  at a given value of the independent variable t = t 0 . We seek a

solution in the range t 0  t  t 1 , where t 0 and t 1 are finite numbers. A numerical solution of
this differential equation involves obtaining the numerical values of y  t  at discrete values of
t , termed grid points. If h represents a for now assumed constant step size, the grid points
are defined by

ti = t0 + i  h where i = 0 1 2  (6.4)

The numerical points of the solution at these points may be denoted by y 0 y 1  y n  .

Therefore the numerical scheme must provide means to evaluate y i + 1 from the given or
computed solution at the preceding grid points.

6.1.1 Computational formula and interpretation of the method

 Method of solution 211

One very general method of attacking the solution of an ordinary differential equation is to
replace the differential equation

y· = f  y u t  y  t0  = y0 (6.5)

by the integral equation

ti + 1 ti + 1
t i
dy = y i + 1 – y i = t i
f  y t  dt (6.6)

or

ti + 1
yi + 1 = yi +  f  y t  dt (6.7)
ti

where  t i t i + 1  is the interval over which we wish to integrate.

This indicates that integration is an important mathematical operation in solving differential

equations. We would start of course, with the interval  t 0 t 1  , then if we could find y 1 we
would consider  t 1 y 1  as the new starting value in order to continue the process over
 t 1 t 2  . Our problem is what to do about the y that appears in the integral because we do not
know the values of y in the interval of integration. Much of the art of solving differential
equations numerically rests on how to approximate f  y u t  in the interval. In other words,
approximating the integral by some scheme based on known or intermediate variables lets us
explicitly compute new values for y i + 1 , where i = 0 1 2  .

If we assume that f is constant in the interval of integration, this yields when inserted into
Eq. 6.7 the computational formula for solving Eq. 6.1 using the Euler method as

y i + 1 = y i + h  f  t i y i  with i = 0 1 2  (6.8)

where h = t i + 1 – t i .

The formula can be utilised to explicitly obtain new values for y i for increasing t , starting
with y  t 0  . This implies that the method is explicit and self-starting.

Euler’s method is one of the simplest methods for solving the initial value problem. It is
therefore very seldom recommended since several more efficient methods are available. The
main reason for studying this method here is that it is simple and provide considerable insight
and understanding of many of the basic features of the numerical solution of ordinary differ-
212 Numerical Solutions

ential equations without the additional complexity of other methods.

In Fig. 6.1 a geometric interpretation of Euler’s method is given. The exact solution is
denoted by y  t  . Starting with the known value of y = y 0 at t = t 0 the tangent of the curve
can be computed at t = t 0 to have a slope of k 0 = f  t 0 y 0  . Thinking of the slope k 0 as an

y
k = fy  t 
0 0 0
y
1
k
0

y yt 
0 i+1

yt
h

t t t
0 1

Fig. 6.1 Eulers method

approximation to the exact curve in the interval t 0  t  t 1 , we can compute an approxima-

tion to the integral in Eq. 6.1 as

ti + 1 ti + 1 tn + 1
t i
f  y t  dt = t i
k 0 dt = k 0 t tn
= k0  tn + 1 – tn  = k0 h (6.9)

Inserting this expression into Eq. 6.7 yields the computational formula of Euler as stated in
Eq. 6.8.

There are other ways of interpreting Euler’s method or to develop computational methods. If
the function y  t  is assumed to be continuous near t i , it may be expanded in a Taylor series
as follows

2
h2
y i + 1 = y i + h  dy
------ + -----  d y where ti    ti + 1 (6.10)
dt ti 2 dt i

Euler’s method is obtained if the last term, which is the principal truncation error for this
computational step, is dropped. Therefor, this formula for deriving Euler’s method allows a
formal determination of the truncation error.

Both the numerical integration procedure and the Taylor series formulation can be employed
to generate more accurate methods.
 Method of solution 213

Euler’s method may easily be extended to yield a solution of a system of first-order equa-
tions. Consider the following system of equations

dy 1
-------- = f 1  t y 1 y 2  y n 
dt y1  t0  = y1  0 
dy 2
-------- = f 2  t y 1 y 2  y n  y2  t0  = y2  0 
dt with (6.11)


yn  t0  = yn  0 
dy n
-------- = f 1  t y 1 y 2  y n 
dt

or in vector form as

dy
------ = f  t y  with y  t0  = y  0  (6.12)
dt

T
where y =  y 1 y 2  y n  , are n dependent variables whose values are given at t = t 0 ,
T
and f  t y  =  f 1  t y 1 y 2  y n  f 2  t y 1 y 2  y n   f n  t y 1 y 2  y n   .

We obtain the numerical solution of these equations from Euler’s method by employing the
computational formula

y 1 i + 1 = y 1 i + h  f 1  t i y i y i  y i 
y 2 i + 1 = y 2 i + h  f 2  t i y i y i  y i 
(6.13)

y n i + 1 = y n i + h  f n  t i y i y i  y i 

in scalar notation, or as

y i + 1 = y i + h  f  t y  (6.14)

in vector form.

6.1.2 Errors and Stability

It is important to examine and be able to control the errors associated with the numerical
solution of a differential equation in order to determine the accuracy of the results obtained.
214 Numerical Solutions

The total error are often said to consist of the round-off error, which arises due to a finite
number of significant figures of mathematical operations in the computer, and the truncation
error which arises due to the finite difference approximation of our integration formula. The
infinite series that often represents the function, is generally truncated after just a few terms
to yield the scheme for numerical solution of the differential equation. This truncation or
dropping of remaining terms leads to the truncation error which for a single step, can be
expressed as the local error of a method, however an even more important feature of integra-
tion methods is the growth or accumulation of errors as computation progresses. This can be
expressed in features of the global error and the stability of the scheme. These considerations
are discussed next.

Definition: The local error of an integration method when performing one single step can be
stated as

li + 1 = yi + 1 – y  ti + 1  (6.15)

where y i + 1 is the integration method approximation to the exact value y  t i + 1  .

For Euler’s method we can introduce the computational formula given in Eq. 6.7 to yield

l i + 1 = y  t i  + h  f  y i t  – y  t i + 1  (6.16)

If the exact solution is continuous near t i we may express y  t i + 1  by a Taylor series as fol-
lows

2 k
h2 hk
y  tn + 1  = y  t i  + h  dy
------ + -----  d y +  + -----  d y , k   (6.17)
dt ti 2 dt ti k! d t ti

Inserting Eq. 6.17 into Eq. 6.16 we obtain

2 k
h2 hk
l i + 1 = – -----  d y – -----  d y , k   (6.18)
2 d t ti k! d t ti

The leading term of the error is of the order of h 2 and is usually denoted as O  h 2  , i.e.
“order of h 2 “, with the remaining terms being represented as O  h 3  .

Using the mean value theorem and truncating the expression after the first term on the right
hand side, gives us a single expression for the local error as
 Method of solution 215

2
h2 d
l i + 1 = -----  y    where ti    ti + 1 (6.19)
2 dt ti

This expression approximates the truncation error per step in Euler’s method.

This expression tells us that when the second derivative of the function is bounded, then the
local error of the method is quadratic in the step length. This means that if the step-length is
cut in half, we obtain on the average, a local error that is four times smaller. In a similar man-
ner the truncation error can be obtained for other integration formulas.

In Fig. 6.2 is shown a geometric interpretation of the local error using the Euler method.

y
i+1
fy  u  t  l
i i i i

y
i yt 
i+1
h
t t
i i+1

Fig. 6.2 Local error of the Euler method

The truncation error per step as expressed in Eq. 6.19 does not however express the total
error of the numerical solution y i at t i . Normally the exact solution is not know or used at
the grid points t i , except for the first step where the initial condition is given. This means that
the values of y i obtained by using Euler’s method contains the accumulated error of all per-
vious steps. The total error at a given value of t i will be the product of the error per step and
the number of steps. This global error can be defined as

Definition: We define the global error of an integration method as

en = yn – y  tn  (6.20)

In Fig. 6.3 is shown a geometric definition of the global error for one-step integration formu-
216 Numerical Solutions

las.

yt 
y yt
i
y
n+1

e
h h h h h n+1
t t t t
0 i+3 n n+1

Fig. 6.3 Geometric definition of the global error

–1
Since the number of steps is proportional to h , the total error is proportional to h , or
O  h  , for the Euler method. Because of the total error is of the first order in h , Euler’s
method is termed a first order method. A detailed analysis of the total error of an integration
formula is generally much more complicated than the derivation of the local error, and nor-
mally it is only possible to create a rough estimate of this error, and is not included here. For
a detailed presentation of this topic see [11]. However the findings from such an analysis can
be easily defined as follows

Definition: Consider a one-step method as Eq. 6.8 and suppose that p is the largest integer
for which the local error satisfies

p+1
ln = O  h  (6.21)

Then the method is said to be of an order of p .

A very important consideration in the numerical solution of differential equations is that of

stability of the numerical method. Several definition of stability are used in the literature,
where the one most commonly employed definition simply considers a numerical method to
be unstable if it yields an unbounded solution when the exact solution is bounded. Instabili-
ties arises due to the amplification of the error, which under certain conditions may cause
unbounded growth. The stability of numerical schemes therefore depends on both the method
and the differential equations or model at hand.

Definition: A numerical integration procedure is said to be stable if the error measured by

the difference between the true solution and the numerical solution decreases in magnitude
on the average as the integration proceeds step by step.

In order to investigate stability properties of a numerical scheme, we normally analyse the

solution of a model equation using the integration formula of interest. The most frequently
used model equations is
 Method of solution 217

y· = y (6.22)

where  is the eigen value of the system.

Suppose that there exists a a function R  h  so that the solution of the differential equation
using a one-step method can be put in the form

y i + 1 = R  h y i (6.23)

where R  h  is called a stability function for the method. Then stability of the numerical
scheme is ensured if

R  h   1 (6.24)

This expression gives several conditions for the time-step and also the location of the eigen-
value  , for the solution to be stable.

Now, inserting Eq. 6.22 into the computational formula for the Euler method, Eq. 6.8, we
obtain

y i + 1 = y i + hy i =  1 + h y i (6.25)

which gives

R  h  =  1 + h   1 (6.26)

This inequality is satisfied if h is found inside a circle with centre in  – 1 0  and with a
radius of 1 in the imaginary plane. For real eigenvalues  , stability is ensured if

– 2  h  0 (6.27)

or equivalently

2
h  – --- (6.28)


As an example let us solve the find the stability requirements for the following system

y· = – y with y0 = 1 (6.29)

218 Numerical Solutions

for 0  t  8 using Euler’s first-order method using different time-steps h=0.5, 0.75, 1.0, 1.5
, 2.0 is shown in Fig. 6.4 . The stability limitation on the time-step h computed from Eq.
6.28 is found to be h  2 . It is clear from the results in Fig. 6.4 that h=2.0 does not accurately
represents the exact solution very good but does not explode either. However, if you conti-
noue to simulate som hundred seconds the results error will grow unbounded. For timesteps
smaller than h=2.0 the solution error gradually decreases and approaches the correct solu-
tion. From Fig. 6.4 it may be concluded by inspection that the timestep should be less than
h=0.5 to yield acceptable accuracy.

1.5
exp(-t)
h=2.0
h=1.5
1
h=1.0
h=0.75
h=0.5
0.5

-0.5

-1

-1.5
0 1 2 3 4 5 6 7 8 9

Fig. 6.4 Euler’s method applied to the system y· = – y for h = 0.5, 1.0,1.5 and 2.0.

For the last example the timestep required for stability was constant in the simulation inter-
val. For the next example we will show that that timestep required for stability may change in
the simulation interval due to nonlinearity of the system.

Consider the system

y· = y + y 2 with y0 = 0 (6.30)

which integrated using Euler’s method gives the computational formula

2
y i + 1 = y i + h  y i + y i  (6.31)

Linearizing the differential equation around ỹ gives

y· = –   + 2ỹ y (6.32)

where the eigenvalue now is expressed as  = –   + 2ỹ  and the stability condition for
the linearized system now becomes
 Method of solution 219

2
h  – ------------------- (6.33)
 + 2ỹ

This shows that for large ỹ small h is required and vice versa, i.e. the timestep must be
altered as a function of the value of the statevariable itself.

Finally, let us demonstrate the total accumulation of errors found in numerical solutions.
Assume that the system

y· = – y with y  0  = 1 and  = 0.5 (6.34)

The exact solution for this system is

– 0,5t
yt = e (6.35)

Computing a numerical solution using Euler’s method using different time-steps h gives at
–5 –5
the right grid point y  10  = e . In the table below is shown the error e x = 10 = y 10 – e .

h e  10 

10 4.0

5 -2.24

2.5 0.00283

1 0.00576

10-1 8.17 10-4

10-2 8.39 10-5

10-3 8.48 10-6

10-4 8.58 10-7

10-5 -1.53 10-6

10-6 -1.83 10-4

Inspecting the size of the error it can be seen that it reduces quickly then to increase again for
the smallest step-sizes. It looks like optimal step-size is approximately 0.0001. For step-sizes
larger than 2.5 the error is several orders of magnitude larger than the smaller values. Com-
paring this to the stability limit h  4 explains the unbounded errors for step-sizes larger than
220 Numerical Solutions

the stability limit and the computer round-off error the growth of the error for small step-
sizes.

A natural question having introduced the error and stability of the Euler method, is how to
obtain methods with smaller error and less restrictive stability requirements. The most obvi-
ous paths to follow to do this is

1. Increase the number of grid-points used to compute y i + 1 .

2. Compute higher order derivatives at t i to compute y i + 1 .

The first option leads to what we call linear multistep methods and the second option leads to
higher-order one-step methods. A particular interesting class of methods here is the Runge-
Kutta methods. These methods will be treated later in the chapter, but first let us look at two
special Euler methods of interest.

6.1.3 The improved Euler method

The improved Euler method contains in addition to a first step using Eulers method to com-
pute a first approximation to the state variable ŷ i + 1 , a second step which is used to compute
an approximation of f i + 1 using ŷ i + 1 . Finally as an improved approximation to f i the aver-
age of f  y i  and f  ŷ i + 1  is used to compute the final approximation to y i + 1 . The method
can be stated as follows

k 1 = f  y i t i 
k 2 = f  y i + hk 1 t i + h 
(6.36)
h
yi + 1 = y i + ---  k 1 + k 2 
2

To establish the order of this method a Taylor series expansion around  y i t i  is used to gen-
erate an expression for the local error. The local truncation error of the scheme is found to be
O  h 3  , meaning that the improved Euler’s method is of order p = 2 .

Applying the scheme to the stability test equation y· = y results in

k 1 = y i
k 2 =   1 + h y i
(6.37)
 h  2
yi + 1 =  1 + h + ------------- y i
 2 
 Method of solution 221

which states that the improved Euler method is stable when

 h  2
R  h  =  1 + h + -------------  1 (6.38)
 2 

On the real axis this corresponds to

2
– ---  h  0 (6.39)


which is the same as for the first order Euler method. In the imaginary axis the area is how-
ever larger as shown in Fig. 6.5 for p = 2 .

6.1.4 The modified Euler method

Another variation of the Euler method is known as the modified Euler method the midpoint
rule, can be written as

k 1 = f  y i t i 

k 2 = f  y i + --- k 1 t i + ---
h h
2 2 (6.40)
h
y i + 1 = y i + --- k 2
2

Here the first step is used to compute an approximation to the state variable ŷ i + 1  2 at
t i + h  2 using Eulers method, for then to use this value to compute a new approximation to
f i at t i + h  2 . Finally this approximation of the function f is used to compute y i + 1 .

To establish the order of this method we proceed as for the improved Euler method. The local
truncation error of the scheme is then found to be O  h 3  , meaning that the modified Euler
method is also of order p = 2 .

Applying the modified Euler method to the stability test equation y· = y yields

k 1 = y i

k 2 =   1 + ---  y i
h
 2  (6.41)
 h 
y i + 1 =  1 + h + ------------- y i
2
 2 
222 Numerical Solutions

which states that the modified Euler method has the same stability condition as the improved
Euler method, and is stable whenever

 h  2
R  h  =  1 + h + -------------  1 (6.42)
 2 

6.2 Explicit Runge-Kutta methods

Explicit Runge-Kutta methods are a set of single-step methods which has been created for
many different orders where the most famous of these is the classical fourth order method.
The Runge-Kutta methods are closely linked to the Taylor series development of a given
function and that the higher order derivatives exists and can be computed. The main idea for
Runge-Kutta methods is however, to evaluate these higher derivatives without actually hav-
ing to compute them.

General explicit Runge-Kutta methods can be defined as

Definition: Let s be an integer (the “number of stages”) and

a 21 a 31  a s1  a s1 a s2  a s s – 1 , b 1 b 2  b s c 2  c s be real coefficients. Then
the method

k 1 = f  t i y i 
k 2 = f  t i + c 2 h y i + ha 21 k 1 
k 3 = f  t i + c 3 h y i + h  a 31 k 1 + a 32 k 2  
(6.43)

k s = f  t i + c s h y i + h  a s1 k 1 + a s2 k 2 +  + a s s – 1 k s – 1  
yi + 1 = yi + h  b1 k1 + b2 k2 +  + bs ks 

is called an s stage explicit Runge-Kutta method (ERK) for the initial value problem.

or expressed in a more compact form as

i–1
 
k i = f  t n + hc i y n + h  a i j k j  i = 1  s
 
j=0
(6.44)
s

yn + 1 = yn + h  bj kj
j=1
 Explicit Runge-Kutta methods 223

An ERK method of s stages then requires s function evaluations. The coefficients c i , a ij

and b j has to be determined such that the computational formula for y n + 1 is equivalent to
the first number of terms of the Taylor series. Determining the coefficients is a huge non-triv-
ial task because the equations which is formed, also called order conditions, are highly non-
linear and extremely difficult to solve. Interested readers will find this thoroughly covered in
[11]. This means that the order of the Runge-Kutta methods are tied to the evaluation of the
constants of the specific method.
s–1 s

Usually, the conditions c i =  a i j and  bi = 1 are satisfied.

j=1 i=1
The maximum order for an s stage ERK method has been found to be as shown in Table 6.1.

Table 6.1 : Order, condition and stages for ERK-methods

Order ( p ) 1 2 3 4 5 6 7 8 ... 10

Minimum no. of 1 2 3 4 6 7 9 11 ... 17

stages s

No. of 1 2 4 8 17 37 85 200 ... 1205

conditions

Observe the relation between the order of the method and the number of stages. This indi-
cates that number of function evaluations, i.e. evaluation of f  t y  is important to bear in
mind when requesting higher orders. This number of stages is also the minimum number of
stages, numerous Runge-Kutta methods exists which have a higher number of stages. This
table clearly shows the tradeoff between computational accuracy of the different methods and
the elapsed computer time required to solve a specific problem. In general a second or fourth-
order method is a very good choise initially.

The general Runge-Kutta method as shown in Eq. 6.43 are often expressed in a more com-
pact form as a table named a Butcher-table [13]. These tables displays the coefficients a i j ,

0
c2 a 21
c2 a 31 a 32
: : : :
cs a s1 a s2  a s s – 1
b1 b2  bs – 1 bs

c i and b j of the specific method according to the definition Eq. 6.43. Some examples of well
known ERK are given in the next section.
224 Numerical Solutions

6.2.5 Some explicit Runge-Kutta methods

Explicit Runge-Kutta methods can be expressed in Butcher tables as introduced in the pervi-
ous section. The Euler’s method which can be denoted RK1 is of order 1 and has the Butcher
table

0
1

The improved Euler method is equivalent to a second order explicit Runge-Kutta method
also denoted RK2 has the Butcher table

0
1 1
1 1
--- ---
2 2

The classical fourth order Runge-Kutta method often denoted RK4 which is of fourth order
has the Butcher table

0
1--- 1---
2 2
1--- 1---
2 0 2
1 0 0 1
1 2 2 1
--- --- --- ---
6 6 6 6

6.2.1 Stability of ERK-methods

Stability of general RK-methods can be established by testing the model equation y· = y ;
y  0  = 1 as for the Euler methods shown previously. A general formula for the stability
function of a Runge-Kutta method in terms of the parameters of the method can be found
from

T
det  I – h  A – 1b  
R  h  = ---------------------------------------------------- (6.45)
det  I – hA 

where A and b is the matrix of a i j and b j coefficients for the method and
T
1 =  1 1  1  . For an explicit Runge-Kutta method the denominator is equal to one due
to non zero elements only below the diagonal, which means that for ERK-methods only the
 Explicit Runge-Kutta methods 225

nominator needs to be evaluated

Example: RK2 or improved Euler method has the Butcher table

0
1 1
1 1
--- ---
2 2

Applying Eq. 6.43 this yields

k 1 = f  t i y i  = y i
k 2 = f  t i + c 2 h y i + ha 2 1 k 1  =   1 + h y i
yi + 1 = yi + h  b1 k1 + b2 k2 
(6.46)
= y i + h  --- y i + ---   1 + h y i
1 1
2 2 

=  1 + h + ---  h  2 y i
1
 2 

which states that the stability condition for the RK2 or improved Euler method is

1
R  h  = 1 + h + ---  h  2  1 (6.47)
2

In Fig. 6.5 is shown the stability area for Runge-Kutta methods of order one to four. The dif-
ference in stability area between the different Runge-Kutta methods is mainly found on the
imaginary axis. For a problem with eigenvalues close to the imaginary axis it can be stated
that the p = 4 order method is to be preferred, while the p = 1 method results in possibly
226 Numerical Solutions

intolerable small step sizes.

Im  h 
3
p = 4
p = 3 2
p = 2
1
p = 1

Re  h 
-3 -2 0

Fig. 6.5 Stability area for ERK-methods of order 1 to 4

6.3 Stiff systems

A differential equation or system of equations with two or more widely varying time con-
stants is known as a stiff system. Stiff systems may occur in chemical kinetics due to variable
reaction speed, in electrical circuits due to variable conductances or in mechanical systems
due to huge differences in inertias. The list of cases for such system can be made huge.

The solution of such systems is difficult due to the fact that in order to get an accurate solu-
tion we must use very or maybe tiny small time steps even if we only are interested in the
large time constant part of the solution. For most physical problems the degree of stiffness
will change along the integration path, and for an explicit variable step size methods the step
size selection procedure will decrease the step size to satisfy the stability condition way
beyond the requirements set by the accuracy condition. This gives problems with simulation
time and accuracy. This means that it is the stability condition of the method which must
allow for large spread in the time constants or eigenvalues to be suitable for such systems.
This leads to a somewhat pragmatic definition of stiff system as

Definition: A problem is stiff for an integration method iff the step length for a requested
accuracy is controlled by the stability condition and not the accuracy condition, i.e.
 Stiff systems 227

  A  max
 = -------------------
-»1 (6.48)
  A  min

where A is the A -matrix of the problem

As class of methods which are of special interest for stiff systems are implicit methods. Com-
pared to explict Runge-Kutta or explicit linear multistep methods these methods have large
stability areas. In the following we will introduce a set of methods which are called A-stable.

Definition: An integration method is A-stable if R  h   1 for all systems y· = y where

Re     0 .

No explicit Runge-Kutta methods are found to be A-stable, but most implicit Runge-Kutta
methods are. For linear multistep methods the same conclusion apply, but for methods of
order p  2 only.

Popular methods for stiff systems are:

• Implicit Runge-Kutta methods

• The Trapezoidal method
• The backward Euler method
• “Backward Differensing Fomula”, or BDF-methods

Implicit methods is dependent on the solution of an implicit system of equations, most often
using the Newton-Raphson method, and the evaluation of the Jacobi-matrix of the system.
All this adds to the computation work using implicit methods compared to explicit ones. Still
the class of implicit methods are very important to be able to cope with stiff systems in par-
ticular.

In the next section implicit Runge-Kutta methods will be presented.

6.3.1 Implicit Runge-Kutta methods

An s -stage implicit Runge-Kutta method can in general be defined as
228 Numerical Solutions

s
 
k i = f  t n + hc i y n + h  a i j k j  i = 1  s
 
j=0
(6.49)
s

yn + 1 = yn + h  bj kj
j=1

As for the explicit Runge-Kutta methods the implicit counterpart can also be expressed in a
Butcher-table.

The stability condition for implicit Runge-Kutta methods are found from

T
det  I – h  A – 1b  -  1
R  h  = --------------------------------------------------- (6.50)
det  I – hA 

as for the explicit Runge-Kutta methods, but now the denominator must be included.

Example 6.2

The implicit Euler method is an implicit Runge-Kutta method with one stage described by
the following Butcher-table

1 1
1

which gives

k 1 = f  t + h y i + hk 1 
(6.51)
y i + 1 = y i + hk 1

The stability function is found to be

1
R  h  = --------------- (6.52)
1 – h

which fulfils the A-stable condition, meaning that the implicit Euler method is an A-stable
implicit Runge-Kutta method (IRK).

Some implicit Runge-Kutta methods can be given using Butcher-tables as

 Multistep methods 229

Implicit mid-point rule:

1
--- 1
---
2 2
1

Trapezoidal rule:

0 0 0
1 1
---
1
---
2 2
1 1
--- ---
2 2

Gauss order 4:

1--- ------3-
– --1- 1
--- – ------3-
2 6 4 4 6
1
1--- ------3-
+
1--- ------3-
+ ---
2 6 4 6 4
1 1
--- ---
2 2

In [11] both explicit and implicit Runge-Kutta methods are treated in detail.

6.4 Multistep methods

A second option to increase the accuracy of integration formulas is to include the values of y
and y· at several grid points from previous time steps. This strategy leads to a class of meth-
ods called linear multistep methods. In Fig. 6.6 is shown an geometric interpretation of a gen-

yt
y y
i+1
i
y y
i–1 i

y
i–2
h h h
t t t t
i–2 i–1 i i+1

Fig. 6.6 Geometric interpretation of linear multistep

eral linear multistep method. Development of linear multistep methods is often be based on
Taylor series development. The number of previous y n values used, denotes the step number
230 Numerical Solutions

of the method. These methods are not self-starting and can be both explicit and implicit.

Definition: A s-step linear multistep method can be defined by the equation

s s

 k yi + k = h   k f  t i + k y i + k  ,  s = 1 , 0 + 0  0 (6.53)
k=0 k=0

This indicates that we need y i  y i + s – 1 to find y i + s .

If  s = 0 then the method is explicit. If  s  0 , then the method is implicit which means
that because the y i + s variable occurs on both sides of the equal sign, an implicit equation has
to be solved at each time step.

Multistep methods which are frequently used is

• Adam-Bashford, a class of explicit methods of variable order p = s + 1

• Adam-Moulton, a class of implicit methods of variable order p = s + 1
• Backward-Difference-Formula (BDF), a class of implict methods of variable order
p = s+1

Example 6.3

The Euler method which is an s = 1 step method, can be given by the coefficient set
 1 = 1 ,  0 = – 1 ,  1 = 0 , and  0 = 1 to yield

y i + 1 – y i = hf  t i y i  (6.54)

Example 6.4

The the well known trapezoidal method which is an s = 1 step method, can be given by the
1 1
coefficient set  1 = 1 ,  0 = – 1 ,  1 = --- , and  0 = --- to yield
2 2

y i + 1 – y i = h  --- f  t i y i  + --- f  t i + 1 y i + 1 
1 1
(6.55)
2 2 

6.4.1 Error and order of linear multistep methods

The order of a linear multistep method can be evaluated by defining a linear difference oper-
ator L defined by
 Multistep methods 231

L  y  t  ;h  =    k y  t + kh  – hk y·  t + kh   (6.56)
k=0

where y  t  is an arbitrary function, continuously differentiable on the interval  a b  .

Expanding the test function and its derivatives as a Taylor series about t and collecting terms
gives

1 q
L  y  t  ;h  = C 0 y  t  + C 1 hy  t  +  + C q hy t +  (6.57)

where the C q are constants.

Definition: The difference operator Eq. 6.18 and the associated linear multistep method are
both said to be of order p if C 0 = C 1 =  = C p = 0 and C p + 1  0 .

A simple calculation yields the following formula for the constants C q in terms of the coeffi-
cients  k ,  k :

C0 =  k
k=0
s s

C1 =  k k –  k (6.58)
k=1 k=0
s s
1 1 q–1
  k k
q
C q = -----  k k – ------------------ q = 2 3 
q!  q – 1 !
k=1 k=1

This formula can be used to construct a linear multistep method of any order.

For a given linear multistep method the first non-vanishing coefficient in the expansion is
p + 1 which then is identified as the error constant of the local truncation error defined by L
often given the symbol T . Hence the truncation error of an explicit linear multistep method is
given by

p + 1 p + 1 p+2
T = y  ti + s  – yi + s = Cp + 1 h y  ti  + O  h  (6.59)

where p is the order of the method.

232 Numerical Solutions

Example 6.5

The local error constant of the linear multistep method defined by  0 = – 1 , and
 0 =  1 = 1  2 which is the trapezoidal rule is then given by

1 3 1 2 1 1 1
T = -----  1 1 – -----  1 1 = --- – --- = – ------ (6.60)
3! 2! 6 4 12

6.4.2 Stability of linear multistep methods

To check for the stability properties of an LMF-method we can use as for the Runge-Kutta
methods the stability test equation y· = y . Inserting this into Eq. 6.53 yields

s s

 k yi + k = h  k yi + k (6.61)
k=0 k=0

or

  k – hk yi + k = 0 (6.62)

k=0

If we define

z = h (6.63)

 k r
k
r = (6.64)
k=0

and

 k r
k
r = (6.65)
k=0

we can define a characteristic polynom for absolute stability of linear multistep methods as
 Multistep methods 233

  k – hk r
k
  r z  = =   r  – z  r  (6.66)
k=0

Then the stability condition for LMF-methods can then be defined as

Definition: Absolute stability of linear multistep methods can be established by solving

  r z  = 0 and then require that r  1

Example 6.6

The stability condition for Eulers method which is an s = 1 step method can be found by
solving

  r z  = r – 1 – z = 0 which yields r = 1 + z (6.67)

The condition r  1 then yields

r = 1+z 1 (6.68)

The solution for this inequality is a circle with radius h and with centre in  – 1 0  which is
as expected.

Example 6.7

The stability condition for Euler implic method which too is an s = 1 step method can be
found from

1
  r z  = r – 1 – rz = 0 whic yields r = ----------- (6.69)
1+z

The condition r  1 then yields

1 1
r  ---------- (6.70)
-
1+z

The solution for this inequality is found for all values of Re  h   0 . Methods satisfying this
condition is said to be A-stable, an advantageous property, but we have to add the cost to
solve an implicit equation for each time step.
234 Numerical Solutions

6.5 High-order systems

In some cases, we might find that part of a model is described by a high-order differential
equation and another part by first-order state equations. To form a unified model we will in
the following describe the relationship between a high-order differential equation and first-
order state equations.

Let us consider that a single-variable, linear time-invariant system is described by the fol-
lowint nth-order differential equation

n n–1 n–2
dy d y d y dy
+ an + an +  + a2 + a1 y = u  t  (6.71)
dt
n
dt
n–1
dt
n–2 dt

The problem is to represent Eq. 6.71 by n state equations. This simply involves the defining
of the n state variables in terms of y  t  and its derivatives. For the present case it is conven-
ient to define the new state variables z k , k = 1 2  m as

0
z1 = y
1 m
z2 = y m d y
where y = m
(6.72)
: dt
m – 1
zm = y

Now take the time derivative of each of the new state variables in Eq. 6.72, realise that
0 1
y· = y and insert the definitions of the z -variables from Eq. 6.71 to yield

0 1
z·1 = y· = y = z 2
1 2
z·2 = y· = y = z 3
(6.73)
:
m – 1 m
z·m = y· = y = – a1 z1 – a2 z2 –  – an zn + r  t  

k
The initial conditions required is y  a  = y k  a  , for k = 0 1 2  m – 1 or expressed
k – 1
as the new state variables z k  a  = y  a  = y k – 1  a  , k = 1 2  m .

The new system of 1st-order ordinary differential equations can now be written as
 High-order systems 235

z·1 = z 2 z1 = y  a 
z·2 = z 3 z2 = y1  a 
(6.74)
: :
z·m = – a 1 z 1 – a 2 z 2 –  – a n z n + r  t  z m = y m – 1  a 

In vector-matrix form, Eq. 6.74 can be written as

d
x  t  = Ax  t  + Br  t  (6.75)
dt

where x  t  is the n  1 state vector and r  t  is the scalar input.The coefficient matrices are

0 1 0 0  0 0
0 0 1 0  0 0
A = 0 0 0 1  0 B = 0 (6.76)
: : : : : :
–a1 –a2 –a2 –a4  –an 1

Example 6.8

Consider the differential equation

3 2
dy dy dy
+5 + + 2y = r  t  (6.77)
dt
3
dt
2 dt

The state variables are defined as

z1 = y  t 
1
z2 = y t (6.78)
2
z3 = y t

The state equations in matrix-vector form then becomes

z·1 0 1 0 z1 0
z·2 = 0 0 1 z 2 + 0 r  t  (6.79)
z·3 –2 –1 –5 z3 1
236 Numerical Solutions

Example 6.9

Given the 3rd-order non-linear initial value problem stated as

y''' + 2xy' +  x 2 + y 2 y = 0 0  x  x end (6.80)

k
and with the initial values y  0  = a k , k = 0 1 2 .

k
Defining a new variable z k = y , for k = 1 2 3 we can write Eq. 6.80 as a system of first
order non-linear ordinary differential equations as

z 1' = z 2 z1  0  = a0
z 2' = z 3 z2  0  = a1 (6.81)
z 3' = – 2xz 2 –  z 12 + x 2 z 1 z3  0  = a2

6.6 Practical considerations

As you may have experienced from what we have discussed up to now, there exists numerous
integration methods and many factors which determines the features of an integration
method. It is important to recognise that no best method can be given in general. The accu-
racy, stability and computational effort is playing a significant role when selecting an integra-
tion method for use on a physical problem, which itself have specific properties and time
constants. This means that the selection of an integration method is highly dependent on the
problem at hand.

A general advice is to start with a simple explicit method such as the Euler O  1  or a Runge-
Kutta O  2  or the classical O  4  method with fixed step size. These methods will with
some testing with variable step length be sufficient for a large number of engineering prob-
lems. A natural second step is to conduct experiments to compare the computational effi-
ciency of variable step methods and in special cases implicit methods when stiffness is
detected.

Below a list of important factors to consider when selecting integration methods is discussed.

Variable step-size and order

As the user of integration methods you are not primarily interested in the step size and order
of the method used to compute the solution. The interesting point is the accuracy and time the
 Practical considerations 237

computer uses to present the solution to you. This means that an integration algorithm which
automatically adjusts the time step and possibly also the order of the actual method used is
very attractive. There are several benefits to the automation of step-size and order selection:

• Integration algorithms which is using the optimal step-size and order minimizes the
number of integration steps, thereby reducing both the computation time and the accu-
mulated round-off error.
• The error is automatically controlled
• Manually choosing the step-size can be extremely difficult due to the fact that the sys-
tem may require different step-sizes along the integration interval and that different
methods also require drastically different step-sizes.
• As the user is primarily interested in the accuracy of the solution, specifying an error
tolerance is more natural than specifying the step-size

A typical step-size control algorithm works as follows. The user supplies an integration error
tolerance  . The integration algorithm suggests a step-size and conducts one step using the
method. Then the state variable is computed and the error is estimated. If the estimated error
is within the bounds requested by the user, then the step is accepted and the integration can be
advanced. If the estimated error on the other hand is larger then requested, then the step-size
is reduced by a fraction and a new step is performed. If the estimated error is much less than
requested, the step-size can be increased by a fraction and a new step performed.

For some integration methods the error can be evaluated by a single formula. For Runge-
Kutta methods a strategy of actually using two different methods to compute the same value
and then comparing the difference can be used to control the step-size. This is called embed-
ded Runge-Kutta methods.

For variable-order algorithms as the linear multistep methods, a similar approach is used to
adjust the order.

Output points

Requesting the solution of the system at a given set of grid points may conflict with the step-
sizes that the integration algorithm uses. If the step-size is much smaller than the interval
between requested output points the integration routine has to perform many steps to reach
the point where it is to store or print the solution and then it has to adjust the step-size of the
last step before reaching the output point to hit the point accurately. On the other hand, if the
user requests very dense output points this limits the step-size considerably and slows down
the computation. The solution to the last case is to use the step-size suggested and then to use
an interpolation routine to obtain the output points in between. To maintain the order and
accuracy of the solution using the interpolation routine special considerations when design-
238 Numerical Solutions

ing the interpolation routine has to be taken.

It is sometimes interesting to integrate the system to a special value of t or also until some
state variable reaches a special condition or value. An example is to integrate a system until
the ball hits the ground. In such situations a combination of the two strategies described
above can be used.

Start-up costs

Most integration algorithms require extra computation at the beginning of an integration.

This overhead, called the start-up cost, stems from the initialisation of multistep methods and
initialization of the step-size selection procedure. For initialising the s step multistep
method, s – 1 grid points must be computed by some means, usually a single step method as
Eulers method. The step-size used to compute these values is not optimal therefore also
requiring a significant cost increasing the step-size. Initialising the step-size selection proce-
dure also involves using secure small step-sizes initially for then to increase the step-size to
the optimal size.

REFERENCES
10. Lambert, J., Numerical methods for ordinary differential equations. The initial value
problem. John Wiley, Chichester, 1991.
11. Hairer, E., Nørsett, S. and Wanner, G., Solving ordinary differential equations I: Nonstiff
problems, 2nd Ed., Springer-Verlag, Berlin, 1993.
12. Hairer, E. and Wanner, G., Solving ordinary differential equations II: Stiff and differen-
tial-algebraic problems, 2nd Ed., Springer-Verlag, Berlin, 1996.
13. Butcher, J. C, Implicit Runge-Kutta integration processes, J. Austral. Math., Soc., Vol.
IV,, Part 2, pp. 179-194.
 Multiport fields 239

Chapter 7

Multiport fields

Up to now we have defined 1-port C, I and R plus Se and Sf elements, 2-port TF and GY ele-
ments and n-port 0 and 1 junction elements. These basic elements each have their specific
constitutive laws relating single power or energy variables and hence may be understood as
general scalar energy operators, hence we often name them scalar bond graph elements.
Extending the application of bond graphs into new areas we, although seldom, find that the
basic bond graph elements do have some limtations when representing physical laws or that a
more compact way of drawing bond graphs is disirable. Extending the bond graph language
to include physical laws described using vector relations is one such extension.

In this chapter, we will introduce multiport generalization of the basic scalar bond graph ele-
ments. These multiport generalization are named fields, and will be used in succeding chapter
to model complex multiport systems. The important message in this text is that the basic
physical characteristics of each of the elements are unchanged, i.e. the C-element whether it
is a scalar element or a field still represents idea energy accumulation and so on.

7.1 Energy storing fields

7.1.1 C-fields
A C-field is an element described as a set of effort-displacement relations at n ports. The
sympol is as for a 1-port C-element, the letter C, but now with n power bonds connected as
shown in Fig. 7.1 . To distinguish between a scalar C-element and the corresponding field,

e fi ei
2
f1
e e
1 n
C
f f
1 n

Fig. 7.1 An n-port C-field

the C-field is often drawn with a bold mnemonic code, i.e. C. The constitutive law for the C-
240 Multiport fields

field can now be written as

–1
e = C  q  q = C  e  (7.1)

in the general nonlinear case, where e =  e 1 e 2  e n  T , q =  q 1 q 2  q n  T and   . 

is any mathematical relations between e and q . For the linear case this can be written in
matrix form as

–1
e = Kq q = K e = Ce (7.2)

where K is the stiffness matrix and C is the corresponding compliance matrix.

The energy accumulated at any given time can be expressed as

t n t n

E =   ei fi dt + E0 =   ei q· i dt + E0
t0 i = 1 t0 i = 1
(7.3)
q n q

=   e i  q  dq i + E0 =  e  q  dq + E 0
q0 i = 1 q0

where the identity f i dt = dq i has been used, e  q  represents the constitutive lows of the
filed and e  q dq represents the vector inner product. The E 0 represents the initial energy
stored at t = t 0 and q 0 = q  t 0  .

A simple example which may enlighten our understanding of C-fields is the bond graph
model of a cantilever beam with a force F and a moment M applied at the free end as
shown in Fig. Neglecting the mass of the beam, it can be represented by constitutive relations

F M
C ·
y· 
a) b)

Fig. 7.2 Cantilever beam. a) schematic and b) the bond graph C-field

among y ,  , F and M as
 Energy storing fields 241

3 2
l l
--------- ---------
y = 3EI 2EI F q = Ce (7.4)
 l
2
l M
--------- ------
2EI EI

These relations beeing the constitutive law of the C-field is in the compliance form, i.e.
q = Ce . Solving for F and M in terms of y and  , we obtain the stiffness form and the
stiffness matrix K as

12EI 6EI
------------ – ---------
F = EI l3 l2 y –1
e = C q = Kq (7.5)
M 6EI- 4EI 
– -------- ---------
l2 l

An element as the beam example which is described from the beginning as a set of effort-dis-
placement relations at n ports is called an explicit field. Generally, it is much more conven-
ient to treat explicit field elements as fields although it may be possible to find and equivalent
bond graph of scalar bond graph elements only. In some cases likewise, subsystems contain-
ing 1-port energy storing elements bounded together with transformators and junction stru-
cure element can be teated as a fields too. Such a field will be called an implicit field.
Reformulating these scalar bond graph structures into explicit fields may yield compact sub-
model representation or even open up new implementation possibilities. In Fig7.3 is shown a
simple example of an implicit field b) rewritten into the equivalent explicit field c). The con-

F 1 x· 1
F1
1
x1 F 1 x· 1

k1 k2 C 0 0 C C

x2 F 2 x· 2
1
F2
a) F 2 x· 2 b) c)

Fig. 7.3 Explicit C-field example. a) schematic b) bond graph c) C-field

stituitive laws of the external ports must be derived from the constituitive laws of the ele-
ments comprising the field. This is why it is called an implicit field. Let us develop the
242 Multiport fields

constituitive laws for the implicit C-field shown in Fig. 7.3c and thus show tht the C-field
representation in terms of x 1 and x 2 is possible and valid.

If we impose integral causality on both C-elements, it is clear that the displacements q 1 and
q 2 will be the state variables and the state equations becomes

q· 1 = x· 1 + x· 2 = external velocity input

(7.6)
q· 2 = x· 1 + x· 2

From time integration of the state equations we obtain

q1 = q2 = x1 + x2 + C (7.7)

With F 1 = F 2 = q 1 = q 2 = 0 when x 1 = x 2 = 0 the constant of integration is equal to

zero. The constitutive laws for the 2-port C-field in stiffness form then becomes

F1 k1 + k2 k1 + k2 x1
= (7.8)
F2 k1 + k2 k1 + k2 x2

which in words looking at Fig 7.3c means: integrating the two flows x· 1 and x· 2 of the power
bonds input to the C-field yields x 1 and x 2 which using Eq. 7.8 enables us to compute both
F 1 and F 2 which is then is the output efforts requested.

After we now have developed the constitutive laws for the C-field, let us see how it is used as
a subelement in a larger system as shown in Fig 7.4.

Ft Se
7

m1 6
1 I
x1
1

k1 k2 C

2
x2
m2 8
1 I
k3 9
a) b)
C
 Energy storing fields 243

Fig. 7.4 Use of C-field example. a) schematic and b) bond graph

According to the bond graph shown in Fig 7.4b, there are five elements with integral causal-
ity which means that we have five state variables y =  x 1 x 2 x 9 p 6 p 8  T . However, x 2 and
x 8 are equal and therefore we may eliminate one of these, or we may just keep both for clear-
ity. The state equations therefore becomes

p
x· 1 = ----6-
I6
p8
x· 2 = -----
I8
(7.9)
p
x· 9 = ----8-
I8
p· 6 = Se 7 – F 1 = Se 7 –  k 1 + k 2   x 1 + x 2 
p· 8 = –  k 3 x 9 + F 2  = – k 3 x 9 –  k 2 + k 2   x 1 + x 2 

where the constituitive laws of the C-field, Eq. 7.8, has been inserted.

An inspection of the constitutive laws in Eqs. 7.4 and 7.5 reveals that both the stiffness
matrix K and the compliance matrix C are symmetric. This can generally be proven for all
conservative energy storing C-fields including nonlinear fields. In order to check a given
field the following arguments apply: a change in any component of q , i.e. q i , will produce
a change in E , i.e. E . By direct inspection of Eq. 7.3 we can deduce that the coefficient
relating E to q i is actually e i . Thus, the partial derivative of E with respect to q i is just
e i or

E
= ei  q  i = 1 2  n (7.10)
 qi

and because E = E  q  is supposed to represent a single valued bounded continous stored-

energy function of q then

2
e i E e
= = j i = 1 2  n (7.11)
 qj  q j q i  qi

which actually displays how e j and e i are constrained by the energy function E . In other
words not every set of constitutive laws expressing effors as a function of displacements can
244 Multiport fields

represent a energy-storing conservative C-field. These relations are often called Maxwell’s
reciprocal relations valid for both the linear and nonlinear case. In the linear case this states
that

T T
K = K C = C (7.12)

both the stiffness and compliance matrices are symmetic.

By inspecting of the consitutive laws for the cantilever beam, i.e Eq. 7.5, we can establish
that

F 6EI M 6EI
= – --------
- = – --------
-
 l2 y l2
(7.13)
F M
=
 y

and consequently the Maxwell reciprocal relation Eq. 7.11 are fullfiled and the constitutive
laws for this C-field yields a energy conservative field.

A more general and also nonlinear example is the 3D massless spring which is frequently
found in many mechanical systems. The spring initially oriented along a streight line from
point x 1 y 1 z 1 to x 2 y 2 z 2 with an initial length l 0 and having a constant stiffness k is
shown in Fig 7.5. At one end of the spring the velocities in all directionss are prescribed as a

Sf Sf Sf

F x1 F y1 F z1
1 1 1
z x· 1 y· 1 z·1
x  y  z 
1 1 1
y
x
k
C: k C: k

x  y  z 
x· 2 y· 2 z·2
2 2 2

F x2 F y2 F z2 1 1 1 Se

I I I

Fig. 7.5 A nonlinear C-field representation of a 3D linear spring.

 Energy storing fields 245

function of time while at the other end the spring is connected to a ball having a mass m . The
constitutive relations of the C-field are

kl 0  x 2 – x 1 
F x2 = k  x 2 – x 1  – -------------------------------------------------------------------------------------
-
 x2 – x1  +  y2 – y1  2 +  z2 – z1  2
2

kl 0  y 2 – y 1 
F y2 = k  y 2 – y 1  – -------------------------------------------------------------------------------------
-
 x2 – x1 2 +  y2 – y1  2 +  z2 – z1  2
kl 0  z 2 – z 1 
F z2 = k  z 2 – z 1  – -------------------------------------------------------------------------------------
-
 x2 – x1  +  y2 – y1 2 +  z2 – z1 2
2
(7.14)
kl 0  x 2 – x 1 
F x1 = – k  x 2 – x 1  + -------------------------------------------------------------------------------------
-
 x2 – x1  2 +  y2 – y1  2 +  z2 – z1  2
kl 0  y 2 – y 1 
F y1 = – k  y 2 – y 1  + -------------------------------------------------------------------------------------
-
 x2 – x1  +  y2 – y1  2 +  z2 – z1  2
2

kl 0  z 2 – z 1 
F z1 = – k  z 2 – z 1  + -------------------------------------------------------------------------------------
-
 x2 – x1 2 +  y2 – y1 2 +  z2 – z1 2

A check of the constitutive laws using Maxwells theorem Eq. 7.5 yields

F x2 kl 0  x 2 – x 1   y 2 – y 1  F y 2
= --------------------------------------------------------------------------------------------- =
 y2   x2 – x1 2 +  y2 – y1 2 +  z2 – z1 2 
3  2  x2

F x2 kl 0  x 2 – x 1   z 2 – z 1  F z2
= --------------------------------------------------------------------------------------------- =
 z2   x2 – x1  2 +  y2 – y1  2 +  z2 – z1  2 
32  x2
(7.15)
F x2 F x1
=  =
 x1  x2
:
F z2 F x2
=  =
 x2  z2

where the reader is invited to complete the missing terms. However, the correct answer is:
Yes, this is a energy conservative C-field. Connecting this 3D spring to the mass m yields the
bond graph shown in Fig. 7.5c.

7.1.2 Causal considerations for C-fields

As for the 1-port C-element, the n-port C-field may display both integral and differential cau-
sality. However, the C-field also have the option to admit mixed integral-derivative causality.
246 Multiport fields

The complete integral causality and complete derivative forms of the bond graph are as
shown i Fig. 7. and the corresponding constitutive laws may be stated as

ej + 1
e2 q· e e e q· · q· j + 1
i i 2 i i ej qj
q· q·
2 2 q· 2
e1 e e e e1 e
n 1 n n
C C q·
C
q· 1 f q· f 1 q·
n 1 n n
a) b) c)

Fig. 7.6 C-field causality options. a) complete integral causality b) complete derivative
causality c) mixed causality

–1
e i =  C i  q 1 q 2    q n  i = 1 2  n (integral causality) (7.16)

q i =  Ci  e 1 e 2  e n  i = 1 2  n (differential causality) (7.17)

The mixed causality bond graph is shown in Fig 7.6c, where the bonds named from 1 to j is
displaying integral causality and the bonds named from j + 1 to n is in derivative causality.
The corresponding constitutive law now becomes

e i =  i  q 1 q 2  q j  e j + 1   e n  i = 1 2   j
(7.18)
q k =  k  q 1 q 2  q j  e j + 1   e n  k = j + 1   n

In general we often start with a predefined and preferred set of constitutive relations. To be
able to convert this initial set to alternative causality options we have to solve for new set of
input/output variables. This will reveal that some solutions are possible and other not. For an
explicit nonlinear field the determination of alternate causality forms of the constitutive laws
may become very hard. To switch from complete derivative to complete integral causality for
a linear consitutive law we have to invert a matrix, and the existence of the inverse will deter-
mine if complete integral causality is possible. For implicit fields we have the option of
applying the causality assignment procedure fixing the causality of the input bonds of the
field and completing the causality as normal, to check if alternative causality options are pos-
sible. For a more thourough discussion of this topic see[Karnopp] .
 I-fields 247

7.2 I-fields
An I-field is an element described as a set of flow-momenta relations at n ports. The symbol
for an I-field is the bold letter I as shown in Fig. 7.7, but now with n power bonds connected.

p· 2 p· f
i
i
f
2
p· 1 p·
n
I
f1 f
n

Fig. 7.7 An n-port I-field.

To distinguish between a scalar I-element and the corresponding field, the I-field is often
drawn with a bond mnemonic code, i.e. I. The constitutive law for the I-field are given as

–1
f = I  p  p = I f  (7.19)

in the general nonlinear case, where f =  f 1 f 2  f n  T , p =  p 1 p 2  p n  T and   .  is

any mathematical relation between f and p . For the linear case this can be written in matrix
form as

–1
f = I p p = If (7.20)

where I is the inertia matrix, also known as the mass matrix in mechanical systems.

The energy accumulated at any given time can be expressed as

t n t n

E =   ei fi dt + E0 =   fi p· i dt + E0
t0 i = 1 t0 i = 1
(7.21)
p n p

=   f i  p  dp i + E 0 =  f  q  dp + E0
p0 i = 1 p0

As for the C-field the relation between a small change in any p i and the corresponding
change in E is actually the flow variable f i , or expressed using partial derivatives
248 Multiport fields

E
= fi  p  i = 1 2  n (7.22)
 pi

and the Maxwell reciprocity relations becomes

2
f i E f
= = j (7.23)
 pj  p j p i  pi

This equation valid for both linear and nonlinear fields implies that for linear cases the inertia
or mass matrix will be symmetric for energy to be conserved.

Typical examples of explicit I-fields are electric circuits having mutually interacting coils as
shown schematically in Fig 7.8. Mutual inductance is the circuit parameter used to relate the

i1 i2
· ·
e1 = 1 e2 = 2
u1 C1 C2 u2 I
i1 i2

a) b)

Fig. 7.8 Coils inductively coupled. a) schematic b) bond graph I-field

voltage induced n one circuit to a time-varying current in the another circuit. This situation
arises whenever two or more circuits are linked by a common magnetic field. The flux linkin
gthe coil C 2 due to the current i 1 in the coil C 1 is

 2 = Mi 1 (7.24)

and the flux linking the coil C 1 due to the current i 2 in the coil C 2 is

 1 = Mi 2 (7.25)

The constant M is called the coefficient of mutual inductance. Thus with self-inductance and
mutual-inductance the constitutive law can be written in matrix form as
 I-fields 249

1 L1 M i1
=  = Li (7.26)
2 M L2 i2

where L 1 and L 2 are the self-inductance coefficients in the two coils. That the constitutive
law is expressed as a matrix equation shows that it must be defined as an I-field. We also note
that the inertia matrix or inductance matrix is symmetric stating that the C-field is conserva-
tive.

As with C-fields, I-fields also occur in both implicit and explicit forms. An example of such
a system is the rigid beam in plane motion. If we assume small rotations and vertical motion
only, the bond graph implicit and explicit fields will be as shown in Fig 7.9b and c. The cor-

I:m I:J

v1 p· c v c p·  
vc
1 1
v1
 F1 F2
m J I
TF: L/2
F2 v1 v2
T F /2
F1 0 -L 0

F1 v1 F2 v2
a) b) c)

Fig. 7.9 Rigid beam bond graph models.a) schematic b) implicit field c) explicit field

responding constitutive law for the I-field becomes

1- L 2
--- L2
1- -----
v1 + ------ --- – - p
–1
= m 4J m 4J 1 v = I p (7.27)
v2 L2
1- ----- L2 p
1- -----
--- – - --- + - 2
m 4J m 4J

following a similar procedure outlined as for the C-field example. First we write the state
equations for the implicit field and secondly we integrate these in time to get

p· c = F 1 + F 2 pc = p1 + p2
L L L L (7.28)
p·  = – --- F 1 + --- F 2 p  = – --- p 1 + --- p 2
2 2 2 2
250 Multiport fields

Inserting the relations between the velocities v 1 v 2 and v c  we arrive at the constitutive
law expressed in Eq. 7.27.

7.2.3 Causal considerations for I-fields

As for the 1-port I-element, the n-port I-field may display both integral and differential cau-
sality and even have the option of mixed integral-differential causality. The complete integral
causality and complete derivative forms of the bond graph I-field are as shown in Fig. 7.10
and the constitutive laws may be stated as

p· j + 1
p· fi p· fi p·
p· 2 i p· 2 i p· 2 f j j
fj + 1
f2 f2 f2
p· 1 p· n p· 1 p· n p· 1 p·
n
f1
I fn f1
I fn f1
I fn

a) b) c)

Fig. 7.10 I-field causality options. a) complete integral causality b) complete differential
causality and c) mixed causality.

–1
f i =  I i  p 1  p 2   p n  i = 1 2  n (integral causality) (7.29)

p i =  Ii  f 1 f 2  f n  i = 1 2  n (differential causality) (7.30)

The mixed causality constituitive laws resulting from the case where bond 1 to bond j is dis-
playing integral causality and bon j+1 to n is having differential causality becomes

f i =  Ii  p 1 p 2  p j f j + 1  f n  i = 1 2   j
–1
(7.31)
p k =  Ik  p 1 p 2  p j f j + 1  f n  k = j + 1  n

As for the C-field the question of existence of different causality options is an interesting
one. For linear fields the conversion between complete integral and complete differential
causality requires that inversion of the inertia matrix is possible. For nonlinear fields a solu-
tion of equation set must likewise be possible for the required set of input/output variables.
Even though now symbolic manipulation software packages like Maple or Mathematica may
be used, the use of Maxwells reciprocity relations should be exercised to check the consist-
ency of constitutive laws of I-fields.
 Mixed Energy-Storing Fields 251

7.3 Mixed Energy-Storing Fields

A field which is not as obvious as the C and I fields just discussed is the CI-field, which is
formally a combination of these, acting as a C-field on some of the ports and as an I-field on
the other. This field are seldom necessary within a single energy domanin, but are often
found advantageous when working with components combing several energy domains. Such
examples are found modelling transducers, rigid body motion or fluid flow. With other
words, a IC-field accumulates total energy where contributions comes from both kinetc and
potential energy. Fig 7.11 shows a general n-port IC-field. The ports are numered such that
the first j-ports, 1  j  n are inertial in character and the ports j + 1 to n are capacitive in
character. The stored energy E is then a sum of the energy flow on each port as

t n

E =   ei fi dt + E0
t0 i = 1
(7.32)
p j q n

=   f i dp i +   e i dq i + E 0
p0 i = 1 q0 k = j + 1

where p represents the vector of momenta and q represents the vector of displacements. The
Maxwell reciprocity condition for the constitutive laws becomes

2
f i E e
= = k 1  i  j j + 1  k  b (7.33)
 qk  q k p i  pi

An example of an IC-field is found modelling an electrical solenoid as shown in Fig. 7.12.

The component consists of a coil where a metallic stick are free to move within. The energy

i
x· ·
 = u F
F IC
u
i v = x·

a) b)

Fig. 7.11 A solenoide. a) schematic b) IC- field

stored at any time may be expressed as

2

E   x  = -------------- (7.34)
2L  x 
252 Multiport fields

where  is the flux linkage and L  x  is the inductance function depending on the positon of
the stick. This yields, using the reciprocity condtion for an conservative field, the constitutive
laws for the IC-field as

2
  d
i = ----------- F = – -------------- L  x  (7.35)
Lx 2L  x  d x

7.4 Resistive fields

A R-field is an element described as a set of effort-flow relations at n ports. The symbol is as
for a 1-port R-element, but to distinguish between 1-port elements and R-fields we draw the
mnemonic code in bold as shown in Fig. 7.12. Both explicit and implicit fields do exist.

e2 fi ei e2 fi ei
f2 f2
e1 e e1 e
n n
R R
f1 f f1 f
n n
a) b)

Fig. 7.12 A geneal R-field. a) resistance form and b) conductance form

The constitutive laws for the R-field becomes

e i =  R i  f 1 f 2  f n  i = 1 2  n (resistance causality) (7.36)

–1
f i =  Ri  e 1 e 2  e n  i = 1 2  n (conductance causality) (7.37)

and in the linear case

–1
e = Rf f = R e (7.38)

where R is the resistance matrix.

As for R-elements which gets their causality assigment as a result of the the causality
assigmenent procedure (SCAP), the individual bonds of a R-field may likewise exhibit any
combination of resistance or conductance causality. This means that in addition to the two
 Multiport transformers 253

fundamental causal forms a large number of mixed causality forms are possible. However,

R R R R
4 2 7 4 2 7
1 5 6 3 1 5 6 3
1 0 1 1 0 1
Impossible !!
a) b)

Fig. 7.13 R-field causality example. a) conductance causality b) resistance causality

even for linear R-fields not necessarily all possible causality options do exist. An example
shown in Fig. 7.13a shows an implicit R-field where the explicit constitutive laws are given
in conductance form as

f1 1  R4 –1  R4 0 e1
f2 = –1  R4 1  R4 + 1  R7 –1  R7 e2 f = Re (7.39)
f3 0 –1  R7 1  R7 e3

Questioning the existance of complete resistance causality of this linear implicit field we
simply reassign the causalities starting with flow input on bond 1 2 and 3 . Completing the
causality assignment we observe that we end up with no effort in causality on the 0-junction!
This is computational impossible, and we must conclude that complete resistance causality
do not exist. However, having the constitutive laws of the R-field in conductance form would
be straightforward to compute the same in resistance form simply by inverting the resistance
matrix. But, surpricingly the inverse do not exist and the bond graph causality assignment
procedure and the algebra both reveal this. In general, assigning causality to an implicit field
nicely reveal which causality option are possible and which are not.

In bond graph modelling a R-element normally signals dissipations of energy. The constitu-
tive laws are however not generally restricted in this way and may even supply energy to the
system. The constitutive laws of a R-field follows this behaviour. However, it can be shown
that if all resistances R i j  0 for a linear R-field, then the R-field is dissipative. In general,
power will be dissipated for any linear field if and only if the resistance matrix is positive
semidefinite. For a general nonlinear R-field there is however no simple way to show that a
R-field is dissipative or not.

7.5 Multiport transformers

In Chapter 2 we introduced the 2-port transformer and gyrator. Now, we formally extend the
254 Multiport fields

characteristics of the 2-port transformer to an n port transformer with no accumulation or dis-

sipation of energy. We select to define port 1 to j to be inflow ports and port j + 1 to n to be
outflow ports displaying the power flow through characteristic of a transformer. Such a field
is shown in Fig. 7.13.

1
M j+1

2 j+2
:
j
TF n
:

Fig. 7.14 Multiport transformer

The constitutive laws for the multiport transformer follows from the definition as

T
e in = Me out M f in = f out (7.40)

where e in =  e 1 e 2  e j  T , e out =  e j + 1  e n  T , f in =  f 1 f 2  f j  T
f out =  f j + 1  f n  and M is a j   n – j  matrix, i.e. the transformer modulus matrix. It
T

is straight forward to prove that the constitutive laws results in power flowing into the multi-
port transformer equals the power flowing out of the transformer. Starting from Eq. 7.40, first
compute the transform of the equation Eq. 7.40b and postmultiply by e out . This yields

T T
f in Me out = f out e out (7.41)

which using Eq. 7.40a can be simplified to

T T
f in e in = f out e out (7.42)

which simply is telling us that the power flowing in and out of the transformer equals at all
times, without any accumulation or dissipation.

From the above we observe that there is no restriction on the number of ports on each side of
the transformer and we also observe that the modulus matrix may not be restricted to a con-
stant matrix. This then includes the multiport modulated transformer where the elements of
the matrix may be any function of state variables.

Although not frequently encountered the multiport gyrator follows from the same formal
 Multiport transformers 255

extension as the multiport transformer. The constitutive laws beeing

T
e in = rf out r f in = e out (7.43)

and the gyrator modulus matrix r is a j   n – j  matrix which may be any function of state
variables, consequently also representing the multiport modulated gyrator.

Multiport transformers are often encountered in dynamics where geometric nonlinearities

arise. In bond graphs we are merely interested in relations between velocities, flows, and cor-
responding efforts. The transformations shown in Fig. 7.14, between rectangular and polar

y · y·
r r· Fr M  r   F x
r· x·
r

P  x y 
M TF Fy
· y·

x

Fig. 7.15 Mechanical multiport transformer

coordinates of point P are developed from the expressions for the displacements. Here the
position of the point P  x y  is

–1
r = x2 + y2 ,  = tan  y  x  (7.44)

The velocity relations are easily derived performing a simple differentiation as

dr r dx r dy d  dx  dy
= + = + (7.45)
dt xdt ydt dt xdt ydt

·
which yields assuming that f in =  r·   T and f out =  x· y·  T

x - -------------------
-------------------
y -
r· x +y
2 2 x + y 2 x·
2 T
= M f in = f out (7.46)
·
 x - y·
y - ---------------
– ---------------
x + y x + y2
2 2 2

Utilizing the bond graph properties of a multiport transformer the relations for the corre-
256 Multiport fields

sponding efforts, i.e. the forces F x F y and the force and moment F r M  acting on point P
becomes

x –y -
-------------------- ---------------
Fx x + y2
2 x + y2
2 Fr
= e in = Mf out (7.47)
Fy y x M
-------------------- ----------------
x + y2
2 x2 + y2

·
The relations between velocities r·  and x· y· can also be represented using an implicit field
as shown in Fig. 7.16. where l 2 = x 2 + y 2 . Here the bond graph is simply generated reading

Fr lx Fx
0 MTF 1
r· x·
ly
MTF
–l 2  y
MTF
M l2  x Fy
· 0 MTF 1
 y·

Fig. 7.16 Multiport transformer in implicit field form

the expressions in Eq. 7.15 which in words states that r· is a sum (which is obtained using a
0-junction) of contributions from x· and y· , each modulated (using a MTF) with a modulus
·
and likewise for  . The expression for the individual modulus are established from the direc-
–1
tion of the power bonds, i.e. the expression beeing m or m in general.

REFERENCES
1.S. H. Crandall, et. al. Dynamics of Mechanical and Electro-Mechanical Systems, McGraw-
Hill, New York,1968.
2. D. C. Karnopp, Power-Conserving Transformations: Physical Interpretations and Appli-
cations Using Bond Braphs, J. Franklin Inst., 288, No. 3, 175-201, Sept 1969.
 Multiport transformers 257

PROBLEMS
7-1 In the scalar bond graphs given below, convert the bond graph to explicit fields where
possible.. State the constitutive laws for the new fields.

C I I

0 0 1 I 0 GY C

7-2 A ball is pinned to two springs as showed below. Each spring is considered unloaded
when the angle between the wall and the springs are 45 degrees. Consider small motions
and construct the bond graph of the system using both implicit and explicit fields.

k y
x
m
k

7-3 A body is partly submerged into a water tank as shown below. The force applied to push
the body into the water is F and the position of the body is x . The pressure at the water
outlet of the tank is P and the flow rate is controlled by a valve. Construct a bond graph
of this system and write down the state equations.

F x

7-4 The pressure of the oil in a hydraulic cylinder is given by the expression

x·
Q – V Q·
P = P 0 +  ------------------ F
V

Construct a bond graph of the hydro-mechanical system.

258 Multiport fields

7-5 A massless rod having a length l is acted apon by three forces F 1 F 2 F 3 where F 3 is
always perpendicular to the rod. The corresponding velocities are V 1 V 2 V 3 . Construct
the MTF-field for this system and write down all the constitutive equations using the
power conservative properties of the transformer.

F2 V2
F3 V3
F1 V1

 
 Multibody dynamics 259

Chapter 8

Multibody dynamics

8.1 Introduction
So far we have used bond graphs to formulate increasingly complex mathematical models of
a wide area of dynamic systems and by now the reader should be convinced that bond graphs
do offer major support in creating such models. Through for most engineering disciplines
simple procedures guides the creation of models and by way of causality assignment, the
state variables and equations are easily assembled. Although algebraic challenges may show
up via non linearity, differential causality or algebraic loops, again simple procedures offer to
guide us through the solution of these problems. Modern software tools even have these or
similar procedures implemented and may offer to avoid or solve these problems. However,
even pretty simple models may turn out to be virtually impossible to complete under the
modelling assumptions used. Multibody dynamics represents an area of models with a class
of problems that in general falls in this class.

This chapter introduces multidimensional rigid-body mechanics, the modelling of systems

including such components, and the assembly of these in overall system models. Classical
examples of such systems are general mechanisms, vehicles both on land and sea, robots and
mechatronic systems. In this chapter we first introduce multidimensional dynamics and the
bond graph representation of such systems and then we will discuss some powerful methods
to create even more complex models maintaining the same level of simplicity that we have
been introduced for using bond graphs.

8.2 Multidimensional dynamics

Fig. 8.1 shows a general translating and rotating rigid body moving in space. The rigid body
has a density  , a total mass m and the motion can be described with the velocity v and
angular velocity  . Let x y z be a coordinate frame with origin o , fixed in the rigid body
and moving with it, and let X Y Z be the corresponding inertial coordinate frame with ori-
gin O from where the body is generally observed. The body’s position, velocity and angular
velocity vector are r o , v =  u v w  T and  =   x  y  z  T respectively. The vector from
the origin o of x y z to a differential element of mass dm = dV defines the centre of
mass through
260 Multibody dynamics

z
P
dm

r
rc
rP y
o
z
rO v
O x
y

Fig. 8.1 Rigid body in 3-D motion

1
r c = ----  r dV (8.1)
m
V

where the integral is understood carried out over the whole volume of the body.

The linear momentum p of the rigid body can hence be expressed as

p =    v o +   r  dV = v o   dV +    r dV
V V V (8.2)
= m  vo +   rc 

where the expression inside the last parenthesis is recognized as the velocity of the center of
mass v c such that the linear momentum of the body can be expressed as
p = mv c (8.3)

Newton’s law which simply states that the net forces acting on a body simply is the time
change of momentum as

n
dp
 Fi =
dt
(8.4)
i=1

or expressing Newton’s law with respect to the rotating frame by inserting Eq. 8.2 into Eq.
8.4 we get
 Multidimensional dynamics 261

n
p
 Fi = d  vo +   rc  = +p (8.5)
dt t rel
i=1

where . rel
indicates the rate of change of momentum relative to the moving frame.

In the same way the angular momentum of the rigid body about o L o is defined to be
Lo =  r   vo +   r  dV
V
(8.6)
= – v o   r dV +  r     r  dV
V V

which yields
L o =  I  + r c  p (8.7)

where the inertia tensor  I  is written as

I xx – I xy – I xz
 I  = – I xy I yy – I yz (8.8)
– I xz – I yz I zz

and where I nn is the moments of inertia and I mn is the products of inertia about the center of
gravity given as

I xx =    y + z  dV I xy = I yx =  xy dV
2 2

V V

I yy =    z + x  dV I xz = I zx =  xz dV
2 2
(8.9)
V V

I zz =    x + y  dV I yz = I zy =  yz dV
2 2

V V

Working with the angular momentum we are faced with selecting the reference point or ori-
gin. The transfer principle for angular momentum states that the angular momentum about
any point P equals the angular momentum about the centre of gravity plus the moment of the
linear momentum vector p about P as

LP = Lc + r  p (8.10)

This relation also applies to a fixed point s anywhere on the body or body extended, where s
merely replaces P . Eq. 8.10 constitutes the transfer theorem for angular momentum. This
theorem allows us to generally select any point of an object as the reference.
262 Multibody dynamics

Newton’s angular momentum law states that the net torque about o acting on the body is
equal to the time rate of change of the angular momentum L o about o as

n
dL o L o
  o i =
dt
=
t
+   Lo (8.11)
rel
i=1

Using the right hand rule, it is straightforward to write down the component equations of Eq.
8.x and Eq. 8.10 as

· ·
m  u· – v z + w y – x G   y +  z  + y G   x  y –  z  + z G   x  z +  y   = F x
2 2

· ·
m  v· – w x + u z – y G   z +  x  + z G   y  z –  x  + x G   y  x +  z   = F y (8.12)
2 2

· ·
m  w· – u y + v x – z G   x +  y  + x G   z  x –  y  + y G   z  y +  x   = F z
2 2

and

· · 2 2 ·
I x  x +  I z – I y  y  z –   z +  x  y I xz +   z –  y I yz +   x  z –  y I xy
+ m  y  w· – u + v  – z  v· – w + u   = M
G y x G x z x
· · 2 2 ·
I y  y +  I x – I z  z  x –   x +  y  z I xy + x
–  z I zx
+   y  x –  z I xy
(8.13)
+ m  z G  u· – v z + w y  – x G  w· – u y + v x   = M y
· · 2 2 ·
I z  z +  I y – I x  x  y –   y +  z  x I yz +   y –  x I xy +   z  y –  x I zy
+ m  x  v· – w + u  – y  u· – v + w   = M
G x z G z y z

where r c =  x G y G z G  T is the position of the centre of mass according to reference frame

o . These equations may be put in matrix form as

Mv· + C  v v =  (8.14)

where v =  u v w  x  y  z  T is the rigid body velocity vector, v· is the corresponding

acceleration,  =  F x F y F z M x M y M z  T is the external force/moment vector and M is
the mass matrix which can be written in component form as
 Multidimensional dynamics 263

m 0 0 0 mz G – m y G
0 m 0 –m zG 0 mx G
0 0 m my G – m x G 0
M = (8.15)
0 – mz G my G I xx – I xy – I xz
mz G 0 – mx G – I yx I yy – I yz
– my G mx G 0 – I zx – I zy I zz

and C  v v is Coriolis-centrifugal forces where the matrix C  v  becomes

0 0 0 m y  + z   –m  x  – w  mx  + v
G y G z G y G z
0 0 0 –m  y  + w  mz  + x   –m  y  – u 
G x G z G x G z
–m  z  – v  –m  z  + u  m x  + y  
C v  =
0 0
m  y  + z   –m  y  + w 
0
–m  z  – v 
G x
0
G z G x
–I  – I  – I  I  + I  – I 
G y (8.16)
G y G z G x G x yz y xz x zz z yz z xy x zz y
–m  x  – w  m  z  + x   – m  z  + u  I  + I  – I  0 –I  –I  + I 
G y G z G x G z yz y xz x zz z xz z xy y xx x
mx  + v – m  y  – u  m  x  + y   –I  – I  + I  I  + I  – I  0
G z G z G x G y yz z xy x zz y xz z xy y xx x

These non linear differential equations are known as Euler’s equations and if solved yields
the linear and angular velocities expressed in a translating and rotating reference frame, a
solution which is extremely difficult to interpret. At the same time we have to express the
attached forces aligned with the rotating reference frame.

Eq. 8.12 and Eq. 8.13 represents the general equation of motion of a rigid body. General in
the sense that the fixed point o may be positioned anywhere relative to the centre of mass of
the rigid body. However, if the point o , i.e. the reference frame x y z , is positioned in the
centre of mass and aligned with the principal axes of the body, which makes the product of
inertia equal to zero, Eq. 8.12 and Eq. 8.13 simplifies considerably to

m  u· – v z + w y  = F x
m  v· – w x + u z  = F y (8.17)
m  w· – u + v  = F
y x z

and

I x  x +  I z – I y  y  z = M x
·
I y  y +  I x – I z  z  x = M y (8.18)
·
I z  z +  I y – I x  x  y = M z

This can according to Karnopp [] elegantly be represented into bond graph form, i.e. as an
264 Multibody dynamics

implicit inertia field, as shown in Fig. 8.2.

I:m I:Ix
p· x p· Ix
Fx Mx
1u 1 x
m x m y Iz z Iy y
MG

MG
Y

Y
MG

MG
Y

Y
Fy v w F
z My  y z M
z
1 MGY 1 1 MGY 1
p·y p· z p·Iy p· Iz
m x Ix x
I:m I:m I:Iy I:Iz

Fig. 8.2 Bond graphs for 3D rigid body motion

Here the bond graph appropriately introduce the state variables p x p y p z p I x p I y p Iz and the
gyrator elements represent the cross-product terms in Eq. 8.11 and Eq. 8.12. The modulated
gyrator moduli m x m y  being introduced are dependent of the angular momentum
state variables. As we observe integral causality are preserved as long as the forces
F x F y F z and moments M x M y M z are input to the bond graph. Reading the bond graph
and writing the state equations yields

p p
p· x = m z -----y – m y ----z + F x
m m
p p
p· y = m x ----z – m z -----x + F y
m m
p p
p· z = m y -----x – m x -----y + F z
m m
pI pI (8.19)
p· Ix = I y  y -----z – I z  z ------y + M x
Iz Iy
pI pI
p· Iy = I z  z ------x – I x  x -----z + M y
Ix Iz
pI pI
p· Iz = I x  x ------y – I y  y ------x + M z
Iy Iz

where
 Multidimensional dynamics 265

 x = p Ix  I x
 y = p Iy  I y (8.20)
 z = p Iz  I z

For reference the kinetic energy of a translating and rotating rigid body can generally be
stated as

1
T = ---      v +   r    v +   r  dV
T
(8.21)
2

T T
where v =  u v w  and  =   x  y  z  is the linear and angular velocities. Assuming
that the velocities are described in the moving reference frame o fixed to the body this yields

1 1
T = --- mv o  v o + v o      r dV  + ---      r      r  dV
2 2
(8.22)
1 1 T
= --- mv o  v o + v o     mr c  + ---   I 
2 2

Coordinate transformations

As stated above solving Eq. 8.12 and Eq. 8.13 yields useless results. To obtain useful results
the position, orientation and velocity, plus forces and moments, of the body at all times must
be referenced to a fixed coordinate frame. It is therefore necessary to transform the body-
fixed results into the inertial frame X Y Z through a series of coordinate transformations.
There are however many possible coordinate transformations which can take us from body-
fixed to inertial coordinates and the selection of transformations often depends on the appli-
cation. The transformation most often encountered is the transformation using Euler angles.

Here we will shown how we can develop the required transformation using the Cardan
angles, which corresponds to the familiar yaw, pitch and roll angles of an vehicle, i.e. a auto-
mobile or a ship.

The transformation can be developed through a set of individual rotations around independ-
ent axes as shown in Fig. 8.3. First we rotate about the Z -axis thought the angle  (yaw),
yielding the intermediate coordinate frame x' y' z' . Next we rotate about the new y' -axis
through the angle  (pitch), yielding the x'' y'' z'' axes. And finally we rotate about the x'' -
axis through the angle  (roll), yielding the body-fixed reference frame x y z . Rewriting
these transformations in matrix form yields
266 Multibody dynamics

Z z’
z’’
z 
 y

y 
z ·
 y’ y’’
· 

Y
·
x 1. Rotation about Z  (yaw)


2. Rotation about y'  (pitch)
X
x’ 3. Rotation about x''  (roll)
x’’ x

Fig. 8.3 Cardan angle coordinate transformation.

 x'' x
c
1 0 0 
X
0
– s
MTF
MTF
s
MTF
1

x'

 y'' = 0 cos  – sin   y 0

c
1 
 MTF
Y y'
 z'' 0 sin  cos   z 
Z
0 1 
z'
(8.23)

 x' cos  0 sin   x''

c
 0 MTF 1 
x' x''
 0 1 
 y' = 0 1 0  y'' y' s
MTF
– s
MTF
y''

– sin  0 cos   z''

c
 z' 
z'
0 MTF 1 
z''
(8.24)

X cos  – sin  0  x'


x''
0
c
1 
z
0 MTF 1
 – s 
Y = sin  cos  0  y' y'' MTF
s
MTF y
c

Z 0 1  z'  0 MTF 1 
0 z'' y

(8.25)

and where the bond graph representation of each transformation is given also.

Then knowing  =   x  y  z  T we can determine the angular velocity components in all

frames including the inertial frame simply using

X x
Y = R Z  R y'  R x''   y (8.26)
Z z
 Multidimensional dynamics 267

where

c – s 0 c 0 s 1 0 0
R Z  = s c 0 R y'  = 0 1 0 R x''  = 0 c – s (8.27)
0 0 1 – s 0 c 0 s c

where c = cos    s = sin    and the corresponding bond graph transformation is

c c
0 MTF 1 0 MTF 1  0 1 
 – s s  x'' z
X MTF MTF x'
c c
 0 MTF 1  0 1  0 MTF 1
Y y' s – s y'' – s s 
MTF MTF MTF MTF y
c  c
z''
 0 1  0 MTF 1 0 MTF 1 
Z z' y

Fig. 8.4 Rigid body to inertia frame transformation in bond graph notation

shown in Fig. 8.4.

It is straight forward to show that the relation between the inertial frame linear velocities and
the body-fixed velocities follows from the same transformation as

uX ux
uY = R Z  R y'  R x''  u y (8.28)
uZ uz

For this transformation consisting of three consecutive power conservative transformations

as discussed in Chapter 7, we can show that even the product of transformations is power
conservative. Hence we can utilize the constitutive laws of a multiport transformer to express
the relations between forces and moments in body-fixed and inertial reference frames respec-
tively or we can utilize the implicit field shown in Fig. 8.4 and write the force/moment equa-
tions directly from the bond graph. However, utilizing the constitutive law of the modulated
transformer we simply get

Fx FX
F y =  R Z  R y'  R x''  T FY (8.29)
Fz FZ
268 Multibody dynamics

Mx MX
My =  R Z  R y'  R x''   T M Y (8.30)
Mz MZ

and where the transformer modulus matrices R Z  R y'  R x''  are dependent of the Euler
angles    only.

Connecting the power bonds representing the body-fixed linear and angular velocities shown
in Fig. 8.2 to the corresponding velocities of the transformation bond graph as shown in Fig.
8.4 now yields a complete bond graph model for a rigid body in 6 degrees of freedom. How-
ever, to close the equation set the Euler angles    must be obtained to evaluate the
actual transformations matrices required. Observing the vectorial relations between the rate
· · ·
of change of the Euler angles    and the body-fixed components of  x  y  z from Fig.
8.3 we may write directly from the figure

· ·
 x =  –  sin 
· ·
 y =  cos  +  cos  sin  (8.31)
· ·
 z = –  sin  +  cos  cos 

From the state equations Eq. 8.13 and Eq. 8.17 we obtain   x  y  z  T as a function of time
· · ·
and simply solving Eq. 8.31 for      T we get

·
 =  x + sin  tan   y + cos  tan   z
·
 = cos   y – sin   z (8.32)
· sin  cos 
 = ------------  y + ------------  z
cos  cos 

Next integrating Eq. 8.32 we obtain the Euler angles    ,i.e. the orientation of the body,
which inserted into the transformations R Z  R y'  R x''  enable us to compute the transforma-
tion matrices and as such complete the bond graph model of a rigid body motion in 6 degrees
of freedom. This now indeed constitute a very nice package for analysis of rigid body motion
by way of simulation, i.e. specifying the forces and moments acting on the rigid body we are
able to predict the position given as  X Y Z  T and orientation      T of the body at all
times.

Here we have not discussed the causality of the transformations or the implicit inertia field,
however as for now let us simply agree to always use complete integral causality on this
model. This will prevent serious complications which we will introduce solutions for in a
 Multidimensional dynamics 269

later chapter.

A nice shorthand bond graph representation using vector bond graph notation, defined in Fig.
8.5c, for the complete 3D mechanics is shown in Fig. 8.5a. Here each of the transformation

a)
 Eq 8.32

 F I M I  R Z  R y'  R x''   F i M i 
1 MTF MTF MTF 1 I:M
u  u  u  u  u  u 
X Y Z x y z
  X  Y  Z       
x y z

F x 1 F x 2
F x 1
u1 R u2
u1
F y 1 F y 2
v1 MTF v1 Vector
F y 1
v1
F z 1 F z 2 bond graph
F z 1
w1 w1
w1
b) = M x 1
M x 1 M x 2  z 1
 z 1 R  z 2 M y 1
M y 1 M y 2 c)  y 1
 y 1 MTF  y 2 M z 1
M z 1 M z 2  z 1
 z 1  z 2

Fig. 8.5 Shorthand bond graph of 3D mechanics including transformations

matrices are shown as individual MTF-fields, but actually being doubled due to transforma-
tions of both linear and angular velocity components. This is illustrated in Fig. 8.5b and the
complete transformation matrix m y'  , i.e. the transformer modulus matrix relating the two
velocity vectors  u' v' w'  x'  y'  z'  T and  u'' v'' w''  x''  y''  z''  T , simply becomes

c 0 s 0 0 0
0 1 0 0 0 0
m y'  –
= s 0 c 0 0 0 (8.33)
0 0 0 c 0 s
0 0 0 0 1 0
0 0 0 – s 0 c

The two additional transformations m Z  and m x''  are established in a similar way. The
I:M field included to the right in Fig. 8.5 is to be understood as a shorthand notation for the
gyrator rings shown in Fig. 8.2. We will later in this chapter discuss a more general imple-
mentation of equation of motion for a rigid body.
270 Multibody dynamics

8.3 A modelling procedure

Modelling mechanical systems we often too quickly experience major problems trying to
extend even simple examples from our basic dynamic courses to include more realistic prob-
lems. However, utilizing bond graph methodology it has proved possible to unify major con-
tributions from many famous scientists as Newton and Lagrange, and to approach such
systems in a convenient and systematic way. The purpose of this section is to extent our basic
bond graph modelling approach to more realistic problems including large-scale motion of
rigid bodies and particles, both linear and non-linear systems and to bring this class of prob-
lems into the multiport-systems context.

Given a general mechanics problem involving rigid bodies, particles and elastic elements the
objective is to create a set of first order differential equations which enables us to study the
dynamic characteristics of the system. The equations which we may write down can gener-
ally be accepted to have the following form:

C-fields: eC = C  qC  (8.34)

I-fields: vI = I  pI  (8.35)

q· C = T CI  q C v I
Junction structure: (8.36)
p· I = – T CI  q C  e C
T

where e C is the forces defining the potential co-energy, q C is the displacements defining the
potential energy, v I is the inertial velocities defining the kinetic co-energy, p I is the
momenta defining the kinetic energy q· C and p· I are the time derivatives of q C and p· I . re Eq.
8.34 which may be both linear or nonlinear arises directly from the constitutive laws of the
compliance elements and Eq. 8.35 is obtained from the inertia elements in the system. The
relations given in Eq. 8.36 gives the nonlinear coupling of the C- and I-fields introduced by
the geometry of large-scale motion of the system. These relations expressed by the transfor-
mation array T CI  q C  and its transpose, are often regarded as the most difficult aspect of
mechanics. Reformulating Eq. 8.36 using the definitions given in Eq. 8.34 and Eq. 8.35 we
obtain

q· C = T CI  q C  I  p I 
(8.37)
p· I = – T CI  q C   C  q C 
T

which can be identified as the state equations of a general linear, or non-linear, conservative
systems expressed in terms of the state variables q C and p I . A bond graph illustrating such a
 A modelling procedure 271

system is shown in Fig. 8.6.

qC
q· C 1 v I 1 p·
I 1
1 1

C :

q·
1
C N
:
MTF 1
v I M
p· I M
I
Fig. 8.6 Bond graph models of general nonlinear conservative mechanical system

Constructing such a bond graph consisting of N compliances and M inertias the tricky thing is
to select and relate the velocities q· C 1 q· C N and v I 1 v I M , which in bond graph terminol-
ogy can be expressed through the N  M port transformation matrix or MTF modulus matrix
m  q C  which is a function of q C .

k2

C 1  r   1 I
g k1 r
y
r· vx
 MTF
m vx
M/r
C 1· 1 I
x F  vy
vy
a) b)

Fig. 8.7 Non-linear mechanical oscillator a) schematic b) bond graph

As an illustration of such a system consider the nonlinear oscillator in Fig. 8.7. Here the
mass, connected to the point of rotation with a spring, is free to slide along a massless rod.
The rod is likewise constrained by a rotational spring at the rotational point. We may easily
write the equivalent to Eq. 8.34 - Eq. 8.36 as

F = k1  r – R 
M = k2 
(8.38)
vx = px  m
vy = py  m

where R is the free length of the rod spring, F is the force in the spring k 1 and M is the
torque of the rotational spring k 1 and k 2 are spring constants. Selecting q C =  r   T and
v I =  v x v y  T , the relation between the velocities q· C and v I may be found be studying the
velocity components in Fig. 8.7 as
272 Multibody dynamics

r· = sin  v x + cos  v y sin  cos 

· cos  sin  T CI  q C  = cos  sin  (8.39)
 = ------------ v x – ----------- v y ------------ – -----------
r r r r

By careful investigation of the force F and torque M in the x and y directions we obtain the
relation for p· I as

cos  cos 
p· x = – sin F – ------------ M – sin  – ------------
r T r
– T CI  q C  = (8.40)
sin  sin 
p· y = – cos F + ----------- M + mg – cos  -----------
r r

where we observe that the negative transpose of T CI , which relates the velocities, defines the
relation in the state equations for p· I . And as so, having obtained T CI relating the velocities,
we can omit deriving the relations between p· I and  F M  T , which in general is much more
complex and error prone. Hence, we get away with having to do only half of the job! This is
due to the conservativeness of the formulation of the state equations, something which the
bond graph method is based on.

In general it is often difficult to relate directly the velocities v I and q· C as we have done here.
It is also often the case that there is more velocities and displacements in the problem than
coupling constraints permit to be independent. For this reason we define a third set of dis-
placements, the kinematic displacements or the generalized coordinates q 1 q 2  q n and
the generalized coordinate vector q k =  q 1 q 2  q n  T . The generalized coordinates are
the smallest number of independent displacement variables which uniquely defines the con-
figuration of the system at all times. This then enables us to formulate the vector q C as a
function of q k , and hence the velocity relations between these two vectors are found by
direct differentiation. Then, by similar reasoning, if q k uniquely defines the motion of the
system, then the vector v I can be found as a function of q· k given q k . Having established the
relations of both q· C and v I as functions of q· k q k , relating q· C and v I can be solved for or
utilized implicitly. The selection of generalized displacements is often not clear or there may
be several possible selections. One selection of generalized variables may end with a largely
simplified relation between q C and q k , which simplifies the work to establish the respective
velocity relations, but may introduce complex relations between v I and q· k on the other hand.
Alternative selections should be checked if too complex relations are experienced. The gen-
eral objective should be to end up with as few and simple velocity transformations as possi-
ble.

A procedure following the approach just discussed is given next followed by an example
worked out in detail clearly indicating the stages of the procedure.
 A modelling procedure 273

Procedure:

1. Determine the generalized variables q k , q C and v I . Then write down the 1-junctions cor-
responding to the q· C , q· k and v I , preferably in columns in that order.

2. Obtain the displacement relating q C and q k .

q C =  Ck  q k 

where  Ck is the set of relations defining each element of q C in terms of q k .

Differentiate the relations with respect to time to obtain the velocity relations as
 Ck  q k  ·
q· C = --------------------
- q k = T Ck  q k q· k
q k

where the i-th relation becomes

N
 Ck i  q k 
q· C i =  ------------------------ q· k j
q k j
j=1

where N is the number of displacements in q k .

Write down the resulting bond graph transformation using MTF’s and 0-junctions.

3. Obtain the velocity transformations relating v I and q· k . Selecting q k with some plan this
can be done directly, i.e. q· k = v I or use differentiation as in 2 which yields

v I = q· I =  Ik  q k 

 Ik  q k  ·
v I = q· I = -------------------
- q k = T Ik  q k q· k
q k

where again the i-th relation becomes

N
 Ik i  q k 
v I i = - q· k j
 ----------------------
q k j
j=1

Write down the resulting bond graph transformation using MTF’s and 0-junctions.
274 Multibody dynamics

4. Now add the C- and I- elements to the corresponding 1-junctions, i.e. connect one I-ele-
ment to each v I 1-junction and connect one C-element to each q C 1-junction.

5. Append diffusion effects, sources etc. connecting R-elements and Se- or Sf-elements to
corresponding 1- or 0-junctions.

6. Simplify the bond graph as normal for bond graph modelling.

Example 8.1

Let us develop a bond graph model of the systems given in Fig. 8.8 walking thought the pro-
cedure just introduced. The origin of the coordinate system is placed in o where the y-axis is
pointing downwards. The individual numbered stages becomes

o x
y
yc k
mb~0
L

m
vx
vy

Fig. 8.8 Example 8.1

1. Determine the variables of interest q k q C v I .

Generalized variables are: (other are possible too)

·
q =  y   T
k c q· =  y·    T
k c

Potential energy variables: (stretch of spring)

q =  y – y T
C c 0 q· =  y·  T
C c

Inertia variables becomes: (displacement and velocity of mass m )

q =  x y  T
I v =  x· y·  T
I

where
q I =  L sin  y + L cos   T
 A modelling procedure 275

We now draw the corresponding 1-junctions as shown below, the q· k 1-junctions in a row
in the middle, the q· I 1-junctions, i.e. v I , to the right and the q· C 1-junctions to the left as

1· 1· 1·
yc yk x

1· 1
k y·

2. Starting from the expression for q C =  y c – y 0  T , where y 0 is the length where the
spring is at rest, we calculate the derivative according to

dq q C dy k q C d k
q· c = --------C- = + = y· k
dt  yk d t  k d t

which gives us the relation for the T Ck matrix as

T Ck =  1 0 

Inserting this relation into the system bond graph simply connecting y· k and y· c using a
single power bond is shown in the figure below. The second element in T Ck which relates
y· c and  k is zero, hence there is no power flow between these 1-junctions and the power
bond is simply omitted.
3. Performing exact the same operation to obtain the relations between q· k and v I yields

· ·
q· I = v I =  y· c – L sin  L cos   T

which may be written in matrix form as

x· = 0 L cos  y· c = T q·
1 – L sin  ·
Ik k
y·

Inserting these relations into the system bond graph below observing that the y· flow is
·
the sum of a contribution from the y· 1-junction and a second contribution from the  1-
c
276 Multibody dynamics

junction. Recognizing that the relation is a flow-flow relation we insert an MTF-element

for each as show in the figure below

1
y· k 0
1· 0 MTF 1 MTF 0 1·
yc x
1  L cos  
MTF MTF
0
MTF 1 MTF 0 1
·
k  – L sin \alphya  y·

4. Now we are ready to inset the representative I- and C-elements attaching each on their
representative 1-junction, i.e C-elements on their q· 1-junctions and the I-elements on
C

their v I 1-junctions as

1
y· k 0
k:C 1· 0 MTF 1 MTF 0 1· I:m
yc x
R 1  L cos  

Se
MTF MTF
0
MTF 1 MTF 0 1· I:m
·  – L sin   y
k

5. Also recognizing the gravity present, we insert a Se-element on the y· 1-junction and also
assuming some friction in the junction where the pendulum is suspended we insert a R-
element here. This makes the acausal bond graph model complete.

6. However it is possible to simplify the bond graph somewhat, recognizing some

“through”-bond situations and also the 0-moduli of the MTF’s which indicates no power
flow and consequently can be removed completely. The complete simplified bond graph
after causality assignment then becomes

1
y· k  L cos  
k:C 1· MTF 1 MTF 0 1· I:m
y c x
Se

 – L sin  
R 1 MTF 0 1 I:m
·
k y·
 A modelling procedure 277

Example 8.2 Moving pendulum

Yet another example may be shown to further emphasize the strength of the procedure. The
example introduced in Fig. 8.9 shows a vehicle with mass m 0 free to move along a horizon-

x· 0
k:C 1 I:m0
x x· o
m0

0 I:m Se:mg
y  R x·
m ·
R
x u = x· 0 + x· MTF 1· MTF 1 I:m
v = y· R cos   R sin  y·
a) b)

Fig. 8.9 Example 2 - Moving pendulum a) schematic b) bond graph

tal path having a pendulum hinged at the centre of mass. The length of the pendulum is R
and with a point mass m . The different variables are defined in Fig. 8.9. Following the proce-
dure we obtain:

1. Variables of interest

q k =  x 0   T qC =  x0 T v I =  x· 0 x· y·  T

2. Establish expression for q· C

q C = 1x 0 + 0 = x 0

q C · q C · x·
q· C = x0 +  = T Ck q· k =  1 0  0
 x0  ·


3. Establish expression for v I

x0 x0 1 0 x·
qI = x = x 0 + R sin  v I = q· I = T Ik q· k = 1 R cos  0
·
y – R cos  0 R sin  
278 Multibody dynamics

Drawing the 1-junctions of q· C q· k v I in rows, adding the respective I- and C-elements, add-
ing possible R-elements (none here) and Se/Sf-elements (remember the gravity) and simplify-
ing the bond graph completes stage 4-6 in the procedure and yields the bond graph shown in
Fig. 8.9b. Straight forward, but continuing assigning causality results in an bond graph model
in differential causality. However, differential causality we do have a strategy to follow and it
does not pose a problem although the algebra may become tricky. Here, the reader is urged to
check out the hurdles ahead. The necessary manipulation yields two implicit state equations
which algebraically has to be solve for to obtain the final state equations. The only comfort is
that using bond graphs and following the procedure you end up with solving the problem
with as little algebra as possible.

Using this procedure modelling most mechanical systems is an extremely efficient way to
develop models of such systems. However, you will often experience that the bond graphs
obtained will be showing derivative causality or having algebraic loops as we just has experi-
enced. This is inevitable, any two inertia coupled together will make one or more velocities
dependent of the other, consequently any model will display derivative causality on one or
another state variable. The field of multibody systems, or robotics, is typically experiencing
derivative causality when connecting bodies.

Lots of solution techniques are available and bond graphs is not necessarily better, but when
it comes to interconnecting multibody systems with other systems and control systems, it has
proved to offer distinct advantages. We therefore continue to show how methods from classi-
cal analytical mechanics may serve alongside bond graphs to yield an especially powerful
method of solution. Next we will introduce methods especially interesting, powerful and yet
intuitive and easy to use, especially when combined with bond graphs.

8.4 Lagrangian or Hamiltonian IC-field modelling

As experienced in the last section although having powerful procedure to establish a bond
graph of most mechanical systems, developing the state equations often prove to be quite a
mouthful due to algebraic loops and differential causality. From analytical mechanics the
Lagrange’s method has proved particularly useful for such situations and in this section we
show how to include the power of the Lagrange’s method in a Hamiltonian form into bond
graph by way of a special type of IC-field. Although not in general showing the explicit cau-
sality options we obtain a field description of the tricky part of the system and still utilize the
causality to define interfaces and to connect systems.

Expressing the kinetic energy T and the potential energy V of a system as functions of a set
of generalized variables q k , (from now on we omit the subscript k) the equation of motions
for the system can be expresses as.
 Lagrangian or Hamiltonian IC-field modelling 279

d T T V 
-----  ·  –  – = Ei i = 1 2    N (8.41)
dt   q i   q i  q i

where q i is the i-th generalized displacement coordinate of the generalized coordinate vector
q k and E i is the generalized forces for the i-th coordinate including those forces which can
be derived from dissipation terms. Deriving the equation of motion using the Lagrange’s
method results in N second order differential equations which has to be solved simultane-
ously. These equations can be transferred to first order equations to be solved, but there are
other disadvantages using the Lagrange’s method directly. However, using a modification of
the Lagrange method, the Hamilton method, directly produces 2N first order equations and
with less algebra, a form which is more known to us as state space equations. Using this
approach we again propose a strict procedure.

1. Assume that expressions for the kinetic T  q·  q  and potential V  q  energy is available as
function of the N generalized displacements q =  q 1 q 2  q N  T . Then define the cor-
responding generalized momentums p i of p =  p 1 p 2  p N  T defined by the expres-
sion

T
pi = i = 1 2  N (8.42)
 q· i

This enables us to write the individual state equations, i.e. Eq. 8.41 as
T V
p· i = – + Ei i = 1 2  N (8.43)
 qi  qi

T V
p· i = e' i + E i e' i = – (8.44)
 qi  qi

or

p· = e'  q·  q t  + E (8.45)

2. The expression for the generalized momentum p i , i.e. from the definition of it, always
shows up a special form when the actual differentiations are carried out, here written in
matrix form as

p = M  q t q· + a  q t  (8.46)
280 Multibody dynamics

where M  q t  is a N  N symmetric matrix of elements each possibly functions of q

and the time t and a  q t  a vector of elements which only occurs if the system include
time-varying velocity sources. Solving for q· gives an expression for the rate of change of
the generalized displacements as

–1
q· = M  q t   p – a  (8.47)

Now observe that the two equations for the rate of change of generalized momentum p·
and rate of change of generalized displacement q· is just the equations that we will search
for when modelling a general mechanical system. Solving for q· do involve a matrix
inversion which in the worst case will have to be performed repeatedly as integration pro-
ceeds, but often is found to be constant or so small that the inverse can be given analytti-
cally.

These two equations, i.e for q· and p· , can according to Karnopp [13] nicely be imple-
mented into a bond graph model using the IC-field as shown in Fig. 8.10.

IC IC
q  i C p  i I

p· 1 p· 2 p· N

e' 1 e' 2 e' N

1 1 1
a)  a)
E 1 q· 1 E 2 q· 2 E N q· N
1

Fig. 8.10 Hamiltonian form of Lagrange’s equations in a bond graph IC-field representa-
tion a) N-size bond graph IC-element and b) {i}-th port name convention

Although the equations for the IC-field may seem complex, we must remember that for
many systems only a few displacements are coupled and the system involved being mod-
elled as an IC-field often is of limited size. This approach solving the differential causal-
ity problem in a local or smaller part of the total system first, may therefore result in a
useful model of a large and complex system without too much algebra.

Implementation
Writing the state equations or implementing the IC-field approach in for example 20-sim it
 Lagrangian or Hamiltonian IC-field modelling 281

has proved clarifying to view the 2N ports of the IC-field as pairs and to define two ports
number i with the names

p  i I and p  i C i = 1 2  N (8.48)

being the port associated with the generalized momentum and the port being associated with
the generalized displacement respectively, and with preferred causality flow out on the I-port
and effort out on the C-port respectively. If more than one generalized coordinate is present
additional pairs of ports are created.

Next, to obtain p and q we simply assign initial conditions of each and integrate the efforts
and flows of the power bonds associated with the momentum and displacement variables
respectively.

Manually or by using symbolic manipulation packages to perform the necessary algebra to

obtain expressions for the momentums and generate the M-matrix symbolically we have to
enter this as procedures into our simulation software. It is often experienced that this matrix
is constant and so small that the symbolic inversion can be performed to yield expression for
the M – 1 matrix. If the matrix is constant of if only a few elements changes as the simulation
continues these elements can be updated while the unchanged elements may be computed in
an initial section of the software used.

Having solved for q· using the inverse of the M -matrix we assign the displacement rates to
the output flows of the I-named ports as

p1I.f q· 1
p2I.f = q· 2 (8.49)
: :
pNI.f ·
qN

Next we implement the expressions for

T V
and (8.50)
 qi  qi

which we have derived analytically to form

T V
ei = – (8.51)
 qi  qi
282 Multibody dynamics

and finally we assign these variables to the output efforts of the N C-named ports as

p1C.e e' 1
p2C.e = e' 2 (8.52)
: :
pNC.e e' N

This then completes the implementation of the IC-field, i.e. being the constitutive law of the
IC-field. Although complex there is a clear structure and having done this once, it should be
no big deal to use this method for new systems provided that the expressions for the kinetic
and potential energy is available.

Example 8.3 Hamiltonian IC-field bond graph model of a moving pendulum.

x· 0 IC
x
m0 p· x p· 
0

e' x e' 
0
y  R
·
m R 1 x· 0 1 · 0
x u = x· 0 + x·
a) b)
v = y·

Fig. 8.11 Example 8.2 Moving pendulum example a) schematic b) bond graph

Now let us show if we can avoid the problems experienced in Example 8.2 by using the IC-
field representation. We start by selecting the generalized coordinates as x 0 and  .

The kinetic and potential energy expressions becomes

1 1 1 2
T = --- m 0 x· 0 + --- m   x· 0 + x·  + y· 
2 2 2
and V = – mgy + --- kx 0
2 2 2

Introducing x = R sin  and y = – R cos  into these expressions yields

1 1 · 2 · 2 1 2
T = --- m 0 x· 0 + --- m   x· 0 + R cos   +  R sin    and V = – mgR cos  + --- kx 0
2
2 2 2
which finally with a little manipulation becomes

1 1 2·2 ·
T = ---  m 0 + m x· 0 + --- m  R  + 2R cos  x· 0 
2
2 2
 Lagrangian or Hamiltonian IC-field modelling 283

Performing the differentiation necessary to define the generalized momentums p x 0 and p 

yields

T ·
p x0 = =  m 0 + m x· 0 + mR cos 
 x· 0

T 2· ·
p = · = mR  + mR cos x 0

which written in matrix form becomes
p x0 m + m 0 mR cos  x· 0
= + 0 p = Mq· + a
mR cos  mR ·
p 2
 0

·
Solving for q· =  x· 0   T requires computing the inverse of the matrix M which we observe
is a function of  , i.e. q 2 and hence must be calculated at each simulation step. This yields

R2 – R cos 
q· x0 x· – R cos  1 + m 0  m p
= 0 = -------------------------------------------------- x0

q·  · 2
mR 2 sin  + m 0 R 2 p 


which also constitute the first part of the state equations.

Next we perform the necessary differentiations to establish e' =  e' x0 e'   T . i.e.

T V
e' x0 = – = 0 – kx 0
 x0  x0
T V ·
e'  = – = – mR sin  2 x· 0 – mgR sin 
 

which completes the derivation of the constitutive laws of the IC-field. The implementation
of this into the bond graph IC-field is given in Fig. 8.11b. As we observe the bond graph
exhibits complete integral causality, the necessary manipulation to bypass the problem has
been conducted while deriving the constitutive laws of the IC-field. The price to pay is hav-
ing to compute the inverse of the M -matrix for each simulation step.

It should be observed that if one of the elements in e' =  e' x0 e'   T equals zero, the corre-
sponding power bond can be deleted, because the state do not change and no power flow
occur.
284 Multibody dynamics

There are no external forces in this case, however adding friction in the bearings of the pen-
·
dulum can easily be carried out adding a R-element to the  1-junction and in a similar way
adding friction towards the road, an external force F  t  or a damping feature of the spring
simply requires a R-element added to the x· 0 1-junction.

The second part of the state equations can be written from

T V
p· i = – + Ei i = 1 2  N (8.53)
 qi  qi

as

p· x0 = – kq x0
– mR sin q   – R cos q  p x0 +  1 + m 0  m p   2  R 2 p x0 – R cos q  p   (8.54)
p·  = ------------------------------------------------------------------------------------------------------------------------------------------------------
2 3
- – mgR sin q 
 mR 2 sin q + m 0 R 2 

·
where the solution for x· 0  has been inserted.

8.5 General rigid body bond graphs

Previously we have introduced a bond graph representation for general rigid body dynamics,
and although the bond graph presented has the capability to represent most cases it is still
interesting to extend the capabilities of bond graphs even further. In this section we discuss
the extension of the bond graph representation to problems where the fixed body reference
frame is not positioned in the centre of mass and the case where the products of inertia is not
equal to zero. These extensions are formulated using vector bonds which enables us to in a
compact and effective way to construct bond graph models where multidimensional rigid
bodies are present. Combining these models as building blocks into system models using tra-
ditional bond graph modelling yields a powerful and efficient method for attaching even the
most demanding systems.

Continuing our development from section 8.2 having established the equation of motion for a
rigid body we may express the kinetic energy of the rigid body as

1
T = ---      v +   r    v +   r  dV
T
(8.55)
2

T T
where v =  u v w  and  =   x  y  z  is the linear and angular velocities respec-
tively. Assuming that the velocities are described in the moving reference frame o fixed to
 General rigid body bond graphs 285

the body this yields

1 1
T = --- mv o  v o + v o      r dV  + ---      r      r  dV
2 2
(8.56)
1 1 T
= --- mv o  v o + v o     mr c  + ---   I 
2 2

Neglecting the potential energy for now, the equations of motion of the rigid body can be
expressed using the quasi-Lagrange equations according to Meirovich [21] as

d  T T
+ = v
d t   v v
(8.57)
d  T  T T
+ +v = 
d t     v

where v = v o and  =  . Utilizing the skew-symmetric matrix form of a vector a

0 –az ay T
 a  = – a 
a = az 0 –ax (8.58)
a  b =  a b
–ay ax 0

enables us to write the necessary equations using simple matrix manipulations. The required
terms to be inserted into Eq. 8.57 renaming v 1 = v , v 2 =  and r G = r c then becomes

T
= mv 1 – mr G  v 2 = mv – m  r G v 2 (8.59)
 v1

T
= – m  v 1  r G  + Iv 2 = m  r G v 1 + Iv 2 (8.60)
 v2

Inserting Eq. 8.59 and Eq. 8.60 into Eq. 8.57 yields

d  T  =  · + v·  r  + v   mv – mr  v 
m v1
d t   v 1 2 2 1 G 1
(8.61)
· ·
= m  v 1 –  r G v 2 –  v 1 v 2 –   v 2 r G v 2 

and
286 Multibody dynamics

d  T  = · ·
mr G  v 1 + Iv 2 + v 2   m  r G  v 1  + Iv 2  +
d t   v 2
v 1   mv 1 – mr G  v 2  (8.62)
·
= m  r G v· 1 +  I v 2 + m   v 1 r G v 2 –  Iv 2 v 2
– m  v 1 v 1 –   v 2 r G v 1

Restating this into the standard form for the equation of motion yields

Mv· + C  v r G v =  (8.63)

where v =  v 1 v 2  T and  =   v1  v2  T is the generalized forces correspondingly, the

mass matrix M is recognized to be

mI –m  rG 
M = (8.64)
m  rG  I

where I is the identity matrix. The remaining terms which are known to contribute the corio-
lis and centripetal forces simply becomes

0 – m  v 1  – m   v 2 r G 
C = (8.65)
– m  v 1  – m   v 2 r G  m   v 1 r G  –  Iv 2 

In this derivation we have neglected the potential energy V of the rigid body and also the
Reyleigh’s dissipation function P , i.e. terms that dissipates energy like friction and damping.
This can however easily be included to form the remaining terms of the equation of motion
for a general mechanical system. However, these effects are easily added at the bond graph
level at a later stage.

The total inertia of the rigid body, i.e. the momentum p =  p 1 p 2  T where  p 1 p 2  T are the
linear and angular momentum respectively, is now expressed via the mass matrix showed in
Eq. 8.64 as p = Mv . The inertia can be divided into the translational and angular inertia cor-
responding velocity v 1 and angular velocity v 2 . These velocities are initially defined in a
multibond bond graph notation as the 1-junctions v 1 and v 2 as shown in Fig. 8.12. The iner-
tia forces corresponding to these velocities are shown added to the 1-junctions using the
multiport I -field, I:M, where both v 1 and v 2 are input ports. The constitutive laws for this
elements then relates the velocities and momentum inverting the mass matrix as
 General rigid body bond graphs 287

–1
v1 mI –m  rG  p1
= (8.66)
v2 m  rG  I p2

and where p 1 and p 2 are the linear and angular momentum of the rigid body. Notice that the
mass matrix is a symmetric almost full matrix for the case where r G  0 , but simplifies con-
siderable when r G = 0 and even to a diagonal matrix when also the products of inertia is
equal to zero. This features will be exploited later.

I:M

v2

1 MGR MGR 1
v1 v2
v1

Se: 1 Se: 2

Fig. 8.12 Rigid body bond graph model for r G  0 and I i j = 0 for i  j .

The remaining terms in the equation of motion as shown in Eq. 8.65 are identified using bond
graph terminology as relations between efforts and flows and hence could be represented by
a R-field. However R-elements or fields generally dissipates and communicate energy dissi-
pation, but there is no energy dissipation in this case. Consequently it would be better to rep-
resent these relations with a modulated gyrator MGY or following Allen [15] a modulated
gyristor MGR. This element nicely represents the gyroscopic relations of Eq. 8.65. Adding
the required MGR-element to the v 1 1-junction we are generally faced with the requirement
to also know v 2 . to compute the efforts. This can be achieved by extracting a multiport signal
from the v 2 1-junction carrying v 2 . Following the same procedure we see that the same situ-
ation occurs at the MGR-element connected to the v 2 1-junction and a multibond signal
introducing v 1 into this element is necessary also. Adding the generalized forces, i.e. forces
 1 and moments  2 as Se-elements to the respective 1-junctions, expressed in the fixed ref-
erence frame, we end up with the bond graph shown in Fig. 8.12. The bond graph given here
is general in the sense that the origin of the coordinate system selected does not need to be
placed in the centre of mass, i.e. r G  0 , and the inertia tensor may include the product of
inertias making the inertia tensor full.

If, however, the origin of the coordinate frame is selected to be at the centre of mass, i.e.
r G = 0 , then the bond graph can be simplified considerably decoupling v 1 and v 2 in the
mass matrix making the constitutive law for the linear momentum I-element simply
288 Multibody dynamics

1
v 1 = ---- I p 1 (8.67)
m

and the constitutive law for the angular momentum I-element is

–1
v2 =  I  p2 (8.68)

Note the mass matrices are constant and computing the inverse only need to be performed
once. If also the product of inertia is zero, the inertia tensor becomes diagonal and computing
the inverse is even further simplified.

Also the constitutive laws of the MGR-element connected to the v 2 1-junction simplifies
considerably. Recognizing that  v 1 v 1 = 0 removes the need for an active signal into the
MGR making it a constant gyristor GR as shown in Fig. 8.13. However, the MGR connected
to the v 1 1-junction is unchanged, still requiring the input signal bond.

I:M I:I
v2

1 MGR GR 1
v1 v2

Sei Seb

Fig. 8.13 Rigid body bond graph model for r G = 0 and I i j  0 for i  j .

These multibond bond graphs representing the equation of motion of a rigid body can be
even further simplified combining v 1 and v 2 into one single 1-junction representing
v =  v 1 v 2  T . Both the inertia elements of Fig. 8.13 then combines into one I-field as in Fig.
8.14 where the constitutive law for the I-element is as given in Eq. 8.67 and where the com-
putation of the inverse follows the simplifications discussed above.

The gyristors connected to the v 1 and v 2 1-junctions are consequently combined into a sin-
gle gyristor field GR connected to the v 1-junction and where the constitutive laws in the
general case becomes

e MGR = C  v r G v (8.69)

where e MGR is the vector of efforts of the gyristor multibonds and C  v r G  is equal to Eq.
8.65 and simplifies according to what was discussed above. The complete bond graph intro-
 General rigid body bond graphs 289

ducing these simplifications in bond graph notation is shown in Fig. 8.14a.

I:M IC:M

P· v
P· v
e' e' v
GR 1
 v 1 v 2 
1
 v 1 v 2 
a) Se b)

Se

Fig. 8.14 Rigid body bond graph model for r G  0 and I i j  0 for i  j .

In modelling mechanical systems in general and also using bond graphs it is often found that
such systems exhibits differential causality or dependant elements. Using Lagrange’s equa-
tions are in such situations often found particularly useful using the Hamiltonian form of the
Lagrange’s equations formulated into a bond graph IC-field. Karnopp outlines this approach
in [6] as a powerful method in multibody systems modelling. Here we have already intro-
duced the Lagrange equation’s to generate the appropriate equations of motion of the rigid
body and it is tempting to use the ideas to further enhance the bond graph notation for our
system. Following this approach first let us define the generalized momentum vector
P =  P v1 P v2  T corresponding to the linear and angular velocities v 1 and v 2 .

T
P v1 =
 v1
(8.70)
T
P v2 =
 v2

Next the equation of motion for this system can be rewritten in momentum form as previ-
ously stated:

dP v1 T
+ v2  = 1
dt  v1
(8.71)
dP v2 T T
+ v2  + v1  = 2
dt  v2  v1

Performing the differentiation of the expression for the kinetic energy T in Eq. 8.70, utilizing
that this differentiation always has a special form we obtain
290 Multibody dynamics

P = M  q t v + a  q t  (8.72)

where q is the vector of generalized displacements corresponding to the v -vector, M  q t  is

a symmetric matrix and a  q t  is a vector which only occurs if the system includes time var-
ying velocity sources, which we do not have and consequently we just neglect. Eq. 8.72 is
recognized as the constitutive law for a bond graph I -element and using the preferred causal-
ity option for I-elements we normally integrate the effort to obtain the momentum, which
here are the P -vector. Inverting the M  q t  matrix which is recognized as the mass matrix
given in Eq. 8.66, we obtain the velocities and displacements of interest as

–1
q· = v = M  q t P (8.73)

The Eq. 8.71 and Eq. 8.73 are elegantly put into bond graph formalism using an IC-field as
shown in Fig. 8.14b. Including the expression for the efforts e' which may be found from Eq.
8.69 or from Eq. 8.71 the representation of the complete inertia of the rigid body is contained
in a single IC-element.

It it emphasized that the use of the IC-field in a better way than the use of the gyristors sig-
nals the conservation of energy which is the situation found in rigid body motion: the energy
is translated between translational and rotational kinetic energy. The current implementation
also removes the need for any active signal bonds which always can be argued to decrease
the clarity of the bond graph.

Now adding the coordinate transformations as introduced previously in Fig. 8.5, here assem-


1 MTF 1 IC:M
vI vi

SeI Sei

Fig. 8.15 Rigid body bond graph model

bled into a single MTF-field, the complete bond graph can be drawn as shown in Fig. 8.15.
Here the multibond 1-junction v I represents the velocities of the body in the inertia coordi-
nate frame and the multibond 1-junction v i represents the velocities in the body-fixed coor-
dinate frame. Integrating the velocities of the v I 1-junction gives the position of the body in
 General rigid body bond graphs 291

the inertial coordinate frame. External forces and moments present in an inertia coordinate
are easily added connecting a multibond Se-element to the v I 1-junction while adding exter-
nal forces in a body fixed reference frame is likevise easily performed connecting a multi-
bond Se-element to the v i 1-junction. In a similar modular approach, damping forces are
easily added, here damping forces in the body-fixed coordinate frame are shown where the
constitutive law of the R-element is of the form

e = g   f  (8.74)

where e is the effort (force, moment) vector and g   f  is a general nonlinear function of
the set of parameters  and the flow (velocity, angular velocity) vector of the rigid-body.

8.5.1 Application to Rotordynamics

Utilizing the framework developed so far a simple example will be given modelling the clas-
sical Jeffcott Rotor which are often used to study whirl phenomena of imbalance rotors.

The rotor consists of a rigidly supported isotropic shaft with finite elasticity, free to rotate
along its bearing axis. The position of the body-fixed coordinate frame which rotates with an
angular velocity  is placed at the origin of the shaft at the rotor position and is represented
by the vector r o . The vector r G fixes the position of the centre of mass of the rotor in the
body-fixed reference frame. The rotor is assumed to be driven using an electric motor sup-
plying a constant torque at one shaft end.
y

c
rG
rc o
z

ro

Fig. 8.16 Jeffcott Rotor

The rotor is free to move in the y and z direction only and the only rotation  is parallel to
the x -axis. Two bond graphs modelling this case using both the previously discussed modi-
fied gyristors and the optional IC-field is found in Fig. 8.16. The derivation of the constitu-
tive equations of each element and also finally the complete state equations is shown below.
292 Multibody dynamics

I:M IC
1 2 3
1 2 3
4 4 5 6
1
MGR 5
1 1 1 1
6 10
1 MGR
10
7 8 9 7 8 9

R
MTF Se MTF Se

R
11 12 11 12

0 0 0 0
13 16 13 16
1
1

1
1
15 18 15 18
14 17 14 17
R
R
C

R
R
C

C
a) b)
Fig. 8.17 Bond graph models of the Jeffcott Rotor using a) modified gyristors and b) the
IC-field implementation

First we define the momentum of the rotor using Eq. 8.59 and Eq. 8.60 using the standard
bond graph variable naming convention according to Fig. 8.17 as

p 1 = mf 1 – mz G f 3
p 2 = mf 2 + my G f 3 (8.75)
2 2
p 3 = – mz G f 1 + my G f 2 +  I x + m  yG + z G  f 3

where p 1 – 3 =  p v p w p   is the linear momentum in y and z coordinate directions plus

the angular momentum, f 1 – 3 =  v w   is the corresponding velocities and angular veloc-
ity and m , I x and  y G z G  is the mass, the moment of inertia and position of the centre of
mass. Solving for the flows f 1 f 2 f 3 gives

2
I x + mz G yG zG zG
f 1 = -------------------
- p 1 – ----------- p 2 + ----- p 3
mI x Ix Ix
2
yG zG I x + my G yG (8.76)
f 2 = – ----------- p + --------------------
- p 2 – ----- p
Ix 1 Ix Ix 3
z yG p
f 3 = ----G- p 1 – ----- p 2 + ----3-
Ix Ix Ix

Utilizing Eq. 8.65 to derive the expressions for the MGR’s according to Fig. 8.12 gives the
expressions for the efforts e 4 , e 5 and e 6 as
 General rigid body bond graphs 293

e 4 = – m  f 2 + f 3 y G f 3
e 5 = – m  – f 1 + f 3 z G f 3 (8.77)
e 6 = – m  – f 2 – f 3 y G f 1 – m  f 1 – f 3 z G f 2

Recognizing that the transformation specified in the modulated transformer [13] relates
efforts and flows according to

f 11
= cos  – sin  f 7 (8.78)
f 12 sin  cos  f 8

and
e7
= cos  sin  e 11 (8.79)
e8 – sin  cos  e 12

we can write the state equations for the rotor dynamic problem directly from the bond graph
in Fig. 8.17 as

q 14 q 17
p· 1 = c ------- - – rf 1 + m  f 2 + f 3 y G f 3
- + s ------- (8.80)
C 14 C 17

q 14 q 17
p· 2 = – s ------- - – rf 2 + m  – f 1 + f 3 z G f 3
- + c ------- (8.81)
C 14 C 17

p· 3 = Se 9 – R 10 f 3 + m  – f 2 – f 3 y G f 1 + m  f 1 – f 3 z G f 2 (8.82)

q· 14 = – cf 1 + sf2 (8.83)

q· 17 = – sf 1 – cf 2 (8.84)

·
 = f3 (8.85)

where C 14 , C 17 is the compliance of the rotor, r is the internal damping of the rotor, Se 9 is
the shaft input torque and R 10 is the rotational damping of the rotor and motor combined. In
the Eq. 8.80 - Eq. 8.85 the expressions for the flows f 1 f 2 f 3 is to be inserted from Eq. 8.76.

In Fig. 8.18 the y -position of the centre of the shaft spinning the rotor from zero angular
294 Multibody dynamics

velocity through its critical speed to constant angular velocity is shown. Simulating a large
number of cases and comparing the actual equation of motions obtained with the same
reported elsewhere [11] proves that the bond graph modelling produces consistent building
block which may serve as a base for further development and analysis.

model

Y
Z
0.002

0.001

-0.001

-0.002

0 2 4 6 8 10
time {s}

Fig. 8.18 Spinup of Jeffcott Rotor. Radial displacement in x- and y- direction.

A second example utilizing the vector bond graph notation is shown next. The Stodola-Green
model is a more realistic model for rotordynamic problems encountered in real systems. This
model includes the gyroscopic effects which occurs when the rotations in the transverse axes
are included. These were absent in the Jeffcott rotor model. A case which often is used to
study gyroscopic effects in rotors are shown in Fig. 8.19. A rotor consisting of a rigid disc
and a shaft with infinite stiffness is mounted in a flexible set of bearings with an assumed
damping present. The disc may be mounted in any position along the shaft axial direction.
The disc is also mounted misaligned with the shaft in the radial direction, causing an imbal-
ance to occur.

The bond graph of this case can be developed based on the framework that has been devel-
oped. Using vector bonds the bond graph is pictured in Fig. 8.20 The inertia of the disc
including the shaft is modelled using the IC-element. The constitutive equations for this ele-
ment is obtained from Eq. 8.64 and Eq. 8.65.

The relation between the velocities v 1 and angular velocities v 2 of the origin of the disc and
the corresponding velocities at the bearings in front or back of the disc can be established
using the transformation
 General rigid body bond graphs 295

Z


z
k B r B
c
y 
o
x Y

k F r F
X

Fig. 8.19 Stodola-Green rotor with rigid shaft and bearing damping and flexibility.

IC:M

 xB   xF 
1 b TF 1 TF b 1
 v 1 B  b b
 v 1 v 2   v 1 F 

MTF  R 
MTF  R   R  MTF
1  v I1 v I2 
I I
1  v 1 B   v 1 F  1
R  R
0 1 C C 1 0

I I I
x  y  z 
Fig. 8.20 Bond graph model of a Stodola-Green rotor

v b = v 1 –  r G v 2 (8.86)

where  r G  is the skew-symmetric matrix of the bearing position vector relative to the body-
fixed coordinate system. This can be represented using a bond graph transformer TF with the
transformer modulus matrix

1 0 0 0 –zG yG
m = 0 1 0 zG 0 –xG (8.87)
0 0 1 – yG x G 0

Performing this transformation yields the translational velocities at the bearing in body-fixed
coordinates. A further transformation have to be carried out using the Euler-angles and the
296 Multibody dynamics

transformation matrices R  R  R  to attain the translational velocities at the bearings in iner-

tia components. Connecting the multiport transformers MTF on both sides performs this
task. Adding the flexibility and damping forces experienced at the bearings is obtained using
I I
the C and R elements connected to the 1-junction’s v 1 B and  v 1 F  at the front and back bear-
I I
ings, here v 1 B and  v 1 F  represents the velocities on both ends (back and front) of the shaft
in inertia frame coordinates. The bond graph of the Stodola-Green rotor is complete. To
attain the position of the centre of the disc it is necessary to connect an additional MTF-trans-
former representing the Euler transformation to the v 1-junction. Integrating the inertia
velocities yields the position of the origin of the disc and the corresponding Euler angles rep-
resents the orientation.

Simulation results of a rotor with the physical data as given in Table A is shown in Fig. 8.21.
The results plotted is the radial displacement and the transverse axis rotation of the disc when
the rotor is spinned from zero through its critical speed. A large set of simulations carried out
reveals that the bond graph model is capable of visualizing all features expected to be present
in the Stodola-Green rotor. Implementing the bond graph models into 20-sim [20] utilizing
its powerful animation features yields a very powerful environment for modelling and visual-
ization of rotordynamic phenomena.

Title

y_g
0.002

0.001

-0.001

-0.002

Vy_g
0.0004

0.0002

-0.0002

-0.0004

0 0.5 1 1.5 2
time {s}

Fig. 8.21 Stodola-Green rotor simulations results. Top: Transverse position of disc and Bot-
tom: Transverse axis rotation.
 General rigid body bond graphs 297

Table 5: Physical data Stodola-Green rotor simulations

Mass of rotor - including shaft m [kg] 33.78
Moment of inertia Ix [kgm2] 0.904
Moment of inertia Iy = Iz [kgm2] 2.620
Bearing stiffness - front kF [N/m] 600 000
Bearing stiffness - back kB [N/m] 800 000
Damping coefficient bearings  [-] 0.01
Misalignment of disc y G z G [m] 0.0001
Disc position from bearings [m] (0.75, -0.15)

REFERENCES
4. J. L. Meriam, L. G. Kraige, Engineering Mechanics - Dynamics, 6th Ed., John Wiley &
Sons, Inc., 2007.
5. D. C. Karnopp, Power-conserving Transformations: Physical Interpretations and Appli-
cations Using Bond Graphs, Journal of the Franklin Institute, 288, No. 3, 175-201, 169.
6. D. C. Karnopp, Lagrange’s Equations for Complex Bond Graph Systems, Trans. ASME
J. Dynam. Syst. Meas. Control, 99, Ser. G, 300-306, 1977.
7. D. C. Karnopp, Understanding Multibody Dynamics Using Bond Graphs, J. Franklin
Inst., 334B, No. 4, 631-642, 1977.
8. Jeffcott N., Lateral Vibration of Laded Shafts in the Neighborhood of a Whirling Speed -
The Effect of Want of Imbalance, Philosophical Magazine, 37, 304-314.
9. Stodola, A., 1924, Dampf- und Gasturbinen, Springer-Verlag, Berlin
10. Green, R., 1948, Gyroscopic Effects f the Critical Speeds of Flexible Rotors, Journal of
Applied Mechanics, 15, 369-376.
11. Childs, D., 1992, Rotordynamics of Turbomachinery, John Wiley & Sons, New York
12. Vance, J. M., 1988, Rotordynamics of Turbomachinery, New York, Wiley, pp. 3-6.
13. Karnopp, D. C., Margolis D. L., Rosenberg C., System Dynamics: Modeling and Simu-
lation of Mechatronic Systems, 4th Ed., John Wiley & Sons, 2006.
14. Karnopp D. C., Understanding Multibody Dynamics Using Bond Graph Models of
Mechanical Systems, J. Franklin Inst., 329, No. 1, 65-75 (1992).
15. Allen R. R, Multiport Representation of Inertia Properties of Kinematic Mechanisms, J.
Franklin Inst., Vol. 308, No. 3, Sept 1979.
16. Bos A. M., Tiernego M. J. L., Formula Manipulation in the Bond Graph Modelling and
Simulation of Large Mechanical Systems, J. Franklin Inst., Vol. 319, No 1/2, pp51-65,
Jan 1985.
17. Breedveld P. C., Proposition for an unambiguous vector bond graph notation, Trans.
ASME J. Dynamic Syst. Measure. Control, Vol. 104, No 3, pp. 267-270, Sept. 1982.
18. Hubbard M., Whirl dynamics of pendulous flywheels using bond graphs, J. Franklin
Inst., Vol 308, no. 4, pp. 505-521., Oct. 1979.
298 Multibody dynamics

19. Compos J., Crawford M, Longoria R., Rotordynamic Modeling Using Bond Graphs:
Modeling
20. 20-sim, Control Labs Products, www.20sim.com.
21. L. Meirovitch, Methods of analytical dynamics, McGraw-Hill, 1970, ISBN 0-486-
43239.
 General rigid body bond graphs 299

PROBLEMS
8-1 The three mass particles are attached to a spring supported rigid massless bar that exe-
cutes small vibratory motion. The vector q C is x 1 x 2 , the vector v I is V 1 V 2 V 3 and
several choices for q k is possible. Assume small displacements and develop a bond
graph and write the state equations using

a) q k =  x 1 x 2  T = q C

b) q k =  x 3   T where  is the angle of rotation of the bar

c) q k =  x 1 x 3  T

l1 l2
m1 m3 m2

x1 k1 x2 k1 x3

Next assume large motion and develop the respective bond graphs as above.

8-2 Develop a bond graph for the 3D flexible pendulum shown below where the mass m is
free to move in 3 dimensions and connected to a spring with stiffness k having an un-
stretch length of L 0 . Also write down the state equations and all constitutive laws.

x
k
y
m
z

8-3 In the figure below two cylinders is shown having masses m 1 , m 2 and polar moment of
inertia J 1 , J 2 . The cylinders each has a fixed rod connected with length l 1 and l 2 re-
spectively. Develop a bond graph model for plane motion for this system and write down
the state equations.

l1
y 1 l2 2

m 1 J 1 m 2 J 2
x
300 Multibody dynamics

8-4 Construct a bond graph for the 2D vehicle model shown below. The suspension of the

L1 L2

Lb Lb
kb cb kf cf

R R
k k

vehicle has an unloaded length L b and the stiffness and damping are k b c b and k f c f for
the front and rear end respectively. The wheels having an initial radius of R can be as-
sumed to have a stiffness coefficient k . You may initially assume that the velocity of the
vehicle is v 0 and constant. The model is supposed to handle the situation when the
wheels leaves the surface!

8-5 Develop a bond graph model of the system shown below and write the state equations
for the system. Write complete state equations using both scalar bond graphs and La-
F(t)

y  L
k
m0
x

grange-Hamiltonian IC-field implementation. Comment any disadvantages you may ob-

serve.

8-6 A linkage moving in the vertical plane is loaded with a torsional spring having a stiffness
k in A and a force F(t) as shown below. The torsional spring is unload when  = 60 deg.
Check the expression for the kinetic energy and develop a bond graph of this system.

C
m AB = m CB = m
· 2
T = --- ml 2  2  --- + 2 sin 
l 1 2

B 2 3 
Ft

 l
A k
 General rigid body bond graphs 301

8-7 Develop a bond graph model of the system shown below using both the scalar bond
graph procedure and the Hamiltonian IC-field approach.
x· 0
x
k1
m0

R
y  k2
m

x u = x· 0 + x·
v = y·

8-8 A sphere with a radius r = 1 cm is rolling along a perimeter with a radius R = 10 cm.
The mass M os the sphere is 0.1 kg and its moment of inertia is 0.4Mr2. Establish the
O

g
R

A C r

equations of motion for this system, develop a bond graph for this system using the Ham-
iltonian IC-field approach. How does the frequency change when R  r ?
·
8-9 A disc having a rotational inertia I z is rotating  around the z-axis due to an small electic
motor supplying the constant torque Tm. In the disc is a slot where a mass m is allowed
to move radially,i.e. the radial position being r . Below the disk another mass M is mov-
ing as a pendulum in a vertical fixed plane with a angle  2 . The masses m and M are
connected to each other with a rope with length l with no flexibility. Develop a bond
graph model of this system using both the scalar bond graphs and the procedural ap-
proach and the Hamiltonian IC-field bond graph implementation. The Lagrangian of the
system is given as
1 · 1 · 1 ·
T = --- m  r· 2 + r 2  2  + --- M  r· 2 +  l – r  2  22  + --- I z  2
2 2 2

8-10 An engine with a mass m is mounted on four engine mounts with vertical and horizontal
stiffnesses k v k h respectively. The engine block has rotational inertias I xx I yy and I zz .
The centre of mass is assumed located in the centre of a cube with dimension l and where
the engine mounts are located in the four lower corners. The moving bodies of the engine
are introduces forces and torques in rigid body coordinate frame as F x F y F z and T x T y
and T z .
a) Develop a bond graph model of this engine and write the state equations and consti-
tutive laws for all elements
302 Multibody dynamics

b) Assume that displacement in the x-direction are can be neglected simplify your bond
graph accordingly.

8-11 A big cylinder with mass m and a radius r is rolling on a horizontal surface. At a point
with radius equal to r  2 , a spring with an unstretched length of 2r is hinged with the
other end fixed at the wall at the vertical position r . Assuming no slip and develop a bond
graph of this system. Secondly, assume that slip may occur and that there is a friction co-
efficient  between the cylinder and the surface, modify the bond graph accordingly.

v = x·
L = 2r + 

r r

8-12 A table rotates in a horizontal plane about bearing A due to a torque T. The mass of the
table is M and its radius of gyration about its centre is  . The slider, whose mass is m ,
moves within a slot BC under the restraint of a pair of springs having a stiffness k that
are unstetched in the position shown. Derive the bond graph of this system and write
down the equations of motion for this system. (Optional: rotate the slot 90 deg, what

C x Tip:
k m 1 · 1 ·
M T = ---  M 2 + mx 2 + ml 2  2 + --- mx· 2 + mlx· 
k 2 2
A B
V = kx 2
·
 = 

now?)

8-13 On the figure below a pendulum consisting of a cylinder with a mass m 2 and rotational
inertia J 2 is connected to a rod with length L and mass m r is shown. The pendulum is
 General rigid body bond graphs 303

hinged at another cylinder with mass m 1 , radius R and rotational inertia J 1 and rolling
without slip on a surface.

x x·
m 1 J 1

  L

m 2 J 2

Show that the kinetic and potential energy of the system can be expressed as
· 1 · · · · ·
T = --- m 1  R  2 + --- J 1  2 + ---  J 2 + --- m r L 2  2 + --- m 2   R + L cos   2 +  L sin   2 
1 1 1 1
2 2 2 3  2
V = m 2 gL  1 – cos  

and a) construct a bond graph of this system and b) write down the equations of motion
of this system.

8-14 Consider the double pendulum hinged at the vehicle moving on the surface as shown be-
low.

x m0

1 L1
x
(x1,y1)
m1 g

2 L2
(x2,y2)
m2
y

a) Set up the multiport displacement-modulated transformer that relates

· ·
x· 0 x· 1 y· 1 x· 2 y· 2  1  2 and exhibit both the velocity and force relations in matrix
form.
b) Using the MTF, show a system bond graph and predict one possible set of state vari-
ables using the integration-causality method.
· ·
c) Use the junction structure of derived in 2) to compute T   1  2  1  2  - the kinetic
energy and generalized torques so that Lagrange equations can be utilized as an
304 Multibody dynamics

alternative procedure to find state equations. What are the state variables if Lagrange
equations are used?

d) Modify each pendulum inserting an axial flexibility making L 1 L 2 changing.

Develop a bond graph of this system using both scalar bond graphs and the Hamilto-
nian IC-field implementation. How is the mass matrix M in this case?

e) Assuming that the vehicle is a barge floating on the water and that the forces exciting
the masses m 1 m 2 can be given according to Morrisons equation for submerged
bodies as F = C 1 x·· + Dx· where F x·· x· is the vectors of forces, the acceleration
and velocity. Now develop or modify you bond graphs accordingly.

8-15 The marine vehicle shown being a rigid body generally having 6 degrees of freedom are
shown below. The reference frame is positioned in x G =  0 0 0  T relative to the centre

X
 xb
v surge
O
Y
o
G
Z


pitch
yaw
yb
sway
zb
heave

of mass and where the mass m and rotational inertias I xx I yy I zz are given. The products
of inertia is however being zero.
a) Construct a bond graph for this system using scalar bond graphs. Include buoyancy
and damping into your model.

b) Assuming that the motions is only in x y  coordinates, now do the same

c) Modify the bond graph in b) to include truster and rudder forces

d) Now assume that the reference frame is positioned in x G =  x G y G z G  T . Use vec-

tor bonds and sketch a bond graph for this situation.
·
8-16 The mechanism shown below spins around the vertical Z-axis with an angular speed 
due to a external torque T . A disc G connected via a rod with length L spins with con-
stant angular velocity p . The angle between the vertical axis and the rod is  . The kinetic
 General rigid body bond graphs 305

and potential energy is given below. Check out the expressions and construct a bond
graph for this system. (Tip: Use the IC-field implementation)

Vertical
· · ·
T = ---  --- m R 2   cos  +   2 + ---  --- m R 2 + m L 2 + --- m L 2   2 +  2 sin  
1 1 1 1 1 2
Z
22 1  1 24 1 1 3 2 
K
V = –  m + --- m  gL cos 
1
 1 2 2

Y
i
9” p X, x

G
70
306 Multibody dynamics
 Thermodynamic systems modelling 307

Chapter 9

Thermodynamic systems modelling

In the field of system theory much attention has been pain to mechanical, electrical and
hydraulic systems, but components of which the behaviour must be described by thermody-
namic relations have been rarely considered. The reason for this fact is partly caused by the
complexity of thermal components and the restricted possibilities of describing thermal proc-
esses in terms of simple analytical relations. Even more important is the difference in frame-
work of classical thermodynamics as a macroscopical physical theory as compared to the
structure of the physical theories underlying the operation of mechanical, electrical and
hydraulic systems. In these systems the physical quantities are functions of space and time
coordinates in general and one speaks of system dynamics in a case where time dependence
is of important major importance. Thermodynamics, however, has nothing to with dynamics
although its name may suggest this. The time dimension does not show up in thermodynam-
ics and the space dependence of thermodynamic functions is a point of no concern.

The broad purpose of this chapter is therefore, to explain modelling and simulation of sys-
tems involving control or use of thermal energy, and which is theoretically based on the sci-
ence of thermodynamics and heat transfer. We will attempt to combine classical and
irreversible thermodynamics with bond graph modelling methodology in order to provide an
organisational framework for modelling complex thermal systems.

In what follows in this chapter we will give particular attention to those situations which
illustrate the interaction between fluid motion, heat transfer and thermodynamics. That is,
between mass flows, momentum changes and pressures on the one hand, and temperature
changes and energy flows on the other. We will consider the effects of energy addition
through heat flow into or out of the fluid and through work transfer. The subject matter of this
chapter, therefore, provides the basis for the analysis of a considerable fraction of the prob-
lems which the engineering profession has to solve.

Modelling of thermal systems is complicated by the non linearity of the individual dynamic
processes, the organizational complexity of the systems as a whole and the intractability of
the describing equations. By exploiting a bond graph framework for thermal systems model-
ling, however, provides a formalism and brings thermodynamics and heat transfer within the
framework and unification of ideas with modern dynamic system theory, thus bringing to
bear a great body of analytical tools on the problem of thermal complexity.
308 Thermodynamic systems modelling

The advantages of the bond graph approach are numerous since engineering experience and
network methods of solving large nonlinear systems can be brought to bear on thermal prob-
lems. For example, complex systems can commonly be thought of as composite systems
made up of interconnected elements of subsystems. In many instances such systems can be
torn or decomposed into their basic elements which are used to generate the model equations
and solution to the whole system. By establishing an isomorphism between the underlying
mathematical structure of bond graphs and thermodynamics, we will formulate and solve
thermofluid problems using network methods.

9.1 Basic principles

Thermodynamics is the science that encompasses the study of energy transformation and the
relationships of a substance which are affected by these transformations. Classical thermody-
namics is limited in that it deals only with the initial and final equilibrium states of a process;
it provides no information about the dynamics behaviour between these states. This means
that it tells us where we have been and where we are, but not how we got there. That is, it
deals with the end states of a process, and provides no information concerning the nature of
the interaction or the time rate at which a process occur. A system undergoes a process when
it changes from one state to another and equilibrium is that state which is characterised by no
change. The equilibrium of a mechanical system is a condition of balance maintained by an
equality of opposing forces. Thermodynamic equilibrium, however, is a broader concept
since it includes not only mechanical forces but also “forces” arising from thermal, electrical,
chemical and other influences. A system in thermodynamic equilibrium is thus in complete
balance with the surroundings. From a practical standpoint, the most important and interest-
ing applications of the concepts, principles and conclusions of thermodynamics involve sys-
tems in states that are not in stable equilibrium.

If the restriction of classical thermodynamics were strictly interpreted to only relate initial
and final equilibrium states, thermodynamics would be of little practical use because real
systems are never exactly in equilibrium. However, during a process which is slow enough,
the state of a system may not differ greatly from an equilibrium state. The system may there-
fore be assumed with reasonably accuracy to pass through a series of equilibrium states dur-
ing the process. This extension and by the definition of the dynamic open system has made it
possible for flow processes or irreversible, non stationary processes to be successfully fitted
into the system theory of bond graphs.

A dynamic open system has a mass transfer across its boundaries, and the mass within the
system is not necessarily constant. Most engineering systems are open. The concept of the
dynamic open system allows analysis to be made of continuous streams which are far from
equilibrium in part of their travels. In this treatment we will assume that the states of the fluid
are described as quasi-static stable states which do no t vary with time during the period to
 Basic principles 309

which the analysis relates. A quasi-static process is an ordered succession of equilibrium

states, whereas a real process is a temporal succession of equilibrium and non equilibrium
states.

In setting up the equations governing the dynamics of the fluid, we need to decide whether
we should use coordinates fixed in space or coordinates that move with the fluid. These two
procedures area known respectively as the Eulerian or Lagrangian specifications. Although
Lagrangian formulations may more closely support understanding and formulation of physi-
cal laws, it is natural to use an Eulerian or control volume approach for components which
have fluids flowing through them. A control volume is any defined region in space, having
boundaries through which the working substance flows. Shape and volume of the control
volume may be changing. The method of analysis is to inspect the control volume boundaries
e and account for quantities of matter and energy transferred through this surface.

9.1.1 Conservation equations

Consider a control volume whose boundaries are fixed in space and stationary having the
volume V . Matter flows across the boundaries of this control volume as indicated in Fig. 9.1

·
W
m· 2

P 1 T 1 h 1 P T u P 2 T 2 h 2
V
m· 1 Q·

Fig. 9.1 A control volume

as m· 1 and m· 2 . In the general case a control volume would be subject to changing conditions
from time to time so that the rates of flow of mass and energy through the control volume
boundaries would change with time and the quantities of mass and energy within the control
volume would hence change with time too.

The flow of matter in Fig. 9.1 is entering the control volume at section 1 and is leaving at sec-
tion 2, and these flows are in general not equal. Our attention is mainly to flows which can be
represented with reasonable accuracy by a single vein of fluid whose fluid geometry is
known in advance and across which the velocity is sensibly constant. This idealization is
referred to as the one-dimensional flow model. The assumption of one-dimensional flow is
justified for most engineering analysis classes.
310 Thermodynamic systems modelling

In addition we allow energy exchange to occur as heat Q· and work flow W · transfer across
the boundaries of the control volume, and to avoid confusion, we shall assume that no energy
is transferred as heat at the inlet and exit stations.

In an open system with variable flow, the equation of continuity of mass is one of the funda-
mental physical laws, often also referred to as the conservation of mass. This equation simply
states that the rate of increase of mass within the control volume is equal to the net rate of
mass of fluid crossing the control volume boundaries. This may be written as

dm cv
= m· 1 – m· 2 (9.1)
dt

where m cv is the mass of fluid within the control volume at any instant. With an average
velocity V , an average density  and cross-sectional area A at the inlet and outlet sections
respectively, the mass flow rate m· s at section 1 or section 2 can be stated as

m· s = VA (9.2)

Here the upstream assumption must be applied. This means that any variable associated with
the flowing fluid must be assigned based on the value of the corresponding variable just
before the fluid passes the control volume boundary, i.e. the upstream value. For example,
the density used for calculating the input mass flow must be the density of the surroundings
of the control volume, while the same for the output mass flow must be the density of the
control volume itself. If the fluid reverses, the “source” of the density must change accord-
ingly.

The next step is then to develop the control volume energy expressions. To do this we begin
with the ideas of the First Law of Thermodynamics which have become the theorem of con-
servation of energy. First-law analysis is actually an accounting procedure in which we take
account of energy transfers to and from a control volume and of changes of energy inside the
control volume. An energy balance on the control volume necessitates not only measurement
of heat and work interactions, but also, an accounting for the energy carried into or out of the
control volume by mass transferred across the control surface. Hence, when matter flows into
a control volume energy is convected in by the matter. We shall restrict the types of energy
associated with the unit mass entering or leaving a control volume to internal energy, linear
kinetic energy and gravitational energy. That is

2
V
e m = u + ----- + gz (9.3)
2

where e m is convected energy per unit mass, u is internal energy, V is fluid velocity, z is
elevation above a fixed datum and g is the acceleration of gravity. Frequently the last two
 Basic principles 311

terms are omitted because their change are negligible when compared to other terms in the
conservation of energy equation.

In addition to the fact that energy is convected in by the matter when it flows into a control
volume, entering matter does work on the matter already within the control volume by push-
ing it out of the way. The amount of energy transfer associated with a unit mass is Pv with P
the pressure and v the specific volume at the control volume boundary. The Pv product is
sometimes referred to as flow work. It does not represent any energy contained by the fluid,
rather it is a form of energy transmitted with the flow which must be credited to the flow to
yield a correct energy balance. Only when matter crosses the boundaries of a control volume
will the additional contribution of the Pv product occur in the energy balance equation.
Thus, the sum of the specific energy quantities of the flowing substance can then be defined
as

2 2
V V
e f = u + Pv + ----- + gz = h + ----- + gz (9.4)
2 2

where h = u + Pv is the enthalpy per unit mass of the fluid. Thus, the conservation of
energy equation for a control volume with one inlet and one outlet can be written as

dE
= Q· – W
· +  m· h  –  m· h  (9.5)
dt in out

where E = m  u + V 2  2 + gz   mu is the total internal thermal energy stored by the fluid

in the control volume where the kinetic and gravitational form of energy most often is
neglected, Q· is the sum of heat transfer rates to the control volume, W · is the sum of power,
i.e. the rate of work done by the control volume system on the surroundings or in other words
output due to normal motion of control volume boundaries. Finally m· and h is the mass
flow rates and enthalpies at the inlet and outlet boundaries respectively. Please observe the
sign conventions used, i.e. positive work is work done by the control volume on the sur-
roundings, while positive heat flow is from the surroundings to the control volume. In words,
this equation simply states that the rate of change of energy inside the control volume is equal
to the heat transfer into minus the rate of work done on the surroundings by the control vol-
ume plus net value of the rate of energy flowing in as a result of mass transfer.

These two equations, Eq. 9.1 and Eq. 9.5, now simply states the conservation of mass and
energy for the control volume. Together with the appropriate additional equations of state
which we will discuss next, we are now ready to develop bond graph elements which will
represent Eq. 9.1 and Eq. 9.5 and which will be used to model a variety of thermofluid sys-
tems.
312 Thermodynamic systems modelling

9.2 Bond graph representation

It has been the subject of much research to find a bond graph representation of thermal
energy systems. One important reason for this is that the exchange of energy is not as
straightforward as in mechanical, hydraulic or electrical systems. Central questions in these
developments have been the choice of variables on the bond and which state variables to
choose. We will explore this further in chapters to come. The true power variables are tem-
perature T for effort and rate of change of entropy S· for flow; the product of these two being
power. However, working with entropy introduces additional levels of complexity and is not
the standard approach for most engineering students. It does though have some advantages in
special applications, but we will not discuss this here. The reader is referred to [5][14].

Since using entropy as a variable is often found difficult, we will here present what is known
as the Karnopp concept of pseudo-bond graphs [5]. This approach deviates from conven-
tional bond graphs in that Karnopp extended the concept of power bonds to that of double
bonds where the standard specifies that the second bond should be drawn using a dashed line.
The effort and flow variables are:

• Efforts:
p - pressure
T - temperature

• Flows:
m· - mass flow rate
·
E· - total energy flow rate, i.e. mu
Q· - heat energy flow rate
V· - rate of change of volume

According to Karnopp we pair P and m· and T and E· or Q· as shown in Fig. 9.2.

A m· T
E·
B
Fig. 9.2 Thermofluid bonding

This pairing is natural and related to true bond graphs for incompressible flow, i.e. as for
hydraulic systems, in which m· would be proportional to volume flow rate V· and to pseudo
bond graphs defined previously, in which T and Q· are used for conduction heat transfer. The
 Bond graph representation 313

use of double bonds with two efforts and two flows allows a close connection to true bond
graphs and permits the use of normal conventions for power sign and causality for thermof-
luid systems. Going back to Eq. 9.5 it shows that the flow of energy now depends on the
amount of flowing matter and enthalpy. Thus, there exists a coupling between the hydraulic
and thermal energy domain. However, although the concept of power ports still may be for-
mally extended, strictly the pseudo-bond graph port now sees two pseudo bonds.

After having selected the bond variables it remains to determine a proper set of state varia-
bles for thermofluid systems. In order to use the state determined system model for ther-
mofluid systems the number of independent variables necessary to fix or establish the state of
the system must be known. This information is provided by a basic law of thermodynamics
known as the state postulate. This is a general rule that allows us to determine the number of
independent state variables for any physical situation. The state postulate states that the
number of independently dynamic thermodynamic properties for a specified substance is
equal to the number of relevant reversible work modes plus one. In the control volume analy-
sis we have considered, only two important reversible work modes - heat input and compres-
sion or expansion as Pdv work. This means that we need three independent state variables.
A more frequently used analogy is the Gibbs phase rule for a single component matter which
simply states that the number of independent intensive state variables necessary to uniquely
fix the thermodynamic state of any matter is equal to two plus one extensive variable deter-
mining the size of the system.

Having elaborated around the required number of state variables, it remains to determine
which three independent variables are possible selections as state variables. As an aid we can
utilize the fact mentioned previously that the state of a system at any given instant is related
to the energy stored in the system at that instant. Thus, the total energy E is therefore a pos-
sible candidate as a meaningful state variable. Likewise the total mass and the volume of the
system are meaningful variables. Hence we initially propose the following state variables:

m - total mass in the control volume [kg]

E - total energy in the control volume [J]
V - total volume of the control volume [ m3 ]

The state variables m , E and V are adequate in the differential equation sense and the ques-
tion is now if they also are adequate state variables in the thermodynamic sense. Using the
postulatory formulations of thermodynamics stated by Callen’s [25] in his first postulate
which simply states:

Postulate I: There exist particular states called equilibrium states of simple systems that
macroscopically are characterized completely by the internal energy U , the volume V
and the mole numbers N 1 , N 2 ,..., N r of the chemical components.
314 Thermodynamic systems modelling

we can conclude that they are. Observing that the suggested total energy E is identical with
the internal energy U since kinetic and potential energy is not accounted for in the control
volume. The choice of mass as a state variable is also justified since mass divided by the
molecular weight of a fluid equals the number of moles present. This means that the choice
of total mass, energy and volume as state variables have a sound basis in the fundamentals of
thermodynamics as justified through Callens Postulate I. This means that the value of any
property of the system in a stable equilibrium state is uniquely determined by the values of
mass, energy and volume. This makes it possible to express any property of a substance as a
function of u and v . As an example, pressure P and temperature T can be determined from
a functional relationship of the form

P = P  u v  T = T  u v  (9.6)

Functional relationships as given above are called equation of states or EOS and are available
for most matters. An example of a simple equation of state which is satisfactory for many
gaseous substances is the perfect gas equation of state which states that

Pv = RT (9.7)

where R is an experimental constant called the gas constant which has different value for
each gas. For a perfect gas the gas constant is related to the molecular weight M as
R = R 0  M , where R 0 is the universal gas constant. Typical values for typical gases are
found in most textbooks on thermodynamics. Eq. 9.7 is satisfactory for most gases at moder-
ate to high temperatures or at low pressures. Over wider temperature ranges, such as are
encountered in combustion processes, account must be taken of the increase in the values of
the specific heat capacities with rise in temperature, so that we must then treat the gas as
semi-perfect or ideal. At high pressures, no gas may be treated as ideal.

A perfect gas not only satisfies the equation of state as in Eq. 9.7, but its specific heats are
constant. A corollary of the perfect gas equation of state is that the internal energy and
enthalpy of a gas is solely a function of state, i.e the internal energy and enthalpy of a gas is
solely a function of temperature and is approximately true for real gases and it follows that

u = c v T or u2 – u1 = cv  T2 – T1  (9.8)

and

h = c p T or h2 – h1 = cp  T2 – T1  (9.9)

where c v and c p are the specific heats at constant volume and constant pressure respectively.

In applications for which the absolute values of properties such as u are not relevant,
 Basic compressible fluid elements 315

namely, applications that require only the evaluation of differences in these properties
between different states, it is conventional therefore to adopt some particular state of a sys-
tem with temperature T 0 as a reference state, the energy of which is arbitrarily taken as zero.
Thus

u  T  = u  T0  + cv  T – T0  (9.10)

We are now ready to develop some bond graph submodels which will be used as basic build-
ing blocks in modelling of thermofluid systems.

Before we start to model thermofluid systems we will define a few bond graph submodels
which we will use as basic building blocks when modelling complete systems. The basic
building block submodels will be developed separately for compressible and incompressible
fluids.

9.3 Basic compressible fluid elements

9.3.2 Thermofluid accumulator

We will start with introducing the basic conceptual device shown in Fig. 9.3, which we will
call a thermofluid accumulator. Gas flows into and out of the fixed volume V and heat Q· is
exchanged with the surroundings. The pseudo bond graph for the accumulator consists of a
0-junction structure representing the common pressure and temperature in the volume. A cru-

C
V p m·
m· in p T m· out T E·
h in h out p p
m· in m· out
T T
(a) · E· in E· out
Q (b)

Fig. 9.3 Accumulator: a) schematic b) Pseudo Bond Graph

cial assumption is made for this to be true and that is that there is sufficient mixing at all
times so that we can assign a single uniform pressure and temperature to the contents of the
accumulator. Yet another approach may be to view the two 0-junctions as the implementation
of the conservation equations for mass and energy respectively. The 0-junctions adding flows
316 Thermodynamic systems modelling

enables us to regard each term in these two equations as flows and hence connect one
pseudo-bond for each term to the corresponding 0-junction. The actual accumulation of mass
and energy now being the flow to the left of the equal sign in Eq. 9.1 and Eq. 9.5 and the inte-
grals, being displacements, now simple are the state variables m and E = mu . Having
defined the efforts as pressure P and temperature T we are faced with the problem of relat-
ing P and T to the state variables E and m . This general relation between efforts and dis-
placements, we recognize as the constitutive relation of a 2-port C-field which for the
thermofluid accumulator nicely displays accumulation in general or especially accumulation
of thermofluid potential energy.

Let us now write the state equations and compute the pressure and temperature in the accu-
mulator based on the causal bond graph shown in Fig. 9.3.The two state equations for this
system becomes

dE
= E· in – E· out + Q· = m· in h in – m· out h out (9.11)
dt

and

dm
= m· in – m· out (9.12)
dt

For a perfect gas assumption, simple relations for pressure and temperature can be deduced
from the perfect gas equation of state in Eq. 9.7 and the relation between the temperature and
internal energy from Eq. 9.8 as

E mRT E
T = --------- and P = ------------ =   – 1  --- (9.13)
mc v V V

using the ratio of specific heats  = c p  c v . These equations then nicely determines the con-
stitutive laws of the C-field as

T =  T  E  m  V    R  c v  
(9.14)
P =  P  E m  V   R c v  

where  V   R c v  are the two parameter sets required, i.e. the geometric parameter vol-
ume V and the working fluid parameters the gas constant and the specific heat at constant
volume.

Extending this basic bond graph model representation of a fluid accumulator to include more
general cases as mixtures and multiple phases will be shown at the end of the chapter.
 Basic compressible fluid elements 317

The accumulator shown in Fig. 9.3 has a constant volume. Extending this case to an accumu-
lator with variable volume and work or energy input due to a piston movement as shown in
Fig. 9.4 we actually introduces a third port where work can pass through. According to the

F x·
TF 1  A
·
V· P V

C
m· in p T m· out
P m·
h in h out T E·
V
p
(a)
Q·
m· out
T T
E· in E· out
(b)
Q·

f
Sf

S
Fig. 9.4 Variable volume accumulator: (a) Schematic, (b) Pseudo Bond Graph

previous discussion of states, the volume V consequently also becomes a state variable. In
this case, we use a 3-port bond graph C-field for representation since we also have a mechan-
ical power port in addition to the m· and E· ports. Observing the conservation of energy equa-
tion Eq. 9.5 we recognize the need to include an additional flow, i.e. the rate of work carried
out by the piston. We therefore to the thermal 0-junction add a flow source representing the
mechanical power PV· from the piston. The value of this flow must as shown in Fig.
9.4extracted from the mechanical power bond as the product of the effort and flow. The rea-
son for adding this source is that the effort and flow variables on the bonds do not represent
complementary power variables. Also, for pseudo C-fields no state function exist, such as the
stored potential energy as in normal C-fields, consequently it is not possible to extract the
relations directly from this as for mechanical systems. This signal interaction is the price we
pay for connecting a pseudo bond graph with a true bond graph.

The three state equations for this system also including the heat flow, which is recognized as
a flow in the conservation of energy equation and as the second Sf-element in Fig. 9.4, then
becomes

m· v = m· in – m· out (9.15)
318 Thermodynamic systems modelling

·
E· v = m· in h in – m· out h out + Q – E· w (9.16)

V· v = Ax· (9.17)

where E· w = PV· = Fx· = PAx·

9.3.3 Pseudo-bond graph restrictor

The next fundamental element shown i Fig. 9.5 is the pipe, duct or element junction named
restrictor to distinguish it from the true bond graph.

A
P1 P2 P1 P2
m· m· m· 1 m· 2
T1
A
T2 T1
E· 1
R T2
E· 2
a) b)

Fig. 9.5 Junction or Restrictor: (a) Schematic, (b) Bond Graph

In general, there is continuity of the flow variables, i.e.

m· 1 = m· 2
(9.18)
E· 1 = E· 2

but the pressure and the temperature are not necessarily the same at the external boundaries,
i.e. T 1  T 2 and P 1  P 2 . We consider the two temperatures and pressures to be determined
upstream and downstream of the junction. With a positive sign convention pointing in the
direction of positive m· as shown in Fig. 9.5the energy flow rate through the junction is deter-
mined by the enthalpy or temperature upstream as

E· = m· h 1 = m· c p T 1 (9.19)

If the flow direction is allowed to reverse, then the energy flow rate through the junction is
determined by the enthalpy or temperature downstream as

E· = – m· h 2 = – m· c p T 2 (9.20)

This flow direction causes a complicated reformulation of the state equations at certain point
in time. To avoid this difficulty, we may introduce a simple fix originally proposed by Kar-
 Basic compressible fluid elements 319

nopp [] where the multiport restrictor element changes output thermal flows when fluid flows
reverse. The constitutive law of the 2-port R-field using this fix takes the form

m· m·
E· = ----  h 1 + h 2  + -------  h 1 – h 2  (9.21)
2 2

or

m· m·
E· = ---- c p  T 1 + T 2  + ------- c p  T 1 – T 2  (9.22)
2 2

This means that a single nonlinear constitutive law remain valid for all flow directions. Then
Eq. 9.22 is a function of T 1 and T 2 and not only their difference. This is the reason why we
must represent the thermal restrictor as a 2-port actually modified R-field, i.e. a MR-element,
and not a 1-port R on a 1-junction. The mass flow has to be added as a signal.

When it is necessary to control the flow of the gas, a valve as a variable area orifice is usually
the type of model used. With the pressure upstream and downstream given, the mass flow
rate is determined by the pressure difference. The basic equation for the flow of a perfect or
ideal gas through a single orifice as a function of orifice or valve area A , nozzle coefficient
C d , upstream pressure P u , and temperature T u , and downstream pressure P d can be com-
puted using the well known relation

2 +1
--- ------------
Pu 2  P  P  
m· = C d A ------------- -----------------   -----d –  -----d  (9.23)
RT u   – 1    P u  P u 

where  is the ratio of specific heats and R is the ideal gas constant.

If however the pressure ratio P d  P u falls below the critical value

 -
-----------
Pd 2 -  + 1
----- =  ----------- (9.24)
Pu   + 1

then we have choked flow and the critical pressure ratio becomes the constant critical value.
For air with  approximately 1.4, the critical pressure ratio becomes

Pd
----- = 0.528 (9.25)
Pu

which means that if the downstream pressure P d falls below 0.528 P u , it no longer affects
the rate of flow in any way. Although the static temperature of the air passing through the
restriction or orifice falls as its velocity increases its stagnation temperature, stagnation
320 Thermodynamic systems modelling

enthalpy and stagnation internal energy remains constant throughout the flow.

Now having established the necessary relations for calculating the mass flow at a restriction
or valve we observe that the preferred causality from the equations are

m·
=  R  P u P d T u T d  A C d   R c v   (9.26)
E·

which we as for the energy flow now recognize as an 4-port R-field with preferred effort in
on all bonds as shown in Fig. 9.5and where  A C d  and  R c v  is the nozzle- and fluid
parameters respectively. This ideal nozzle model is frequently used modelling valves or other
flow restrictions and the selection of parameters are often considered difficult. However,
experience has shown that an initial selection of C d close to 0.7-0.9 yields favourable results.

9.4 Thermal Incompressible Fluid Models

In the preceding sections, we have considered thermal compressible fluid models. In what
follows, we will model thermal incompressible fluid systems where we consider the density
of the fluid to be constant. In spite of their great similarity we find it convenient to treat com-
pressible and incompressible thermal fluid system separately, although it is possible to use
the same approach.

As for the compressible fluid case, a flow model will be coupled to a thermal transport model
which account for internal energy transport and hence, will predict fluid temperatures. The
hydraulic model will account for energy transport in the flowing fluid which often is called
“flow work”.

9.4.4 Basic element Set

In the incompressible case where fluid density is considered constant the basic effort and
flow variables will be:

• Efforts:
T - temperature
p - pressure
• Flows:
V· - volume flow rate
·
Q - heat flow rate

The product PV· represents power in the hydraulic model when convection of kinetic energy
 Thermal Incompressible Fluid Models 321

is neglected. The product TQ· does not represent power since Q· is a power term itself, hence
this will yield a pseudo bond graph case.

As for the compressible case we start with defining a flanged joint or flux boundary with two
lumps of fluid as shown in Fig. 9.6. The temperatures of the associated lumps are considered
T 1 and T 2 , unequal across the joint. The pressure, the volume flow and heat flow however is

1 2

P1 = P2
T1  T2
V· = V·
1 2

Q 1 = Q· 2
·

Fig. 9.6 Pipe port or flux boundary

considered equal at the joint.

P1 P2
1
V· V·
P1 P2
V· 2
2
V· 1
1

T1
Q· 1
R T2
Q· 2
T1
Q· 1
R
T2
Q· 2

Fig. 9.7 Incompressible Pipe port Bond Graphs

In determining the heat flow rate through the joint, case must be exercised when it is possible
for flow direction to reverse. For a flow from section 1 to section 2 the heat flow rate become

Q· 1 = Q· 2 = cV· T 1 (9.27)

where  is mass density and c is the specific heat of the fluid. With a flow from 2 to 1 the
heat flow rate become

Q· 1 = Q· 2 = –  cV· T 2 (9.28)

This problem can be avoided by representing the heat flow rate through the joint by the fol-
322 Thermodynamic systems modelling

lowing constitutive law

· · c
Q 1 = Q 2 = ------  V·  T 1 + T 2  + V·  T 1 – T 2   (9.29)
2

This equation cannot be represented by a 1-port R on a 1-junction since it a function of both

the sum and the difference of the two temperatures T 1 and T 2 . The bond graph model for the
joint is shown in Fig. 9.7with a given causality which will be used permanently, i.e. fixed
causality.

The next element to be defined is the pipe or duct segment shown in Fig. 9.8.

I R

P1 P2
V· V·
L 1
1 1

1 2 C
 T Q· 3 
T1 T T T2
Q· 1
R · 0 · R ·
Q 1 Q 2 Q2
a) b)

Fig. 9.8 Thermal pipe segment model a) schematic b) bond graph

The hydraulic model is shown to have inertia and resistance elements which the modeller
may or may not choose to include in his model. The thermal capacitance stores the energy
difference between incoming and outflowing energies from the element. The temperature
within the element can in the linear case be determined from

Q3
T = T 0 + -------------
- where Q 3 =  Q· dt (9.30)
ALc

where A is the cross-sectional area of the pipe and T 0 is the initial temperature of the fluid.

Using a single temperature to characterise the entire slug of fluid means that we are assuming
instantaneous mixing which might be questionable for very long lines. To remedy this, one
has to break up long pipes into a series of short segments for which the instantaneous mixing
assumption is reasonable.
 Thermal Incompressible Fluid Models 323

The last element we will consider here is the open tank shown in Fix FF

P P
·
V1
1 ·
V3
P V· 2
C
 T 2 Q· 2 
2
T1 T2 T2 T3
3
1
· R · 0 · R ·
Q1 Q1 Q3 Q3
a) b)

Fig. 9.9 Open tank a) schematic b) bond graph model

For the tank the hydraulic capacitance is

A
C H = ------ (9.31)
g

where g is the gravity and A is the tank cross-sectional area. The thermal capacitance is
given as

C T = mc = V  t c (9.32)

The thermal capacitance is shown to depend on the fluid volume in the tank V  t  , which is
the state variable on the hydraulic side. Therefore, the tank capacitance has to be modelled as
a 2-port C-field as shown in Fig. 9.9.

As shown in Fig. 9.9a single temperature is assigned to the tank content. Whether this is a
reasonable assumption might be problematic in some applications. However, with low flows
and slow changes, good mixing may occur. It is not as easy to break up tanks into many
lumps as it is to segment a pipe, i.e. it is a distributed problem.

Example 9.1

In the first example of modelling incompressible thermofluid systems we consider a pipe

324 Thermodynamic systems modelling

delivering hot water from a tank through a double-pipe heat exchange as shown in Fig. 9.10

xu

V· 1
xd
h

V· 2

Fig. 9.10 Sketch of thermofluid system

The hot water flowing from the tank is heated from a secondary fluid through a counterflow
double-pipe heat exchanger. The flow into and out of the tank can be varied by adjusting the
position x u and x d of the valves shown in Fig. 9.10. For illustrative purposes the heat
exchange pipes will be subdivided into only two segments where convective and conductive
bond graphs will be coupled. The hydraulic model of the pipes will be non-dynamic in that
the volume flow rate is set by pressure and valve position, i.e. no inertia are present.

xu
Inlet R Tank Pipe I R xd
(heated fluid)
2 6
1 3 5 P P 7
Se 1 0 1 Se:Pd
V· 2
4
V· u
C C C
 12
 16

9

V· u cT u :Sf 8
0
10
R
11
0
14
R
15
0
17
R
13 18

26 21
0 R 0 R

25 20

Sf:V· 1 cT 1
28 24 23 18
R 0 R 0
27 22
V· 1 V· 1
C C
Pipe II (heating fluid thermal model)

Fig. 9.11 Tank heat-exchanger pseudo bond graph

 Thermal Incompressible Fluid Models 325

By a proper assembly of the elements shown in Fig. 9.9 and Fig. 9.10 we obtain a system
model as shown in Fig. 9.11. The hydraulic models are drawn by solid lines and the thermal
models are drawn by dotted lines. From bottom towards the top, we have first the convective
model for the heating fluid, then the conductive part representing the heat transfer between
the two fluids and so the convective model for the heated fluid. The thermal models consists
of a single C-element within each pipe segment representing the thermal capacitance of the
fluids and connected to two R-elements representing the ports at either end. At the pipe exit is
used only half of a 2-port R since only the energy flow out of the C-element is important. The
tank model is shown as a 2-port C-field.

As an instructive exercise let us write the system state equations where the desired output is
T 17 and T 28 .

According to the bond graph shown in Fig. 9.11 there are six state variables which yields the
following state vector.

T
X =  q 4 q 9 q 12 q 16 q 22 q 27  (9.33)

Writing the state equations in the order of the variables in the state vector we obtain

q· 4 = f 4 = f 3 – f 5 = f 2 – f 6 (9.34)

With the assumption that the flow-pressure drop relation for the two valves can be expressed
by Eq. 3.82 we obtain

q 2 q
q· 4 = C d A  x u  ---  Se 1 – -----4- – C d A  x d  ---  -----4- – P d
2
(9.35)
  C4    C4 

where C 4 = A T   g  .

q· 9 = f 9 = f 8 – f 10 = Sf 8 – cV· 6 T 9 (9.36)

2 q q
where V· 6 = C d A  x d  ---  -----4- – P d and T 9 = T 0 + -----9- plus C 9 = A T hc .
  C4  C9

q· 12 = f 12 = f 11 + f 13 – f 14 (9.37)

f 11 = f 10 = cV· 6 T 9

f 13 = f 26 = UA p  T 27 – T 12 
326 Thermodynamic systems modelling

where A p is the surface area inside the pipe segment.

q 12
T 12 = T 0 + -------
-
C 12

q 27
T 27 = T 0 + -------
-
C 27

C 12 = A in L seg c

C 27 =   A out – A in L seg c

where A in is inside pipe cross-sectional area, A out is outside pipe cross-sectional area, L seg
is the length of the pipe segment.

q· 16 = f 16 = f 15 + f 18 – f 17 (9.38)

f 15 = f 14

f 14 = cV· 6 T 12

f 17 = cV· 6 T 16

q 16
T 16 = T 0 + -------
-
C 16

C 16 = C 12

f 18 = f 21 = UA p  T 22 – T 16 

q 22
T 22 = T 0 + -------
- C 22 = C 27
C 22

q· 22 = f 22 = f 19 – f 23 = Sf 19 – f 23 – f 20 (9.39)

f 23 = cV· 1 T 20
 Multicomponent two-phase thermodynamic systems using pseudo bond graphs 327

q 22
T 22 = T 0 + -------
- C 22 = C 27
C 22

f 20 = f 21

q· 27 = f 27 = f 24 – f 25 – f 28 (9.40)

f 24 = f 23 f 25 = f 26 f 28 = cV· 1 T 27

q 27
T 27 = T 0 + -------
-
C 27

We have here demonstrated how we can write a set of nonlinear state equations in a system-
atic manner with the use of the causal assignment, power directions and numbered bonds.

We have now reached a stage where we have a set of basic building blocks for thermofluid
modelling. The bond graph representation yields a clear definition of interfaces and require-
ments for assembling these basic elements into larger models for simulation. Working with
thermofluid models you may face situations where the working fluid is no longer a perfect
gas, a multi phase fluid or a multi component mixture. However, the framework which we
have developed so far can be nicely extended to include these situations and the objective of
the next few sections is to introduce extensions which enables us to handle multi component
mixtures, multi phase fluids and real gas thermodynamics.

9.5 Multicomponent two-phase thermodynamic systems using

pseudo bond graphs
Thermofluid processes are found in many engineering systems and the number of applica-
tions where a thorough analysis of both the steady-state and dynamic behaviour of such sys-
tems are required has been increasing during the last decades.

Modelling thermofluid systems are however not a trivial task since the choice of governing
equations and state variables, and calculation of thermodynamic properties are closely
related to each other. The calculation of thermodynamic properties for multi component mul-
tiphase fluids is in itself a research topic within the chemical engineering community. For
simple pure species and mixtures methods and data are however available which can be uti-
lized in modelling most thermofluid systems of interest to the industry.
328 Thermodynamic systems modelling

The rapid development of computer power has also made it possible to analyse real problems
with a reasonable use of resources. The result is that there nowadays exists computer pro-
grams for almost any analysis of interest to the engineer. However, often the system to be
modelled contains components or submodels from different engineering disciplines, and the
required flexibility necessary to analyse the system is not available using a specific tool. It is
therefore mandatory to have available an modelling technique or framework which supports
such a flexibility and which supports the modelling process especially.

A modelling technique which answers most of the requirements put forward by engineers
modelling complex systems is the bond graphs method. Bond graphs have since they were
invented by professor H. M. Paynter at MIT in the late fifties and brought into applications
by professors Karnopp and Rosenberg [5], proved themselves within different areas such as
mechanical, electrical and hydraulic systems. An extensive list of publications can be found
in [6][7]. Within the thermofluid system domain the number of publications from research
and applications has however been more scarce.

Using bond graphs simplifies modelling of systems of mechanical, hydraulic, electrical and
thermodynamical energy domains, because it uses the same unifying graphical notation in all
cases. It clearly supports both an inter and intra disciplinary modelling philosophy.

As soon as the bond graph system model is established, a set of first order ordinary differen-
tial equations can be automatically derived using preferably a computer program [8][9].
Solving the equations are likewise preferably performed using an ordinary differential equa-
tion solver as ACSL [10] or Matlab [11].

Despite the fact that the research on bond graph representation and applications to thermof-
luid systems is somewhat limited, some important contributions deserve to be mentioned.

One of the first researchers to model thermofluid systems by the way of bond graphs was
Thoma [12][13]. He proposed to use temperature as the effort variable and entropy flow as
the flow variable, a natural choice since their product gives power just as true bond graphs.
Although this approach has many advantages especially when systems of different energy
domains are present, thermodynamic relations are complex and entropy flow and accumula-
tion is not very well understood by most engineers. For some applications and analysis it may
still be advantageous.

Other suggestions for modelling thermofluid systems is given by Brown [14]. He proposes to
include a convection bond for the flow of a pure substance with two effort variables, stagna-
tion enthalpy and stagnation pressure, and one flow variable, mass flow. His motivation for
introducing this additional convection bond is that two variables, as in true bond graphs, is
not sufficient to define the thermodynamic state of the fluid. The result of this new represen-
 Multicomponent two-phase thermodynamic systems using pseudo bond graphs 329

tation is a notation that differs significantly from conventional bond graphs, does not contrib-
ute to the understanding of the underlaying physical process and are not easily incorporated
into most bond graph modelling environments.

Yet another contribution is given by Shoureshi [15]. He proposes a bond graph notation for
modelling two-phase variable density systems using the same effort and flow variables as
Thoma. The result of the choice of variables is a very complicated bond graph with several
signal bonds, modulated transformers and gyrators. In addition to a complex bond graph the
choice of variables led to additional complications computing the state of the fluid from
known thermodynamic tables. He solves this problem by developing a computer program
which transforms the known thermodynamic tables to comply to his selection of state varia-
bles selected. The work involved in developing such transformations are in general not
straight forward.

In another approach Karnopp [16] introduces the pseudo bond graph concept for modelling
thermodynamic systems. He introduces two bonds to represent the energy flow between sys-
tems, and his selection for effort, flow and state variables are pressure and temperature
( P T ), mass and energy flow (m·  E· ) and mass and internal energy (m E ) respectively. These
selections do not multiply to yield power as required for true bond graphs and as such repre-
sents a special class within the bond graph method. For thermodynamic systems his selec-
tions however results in state variables and effort and flow variables which are well known
for most engineers and easy to understand, something which is not the case using true bond
graphs. Utilizing thermodynamic tables and iteration techniques he also shows how to extend
the pseudo bond graph concept to two-phase thermodynamic systems. The pseudo bond
graph concept are presented in more detail in the next section as our extension for multi com-
ponent fluids is based on this concept.

A final contribution to the modelling of two-phase thermofluid systems which deserves to be

mentioned are given by Moksnes [17]. Based on the pseudo bond graph concept as proposed
by Karnopp he formally merges pseudo bond graphs and thermodynamic fundamental rela-
tions and equations of state. In his thesis thermodynamic properties are calculated based on
the state variables and this fundamental relation. For fluids where a fundamental relation is
not available, he shows how to establish the relation from a general volumetric equation of
state and heat capacity data. This is developed into an flexible and powerful framework for
modelling single component one and two-phase thermofluid systems.

Despite the fact that pseudo bond graphs do not comply with true bond graphs, they have
been applied in a number of engineering systems. Examples of use are in modelling of com-
plete diesel engine systems by Engja [18], Granda [19] and the author [20], one-dimensional
compressible fluid flow by Strand [21], building simulation by Thoma [22], multi component
gas mixtures by the author [23] and two-phase systems by Moksnes [17].
330 Thermodynamic systems modelling

Reviewing the list of applications mentioned, a natural next step would be to extend the tech-
nique to include systems where the working medium is a mixture of different components
and with single or multiple phases. This is the purpose of the following sections of this paper.

The paper is organized into four sections. The first section presents the fundamentals of mod-
elling thermofluid systems using the pseudo bond graph concept. The second section dis-
cusses and presents a proposed extension of the pseudo bond graph concept for multi
component single and multi phase thermofluid system. The third section shows how thermo-
dynamic relations for the constitutive laws can be derived from the selected state variables
and a fundamental relation for multi component fluids. The last section shows how to utilize
the extensions introduced to model a simple thermofluid system.

9.5.5 Fundamentals
Modelling thermofluid systems we are faced with the challenge of selecting both which state
equations or state variables to represent the system, and a model for calculation of the ther-
modynamic properties. Thermodynamic data for most fluids are available as tables, equation
of states or as fundamental relations. The selection of state variables should reflect this and
should support the use of thermodynamic data in its original format.

Such a selection of state variables is used by Karnopp [16] in his concept called: pseudo bond
graphs.

Following his proposition, the exchange of mass and energy between two systems can be rep-
resented schematically using a double bond representation as shown in Fig. 9.12. The pseudo
bonds do not carry variables which multiply to yield power as in true bond graphs. A “ther-
mal” bond is drawn with a dashed line and the efforts and flow on both bonds as shown in
Fig. 9.12.

A B
p
m·
T
E·

Fig. 9.12 Energy exchange between system A and B. Pseudo bond graph representation.
 Multicomponent two-phase thermodynamic systems using pseudo bond graphs 331

The effort and flow variables he proposes are:

• Efforts:
T - temperature
P - pressure
• Flows:
m· - mass flow rate
E· - total energy flow rate ( m· H )
· - heat energy flow rate
Q
V· - rate of change of volume
• State variables:
m - mass
E - total energy ( mU )
V - volume

Example 9.2 Thermofluid accumulator

Modelling a single component thermofluid accumulator using the pseudo bond concept is
shown in Fig. 9.13a and b. The pseudo bond graph for the system consists of a 0-junction

C
P m·
V T E·
m· in P T m· out P P
m· in m· out
H in T T
H out
E· in E· out

(b) Q·
(a)
Q·
Sf

Fig. 9.13 Accumulator a) schematic b) pseudo-bond graph model

structure representing the continuity of mass and energy, the two state variables selected, and
a C-field giving the relations between the displacements and the efforts. The fixed causality
caused by the selection of variables on the bonds in the pseudo bond graph concept, is shown
on the bond graph in Fig. 9.13b. This fixed causality is the price we have to pay for the selec-
tion of variables.

The state equations for the system can be derived directly from the pseudo bond graph to
give:
332 Thermodynamic systems modelling

m· = m· in – m· out
· (9.41)
E· = m· in H in – m· out H out + Q

where m· in is the mass flow into the system, m· out is the mass flow leaving the system, H in is
the enthalpy of the fluid entering the systems and H out is the enthalpy of the fluid leaving the
system. The heat added to the system is represented by Q· .

The constitutive laws of the C-field is given by

P =  P  m E V 
(9.42)
T =  T  m E  V 

where m and E are the state variables mass and energy and V , here given as a parameter, is
the volume of the accumulator. The particular laws to be applied depends on the type of sys-
tems modelled, i.e. number of phases and the thermodynamic model for the fluid, i.e ideal
gas, perfect gas or real fluid. For a perfect gas assumption simple relations can be deduced
from the thermodynamic equation of state and the relation between the temperature and inter-
nal energy, E = mC V T .

1 E
T = ------ ----
CV m
(9.43)
mRT RE
p = ------------ = ------ ---
V CV V

Assuming that the fluid can be represented by a ideal gas law, i.e. where the specific heats are
functions of temperature only and that an ideal equation of state is valid, the temperature can
be computed by a simple iterative scheme given by:

1 E
T = --------------- ---- (9.44)
CV  T  m

and the pressure explicitly derived using

T
P = mR --- (9.45)
V

Extending the pseudo bond graph to multi component two-phase systems is the topic of the
next section.

9.5.6 Modeling multicomponent two-phase systems

The pseudo bond graph concept as used for modelling single component one or two-phase
 Multicomponent two-phase thermodynamic systems using pseudo bond graphs 333

thermodynamic systems uses three state variables, mass m , total internal energy E and vol-
ume V , or equivalently m U V . Although there are only two state variables in the pseudo
bond graph, also the volume often, presented as an parameter, is a thermodynamic state vari-
able and appears as such for variable volume systems. These 3 variables are found to com-
pletely specify the thermodynamic state of the system. Extending the model to include multi
component two-phase thermodynamic systems the amount of state variables and type of
information necessary to fix the state of the system has to be considered carefully. However,
starting from the classical pseudo bond graph concept, a preferred selection of state variables
has to maintain as much as possible of the origin and only add the additional features neces-
sary. This means that the total mass m and internal energy U are preferred as state variables
in our new model, in addition to some new variables.

State variables
Gibbs phase rule [24] states the number of independent, intensive state variables F necessary
to describe the thermodynamic state of each phase in a non-reacting multi component multi
phase fluid in equilibrium is given by

F = C–P+2 (9.46)

where C is the number of components of the mixture and P is the number of phases present
in the system. By independent state variables in this respect we mean non redundant pieces of
information about the thermodynamic state of the system. Temperature, pressure and C-1
mole fractions forms a set of such variables, but temperature and C mole fractions does not,
since the sum of fractions is unity, so only C-1 mole fractions are independent. In addition we
need P – 1 thermodynamic properties to fix the mass distribution between the phases and one
extensive variable to determine the total size of the system.

For a multi component single-phase fluid consisting of C components, the number of state
variables necessary to specify the thermodynamic state of the fluid is F = C + 1, and to spec-
ify the total size of the system one extensive property is needed.

For a multi component two-phase fluid consisting of C components the number of intensive
variables necessary to fix the state of the mixture is: F = C for each individual phase. To fix
the mass distribution between the phases one additional variable has to be fixed and one
extensive variable needs to be specified to determine the total size of the system.

The difference between these two systems is that for the multi component two-phase system
one less intensive parameter has to be specified but now the number of state variables has to
be given for each individual phase. In addition one additional variable has to be specified to
fix the mass distribution, making the total number of parameters equal if the selection of var-
iables is done with some care.
334 Thermodynamic systems modelling

Applying these requirements to pseudo bond graph modelling of thermofluid multi compo-
nent single- and two-phase systems we conclude that the number of state variables necessary
are C+2. In addition the selected state variables must be independent, and at least one exten-
sive variable must be given to fix the total size of the system. The selection of variables must
also be independent of number of phases.

To support our selection of state variables we may look into fundamental thermodynamics. In
[25] Callen states that the equilibrium state for a thermodynamic system is completely char-
acterized by the extensive variables internal energy, U , volume V and number of moles
N 1 N n of each chemical component present in the mixture.

He also states that for any equilibrium state of the thermodynamic system there exists a func-
tion called entropy S , given by the extensive parameters U , V and N 1 N n . From this func-
tion also known as a fundamental relation, all thermodynamic properties can be derived. This
means that it exists a unique relation between the state variables U , V and N 1 N n , and the
fundamental relation, and if such a relation had been available all thermodynamic properties
could have been calculated.

An important property with the fundamental relation is also stated by Callen. The entropy for
a mixture of n components is a continuous, differentiable and monotonically increasing func-
tion of the energy U , volume and number of moles of each component. This means that by
use of Lagrange transformations the fundamental relation S , can be transformed into other
relations with the same properties. Also from these fundamental relations all thermodynamic
properties can be calculated. Well known transformations are

S  U V N 1 N n  H  S P N 1 N n 
U  S V N 1 N n  G  T P N 1 N n 
A  T V N 1 N n 

where H is the enthalpy, U is the total internal energy, G is Gibbs energy and A is the
Helmholtz energy.

Unfortunately there exists no fundamental equation in temperature as

T = T  U V N 1 N n  which is the relation corresponding to our preferred selection of state
variables for use in modelling multi component two-phase thermodynamic systems. Even if
this is not true looking at the selection of fundamental relations available, we find that com-
pared to this the Helmholtz energy only differs having replaced internal energy with temper-
ature. Knowing internal energy, volume and number of moles of each component which is
equivalent to mass of each component, the temperature is available as an implicit relation
which must be solved by an iterative procedure. Once having the temperature all the other
thermodynamic properties can be calculated directly from the Helmholz energy fundamental
 Multicomponent two-phase thermodynamic systems using pseudo bond graphs 335

relation. Utilizing the monotone properties of the fundamental relations, an iterative solution
should give a fast and unambiguous solution for any number of components and any number
of phases.

Summarizing our discussion we have argued that to extend the pseudo bond graph concept to
model multi component two-phase systems we need C+2 state variables, and equivalently
C+2 differential equations. A selection of state variables which is consistent with fundamen-
tal thermodynamics are internal energy U , volume V and number of moles N 1 N n of
each component. As we can see the selection of state variables for single component one and
two-phase fluids falls within this selection of state variables, because the mass divided with
molar weight gives number of moles.

Following these findings we propose an extension to the pseudo bond graph concept which is
capable of modelling multi component two-phase thermofluid systems. Based on the discus-
sions above the following effort, flow and state variables are proposed for modelling a two-
phase multi component thermofluid thermodynamic systems:

• Efforts ( e ):
P - pressure [Pa]
c2 - mass fraction for component 2 [-]
: :
cn - mass fraction for component n [-]
T - temperature [K ]
• Flows ( f ):
m· - total mass flow rate [kg/s]
m· 2 - mass flow rate component 2 [kg/s]
: :
m· n - mass flow rate component n [kg/s]
E· - total energy flow rate ( mH ) [J/s]
• State variables (displacements) ( q ):
m - total mass [kg]
m 2 - total mass of component 2 [kg]
: :
m n - total mass of component n [kg]
E - total energy (i.e. E = mU) [J]
V - total volume [m3]

The pseudo bond graph model for a thermofluid accumulator, as in Fig. 9.13, is shown in Fig.
9.14. The n – 1 added pseudo bonds with the mass fractions of component i as effort and
mass flow rate of component i , are inserted between the pseudo bonds for pressure and tem-
perature as given by Karnopp. These new pseudo bonds are shown as dash-dot lines to distin-
336 Thermodynamic systems modelling

guish them graphically from the other bonds. This extended graph maintains the structure of
the classical pseudo bond graph concept for single-component one and two-phase mixtures,
only adding the additional features making it easy to implement into any bond graph model-
ling environment.

State equations for the thermofluid accumulator are given as:

C
P m·
c m· c m·
2 2 n n
...
T E·

P P
m· in c 2 in m· out c
2 out
m· 2 in m·
2 out
....

...

....
.

c c
n in n out
m· n in m· n out T out
T in
· ·
E in E out

Q·

Sf

Fig. 9.14 Pseudo bond graph model of multicomputer gas accumulator

m· = m· in – m· out
m· 2 = m· 2 in – m· 2 out
:
(9.47)

m· n = m· n in – m· n out
E· = m· H in – m· H out

and the constitutive laws for the C-field are given as:
P =  P  m  m 2   m n E  V 
c i =  c i  m  m 2   m n E  V  i = 1n (9.48)
T =  T  m m 2  m n E V 

The selection of the mass fractions as effort variables on the pseudo bonds are only one of
several possible choices. Any other variable fixing the mixture composition could also be
used. For a gas phase the partial pressures defines the molar compositions of the fluid. This is
however not the case in multi phase systems, and the proposition of effort variables from [23]
 Multicomponent two-phase thermodynamic systems using pseudo bond graphs 337

must therefore be disregarded in this case. Using mass fractions (or molar fractions) are more
general selections than partial pressures.

In the next section procedures for calculating the constitutive laws of the C-field are outlined,
but first let us discuss in detail an example modelling multi component perfect- or ideal gas
systems.

Example 9.3 Gas mixing

A simplified gas mixer found in a combustion experimental rig for testing combustion prop-
erties of various natural gases is shown in Fig. 9.15. The two bottles B1 and B2 contain meth-
ane and air, delivered through non-return valves to a mixing chamber. The resulting gas
composition are leaving the mixing chamber through an exit valve. Building a bond graph

Air Methane
bottle bottle
B1 B2
Mixing
valves

Mixing chamber:MC
Exit valve

Fig. 9.15 Gas mixing system

model of this system we may observe that the classical single component pseudo-bond graph
representation do not handle this case, i.e. we have to extend the pseudo-bond graph to be
able to represent multi component systems in accordance with the discussion of the previous
section. However, now we restrict ourselves assuming that the gas mixture will follow a per-
fect gas assumption.

According to the discussion in the previous section the number of independent intensive var-
iables necessary to calculate the thermodynamic state of a fluid in equilibrium showing only
one phase, i.e. the gas phase, and two constituencies is given by the Gibbs phase rule to be 3
plus one extensive variable fixing the size of the problem. This means that we have to adopt a
pseudo-bond graph using 3 bonds with the pseudo power variables pressure and mass flow
P m· , mass fraction and mass flow rate of constituent 2 c 2 m· 2 plus temperature and total
energy flow rate T E· = m· h . This then yields the state variables mass m , total mass of con-
stituent 2 m 2 and total energy E . The constitutive laws for the C-field representing the fluid
accumulator then in general becomes
338 Thermodynamic systems modelling

P =  p  m  m 2 E 
c 2 =   m  m 2 E  (9.49)
T =  T  m  m 2 E 

where fluid parameters and total volume has to be considered also. For a perfect gas assump-
tion this becomes using the perfect gas equation of state and the relation between the internal
energy and temperature

E
T = ---------------------------
2
-
mi
m  ----- c v i
m
i=1

 m T
2 where m1 = m – m2 (9.50)
P = m   -----i R i ---
i = 1 m  V
m–m
c 2 = ----------------i
m
Adopting an ideal gas assumption the only modification necessary is to modify the expres-
sion for the temperature in Eq. 9.50 due to the temperature dependence of the specific heat as

E
T = -----------------------------------
2
- (9.51)
mi
m  ----- c v i  T 
m
i=1

which means you have to solve an implicit equation using iteration techniques as Newton-
Raphson iteration scheme to obtain a solution for the temperature. However, for most species
the temperature function of the specific heat is a monotone function of temperature and the
iteration will converge quickly.

The second fundamental bond graph element in fluid flow, the restrictor, also has to be mod-
ified according to the new selection of pseudo-bond variables. Using a bond graph R-field for
representing the restriction the constitutive laws using an ideal nozzle flow model becomes

Pu
-    u sign  P a – P b 
m· a = m· b = C d A ---------------
Ru Tu
m· 2a = m· 2b = m· c 2 u (9.52)
2
·
E·a = E b = m·  c i u h i  T u 
i=1

where the a and b subscripts represents the in- and outflow side of the R-field and the added
subscript u indicates that this variable has to be selected as the upstream variable according to
 Multicomponent two-phase thermodynamic systems using pseudo bond graphs 339

the flow direction, and where

1 
2 +1 --- ------------
2  --- ------------ 2
2 –1
    = -----------
–1 
-   –     1     ------------
  + 1

(9.53)
1
+1 --- 
------------ 2 ------------
–1 –1
  ------------     ------------
2 2
=
  + 1   + 1

where  = P d  P u and  = c p u  c v u .

The pseudo bond graph for the system in Fig. 9.15 is shown in Fig. 9.16. The bottles and the
mixing chamber are modelled using the fluid accumulator developed and the valves at the
mixing chamber inlet and outlet are modelled using the ideal nozzle model developed. The
opening area of the mixing valves are 0.785 10-4 m2. The area of the exit valve of the mixing
chamber is 1.0 10-4 m2. In this example no heat loss from the bottles or from the mixing
chamber is assumed. The nozzle flow coefficient for all the valves are set to unity. Using a
perfect gas assumption for the working media the thermodynamic properties necessary are.:

Properties Air Methane

R [KC/kg] 287.0 518.35
c v [KC/Kg] 716.5 1735.4

The total pressure and temperature in the bottles are initially between 100 and 300 bar and
300 K. The partial pressure p 2 in the air and methane bottles is 0 and 300 bars respectively.

The initial conditions in the mixing chamber is 1 bar and 300 K, completely filled with air,
i.e. the partial pressure p 2 is 0. The volume of the air bottle is 2 m3 and the volume of the
methane bottle is 0.5 m3.

Deriving the model equations from the bond graph in Fig 6 is shown next, although this is a
process which efficiently can be given to computer programs like MS1 [13] or 20Sim [15].
Here we only give an outline of the equation generation process utilizing standard bond
graph techniques.

The state variables of the model are the displacements q 1 , q 2 , q 3 , q 13 , q 14 , q 15 , q 25 , q 26

and q 27 , or m B1 , m 2 B1 , E B1 , m MC , m 2 MC , E MC , m B2 , m 2 B1 and E B2 as named in the text
above. Using the constitutive laws for the mixing chamber or C-field MC, the efforts of each
bond can be evaluated using
340 Thermodynamic systems modelling

C :B1 C :B2
1 2 3 25 26 27

A1 C :MC A2
4 5 6 13 14 15 22 23 24

7 10 16 19

9 R 11

12
17

18 R 20

21

28

f
29 30 31

R S A3

32 33 34

Se :atm
Fig. 9.16 Complete pseudo bond graph model for the simple gas mixer.

e 13 

e 14  = PCT  q 13 q 14 q 15 V Y  (9.54)

e 15 

where PCT is a function returning the efforts on each power bond of the C-field, i.e. the pres-
sure, the mass fraction of methane and the temperature, and Y is a vector identifying the con-
stituents of the mixture. Repeating the use of the PCT-function on C-field B1 and B2 gives
the efforts e 1 , e 2 , e 3 , e 25 , e 26 and e 27 . The effort source Se sets the pressure, the mass frac-
tion of methane and temperature at ambient conditions. Using the information present in the
bond graph all efforts can now be assigned.The state equations for the mixing chamber are:

q· 13 = f 13 = f 10 + f 16 – f 29
q· 14 = f 14 = f 11 + f 17 – f 30 (9.55)
q· 15 = f 15 = f 12 + f 18 – f 31 – f 28

and the flows of each bond is calculated using

f 10 

f 11  = Nozzle A1  e 7 e 10 e 8 e 11 e 9 e 12 A C d Y  (9.56)

f 12 
 Multicomponent two-phase thermodynamic systems using pseudo bond graphs 341

f 16 

f 17  = Nozzle A2  e 19 e 16 e 20 e 17 e 21 e 18 A C d Y  (9.57)

f 18 

f 29 

f 30  = Nozzle A3  e 29 e 32 e 30 e 33 e 31 e 34 A C d Y  (9.58)

f 31 

where Nozzle A1 , A2 and A3 is a function implementing Eq. 9.52 returning the total mass
flow, the 2nd constituent mass flow and the total energy flow. Writing the state equations for
the bottles B1 and B2 similar to Eq. 9.55 closes the equation writing process.

Fig. 7 shows the simulated pressures in the bottles and in the mixing chamber as a function of
time for the case where the methane bottle initially is at 100 bar. The molar fraction of meth-
ane in the gas leaving the mixing chamber is also shown in the same figure. The results show

Pressure B1
Pressure B2
Pressure MC

Mass fraction CH4.

Molar fraction CH4 [-]

in mixing chamber
Pressure [Pa] 107

Time [s]

Fig. 9.17 Pressure in the bottles and mixing chamber, and mass fraction of methane leaving
the mixing chamber for an initial pressure of bottle B2 of 100 bar.

that the lowered total pressure of the methane bottle results in a cut-off period where pure air
is delivered and no ignition of the mixture delivered will be possible.

In Fig. 9.18 the simulated molar fraction of methane leaving the mixing chamber is shown as
a function of time for initial pressure of the methane bottle of 100, 200 and 300 bars respec-
tively.

The mixture compositions dependence on the pressure in the bottles are clearly documented
with the use of the model developed, and the methane cut-off period is nicely captured for the
342 Thermodynamic systems modelling

low initial pressure case.

0.6
Pressure : 100 bar
InitialPressure
pressure: 200300
bar bar
InitialPressure
pressure: 300200
bar bar

0.5 Initial pressure 100 bar

fraction [-]CH4 [-]

0.4
Molarfraction

0.3
Molar

0.2

0.1

0
0 20 40 60 80 100 120 140 160 180 200
Time [s]

Fig. 9.18 Mixture composition in mixing chamber for methane bottle initial pressure of 100,
200 and 300 bar.

9.5.7 Constitutive laws for multicomponent mixtures

Extending our modelling efforts to an increasingly larger number of systems we also face the
need for not only perfect or ideal gas thermodynamics, but also real gas thermodynamics and
even multi phase and multi component thermodynamics. In our development so far we have
discussed the required modifications to the bond graph elements to be able to model such
systems. These modifications are such that having extended the bond graph with additional
bonds to represent multiple components, the bond graph accumulator model in itself is gen-
eral in the sense that the differential states is valid for any thermodynamic assumption. How-
ever, the calculations obtaining the pressure P , the mass fractions c i and the temperature T ,
knowing the total mass of each constituent m i , the total internal energy E = mu and the
volume V is increasingly complex as we introduce non-ideal thermodynamics and multiple
phases. This is the topic on the following sections.

Single component one and two-phase systems

In [17] a solution procedure for single component single and two-phase systems using the
state variables mass m , internal energy U and volume V is outlined. The computation of
thermodynamic variables is based on using the Helmholtz energy fundamental relation. From
this fundamental relation, i.e. a equation of state or an EOS, for a single species all necessary
 Multicomponent two-phase thermodynamic systems using pseudo bond graphs 343

thermodynamic properties can be evaluated at any phase. The solution procedure then
becomes:

At the starting point the differential state variables m , E and V are known. From these the
mixture molar volume V and the mixture molar internal energy U can be found knowing
the molecular weight of the mixture. After guessing a temperature the specific volume of the
liquid and vapour phases, and corresponding pressure are calculated by an iterative solution
of the nonlinear system of equations given by

L V
T = T = T guessed
L
P  T V  = P  T V 
V (9.59)
L V
G  T V  = G  T V 

which expresses the criteria for equilibrium between the phases. Having found the pressure
and specific volumes at this temperature, the internal energy U for each of the phases are
found from the Helmholtz energy relation as

II
A  T  V 
U  T V  = A  T V  – T  -------------------------
II II
(9.60)
 T V

(where II represents L or V respectively). The mixture internal energy is calculated using

the lever rule as

L V
U = U  T V   1 – x  + U  T V x (9.61)

where the vapour quality x is given by

L
V–V
x = -----------------
V
-
L
(9.62)
V –V

Iterating this calculation sequence for a new guessed temperature until the known internal
energy state variable U and the calculated internal energy U is within a given solution tol-
erance produces a solution when the mixture is within the two-phase region.

In the single phase region the specific volume is known and the molar internal energy U for
the guessed temperature can be calculated directly using Eq. 9.60.

Subsequently all necessary thermodynamic properties can be found from the Helmholtz
energy fundamental relation for both the single and two-phase region.

For multi component mixtures the situation is somewhat more complicated, and this is out-
344 Thermodynamic systems modelling

lined in the next section.

Multicomponent one and two-phase mixtures

In order to use U , m i and V state variables for modelling multi component one and two-
phase mixtures, the relation between these and other thermodynamic properties such as tem-
perature and pressures must be established. This is the purpose of this section.

For any substance, regardless of whether it is pure or a mixture, most thermodynamic proper-
ties of interest can be calculated from thermal and volumetric measurements. Volumetric data
are most often available as volumetric equations of state that uses temperature and volume as
independent variables, P = P  T V  . Thermal data are often presented as ideal gas heat
capacity data as function of temperature. Both given for pure substances only.

In extending these data to mixtures of substances it is assumed that the same equations can be
used if it is possible to establish specific mixture rules to calculate the parameters of the
equation of state. No unified set of mixing rules exists, instead different sets are normally
applied for different equations of state and also dependent on the type of fluid. For dissimilar
fluids these rules can become quite complex.

In this work the Peng Robinson [26] equation of state is selected as basis for the calculation
of thermodynamic properties. Although other equations of state exists that are more accurate,
we have selected this equation because of its widespread use and modest complexity. Mixing
rules for the Peng Robinson equation can be found in [26] giving the Peng Robinson volu-
metric equation of state for multi component fluids as:

RT aT
P = ------------ – -------------------------------------------------
V – b VV + b + bV – b
2 2
R T c i RT c i
a i = 0,45724 ----------------
- b i = 0,07780 ------------
P c i P c i
n

b=  xi a ii = a i  i
i=1 (9.63)
T -  2
 i  T  =  1 + m i  1 – -------
  T c i 
m i = 0,37464 + 1,54226 i – 0,26992 i2
n n

aT =   xi xj  a ii   a jj   1 – k ij 
i = 1j = 1
where x i is the molar fraction N i  N of component i ,  i is the acentric factor and k ij is the
binary interaction coefficient characterizing the binary mixture formed by component i and
 Multicomponent two-phase thermodynamic systems using pseudo bond graphs 345

j , assuming that ternary or higher effects are negligible. Values for the binary interaction
coefficient has been published for a number of binary mixtures in [27]. This now makes the
equation of state dependent of the mole fractions in addition to mixture temperature and
molar volume, i.e P = P  T V x 1 x 2  x n  .

This enables us to establish a reasonable accurate volumetric equation of state for a mixture,
knowing only the pure component critical temperature, critical pressure and acentric factor,
and the binary interaction coefficients for the mixture. This last coefficient is however often
not available and must be found by experiments, or estimated from similar species.

Likewise the heat capacity data for an ideal gas mixture can be found from

n p
 m
 
m
C V  T x 1 x 2  x n  = xi  C V i T  (9.64)
 
i=1 m=0
m
where is the m -th coefficient in the p -th order polynom representing the heat capacity
C V i
data for component i . Data for many components have been published in [28].

In deriving explicit expressions for calculating particular thermodynamic properties, basi-

cally two approaches is possible. The first is to establish a fundamental relation as the Helm-
holtz energy based on the volumetric equation of state and heat capacity data for a mixture as
shown above, or to develop specific functions for each of the thermodynamic properties of
interest. Which option to select depends on the application. In this work the first option is
selected.

9.5.8 Calculation of thermodynamic data for multicomponent

mixtures from a fundamental relation.
The Helmholtz energy relation is in differential form given [29] as

dA = – P dV – S dT (9.65)

Integrating this relation from a reference state given as ( T ref V ref ) to an arbitrary state
( T V ), recognizing that the Helmholz energy is a state variable an hence the integration path
can be selected in an arbitrary manner, and combining this with the corresponding equation
for the ideal fluid, gives

T V   T V   T V 
ref  

A = –   ------- – P dV –  S dT –   ------- – P dV
RT RT
(9.66)

T  VV  T  V 
 
 T V V

ref ref ref  
346 Thermodynamic systems modelling

In order to evaluate this equation a volumetric equation of state for the mixture and the
entropy for the ideal fluid mixture as a function of temperature have to be established.

The volumetric equation of state selected is the Peng Robinson equation of state together
with the proper mixing rules as given above.

The entropy relative to a reference point can be obtained by valuating

 T V 
C P
S  T V  – S  T ref V ref  =  -----V- dT +  ------ dV (9.67)
T

 V T
 T V 
ref ref

Entropy is a state function and the integration can be performed along an arbitrary integration
path. The specific heat at constant volume for ideal fluid is given as Eq. 9.64.

The change of entropy for an ideal fluid mixture relative to a chosen reference state can be
written as

T V 
ref 
V
   x i C V i  dT
IGM IGM
S  T V x  – S  T ref V ref x  = + R ln --------- (9.68)
T V 
ref ref
V ref

which when inserted into Eq. 9.66 and evaluated gives an expression for the change of Helm-
holtz energy between two arbitrary points T V .

This result can be written in a more compact form as

IGM IGM
A  T V x  – A  T ref V ref x  =  A  T V x  – A  T ref V ref x   +
 A  T  V  x  – A  T  V  x   T  V x – (9.69)
 A  T V x  – A  T V x   Tref Vref x

which expresses that the change in the Helmholtz energy between two points is equal to that
of an ideal fluid mixture undergoing the same change of state plus the departure of the fluid
from ideal fluid behaviour at the end state minus the departure of the fluid from ideal fluid
behaviour at the initial state.

The first term in this equation is the ideal fluid contribution and can be expressed making a
summation over each of the contributing components as

IGM IGM
A  T V x  – A  T ref V ref x   =
N
IGM IGM
(9.70)
 xi  Ai  T V x  – A i  T ref V ref x  
i=1
 Multicomponent two-phase thermodynamic systems using pseudo bond graphs 347

where the summation is of partial Helmholtz energies valid for both real fluid and ideal fluid.
This can be further rewritten as

IGM IGM
A  T V x  – A  T ref V ref x   =
N
 A  T V i  – A i  T ref V i ref  + 
IG IG (9.71)
 xi  i
 RT ln  x i  – RT ref ln  x i 


i=1
IG IG
and using that A i  T V i  = A i  T V  – RT ln  x i  this is finally reduced to:

IGM IGM
A  T V x  – A  T ref V ref x   =
N
(9.72)

IG IG
x i  A i  T V  – A i  T ref V ref   +
i=1

which expressed that the ideal gas Helmholtz energy for each component can be summed
using the temperature and volume of the composite fluid. This simplifies the computation of
the expression because the same expression that is used for the single component Helmholtz
energy function can be used here too.

The second and third term in Eq. 9.70 can be identified as the departure function for the
Helmholtz energy given as:

 A  T V x  – A IGM  T V x   T V x =
 T V 

(9.73)
 RT
------- – P dV
–  V 
 T V 

Evaluating the integral after inserting a volumetric equation of state for P in this equation,
gives us the final expression for the departure function for computation. The evaluation of
the departure function using the Peng Robinson equation of state for single component mix-
tures is given in [17].

The calculation of the Helmholtz energy at a specific temperature and volume, requires the
knowledge of the Helmholtz energy value at a reference point. In this work the ideal gas state
at some temperature and pressure are selected as reference state cancelling the need for eval-
uation of the departure function at the reference temperature.

Most thermodynamic properties is then available as derivatives of the Helmholtz energy fun-
damental relation [29] as:
348 Thermodynamic systems modelling

A
P = – ------
V T x i

A
S = – ------
T V x i

A
U = A + TS = A – T ------
T V x i

A A
H = U + PV = A – T ------ – V ------
T V x i
V T x i (9.74)
2
A
C V = – T -------2-
T V x i
2
 2A  2A
C P = C V + T  -------------   --------2-
 T V xi V T x i

A
G i = --------
N i
T V N i  j

Having available a fundamental equation for the multi component mixture valid for both
phases gives a framework to derive all thermodynamic properties necessary to evaluate the
constitutive laws for the pseudo bond graph of an fluid accumulator.

The evaluation of the output variables P c 1 c 2  c n T is however not trivial. The key
component in the procedure is a multi component vapor-liquid equilibrium computation
(VLE), also known as a flash-calculation. Fig. 9.19 gives the P-V-T and x-P diagrams for a
two-phase multi component mixture respectively. In the P-V-T diagram the phase envelope

Tc (Critical isotherm)
P P
T1, z T=T1
L

L+V
PB

PVL V
PD T1, z
PVT phase envelope

L V
V V V 0.0
x1
0.2
z1
0.4 0.6 0.8
y1
1.0

Fig. 9.19 P-V-T and x-P diagrams for a multi component two-phase mixture.
 Multicomponent two-phase thermodynamic systems using pseudo bond graphs 349

and the critical isotherm is shown. For a fluid at a specific temperature T 1 and composition
z the  T 1 z  -isotherm gives the P-V solution as shown in Fig. 9.19. In the two-phase region
L V
between V and V the solution follows the P-V-line dawn, but now the vapour and liquid
concentration changes to y 1 , x 1 respectively as shown here for the point P VL marked on
both diagrams. This is the equilibrium concentration at this particular temperature and pres-
sure.

A solution procedure assuming simple fluids is outlined in Fig. 9.20. The starting point is the
known state variables internal energy E , total mass m , mass of components 2 to n and vol-
ume V . Computing the molecular weight based on the mass fractions enables us to calculate
the molar values of the internal energy U , volume V and vector of mixture molar fractions
z.

The outer iteration loop is started by guessing an initial temperature T . Calculating the dew-
point and bubble point pressures of this mixture at this guessed temperature and composition
[29], gives the molar specific volume of the mixture at the dew-point and bubble points
respectively. If the given specific molar volume V is larger than the computed specific
molar volume at the dew-point V DP , the solution is found in the vapour-phase region. Like-
wise if the given specific molar volume is less than the computed specific volume at the bub-
ble point V BP , the solution is found in the one-phase liquid region. In both these cases the
composition of the one phase is equal to z and the internal energy can be found directly
using
350 Thermodynamic systems modelling

Known
E  z V 

Guess T

Calulate
Calculate
V P
DP DP
 V   VDP 
Calculate
Calulate
V P
BP BP
 V   V BP 
T V  z known

Iterate on p
Estimate K i L
Calculate x i y i

Calculate Single phase

L L liquid
f f x = z
i i
Solve
P = P  T V x 

Is Calculate
Update K
i U

Solve for L V L
Calculate x i y i fi – fi  f

Single phase
L V vapor
V = V  V – 1 + V V y = z
Solve
P = P  T V y 
Is Calculate
U
V – V   V

Is
U – U   U

Solution found
L V
p T x y V  V  X
U H S 

Fig. 9.20 Solution procedure for calculating constitutive laws for multi component two-
phase systems using the extended pseudo bond graph concept
 Multicomponent two-phase thermodynamic systems using pseudo bond graphs 351

A  T V z 
U = A  T V z  – T --------------------------- (9.75)
T V x i

If U – U   U where  U is some iteration tolerance, the guessed temperature is the solu-

tion sought, and all thermodynamic variables can be computed using Eq. 9.74.

If the known specific molar volume V is between the bubble and dew point specific molar
volumes the solution is found to exist in the two-phase region. This case is more complex
involving a flash calculation [29].

For phase equilibrium to exist in a closed multi component two-phase non-reacting system,
the temperature, pressure and partial molar Gibbs energy for each component must be the
same in each phase. This is expressed by the following system of n + 2 equations

L V
T = T
L
P  T V  x  = P  T V  y 
V
(9.76)
L V
G i  T V  x  = G i  T V  y  i = 1  n

where the vectors x and y represents the molar fractions of the liquid and vapour phases
respectively. The third equation in this nonlinear system of equations is more often given as

L V
f i  T P x  = f i  T P y  (9.77)

which is equivalent, and expresses that the fugacity of each component must be equal in
vapour and liquid phases respectively. As indicated by the system of equations, the solution
is dependent of the composition of the individual phases. The general solution of this system
of equations is not a trivial task. This is itself a large research area and no general procedure
of solution exists that is able to solve this equations in all its applications. The main reason
for this is the behaviour in the critical area, and mixtures showing azeotropic behaviour.
However, for simple fluids solutions for this set of equations can be found using iterative pro-
cedures.

The direct substitution method [29] is the traditional method of solution for two-phase
vapour-liquid equilibrium (VLE) calculations at given temperature and pressure. Assume a
closed two-phase system in equilibrium and at given temperature and pressure with composi-
tion z =  z 1 z 2  z n  and where the liquid and vapour composition are given as
x =  x 1 x 2  x n  and y =  y 1 y 2  y n  respectively. The equilibrium factor is
defined by K i = y i  x i . Material balance of each phase gives
352 Thermodynamic systems modelling

x i = z i   1 +  K i – 1 V 
(9.78)
y i = z i K i   1 +  K i – 1 V 

where V is the vapour phase fraction.

The value of V corresponding to an assumed set of K -factors is found solving the flash
equation

F V =   yi – xi  =   zi  K i – 1     1 +  Ki – 1 V  = 0 (9.79)
i i

F  V  is a monotone function and a root 0  V  1 exists provided that  Ki zi  1 and

 zi  Ki  1 which is the bubble point and dew point criteria respectively. i

The selection of initial values for K i at the given temperature and pressure, is particularly
important near the critical region. Initial values at low pressures can be computed from Anto-
ines equation [29], but for pressures near the critical region the estimation of initial values
and subsequent solution of Eq. 9.79 is difficult and the solution may easily divergence. How-
ever given that a solution of Eq. 9.79 is found, the composition of the liquid x i and vapour y i
can be calculated using Eq. 9.78. Normalizing each composition vector, and calculating new
K -factors using

L
f i  T P x 
K i = -------------------------
V
- (9.80)
f i  T P y 

a new solution of the vapour fraction V is done solving Eq. 9.79. Repeating this procedure
iteratively gives a solution for the composition of each phase respectively for this tempera-
ture and pressure.

Knowing the temperature, pressure and composition of each phase, the molar volume of each
phase is solved using the volumetric equation of state for each phase respectively. The mix-
ture molar volume is calculated using

L V
V =  1 – V V + VV (9.81)

This is then compared to the known volume, and utilizing an iterative solution procedure iter-
ating on pressure until V – V   V .

Based on this pressure, temperature, vapour fraction and composition of the liquid and
vapour phases respectively, the internal energy of the mixture can be found as
 Multicomponent two-phase thermodynamic systems using pseudo bond graphs 353

L L V V
U = U  T V  x   1 – V  + U  T V  y V (9.82)
L L V V
where U  T V  x  and U  T V  y V are calculated from the Helmholtz energy funda-
mental relation using Eq. 9.75.

If E – U   U the guessed temperature is the solution sought, and all thermodynamic vari-
ables can be calculated for each phase respectively using Eq. 9.74.

Although this solution procedure is not optimized for fast computations it is found to be
robust for most cases except close to the mixture critical area. However there is no reason for
not including such robustness into the solution procedure. The balance between computa-
tional speed and algorithmic robustness for more special cases must however be considered
carefully.

Example 9.4 Multicomponent two-phase boiler

As an example of the use of the extended pseudo bond graph concept we show how to model
a closed multi component two-phase system. The system shown in Fig. 9.21a initially con-

Vapor
C
T = 280 K
m = 1 kg P m·
c i m· i
V = 0.1 m3 T E·
0
c2 = 0.1 0
Liquid 0

Q·
Sf

Fig. 9.21 Carbon dioxide/n-Butane heating test example

tains a mixture of carbon dioxide CO 2 and n-butane C 3 H 8 at equilibrium conditions. The

volume of the system is 0.1 m3, and the total mass of the mixture is 1 kg. The equilibrium
temperature is 280 K. The heat supplied to the system is 2.5 kJ/min, which is aimed at bring-
ing the system into the super-heated region. The volumetric equation of state selected is the
354 Thermodynamic systems modelling

Peng Robinson equation and the thermodynamic data required is given in Table 6 as

Table 6: Thermodynamic data for simulation example [Ref]

Carbon dioxide CO 2 n-Butane C 3 H 8

Crital pressure [bar] 73.76 38.0

Critical temperature [K] 304.2 425.2

Acentric factor 0.225 0.193

Binary mixing parameter 0.135

The pseudo bond graph of the model is given in Fig. 9.21b.

The differential state equations for this system is derived directly for the bond graphs as

m· = 0
m· 2 = 0
(9.83)
E· = Q·
V· = 0

The initial conditions of the differential state variables are calculated using the given temper-
ature, total mass, mixture composition and volume. This calculation which involves a flash
calculation gives the total internal energy for the mixture. Numerous other options exists for
specifying consistent initial conditions, which to use is a matter of convenience. Another use-
ful option is giving temperature, pressure and total volume.

In the following, the results from the simulations are shown:

In Fig. 9.22 the temperature history is shown. The discontinuous slope of the temperature
which represents the transition from the two-phase region to the superheated vapour phase is
clearly indicated.

In Fig. 9.23 the composition change in the vapour and liquid phases represented by the molar
fraction of the CO 2 component are shown as a function of pressure for the simulation exam-
ple. In Fig. 9.24 the pressure history of the simulation is plotted together with the dew- and
 Multicomponent two-phase thermodynamic systems using pseudo bond graphs 355

400

380
Vapor + Liquid Vapor
360
region region
Temperature [K]

340

320

300

280

260
0 20 40 60 80 100 120 140
Time [min]

Fig. 9.22 Temperature time history for the simulation example

5
x 10
5

V region
4
V + L region
Pressure [Pa]

3
Liquid Vapor
2

1
0 0.05 0.1 0.15 0.2 0.25
X/Y molefractions

Fig. 9.23 Pressure/Concentration history for liquid and vapor phases for the simulation
example
356 Thermodynamic systems modelling

bubble point pressures at the corresponding temperature.

5
x 10
14

12 Bubble point

10

Two-phase Superheated
Pressure [Pa]

8 region vapor region

4 Mixture pressure

Dew point
0
0 20 40 60 80 100 120 140
Time [min]

Fig. 9.24 Pressure and corresponding dew- and bubble point pressures for the simulation
example.

Fig. 9.25 shows the vapour fraction time history for the simulations example. Here note that
the value 1 is reached representing saturated vapour. Beyond that the vapour fraction is actu-
 Summary 357

ally not defined, but the computer routines just puts it equal to 1 if within the superheated

1.2

0.8
Vapor fraction

Liquid Vapor
region region
0.6

0.4

0.2

0
0 20 40 60 80 100 120 140
Time [min]

Fig. 9.25 Vapor fraction time history for the simulation example

region. The results are compared to calculations using SUPERTRAPP [31]. The relative error
for the results shown in this example are found to be in accordance with thermodynamic data
obtained using this package.

9.6 Summary
An extension to the pseudo bond graph concept for modelling multi component two-phase
thermofluid systems has been discussed. The extension introduces additional bonds and state
variables to the pseudo bond graph representation, but maintains the basic features of the
classical pseudo bond graph concept.

In this work it has been showed how a solution procedure based on using a thermodynamic
fundamental relation for multi component two-phase real fluids can be utilized to give the
constitutive laws for the extended pseudo bond graph model of a fluid accumulator.

A simple example using the features developed is also shown and demonstrates clearly how
358 Thermodynamic systems modelling

bond graphs can be extended and offer its advantageous features also for modelling multi
component two-phase thermofluid systems.

A natural extension to this work is to adapt existing routines or to develop new routines for
mixtures showing azeotropic and other complex behaviour, and also to extend their capabil-
ity to handle equilibrium calculations within the critical area. Although computers are faster
than ever, effective algorithms and implementations are still to be a key question for this kind
of calculations.

REFERENCES
5. Dean C. Karnopp, Donald L. Margolis and Ronald C. Rosenberg, System Dynamics: A
Unified Approach, John Wiley & Sons, Inc, 2nd ed., 1990.
6. P. C. Breedveld, Current topics in bond graph related research, Journal of The Franklin
Institute, 328(5/6), 1991.
7. Francois E. Cellier, The Bond Graph Compendium, WWW-address: http://www.ece.ari-
zona.edu/~cellier/bg.html, 1997.
8. 20-Sim Reference Manual, Controllabs Products Inc., P.O.Box 217, 7500 AE Enshede,
The Netherland.
9. MS-1 Reference Manual, Lorenz Simulation, SOCRAN Centre, Scientific Park, Avenue
Pre-Aily, B-4031, Liebe, Belgium
10. ACSL 11 Reference Manual, MGA Software, 200 Baker Avenue, Concord, MA. 01742,
USA
11. MATLAB Reference Manual,
12. Jean U. Thoma, Entropy and mass flow for energy conversion, Journal of the Franklin
Institute, 299(2), p 89-96, February 1975.
13. Jean U. Thoma, Network thermodynamics with entropy stripping, Journal of the Frank-
lin Institute, 303(4), p 319-328, 1977.
14. Forbes T. Brown, Convection bonds and bond graphs, Journal of the Franklin Institute,
325(5/6), p 871-886, 1991.
15. R. Shoureshi, K. McLaughlin, Application of bond graph to thermo-fluid processes and
systems, Trans. ASME J. Dynamic Syst. Measure. Control, 107, p 241-245, December
1985
16. Dean C. Karnopp, State variables and pseudo bond graphs for compressible thermo-fluid
systems, Transactions of ASME, Journal of Dynamc Systems, Measurement and COn-
trol, 101(3), Sept. 1979.
17. Paul O. Moksnes, Modeling two-phase thermo-fluid systems using bond graph, Dr.ing
thesis, Department of Marine Engineering, Norwegian University of Science and Tech-
nology, 1997.
18. H. Engja, K. Strand, Modeling for transient performance of diesel engines using bond
graphs, In ISME, Tokyo, 1983.
19. Jose J. Granda, G.R. Channel, V-8 Internal Combustion Engine bond graph model - A
 Summary 359

Detailed Modeling Procedure, In Proc. of the ICBGM’97 Conference, Ed. J.J. Granda
and G. Dauphin-Tanguy, SCS Simulation Series, Vol. 29, no. 1, 1997
20. E. Pedersen, Ø. Bunes, Modeling and Simulation of the Ulstein Bergen BR-3 Engine -
Comparison with measured data, MARINTEK Report MT222502, June 1998
21. Kurt A. Strand, A System Dynamic Approach to One-Dimensionl Fluid Flow, Dr.ing
thesis, Department of Marine Engineering, Norwegian University of Science and Tech-
nology, 1986
22. Jean Thoma et. al., Building Simulation with Convection and Conduction, In
ICBGM’97, Simulation Series, Vol. 29, no 1, SCS 97
23. Eilif Pedersen, Modelling Thermodynamic Systems with Changing Gas Mixtures, In
Proc. of the ICBGM’99 Conference, Ed. J. J. Granda and F. E. Cellier, SCS Simulations
Seriec, Vol. 31, no. 1, 1999.
24. R. E. Sonntag and G. J. Van Wylen, Introduction ot Thermodynamics- Classical & Statis-
tical, John Wiley & Sons, 3rd. edition, 1991.
25. Herbert B. Callen, Thermodynamics and an Introduction to Thermostatistics, John Wiley
& Sons, 1985.
26. Ding-Yu Peng and Donald B. Robinson, A new two-constant equation of state, Ind. Eng.
Chem. Funaam., 15(1):59-64, 1976.
27. H. Knapp, R. Doring, L. Oellrich, U. Plocker and J. M. Prausnitz, Vapor-Liquid Equilib-
ria for Mixtures of Low-Boiling Substances, DECMA Chemistry Data Series, Vol. VI,
Frankfurt/Main, 1982.
28. C. L. Yaws, R. W. Gallant, Physical proprties of hydrocarbons, Vol 1-4, Gulf Publishing
Company,1980.l
29. Steven I. Sandler, Chemical and Engineering Thermodynamics, 3rd ed., John Wiley &
Sons, Inc., 1999.
30. R. C. Reid, J. M. Prausnitz and B. E. Poling, The Properties of Gases and Liquids, 4th
ed., McGraw-Hill, New York, 1987.
31. NIST Thermophysical Properties of Hydrocarbon Mixtures Database (SUPERTRAPP),
Version 3.0, October 1999, National Institute of Standards and Technology, Gaithers-
burg, MD 20899.
360 Thermodynamic systems modelling

PROBLEMS
9-1 A large high pressure air tank is used for filling up air bottles for scuba divers. The air in
the tank is at 300 bars and 25 C . The bottles are 15 dm 3 each and initially at ambient
conditions. Make a bond graph model and write down the state equations and all consti-
tutive laws of individual elements of this system assuming perfect gas and no heat loss
from the bottles during the filling process.

9-2 An pneumatic actuator shaking the body m is shown below. The pneumatic cylinder is

Leakage area Al
d
Ap m
Ar
P 1 T 1 P 2 T 2

actuated using the 3-way valve where input air supply is at 7 bars and 25 C . The pipes
connecting the supply, the valve and the cylinder can be neglected. The cylinder is as-
sumed to have a small leakage between the piston and the cylinder walls and the cylinder
friction has to be included having a friction coefficient  d . Construct a bond graph of this
component and write the state equations necessary for simulation. Missing variables may
be assumed.

9-3 An two stage variable displacement reciprocating compressor with an cooling stage be-
tween the first and second stage and with the second stage piston 180 degrees after the
first stage is to be modelled using pseudo bond graphs. Construct the bond graph and dis-
cuss how the causality of the model may guide you to modify your model.

9-4 In internal combustion engine modelling we find that the control volume is changing its
mixture as the combustion takes place. The fuel factor which is F =  m b  m a   F s
where m b is the amount of fuel burnt, m a is the amount of air and F s is the stoichiomet-
ric fuel/air ratio. Develop a pseudo-bond graph model for the combustion chamber using
the fuel factor as the effort variable on the mixture pseudo bond.

9-5 Develop a bond graph model for simulation of the human cannonball problem. How do
the heat release rate influence the exit velocity of the man in the cannon?

9-6 Tuned vibration absorbers (TVA) are devices which may promise virtually cancellation
of vibration at their attachment points at specific frequencies. A pneumatic TVA uses air
 Summary 361

chambers to create the compliance element needed in a TVA. In the figure below you’ll
find a schematic drawing of the TVA. Construct a bond graph of this system and convert
this bond graph to a block diagram.

P2,T2,V2 Vs
g
Ls ms
Lc
P1,T1,V1 ks << 1
V

Vin
b k

V in = 2f0.01 sin  2ft 

9-7 The figure shown a schematic drawing of the skirt of an air cushing vehicle. A fan com-
presses air at pressure P s and temperature T s and fills the skirt volume so that the vehicle
“floats” on a cushion of air, although some air escapes m· l under the skirt to the surround-
ings. The relation between the mass flow of the fan and it angular velocity is shown be-
low. The mass of the vehicle is m and we may assume that the vehicle only moves in the
vertical direction. Construct a bond graph of this system.

x x·
m
m· a = a b
m· a
m· l

9-8 A lown mower two stroke engine is shown below. Construct a bond graph for this system.
Include the heat transfer through the cylinder liner and the dynamics of the crank mech-
anism.

P H T H
Exhaust port, Ae
Transfer port Intake port port, Ai
P c T c

Crank case
Rotory inertia, J
362 Thermodynamic systems modelling
Index conduction 93
constitutive laws 239
constitutive relation 37
constraints 8
construction of models 5
Numerics convection 93
0-junction 29 cross-sectional area 37
1-junction 29
1-port 36 D
1-port capacitor 21 decomposition 13
1-port elements 20 deformation 39
1-port inertia 23 design mode 2
1-port resistor 24 development of models 6
2-port elements 25 direct inspection method 50
displacement 17, 39
A dynamic systems 2
active bond 15
analysis mode 2 E
Analytical methods 8 effort 15
angle 36 effort sources 36
angular velocity 36, 44 electrical network formalism 6
Electrical systems 60
B electrical transformer 64
Behavioural theory 3 elongation 37
block diagram 15 energetic structure 7
bond 14 energy dissipation 44
bond graph 14, 37 energy domain 17
bond graph language 6 energy domains 35
bond graph symbol 37 energy port 14
building blocks 13 energy variables 16
bulk modulus 77 engineering systems 2
Excitation forces 36
C Excitation torques 36
cantilever beam 240 explicit 249
capacitance 62
capacitor 36, 62 F
Castigliano 38 fields 239
causality 7 flexibility 37
C-element 21 flow 15
centrifugal pump 75 fluid capacitor 76
C-field 239 Fluid inertance 84
chemical kinetics 92 fluid inertance 86
coefficient of discharge 84 fluid mechanics 92
coil spring 36 fluid power systems 70
compliance 37 fluid resistance 82
compliance matrix 240 fluid transfer systems 70
Computational methods 8 formal models 4
computing power 9 forms 37
friction 45 Models 4
modulated transformers 28
G modulus of elasticity 38
generalised displacement 16 multi disciplinary systems 6
generalised momentum 15 multiport 239
geometric compatibility 29 mutual inductance 248
Gravity forces 36
gyrator 27 N
gyrator modulus 27 non-linear systems 7

H O
heat flux 94 one-dimensional 94
heat transfer 92 onservation of energy 39
heat transfer coefficient 95
hydraulic actuators 79 P
hydraulic capacitor 78 physical models 4
physical properties 9
I physical structures 37
ideal elements 13 physical system 1
ideal junction 29 Physical system modelling 5
idealised sources 20 polar moment of inertia 38
I-element 23 potential energy 17
I-field 247 power 15
implicit 249 power bond 14
inductance 63 power interaction 15
inertia 36 power ports 14
interdisciplinary approach 6 power postulate 15
internal energy 97 power variables 15

K R
kinetic energy 17 radiation 93
reciprocating pumps 71
L R-element 24
linear 7 resistance 63
lumped ideal inertia 43 resistance parameter 25
resistor 64
M reticulation 13
mass moment of inertia 44 reticulation principle 15
mass transfer 92 rotary pumps 71
Maxwell reciprocity 248 rotational 35
mechanical capacitors 36
mechanical elements 35 S
mechanical systems 35 scalar energy operators 239
Mechanical transformers 47 scaled model 4
Modelling 5 shapes 37
modelling 5 shock absorbers 45
modelling electrical systems 65 sign convention 14
Modelling theory 3 signal flow 15
simulation 8
simulation programs 9
slender bar 39
Source of effort 20
Source of flow 21
spring 36
spring constant 37
Static friction 45
stiffness matrix 240
strain energy 39
subsystem 2
synthesis 2
system 1
system aspect 2
systematic modelling procedure 50

T
teaching of models 5
tetrahedron of state 25
thermal conductivity 94
thermal resistance 94
Thermal systems 92
thermodynamics 92
thermofluid systems 70
torque 36
transformer 27
transformer modulus 27
translational 35
translational spring 36
turbo machine 74

V
valve 84
velocity sources 36
Viscous friction 45

W
word bond graph 17

Z
zero-junction 29