Introduction to Modern Controls, with Illustrations in MATLAB and Python

Introduction to Modern Controls

– with Illustrations in MATLAB and Python

Xu Chen and Masayoshi Tomizuka

October 31, 2023

University of Washington
University of California, Berkeley
Introduction to Modern Controls, with Illustrations in MATLAB and Python

Copyright

Copyright ©Xu Chen and Masayoshi Tomizuka, 2023-

License for Code

The code/software in this book is contained under the c zMIT License.

To view a copy of the CC0 code, visit:
http://creativecommons.org/publicdomain/zero/1.0/
Publisher

To cite this material: Xu Chen and Masayoshi Tomizuka, “Introduction to Modern Controls – with Illustrations
in MATLAB and Python,” 2023, ISBN: 9798860587496.
Preface

This book introduces the theory and practice of modern control systems. The emphasis is on using state-space
methods to model, analyze, and control dynamic systems. Topics include state-space modeling and solutions,
stability, controllability and observability, state-feedback control, observers, observer state feedback controls,
least square estimation, Kalman filter, and Linear Quadratic Gaussian optimal control. These topics are discussed
in both continuous- and discrete-time settings throughout the book.

The material in this book is based on many years of teaching experience at the University of Washington and the
University of California, Berkeley. The main sources of the material are:

▶ ME 232 and ME 233 at the University of California, Berkeley, and

▶ ME 547 at the University of Washington, Seattle.

This book consists of four parts. Part I introduces the basics of dynamic systems modeling, such as Laplace and Z
transforms, state-space descriptions and realization theory, and how to solve the state equation. Part II examines
the properties of dynamic systems, such as classic and Lyapunov stability theories, controllability, observability,
and the decomposition of an uncontrollable and/or unobservable system. After understanding these system
properties, in Part III, we cover estimation and controls for state-space systems. Chapter 11 centers on the power
of state feedback. Then in Chapter 12, we discuss state observers and observer-state feedback. As a powerful
state-feedback control method, Chapter 13 covers the linear quadratic optimal control algorithm. Part IV is
dedicated to estimation and control of stochastic systems, where the state-space system equations are subject to
input and output stochastic noises. We review first relevant results in probability theory in Chapter 14, building
on which we derive the least square estimation in Chapter 15 and then the discrete- and continuous-time Kalman
filters in Chapter 16. Chapter 17 integrates linear quadratic optimal control with the Kalman filter, to provide the
celebrated Linear Quadratic Gaussian (LQG) Optimal control. At the end of the book, we provide a review of
related linear algebra for controls.

Over three hundred examples, figures, table summaries, and exercises distilled from physical systems supplement
the learning. Perhaps unique in this book is the modernization of tools we use to illustrate controls. MATLAB
and Python are the primary tools for our numerical demonstrations. Whenever MATLAB examples appear,
complementary Python codes will follow to provide results as equivalent as possible in the more nascent
computation environment.

All the main codes are available for download on the book website https://mcimp-book.github.io/mcimp/. We
have also provided accompanying slides and lecture recordings – all accessible from the book website.

All the MATLAB demonstrations were performed in MATLAB 2022b, and the Python demonstrations in
Python 3.9.13, using toolboxes SymPy v1.11.1 and python-control v0.9.2. For simple calculations and graphical
illustrations, we use gnuplot, a light-weight command-line driven graphing utility across different operation
systems. The coding commands and results are all provided in an “in-line” fashion, directly embedded in the
text materials. Appendix “How to Install and Run Python” provides a summary of ways to configure
Python in different operation systems.

We are grateful to the many teaching assistants and students who helped typeset problems and proofread course
contents over the years. In particular, we would like to thank Liting Sun for creating some of the drawings
in LATEX and Lingfeng Sun for typesetting the Berkeley ME 233 course reader. Jonas Beachy, Xiaohai Hu, and
Marina Ruediger helped with a few case studies. Their contributions have greatly enhanced the quality of this
book.
We hope this book will serve as a useful reference for students and researchers interested in the field of dynamics
and control.

October 31, 2023

Seattle, Washington
Berkeley, California
Contents

Preface iii

Contents v

1 Introduction 1
1.1 The Power of Controls . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Relevant Terminologies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.3 The Objectives and The Means of Controls . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.4 Societies to Learn More about Controls . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

System Description 7
2 Modeling 9
2.1 Methods of Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.2 Continuous-Time Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.3 Discrete-Time Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.4 Example: Atomic Force Microscopy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.5 Example: Hard Disk Drive and Information Storage . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.6 Model Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.7 Nonlinear Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.7.1 Example: Magnetically Suspended Ball . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.7.2 Example: Water Tank . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.7.3 Example: Pendulum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.7.4 Example: Vehicle Steering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.8 “All Models are Wrong, but Some are Useful” . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.9 Exercise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

3 Laplace and Z Transforms 29

3.1 The Laplace Transform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.1.1 The Laplace Approach to ODEs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.1.2 Relevant Properties of the Laplace Transofrm . . . . . . . . . . . . . . . . . . . . . . . . 36
3.2 Inverse Laplace Transform and Partial Fraction Expansion . . . . . . . . . . . . . . . . . . . . . 39
3.3 From Laplace Transform to Transfer Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
3.4 The Z Transform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.4.1 Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.4.2 Relevant Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.4.3 Applications to Dynamic Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
3.5 From Difference Equation to Discrete-Time Transfer Functions . . . . . . . . . . . . . . . . . . . 51
3.6 Recap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
3.7 Exercise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54

4 State-Space Description of a Dynamic System 57

4.1 The Concept of States . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
 4.2 General State-Space Descriptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
4.3 From the State Space to Transfer Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
4.4 Linearization and State-Space Representation of Nonlinear Systems . . . . . . . . . . . . . . . . 66
4.4.1 State-Space Representation of General Nonlinear Systems . . . . . . . . . . . . . . . . . 66
4.4.2 Equilibrium Point and Linearization around an Equilibrium Point . . . . . . . . . . . . 67
4.4.3 Multivariariate Partial Derivative . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
4.4.4 Example: Water Tank . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
4.4.5 Example: Vehicle Steering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
4.5 Recap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
4.6 Exercise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81

5 State-space Realizations: The Canonical Forms 83

5.1 Controllable Canonical Form . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
5.2 Observable Canonical Form . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
5.3 Diagonal and Jordan Canonical Forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
5.4 Discrete-Time LTI Systems and Their State-Space Canonical Forms . . . . . . . . . . . . . . . . 92
5.5 Similar Realizations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
5.6 Recap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
5.7 Exercise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97

6 Solution of Time-Invariant State-Space Equations 99

6.1 Continuous-Time State-Space Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
6.1.1 The Solution to 𝑥¤ = 𝑎𝑥 + 𝑏𝑢 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
6.1.2 Fundamental Theorem of Differential Equations . . . . . . . . . . . . . . . . . . . . . . . 101
6.1.3 The Solution to 𝑛 th -order LTI Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
6.2 Discrete-Time LTI State-Space Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
6.3 Explicit Computation of the State Transition Matrix 𝑒 𝐴𝑡 . . . . . . . . . . . . . . . . . . . . . . . 108
6.3.1 The Case with Distinct Eigenvalues (Diagonalization) . . . . . . . . . . . . . . . . . . . . 109
6.3.2 Physical Interpretations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
6.3.3 The Case with Complex Eigenvalues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
6.3.4 The Case with Repeated Eigenvalues, via Generalized Eigenvectors . . . . . . . . . . . . 112
6.3.5 Physical Interpretation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
6.4 Explicit Computation of the State Transition Matrix 𝐴 𝑘 . . . . . . . . . . . . . . . . . . . . . . . 118
6.5 Transition Matrix via Inverse Transformation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
6.6 Solutions of Time-Varying State Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
6.6.1 Continuous-Time Case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
6.6.2 Discrete-Time Case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
6.7 Recap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
6.8 Exercise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122

7 Discrete-Time Models of Continuous Systems 125

7.1 State-Space Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
7.2 Transfer-Function Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
7.3 Exercise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
System Properties 135
8 Stability 137
8.1 Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
8.1.1 Review of Relevant Functional Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
8.1.2 Lyapunov’s Definition of Stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
8.2 Stability of LTI Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
8.2.1 Method of Eigenvalue Locations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
8.2.2 Routh-Hurwitz Criterion for Continuous-Time LTI Systems . . . . . . . . . . . . . . . . 142
8.2.3 Routh-Hurwitz Criterion for Discrete-Time LTI Systems . . . . . . . . . . . . . . . . . . 144
8.3 Lyapunov’s Approach to Stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
8.3.1 Stability from an Energy Viewpoint . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
8.3.2 Relevant Mathematical Tools . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
8.4 Lyapunov Stability Theorems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
8.4.1 Lyapunov Stability for Continuous-Time Systems . . . . . . . . . . . . . . . . . . . . . . 155
8.4.2 Lyapunov Stability for Discrete-Time Systems . . . . . . . . . . . . . . . . . . . . . . . . 160
8.5 Recap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
8.6 Exercise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164

9 Controllability and Observability 167

9.1 Basic Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
9.2 The Case for Discrete-Time Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
9.2.1 Controllability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
9.2.2 Observability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
9.3 The Case for Continuous-Time Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
9.3.1 Controllability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
9.3.2 Observability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180
9.4 Transforming Controllable Systems Into the Controllable Canonical Form . . . . . . . . . . . . 186
9.5 Transforming Observable Systems Into the Observable Canonical Form . . . . . . . . . . . . . . 188
9.6 Recap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
9.7 Exercise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191

10 Kalman Decomposition 195

10.1 Basic Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
10.2 Kalman Decomposition of Uncontrollable Systems . . . . . . . . . . . . . . . . . . . . . . . . . 196
10.3 Kalman Decomposition of Unobservable Systems . . . . . . . . . . . . . . . . . . . . . . . . . . 204
10.4 General Kalman Decomposition, Stabilizability, and Detectability . . . . . . . . . . . . . . . . . 206
10.5 Recap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
10.6 Exercise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208

Estimation and Control 209

11 State Feedback 211
11.1 State Feedback and Eigenvalue Assignments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211
11.1.1 Dynamic Properties of the Closed-Loop System . . . . . . . . . . . . . . . . . . . . . . . 212
11.1.2 Eigenvalue Assignment for Single-Input Systems . . . . . . . . . . . . . . . . . . . . . . 213
11.1.3 Multi-Input Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216
11.2 Output Feedback . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216
 11.3 Recap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216
11.4 Exercise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217

12 Observers and Observer-State Feedback 219

12.1 Open-Loop Observer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219
12.2 Continuous-Time Luenberger Observer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220
12.3 Observer State Feedback Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 228
12.4 Discrete-Time Observers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229
12.4.1 Discrete-Time Full State Observer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229
12.4.2 Discrete-Time Full State Observer with Predictor . . . . . . . . . . . . . . . . . . . . . . 229
12.5 Reduced-Order Observer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232
12.6 Recap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237
12.7 Exercise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 238

13 Linear Quadratic Optimal Control 241

13.1 Problem Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241
13.2 Solution of the Continuous-Time LQ Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242
13.3 Stationary Continuous-Time LQ Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 250
13.4 Application and Practice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 256
13.4.1 Choosing 𝑄 and 𝑅 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 256
13.4.2 MATLAB and Python Commands . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 256
13.4.3 Obtaining 𝑃+ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 256
13.4.4 Example: Inverted Pendulum on a Cart . . . . . . . . . . . . . . . . . . . . . . . . . . . . 258
13.4.5 Robustness of Stationary LQ Regulators . . . . . . . . . . . . . . . . . . . . . . . . . . . 271
13.5 Further Development of the Continuous-Time LQ Regulator . . . . . . . . . . . . . . . . . . . . 271
13.5.1 Return Difference Equality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 271
13.5.2 Stability Margins of Scalar LQR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 272
13.5.3 The Closed-Loop Eigenvalues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 273
13.5.4 Symmetric Root Locus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 275
13.5.5 Steady-State Property . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 281
13.6 Discrete-Time LQ Optimal Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 283
13.6.1 Introduction of Dynamic Programming . . . . . . . . . . . . . . . . . . . . . . . . . . . . 283
13.6.2 Dynamic Programming for General Optimal Control Problems . . . . . . . . . . . . . . 284
13.6.3 Solving Discrete-Time Finite-Horizon LQ Problems via Dynamic Programming . . . . . 286
13.6.4 Discrete-Time Stationary LQ Optimal Control . . . . . . . . . . . . . . . . . . . . . . . . 289
13.7 Further Development of the Discrete-Time LQ Regulator . . . . . . . . . . . . . . . . . . . . . . 293
13.7.1 Obtaining 𝑃+ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 293
13.7.2 Return Difference Equality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 293
13.7.3 The Closed-Loop Eigenvalues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 295
13.8 Recap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 297
13.9 Exercise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 302

Stochastic Estimation and Control 305

14 Review of Probability Theory 307

14.1 Sample Space, Events, and Probability Axioms . . . . . . . . . . . . . . . . . . . . . . . . . . . . 307
14.2 Random Variables, Probability Density, and Moments of Distributions . . . . . . . . . . . . . . 308
 14.3 Example Distributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 309
14.4 Random Vector, Joint Probability and Distribution, Conditional Probability . . . . . . . . . . . 313
14.5 Discrete-Time Random Process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 316
14.5.1 Filtering a Random Process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 319
14.5.2 Filtering a Random Process in the State Space . . . . . . . . . . . . . . . . . . . . . . . . 320
14.6 Continuous-Time Random Process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 321
14.7 Recap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 325
14.8 Exercise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 327

15 Least Square Estimation 329

15.1 General Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 329
15.2 Solution in the Gaussian Case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 331
15.3 Properties of Least Square Estimate (Gaussian Case) . . . . . . . . . . . . . . . . . . . . . . . . 332
15.4 Example Application of Least Square Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . 336
15.5 Recap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 341
15.6 Exercise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 342

16 Stochastic State Estimation and Kalman Filter 343

16.1 Review of State Observers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 343
16.2 Discrete-Time Stochastic State Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 343
16.2.1 Problem Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 343
16.2.2 Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 345
16.2.3 Discrete-Time Kalman Filter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 347
16.2.4 Steady-State Kalman Filter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 348
16.2.5 Return Difference Equality for Equation 16.29 . . . . . . . . . . . . . . . . . . . . . . . . 358
16.3 Continuous-Time Stochastic State Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 362
16.3.1 General Continuous-Time Kalman Filter . . . . . . . . . . . . . . . . . . . . . . . . . . . 362
16.3.2 Steady-State Kalman Filter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 365
16.3.3 Return Difference Equality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 365
16.4 Kalman Filter Application: Kinematic Kalman Filter . . . . . . . . . . . . . . . . . . . . . . . . . 367
16.5 The Kalman Filter Equations using Other Notation Systems . . . . . . . . . . . . . . . . . . . . 368
16.6 Recap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 371
16.7 Exercise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 374

17 Linear Quadratic Gaussian (LQG) Optimal Control 377

17.1 Stochastic Control with Exactly Known State . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 377
17.1.1 Stochastic Control with Inexactly Known State . . . . . . . . . . . . . . . . . . . . . . . . 381
17.1.2 Stationary LQG Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 384
17.2 Continuous-Time LQG Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 385
17.3 Recap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 386
17.4 Exercise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 389

18 Further Readings 393

Review of Relevant Linear Algebra 395

1 Basic Concepts of Matrices and Vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 395
1.1 Matrix Addition and Multiplication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 396
1.2 Matrix Transposition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 397
2 Linear Systems of Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 397
 3 Vector Space, Linear Independence, Basis, and Span . . . . . . . . . . . . . . . . . . . . . . . . . 400
4 Matrix Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 401
4.1 Rank . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 401
4.2 Range and Null Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 401
4.3 Determinants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 401
4.4 Matrix and Linear Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 402
5 Eigenvalues and Eigenvectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 403
5.1 Matrix, Mappings, and Eigenvectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 403
5.2 Computation of Eigenvalue and Eigenvectors . . . . . . . . . . . . . . . . . . . . . . . . 404
5.3 Eigenbase and Diagonalization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 405
5.4 Similarity Transformation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 407
6 Matrix Inversion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 409
6.1 Block Matrix Decomposition and Inversion . . . . . . . . . . . . . . . . . . . . . . . . . . 410
6.2 LU and Cholesky Decomposition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 412
6.3 Determinant and Matrix Inverse Identity . . . . . . . . . . . . . . . . . . . . . . . . . . . 413
7 Spectral Mapping Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 416
8 Matrix Exponential . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 416
9 Inner Product . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 417
9.1 Inner Product Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 417
9.2 Trace (Standard Matrix Inner Product) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 418
10 Vector Norms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 418
11 Symmetric and Orthogonal Matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 419
12 Positive-Definite Matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 421
12.1 Definitions and Basic Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 421
12.2 General Positive-Definite Matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 422
12.3 Positive-Definite Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 422
13 Singular Value Decomposition (SVD) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 423
13.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 423
13.2 Main Result . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 423
13.3 Properties of Singular Values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 425
14 Induced Matrix Norm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 425

How to Install and Run Python 427

Alphabetical Index 431

About the Authors 433

List of Figures

1.1 A general block diagram. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

1.2 An open-loop control system. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 A closed-loop heating control example. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.4 Control system with feedforward control. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.5 Overview of book topics from system description to estimation and controls. . . . . . . . . . . . . 4
1.6 Structure of digital control systems. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.7 Design approaches of digital control systems. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

2.1 Major components in a hard disk drive. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

2.2 Example frequency response of the voice coil motor in a hard disk drive. . . . . . . . . . . . . . . . 10
2.3 Isaac Newton. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.4 Schematic of the AFM. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.5 Mechanical models of an AFM. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.6 Control-related components in a hard disk drive. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.7 Baseline model of the VCM component in an HDD. . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.8 Schematic structure of dual-stage HDDs. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.9 Baseline model of the PZT component in an HDD. . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.10 Kinematic modelling of the four-wheel vehicle steering system. . . . . . . . . . . . . . . . . . . . . 22
2.11 The bicycle model of the vehicle steering system. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.12 Example frequency responses of the voice coil motor stage in a batch of hard disk drives. . . . . . 24

3.1 The Laplace approach to ODEs. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

3.2 Pierre-Simon Laplace. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.3 Illustration of piecewise continuous functions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.4 Illustration of exponential order. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.5 Illustration of the Dirac impulse function. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.6 Approximation of the unit-step function. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.7 Approximation of the Dirac-delta function. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.8 Toward smoother approximations of the unit-step function. . . . . . . . . . . . . . . . . . . . . . . 35

5.1 Block diagram of the controllable canonical form for third-order systems. . . . . . . . . . . . . . . 85
5.2 Block diagram of the observable canonical form for third-order systems. . . . . . . . . . . . . . . . 87
5.3 Block diagram of the diagonal realization for third-order systems with distinct poles. . . . . . . . 90
5.4 Block diagram of the Jordan form realization for third-order systems with repeated poles. . . . . . 91
5.5 Block diagram of the modified canonical form for third-order systems with complex poles. 𝑘2 =
(𝛽 + 𝛼𝜎)/𝜔 and 𝑘3 = 𝛼 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92

6.1 The concept of the time constant. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101

6.2 The unit-step response of a first-order system. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
6.3 Example state trajectory for a second-order system under different initial conditions. . . . . . . . . 111
6.4 Map of eigenvalue and associated response mode (continuous-time case). . . . . . . . . . . . . . . 116
6.5 Map of eigenvalue and associated response mode (discrete-time case). . . . . . . . . . . . . . . . . 117

7.1 A continuous-time state-space system preceded by zero order hold. . . . . . . . . . . . . . . . . . 125

7.2 Illustration of the input and output of a zero order hold. . . . . . . . . . . . . . . . . . . . . . . . . 126
7.3 Continuous-time transfer function preceded by a zero order hold. . . . . . . . . . . . . . . . . . . 129

8.1 Example continuous functions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138

8.2 A uniformly continuous function. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
8.3 A continuous but not uniformly continuous function. . . . . . . . . . . . . . . . . . . . . . . . . . . 138
8.4 Stability concepts in the sense of Lyapunov. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
8.5 Time-domain plot of function 𝑡𝑒 𝜆𝑡 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
8.6 The bilinear transform maps the closed left half plane to the closed unit disk. . . . . . . . . . . . . 144
8.7 A positive-definite function 𝑊 (𝑥 1 , 𝑥 2 ) = 𝑥 12 + 𝑥 22 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
8.8 A locally positive-definite function 𝑊 (𝑥 1 , 𝑥 2 ) = 𝑥 12 + sin2 𝑥 2 . . . . . . . . . . . . . . . . . . . . . . . 154

9.1 Arthur Cayley. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169

9.2 William Hamilton. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170

11.1 Illustration of state-feedback control. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211

12.1 Concept of a closed-loop observer. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220

12.2 Realization of the Luenberger observer. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220
12.3 Block diagram of observer state feedback control. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 228
12.4 Block diagram of the continuous-time reduced-order observer. . . . . . . . . . . . . . . . . . . . . 234
12.5 Block diagram of reduced-order observer: implementation form. . . . . . . . . . . . . . . . . . . . 235
 
0 1
13.1 LQ example with small penalty on control. 𝑃 ∗ (0) = , 𝑅 = 0.0001 . . . . . . . . . . . . . . . . 247
1 0
 
1 0
13.2 LQ example with medium penalty on control. 𝑃 (0) =∗ ,𝑅=1 . . . . . . . . . . . . . . . . . 248
0 1
 
1 0
13.3 LQ example with large penalty on control. 𝑃 ∗ (0) = , 𝑅 = 100 . . . . . . . . . . . . . . . . . 248
0 1
 
20 0
13.4 LQ example with large penalty on control. 𝑃 (0) =
∗ , 𝑅 = 100 . . . . . . . . . . . . . . . . . 248
0 2
13.5 State trajectories of an inverted pendulum on a cart under full-state LQ optimal control. . . . . . . 262
13.6 State trajectories of a linearized inverted pendulum on a cart under full-state LQ optimal control. . 264
13.7 State trajectories of a linearized inverted pendulum on a cart under observer-state LQ control. . . 267
13.8 State estimation errors of a linearized inverted pendulum on a cart under observer-state LQ control. 268
13.9 State estimation errors of a linearized inverted pendulum on a cart under observer-state LQ control. 270
13.10 Block diagram of the optimal LQ system. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 271
13.11 Frequency response of the continuous-time LQ loop transfer function. . . . . . . . . . . . . . . . . 273
13.12 Frequency response of the discrete-time LQ loop transfer function. . . . . . . . . . . . . . . . . . . 295

14.1 Probability density function of a uniform distribution. . . . . . . . . . . . . . . . . . . . . . . . . . 309

14.2 Probability density function of a Gaussian distribution. . . . . . . . . . . . . . . . . . . . . . . . . 310
14.3 Distribution of the addition of two uniformly distributed random variables. . . . . . . . . . . . . . 311
14.4 Distribution of the addition of three uniformly distributed random variables. . . . . . . . . . . . . 312
14.5 Illustration of a discrete-time random process. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 317

15.1 Geometric interpretation of the least square estimator 𝐸 [𝑥|𝑦]. . . . . . . . . . . . . . . . . . . . . . 333

15.2 Geometric interpretation of the least square estimator 𝐸 [𝑥|𝑦, 𝑧] where 𝑦 and 𝑧 are uncorrelated. . 334
15.3 Geometric interpretation of the least square estimator 𝐸 [𝑥|𝑦, 𝑧] where 𝑦 and 𝑧 are correlated. . . 336
16.1 Block-diagram representation of the corrector and predictor form of the Kalman filter. . . . . . . . 347
16.2 State estimation by Kalman filter and finite settling time observer (FSTO) with 𝑟 = 2. . . . . . . . . 352
16.3 Performance of the Kalman filter and finite settling time observer (FSTO). . . . . . . . . . . . . . . 353
16.4 A closed-loop return-difference representation of the Kalman filter. . . . . . . . . . . . . . . . . . . 359
16.5 Block diagram of the return difference in a Kalman filter (𝑢(𝑡) = 0 or the forced response has been
subtracted). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 365
16.6 Example eigenvalues of a second-order Kalman filter. . . . . . . . . . . . . . . . . . . . . . . . . . . 367

17.1 Structure of the LQG control scheme. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 384

1 Example relationship between 𝑥 and 𝐴𝑥 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 403

2 Decomposition of 𝑥 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 403
3 Construction of 𝐴𝑥 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 404
4 Example points in vector space 1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 408
5 The same points in vector space 2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 408

List of Tables

3.1 Common Laplace transforms. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

3.2 Common properties of the Z transform. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
3.3 Table of Laplace and Z transforms [5]. 𝑥(𝑡) = 0 for 𝑡 < 0. 𝑥(𝑘𝑇) = 𝑥(𝑘) = 0 for 𝑘 < 0. Unless otherwise
noted, 𝑘 = 0 , 1 , 2 , 3 , . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.4 Comparison of continuous- and discrete-time transfer functions. . . . . . . . . . . . . . . . . . . . . 52

6.1 Common pairs of 𝐽 and 𝐽 𝑘 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118

6.2 Summary of solutions to state equations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
6.3 Common state-transition matrices. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122

7.1 Zero order hold equivalents of continuous-time transfer functions. . . . . . . . . . . . . . . . . . . . 130

8.1 Stability of the origin for 𝑥¤ = 𝐴𝑥 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141

8.2 Stability of the origin for 𝑥 (𝑘 + 1) = 𝐴𝑥 (𝑘). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
8.3 Analogy between symmetric, skew-symmetric, and orthogonal matrices and the complex plane. . . 149
8.4 Summary of Lyapunov equations in continuous- and discrete-time dynamics. . . . . . . . . . . . . 164

13.1 Summary of continuous- and discrete-time LQ optimal-control properties. . . . . . . . . . . . . . . 301

16.1 Comparison of the estimation error covariances (diagonal entries) between a Kalman filter and a finite
settling time observer. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 351
 Introduction 1
1.1 The Power of Controls . . . . 1
1.1 The Power of Controls
1.2 Relevant Terminologies . . . 1
1.3 The Objectives and The
Our internal body temperature is regulated around the normal value of Means of Controls . . . . . . 3
about 37◦ C or 98.6◦ F. A part of our brain called the hypothalamus checks
1.4 Societies to Learn More
our current temperature and compares it with the normal value. In a sauna about Controls . . . . . . . . . 6
room where the temperature is too high, sweat is produced to cool the
skin,1 and the blood vessels under our skin get wider to increase the blood 1: The middle layer of the skin, or dermis,
flow to the skin.2 On the other hand, when building a snowman outside, stores most of the body’s water. When the
temperature is too high, that water, along
the blood vessels under our skin become narrower to decrease blood flow with the body’s salt, are brought to the
to the skin, retaining heat near the warm inner body; muscles, organs, and surface of the skin as sweat. On the skin
brain produce heat (e.g., muscles can produce heat by shivering); and our surface, the water evaporates and cools
thyroid gland releases hormones to increase our metabolism.3 4 That is the the body.

power of feedback controls: it allows us to make a precision device out of a 2: This process is called vasodilatation.

crude one that works well even in changing environments. 3: The three mechanisms are called vaso-
constriction, thermogenesis, and hor-
We also use prediction and feedforward controls in our regulation of body monal thermogenesis, respectively.
temperature: as kids, we had learned to wear T-shirts in summer, long 4: We are continuously losing heat:
The Basal Metabolic Rate (BMR) is the
sleeves and coats in winter. With such predictive and feedforward controls,
number of calories we burn as our body
the burden of feedback control is greatly lifted. performs basic (basal) life-sustaining
function. An average man has a BMR
Using these temperature-control activities, fine-tuned naturally as we of around 7,100 kJ per day, while an
grow, our body can respond to internal and external stimuli and make average woman has a BMR of around
adjustments to keep the body within one or two degrees of the normal 5,900 kJ per day. See more at https:
//www.betterhealth.vic.gov.au/health/
temperature, whether in summer or winter, at the north pole or in the
conditionsandtreatments/metabolism.
Sahara Desert!

1.2 Relevant Terminologies

More formally, feedback is the use of information of the past or the present
to influence behaviors of a system. A system is an interconnection of
elements and devices for a desired purpose. In your undergraduate control
course, you have obtained basic understandings of a control system with
concepts such as transfer functions, proportional integral derivative (PID)
controllers, and frequency responses in classic control. Below, we provide
a brief review of the key concepts and terminologies.

We use a block diagram such as the one in Figure 1.1 to visualize the
system structure and the interconnection of system components. Here, the
plant consists of: (i) a process whose output is to be controlled and (ii) an
actuator – a physical device capable to influence the controlled variable
of the process. The sensor measures the output of the plant and feeds it
back to be compared to a reference signal. The error after the comparison
then drives the controller to generate the command for the actuator. It
is not uncommon to have an input filter that shapes the reference signal:
 2 1 Introduction

Disturbance
Plant
Reference Output
+
Input filter Controller Actuator Process
−

Sensor

Figure 1.1: A general block diagram. Sensor noise

for example, in controlling the body temperature, we wear clothes of

different thickness and warmth that build a buffer between our skin and
the environment.

In our introductory example, the body cannot immediately adjust the

temperature. Instead, it is an intricate dynamic process – blood vessels
expand and contract to move blood and heat closer to or further away from
the skin, thus releasing or conserving warmth. Dynamic systems do not
show the effect of the input immediately but after some transient and/or
If the mechanism of transient and delay delays.
is not correctly taken into account, the
feedback controller may over compensate The inputs and outputs in a dynamic system are signals, i.e., they are
or pump in an excessive amount of control functions of time, e.g., speed of a car, temperature in a room, voltage applied
energy, resulting in system instability.
to a motor, price of a stock, and electrical-cardiograph. In practice, there will
be disturbances to the system and noises in the sensor measurement signals.
These are also signals and will negatively influence the performance of the
control system.

An open-loop control system (cf. Figure 1.2) is one where the output of the
plant does not influence the input to the controller.

Disturbance

Desired Output 𝑢(𝑡) 𝑦(𝑡)

Controller Controlled System
Figure 1.2: An open-loop control system.

A closed-loop system is one where multiple components (plant, controller,

etc) have a closed interconnection. For exampel, in the heating control
system for a house, the thermostat will measure the room temperature,
compare it to the set value, and turn on or off the furnace to keep the house
warm in a closed loop. There is always feedback in a closed-loop system.

Closed-loop (feedback) controllers provide a more robust performance

than open-loop controllers in the presence of disturbances and plant
uncertainties. If the reference signal varies fast and its variation is known
in advance, then feedforward control based on information about the
 1.3 The Objectives and The Means of Controls 3

Heat Loss

Desired 𝑇 − Room 𝑇
Thermostat Gas Valve Furnace House
+

Figure 1.3: A closed-loop heating control

example.

reference is useful. This is often the case in machine tool control and robot
control. Feedforward control is also effective if the disturbance signal can
be measured or estimated. Figure 1.4 shows a control system that integrates
these benefits of feedback and feedforward controls.

Feedforward
Controller
Disturbance

Desired Error
output + 𝑒(𝑡) Feedback + Controlled 𝑦(𝑡)
− Controller + Plant
𝑢(𝑡)

Sensors
Figure 1.4: Control system with feedfor-
ward control.

1.3 The Objectives and The Means of Controls

A few aspects of control objectives are universal. For example, we would

always want our control system to result in closed-loop dynamics that
are stable and insensitive to disturbances. These form the stabilization
problem and the disturbance rejection problem, respectively. When the
reference is a fixed point such as the normal temperature of our body, the
control objective is the regulation of output in the presence of disturbances
and noises. When the reference changes and is time-varying, a tracking
problem is formed.

To achieve the control objectives, the control engineer must model the
controlled plant, analyze the characteristics of the plant, design control
algorithms (controllers), analyze performance and robustness of the control
system, and implement the controller.

To model, analyze, and design control systems, we must be able to describe

systems formally, understand their key properties, and then design esti-
mation and control algorithms. Figure 1.5 provides the essential workflow
of this book covering these topics. In describing a system, we cover topics
from state-space description and system realization theory to linearization
 4 1 Introduction

and discretization of system models. We will provide solutions to con-

trol systems from the transfer-function domain to the state-space domain.
Build on the system description and solution concepts, we analyze sys-
tem properties and exploit the importance of stability, controllability, and
observability, along with foundational properties such as causality and
linearity. In estimation and controls, we integrate all topics and discuss
the power of control design when certain system properties are met. We
cover state estimation in both the deterministic and the stochastic cases.
We discuss the power of feedback control from arbitrary pole placement of
closed-loop eigenvalues to LQR and LQG optimal feedback control.

fer func
ns

Tra

tio
Discreti-

n
zation
ate e LQG

St
spac
Block Z
Diagram Kalman
Probability
Transform
State Theory Filter
Lap e
Space
l ac LQR
System Causality Least
Realization Linear &
Nonlinear Square
State Estimate
Estimation
Static &
Dynamic
Stability
Controll-
ability

Observa-
bility
Figure 1.5: Overview of book topics from
system description to estimation and con-
trols.

For implementation of controls, control engineers use computers extensively

in both (off-line) analysis/design and (real-time) implementation. Com-
puters are inherently discrete devices. If a computer is used in real-time
control, it receives the output of the controlled plant as a sensor signal
intermittently often at a fixed sampling frequency after an analog to digital
(A/D) converter, computes the right control input and send it out to the
controlled plant, as depicted in Figure 1.6. The discrete control sequence
is then sent to a digital to analog (D/A) converter to form the continuous-
time control command to the plant. From the viewpoint of the computer,
the plant is a discrete-time device that produces a discrete-time output
sequence in response to a control sequence provided by the computer.
While computers are used in almost all implementation of control, the
controller design, however, may be carried out in the continuous-time
domain or in the discrete-time domain. The choice of the time domain may
depend on many factors such as target systems, control methodologies,
and personal taste. It is important that the control engineers have a broad
knowledge of the analysis and design tools for control systems. For example,
there are many design approaches to digital control systems. Figure 1.7
summarizes typical design approaches that we may follow given a physical
plant. The continuous-time linear control theory and the discrete-time
control theory that we will study will provide us an important and useful
set of tools.
 1.3 The Objectives and The Means of Controls 5

Figure 1.6: Structure of digital control sys-

tems.

Figure 1.7: Design approaches of digital control systems.

6 1 Introduction

1.4 Societies to Learn More about Controls

Founded in Paris in 1957, the International Federation of Automatic Control

(IFAC, website: https://www.ifac-control.org) is the worldwide organi-
zation tasked with promoting the science and technology of automatic
control in both theory and application. IFAC also disseminate the impact of
control technology on society through its conferences, publications, techni-
cal committees, and journals. IFAC is well known through the editorship
of eight archival journals:
▶ Automatica,
▶ Control Engineering Practice,
▶ Annual Reviews of Control,
▶ Engineering Applications of Artificial Intelligence,
▶ Journal of Process Control,
▶ Mechatronics,
▶ Nonlinear Analysis: Hybrid Systems, and
▶ IFAC Journal of Systems and Control.
These are published in partnership with the official IFAC publisher, Elsevier.
Control Engineering Practice and Annual Reviews of Control, for example,
are good starting points that are rich in examples and reviews of recent
advances in controls.
The American Automatic Control Council (AACC, website: www.a2c2.org)
represents the United States to the global control community and is the US
National Member Organization (NMO) of IFAC. AACC helps arrange for
IFAC events in the U.S. and is an association of nine professional societies:
▶ American Institute of Aeronautics and Astronautics (AIAA)
▶ American Institute of Chemical Engineers (AIChE)
▶ American Society of Civil Engineers (ASCE)
▶ American Society of Mechanical Engineers (ASME)
▶ Institute of Electrical and Electronics Engineers (IEEE)
▶ Institute for Operations Research and the Management Sciences
Applied Probability Society (INFORMS APS)
▶ International Society of Automation (ISA)
▶ Society for Industrial and Applied Mathematics (SIAM)
▶ Society for Modeling and Simulation International (SCS)

AIAA, ASME, and IEEE, for example, publish journals such as

▶ IEEE Control Systems Magazine
▶ IEEE Transactions on Control Systems Technology
▶ IEEE Transactions on Automatic Control
▶ AIAA Journal of Guidance, Control and Navigation
▶ ASME Journal of Dynamic Systems, Measurement and Control
These are good starting good points at AACC to learn more about controls
in both theory and applications.
System Description
2 Modeling

Modeling of physical systems is a vital component of modern engineering.

2.1 Methods of Modeling . . . 9
After we understand the governing dynamics of a system, we can simulate
and predict system responses, design model-based controllers, and evaluate 2.2 Continuous-Time Systems 10
system properties. 2.3 Discrete-Time Systems . . 11
2.4 Example: Atomic Force
Microscopy . . . . . . . . . . 11
2.5 Example: Hard Disk Drive
2.1 Methods of Modeling and Information Storage . 14
2.6 Model Properties . . . . . . 19
The dynamics of many systems often consist of complex coupled differential 2.7 Nonlinear Systems . . . . . 20
or difference equations. Two general approaches exist to extract these system 2.8 “All Models are Wrong, but
models. The first and more physics-based approach capitalizes on principles Some are Useful” . . . . . . 24
of physics such as Newton’s laws and energy conservation. The second and 2.9 Exercise . . . . . . . . . . . . 28
more data-centric approach integrates input-output responses to extract
the system dynamics.1 1: The field of system identification and
adaptive control is dedicated to such data-
The most successful modeling often integrates both physics- and data-based based approach to model and control dy-
namic systems.
modeling and analysis techniques. We provide an example below.
Spindle Head
In a hard disk drive storage system, the main dynamics of the voice coil Platter
Actuator Arm
motor that rotates the read/write head are governed by Newton’s second Actuator Axis
law for rotation:

𝜏 = 𝐽 𝛼 .
|{z} |{z} |{z}
net torque moment of inertia angular acceleration Power Connector
Jumper Block
Actuator
Let the angular position 𝜃 be the output and 𝜏 be the input. Then the IDE Connector

input-output dynamics follow the formula: Figure 2.1: Major components in a hard
disk drive.
1 https://en.wikipedia.org/wiki/
𝜃¥ = 𝛼 = 𝜏.
𝐽
Hard_disk_drive#/media/File:
Hard_drive-en.svg
 10 2 Modeling

However, hard disk drives are high-speed high-precision (nanometer-

scale!) mechatronic systems. In addition to the above fundamental mode,
at high angular speeds and frequencies, the rotating disks and actuator
arms become no longer rigid, but instead will bend and exhibit high-
order vibration modes. Figure 2.2 shows the frequency response of a
typical voice coil motor in modern hard disk drives. Many vibration modes
appear at high frequencies. The parameters of these modes are not as easy
to obtain analytically. Finite element methods and system identification
become useful here. However, the analysis from physics is still critical to
In section 2.5, we discuss details about understand the shapes of these vibration modes.
how to construct the model in Figure 2.2.

P cv
100

Gain [dB] 50

-50

10 1 10 2 10 3 10 4
Frequency [Hz]

-90
Phase [deg.]

-180

-270

-360
10 1 10 2 10 3 10 4
Figure 2.2: Example frequency response Frequency [Hz]
of the voice coil motor in a hard disk drive.

2.2 Continuous-Time Systems

Mathematical models of continuous dynamic systems are differential

equations. Here, inputs and outputs of the continuous-time systems are
defined for all 𝑡 , i.e. 𝑢(𝑡) and 𝑦(𝑡). Continuous linear dynamic systems are
described by linear differential equations in the form of

𝑑 𝑛 𝑦(𝑡) 𝑑 𝑛−1 𝑦(𝑡) 𝑑 𝑚 𝑢(𝑡) 𝑑 𝑚−1 𝑢(𝑡)

𝑛
+𝑎 𝑛−1 +· · ·+𝑎0 𝑦(𝑡) = 𝑏 𝑚 𝑚
+𝑏 𝑚−1 +· · ·+𝑏0 𝑢(𝑡),
𝑑𝑡 𝑑𝑡 𝑛−1 𝑑𝑡 𝑑𝑡 𝑚−1
(2.1)
(𝑛)
with the initial conditions 𝑦(0) = 𝑦0 , . . . , 𝑦 (0) = 𝑦0 , where 𝑦 is a
(𝑛) (𝑛)
𝑑𝑛 𝑦
shorthand of 𝑑𝑡 𝑛 .

Example 2.2.1 (Mass spring damper) Consider a mass spring damper

system:
 2.3 Discrete-Time Systems 11

position: 𝑦(𝑡)
𝑘

𝑚 𝑢=𝐹

Newton’s second law gives

𝑚 𝑦¥ (𝑡) + 𝑏 𝑦¤ (𝑡) + 𝑘 𝑦 (𝑡) = 𝑢 (𝑡) , 𝑦(0) = 𝑦0 , 𝑦(

¤ 0) = 𝑦¤0 . (2.2)

The system is modeled as a second-order ordinary differential equation

(ODE) with input 𝑢(𝑡) and output 𝑦(𝑡).

2.3 Discrete-Time Systems

Inputs and outputs of discrete-time systems are defined at discrete-time

points, i.e. 𝑢(𝑘) and 𝑦(𝑘), where 𝑘 = 0 , 1 , 2 , . . .. Models of discrete dynamic
systems are difference equations in the form of:

𝑦(𝑘 + 1) = 𝑓 (𝑦(𝑘), 𝑦(𝑘 − 1), . . . , 𝑦(𝑘 − 𝑛), 𝑢(𝑘), 𝑢(𝑘 − 1), . . . , 𝑢(𝑘 − 𝑛)),

where the output at 𝑘 + 1 depends on the input and output at 𝑘, 𝑘 −

1, . . . , 𝑘 − 𝑛 .

Example 2.3.1 (Banking and Interest Rate) In a bank account, let 𝑥(𝑘)
Figure 2.3: Isaac Newton (1642-1726)
denote the beginning balance at the 𝑘 -th month, and let 𝑢(𝑘) denote the developed Newton’s laws in 1686. He is
accumulated deposit/credit or payment/debit during the 𝑘 -th month). an extremely brilliant scientist and in the
A discrete dynamic model to describe the balance at the beginning of meantime often known to be eccentric.
He was described as "...so absorbed in his
every month is studies that he forgot to eat".
https://en.wikipedia.org/wiki/
𝑥(𝑘 + 1) = (1 + 𝑎(𝑘))𝑥(𝑘) + 𝑢(𝑘), Isaac_Newton#/media/File:
Portrait_of_Sir_Isaac_Newton,_1689.jpg

where 𝑎(𝑘) is the interest rate of the 𝑘 -th month. This model is used for a
variety of purposes, for example, to predict the balance after 12 months
assuming the interest rate and the pattern of deposit during the period.

2.4 Example: Atomic Force Microscopy

One of the most powerful techniques for imaging nanoscale objects is

Atomic Force Microscopy (AFM). It can reveal the fine details of surfaces
ranging from single molecules to the uneven texture of a glass pane. AFM
has many applications in various fields of research, such as cell biology,
semiconductors, thin film and coatings, tribology (surface and friction
interactions), molecular biology, and energy storage and energy generation
(photovoltaic) materials. The key to the high performance of AFM is the
 12 2 Modeling

feedback control system that enables precise scanning at the nanometer

level.

The AFM system has two working modes: the tapping mode and the contact
mode. A contact mode AFM system that images a sample surface is shown
in Figure 2.4. The system consists of a cantilever with an atomic-point
needle that scans the sample’s surface. The contact point follows the surface
topology by moving up and down. A laser beam is directed at the cantilever
and is reflected onto a photodiode that measures the beam’s exact location.
Based on this information, a control system can adjust the position of the
cantilever (or the sample under it). The height of the point is recorded as
the surface height at that location. A map of the surface can be created by
combining the heights from the whole scan.

Figure 2.4: Schematic of the AFM.

The control system aims to maintain a constant force on the sample surface
by the needle tip. The cantilever deflection changes with the force on the
needle, and the photodiode can sense the laser movement. This information
is fed into the control system, which adjusts the sample height to keep the
cantilever deflection at the desired level.

The sample height is controlled by a piezoelectric stack actuator element

(piezo) under the sample. Piezo elements are crystals that deform according
to the electric charge they receive. This deformation is used to move the
sample up or down in the z-direction, to regulate the force on the needle.
Usually, piezoelectric stack actuators are also used to move the sample
in the 𝑥 and 𝑦 directions, but since scanning is done at a constant speed
and pattern, these piezo elements do not require the same precision and
bandwidth in their feedback control as the z-direction piezo.

The needle encounters forces on the nanoscale that are not obvious to us.
Forces such as the attractive van der Waals force, which pulls molecules
together, and the repulsive Pauli force, which pushes molecules away,
influence the needle as it moves across the surface, in addition to the
reaction force from the surface. These forces are nonlinear. However,
[1]: Schitter et al. (2007), Design and Model- the major mode of AFM is a spring-mass-damper system. Reference [1]
ing of a High-Speed AFM-Scanner created two models, one second-order and one fourth-order, to describe
the dynamics of the AFM system simply, and then more accurately.

Figure 2.5 shows the second-order system. Here, 𝑀1 models the sample
and the upper part of the piezo element, and has a combined mass of
𝑚1 . 𝑀2 models the lower part of the piezo element and the mass of the
 2.4 Example: Atomic Force Microscopy 13

supporting element beneath, with a combined mass of 𝑚2 . The piezo

element resizes due to force 𝐹 , which affects the two masses at the center
of the piezo element. The supporting element has a spring constant of 𝑘 2
and a damping coefficient of 𝑏 2 . The input to the system is force 𝐹 , which
is generated by a voltage signal that causes the piezo element to expand
and contract.

Figure 2.5: Mechanical models of an AFM.

Based on Newton’s law, the governing equation of the second-order model

of the AFM here is

𝑑2 𝑥1
𝑚1 = 𝐹,
𝑑𝑡 2
𝑑2 𝑥2 𝑑𝑥 2 (2.3)
𝑚2 2 = −𝑏 2 − 𝑘 2 𝑥2 − 𝐹,
𝑑𝑡 𝑑𝑡
𝑙 = 𝑥1 − 𝑥2 ,

where 𝑙 is the distance between 𝑥 1 and 𝑥 2 as shown in Figure 2.5. Fourth order is known to be the highest-
order model for an AFM system that pro-
vides benefits in control design. Models
with higher orders do not enhance the
To account for the piezo element dynamics, we need to modify the system precision much, but instead will increase
to a fourth-order model. This is done by adding another spring-damper the computational cost considerably.
component. The final schematic model in Figure 2.5 shows 𝑘 1 as the effective
spring constant and 𝑏 1 as the effective damping coefficient of the piezo
element. Force 𝐹 still affects both masses. The fourth-order model includes
the dynamics of the second-order model and also captures high-order
dynamics observed in the frequency response of a real-world AFM system
[2]. [2]: Schitter et al. (2001), High performance
feedback for fast scanning atomic force micro-
scopes

Using first principles in Figure 2.5, when spring and damping effects are
considered between the two masses, the governing equations become:

𝑑2 𝑥1 𝑑𝑥1 𝑑𝑥 2
 
𝑚1 2 = −𝑏1 − − 𝑘1 (𝑥1 − 𝑥2 ) + 𝐹,
𝑑𝑡 𝑑𝑡 𝑑𝑡
𝑑2 𝑥2 𝑑𝑥1 𝑑𝑥 2 𝑑𝑥2
   
𝑚2 = 𝑏1 − + 𝑘1 (𝑥1 − 𝑥2 ) − 𝑏2 − 0 − 𝑘 2 𝑥2 − 𝐹.
𝑑𝑡 2 𝑑𝑡 𝑑𝑡 𝑑𝑡
(2.4)
 14 2 Modeling

2.5 Example: Hard Disk Drive and Information

Storage
(millions of them on one disk )
Disk Data track Hard disk drives (HDDs) are amazing mechanical systems that store and
retrieve digital data using magnetic storage. They consist of one or more
Read/write heads
rigid rapidly rotating platters coated with magnetic material, and a read-
write head that moves across the platter surface to access the data. HDDs
were the standard storage system for personal computers for over 30 years
and have been the main storage element in data centers. Their history goes
back to the 1950s when IBM invented the first HDD. Since then, HDDs
have undergone tremendous improvements in terms of capacity, speed,
size, power consumption and reliability.
VCM

Figure 2.6: Control-related components HDDs are examples of high-precision engineering, as they operate at
in a hard disk drive. nanometer scales and millisecond speeds, while being mass-produced and
affordable. HDDs are also versatile, as they can store any kind of digital
data, from text and images to audio and video. They are truly remarkable
devices that have revolutionized the field of data storage and enabled the
Read more about the HDD mechatronics development of modern computing.
at, e.g.,

1. Hard disk drive - Wikipedia. In a modern HDD, data is stored in circular patterns of magnetization
https://en.wikipedia.org/wiki/ known as data tracks or simply, tracks (Figure 2.6). During reading and
Hard_disk_drive. writing of the data, the disk spins and the read/write head is controlled to
2. Anatomy of a Storage Drive: Hard
follow the circular tracks. This creates the track-following problem, where
Disk Drives | TechSpot. https:
//www.techspot.com/article/ the servo system performs regulation control to position the read/write
1984-anatomy-hard-drive/ head at the desired track, with as low variance as possible. During track
Accessed 6/7/2023. following, the position errors are measured periodically at servo sectors
3. Hard disk | Definition
& Facts | Britannica.
that are embedded uniformly over one period of rotation of the disk.
https://www.britannica.com/ Suppose a disk has a rotational speed of 7200 revolutions per minute
technology/hard-disk Accessed (rpm) and the number of servo sectors are 220. Then at every revolution
6/7/2023. of the disks, 220 measurements are obtained, at a sampling frequency of
220 × 7200/60 (= 26 , 400) Hz.

The actuator in a single-stage HDD is powered by a voice coil motor (VCM).

The dynamics between the input current to the voice coil motor and the
output position error signal of the read/write head include the effects of
inertia, damping, spring constant, and resonant modes of the head assembly.
In Section 2.1, we have seen that by Newton’s law, the dynamics again has
a nominal response of a double integrator. Sometimes, the nominal model
also considers friction effects and is written as a second-order damped
The example is based on: Takenori system instead of a pure double integrator. In high precision control,
Atsumi (2023), Magnetic-head multiple high-frequency modes are typically present due to structural
positioning control system in
HDDs (https://www.mathworks.
resonances (Figure 2.7) and the full-order model of the VCM system is in
com/matlabcentral/fileexchange/ the form of:
𝑛
X 𝜅𝑖
111515-magnetic-head-positioning-\
𝐺(𝑠) = 𝐾 . (2.5)
𝑖=1 𝑠 + 2 𝜁 𝑖 𝜔 𝑖 𝑠 + 𝜔 𝑖
control-system-in-hdds), MATLAB 2 2
Central File Exchange. Retrieved June 9,
2023.
The following codes establish a VCM model of a 7200 rpm HDD with 420
servo sectors.
 2.5 Example: Hard Disk Drive and Information Storage 15

P cv
100

50
Gain [dB]

-50

10 1 10 2 10 3 10 4
Frequency [Hz]

-90
Phase [deg.]

-180

-270

-360
10 1 10 2 10 3 10 4
Figure 2.7: Baseline model of the VCM
Frequency [Hz]
component in an HDD.

% modeling/hddvcm.m
% MATLAB code to generate a single-stage HDD model
num_sector=420; % Number of sector
num_rpm=7200; % Number of RPM
Ts = 1/(num_rpm/60*num_sector); % Sampling time

% VCM
Kp_vcm=3.7976e+07; % VCM gain
omega_vcm=[0, 5300 ,6100 ,6500 ,8050 ,9600 ,14800 ,17400 ,21000 ,26000
↩→ ,26600 ,29000 ,32200 ,38300 ,43300 ,44800]*2*pi;
kappa_vcm=[1, -1.0 ,+0.1 ,-0.1 ,0.04 ,-0.7 ,-0.2 ,-1.0 ,+3.0 ,-3.2
↩→ ,2.1 ,-1.5 ,+2.0 ,-0.2 ,+0.3 ,-0.5 ];
zeta_vcm =[0, 0.02 ,0.04 ,0.02 ,0.01 ,0.03 ,0.01 ,0.02 ,0.02 ,0.012
↩→ ,0.007 ,0.01 ,0.03 ,0.01 ,0.01 ,0.01 ];

Sys_Pc_vcm_c1=0;
for i=1:length(omega_vcm)
Sys_Pc_vcm_c1=Sys_Pc_vcm_c1+tf([0,0,kappa_vcm(i)]*Kp_vcm,[1,
↩→ 2*zeta_vcm(i)*omega_vcm(i), (omega_vcm(i))^2]);
end

%% Frequency response
f=logspace(1,log10(60e3),3000);
Fr_Pc_vcm_c1=squeeze(freqresp(Sys_Pc_vcm_c1,f*2*pi)).';

figure
subplot(211)
semilogx(f,20*log10(abs(Fr_Pc_vcm_c1)))
title('P_{cv}');xlabel('Frequency [Hz]');ylabel('Gain
↩→ [dB]');grid;axis([1e1 f(end) -90 100])
subplot(212)
semilogx(f,mod(angle(Fr_Pc_vcm_c1)*180/pi+360,360)-360)
xlabel('Frequency [Hz]');ylabel('Phase [deg.]');grid;axis([1e1 f(end)
↩→ -360 0]);yticks(-360:90:0)
% if you want, you can save the images as follows:
16 2 Modeling

% saveas(gcf,'images/hdd_pcvm_baseline.pdf')
% saveas(gcf,'images/hdd_pcvm_baseline.png')

The Python version of the code is as follows:

# modeling/hddvcm.py
import numpy as np
import matplotlib.pyplot as plt
from scipy import signal
import control as ct

num_sector = 420 # Number of sector

num_rpm = 7200 # Number of RPM
Ts = 1 / (num_rpm / 60 * num_sector) # Sampling time

# VCM
Kp_vcm = 3.7976e+07 # VCM gain
omega_vcm = np.array([0, 5300, 6100, 6500, 8050, 9600, 14800, 17400,
21000, 26000, 26600, 29000, 32200, 38300, 43300,
↩→ 44800]) * 2 * np.pi
kappa_vcm = np.array([1, -1.0, +0.1, -0.1, 0.04, -0.7, -
0.2, -1.0, +3.0, -3.2, 2.1, -1.5, +2.0, -0.2, +0.3,
↩→ -0.5])
zeta_vcm = np.array([0, 0.02, 0.04, 0.02, 0.01, 0.03, 0.01,
0.02, 0.02, 0.012, 0.007, 0.01, 0.03, 0.01, 0.01,
↩→ 0.01])

Sys_Pc_vcm_c1 = ct.TransferFunction(
[], [1]) # Create an empty transfer function
for i in range(len(omega_vcm)):
Sys_Pc_vcm_c1 = Sys_Pc_vcm_c1 + ct.TransferFunction(np.array(
[0, 0, kappa_vcm[i]]) * Kp_vcm, np.array([1, 2 * zeta_vcm[i] *
↩→ omega_vcm[i], (omega_vcm[i]) ** 2]))

# Frequency response
f = np.logspace(1, np.log10(60e3), 3000)
w = f * 2 * np.pi
magPc_vcm, phase_Pc_vcm, omega_Pc_vcm = ct.freqresp(
Sys_Pc_vcm_c1, w) # Get the frequency response

plt.figure()
plt.subplot(211)
plt.semilogx(f, 20*np.log10(magPc_vcm))
plt.title('$P_{cv}$')
plt.xlabel('Frequency [Hz]')
plt.ylabel('Gain [dB]')
plt.grid()
plt.axis([10, f[-1], -90, 100])
plt.subplot(212)
plt.semilogx(f, np.mod(phase_Pc_vcm*180/np.pi+360, 360)-360)
plt.xlabel('Frequency [Hz]')
plt.ylabel('Phase [deg.]')
plt.grid()
plt.axis([10, f[-1], -360, 0])
plt.yticks(np.arange(-360, 90, 90))

With the ever increasing demand of larger capacity in HDDs, dual-stage

actuation using both a VCM and a piezoelectric actuator has become an
essential technique to break the bottleneck of the servo performance in
single-actuator HDDs. A dual-stage HDD applies an additional piezoelec-
tric microactuator (MA) at the end of the VCM actuator, as shown in Figure
 2.5 Example: Hard Disk Drive and Information Storage 17

2.8. We have discussed how piezoelectric microactuators are useful for

high-precision control in AFMs. The MA stage has much smaller moving
range but greatly improved positioning speed and accuracy. Its dynamical
response is also much easier to control, with the low-frequency dynamics
governed simply by a DC gain (Figure 2.9). Compared to the VCM actuator,
the MA has enhanced mechanical performance in the high-frequency
region, providing the capacity to greatly increase the servo bandwidth and
disturbance-attenuation capacity.

Only the position error of the read/write head is measurable in practice.

The plant is hence a dual-input-single-output system.
D
P cp +
30
sD

20
-
Gain [dB]

10

0 Figure 2.8: Schematic structure of dual-

stage HDDs.

-10
10 3 10 4
Frequency [Hz]

180

90
Phase [deg.]

-90

-180
10 3 10 4 Figure 2.9: Baseline model of the PZT
Frequency [Hz] component in an HDD.

The following codes establish a normalized MA model of a 7200 rpm HDD

with 420 servo sectors.

% modeling/hddpzt.m
% MATLAB code to generate the pzt-stage HDD model
num_sector=420; % Number of sector
num_rpm=7200; % Number of RPM
Ts = 1/(num_rpm/60*num_sector); % Sampling time

% PZT
omega_pzt=[14800 ,21500 ,28000 ,40200 ,42050,44400,46500
↩→ ,100000]*2*pi;
kappa_pzt=[-0.005,-0.01 ,-0.1 ,+0.8 ,0.3 ,-0.25 ,0.3 ,10.0 ];
zeta_pzt =[0.025 ,0.03 ,0.05 ,0.008 ,0.008 ,0.01 ,0.02 ,0.3 ];

Sys_Pc_pzt_c1=0;
for i=1:length(omega_pzt)
Sys_Pc_pzt_c1=Sys_Pc_pzt_c1+tf([0,0,kappa_pzt(i)],[1,
↩→ 2*zeta_pzt(i)*omega_pzt(i), (omega_pzt(i))^2]);
end
Sys_Pc_pzt_c1=Sys_Pc_pzt_c1/abs(freqresp(Sys_Pc_pzt_c1,0));
18 2 Modeling

%% Frequency response
f=logspace(1,log10(60e3),3000);
Fr_Pc_pzt_c1=squeeze(freqresp(Sys_Pc_pzt_c1,f*2*pi)).';

figure
subplot(211)
semilogx(f,20*log10(abs(Fr_Pc_pzt_c1)))
title('P_{cp}');xlabel('Frequency [Hz]');ylabel('Gain
↩→ [dB]');grid;axis([1e3 f(end) -10 30])
subplot(212)
semilogx(f,angle(Fr_Pc_pzt_c1)*180/pi)
xlabel('Frequency [Hz]');ylabel('Phase [deg.]');grid;axis([1e3 f(end)
↩→ -180 180]);yticks(-180:90:180)

Here is the model construction in Python:

# modeling/hddpzt.py
import numpy as np
import matplotlib.pyplot as plt
import control as ct

num_sector = 420 # Number of sector

num_rpm = 7200 # Number of RPM
Ts = 1 / (num_rpm / 60 * num_sector) # Sampling time

# PZT
omega_pzt = np.array([14800, 21500, 28000, 40200, 42050,
44400, 46500, 100000]) * 2 * np.pi
kappa_pzt = np.array([-0.005, -0.01, -0.1, +0.8, 0.3, -0.25, 0.3,
↩→ 10.0])
zeta_pzt = np.array([0.025, 0.03, 0.05, 0.008, 0.008, 0.01, 0.02,
↩→ 0.3])

s = ct.TransferFunction.s # Create a variable for the differentiation

↩→ operator
Sys_Pc_pzt_c1 = 0 # Create an empty transfer function
for i in range(len(omega_pzt)):
Sys_Pc_pzt_c1 += kappa_pzt[i] / (s**2 + 2 * zeta_pzt[i] *
↩→ omega_pzt[i]
* s + (omega_pzt[i]) ** 2) # Add
↩→ the transfer functions
Sys_Pc_pzt_c1 /= Sys_Pc_pzt_c1(0) # Normalize the gain at zero
↩→ frequency

# Frequency response
f = np.logspace(1, np.log10(60e3), 3000)
w = f * 2 * np.pi

magPc_pzt, phase_Pc_pzt, omega_Pc_pzt = ct.freqresp(

Sys_Pc_pzt_c1, w) # Get the frequency response

plt.figure()
plt.subplot(211)
plt.semilogx(f, 20*np.log10(magPc_pzt))
plt.title('$P_{cp}$')
plt.xlabel('Frequency [Hz]')
plt.ylabel('Gain [dB]')
plt.grid()
plt.axis([1000, f[-1], - 10, 30])
plt.subplot(212)
plt.semilogx(f, phase_Pc_pzt*180/np.pi)
 2.6 Model Properties 19

plt.xlabel('Frequency [Hz]')
plt.ylabel('Phase [deg.]')
plt.grid()
plt.axis([1000, f[-1], - 180, 180])
plt.yticks(np.arange(-180, 270, 90))

With the VCM and PZT models, a dual-stage HDD model can be formed
as follows.

% MATLAB:
Sys_Pc = [Sys_Pc_vcm_c1; Sys_Pc_pzt_c1];

# Python
Sys_Pc = ct.append(Sys_Pc_vcm_c1, Sys_Pc_pzt_c1)

2.6 Model Properties

We now formalize key properties of system models for control purpose.

Consider a general system M with input 𝑢(𝜎) ( 𝜎 = 𝑡 or 𝑘 depending on
whether the signal is in the continuous- or discrete-time domain) and
output 𝑦(𝜎):
𝑢 /M /𝑦

M is said to be

▶ memoryless or static if 𝑦(𝜎) depends only on 𝑢(𝜎),

▶ dynamic (has memory) if 𝑦 at the current time depends on input
values at other times.
∫𝑡
For example, 𝑦(𝑡) = M(𝑢(𝑡)) = 𝛾𝑢(𝑡) is memoryless; 𝑦(𝑡) = 0
𝑢(𝜏)𝑑𝜏 and
P𝑘
𝑦(𝑘) = 𝑖=0
𝑢(𝑖) are dynamic.

The system M is linear if it satisfies the superposition property:

M(𝛼 1 𝑢1 (𝜎) + 𝛼2 𝑢2 (𝜎)) = 𝛼1 M(𝑢1 (𝜎)) + 𝛼 2 M(𝑢2 (𝜎))

for any input signals 𝑢1 (𝜎) and 𝑢2 (𝜎), and any real numbers 𝛼 1 and 𝛼 2 . If
not, the system is nonlinear.

M is time-invariant if its properties do not change with respect to time.

Assuming the same initial conditions, if we shift 𝑢(𝜎) by a constant time
interval, then M is time-invariant if the output M(𝑢(𝜎 +𝜏0 )) = 𝑦(𝜎 +𝜏0 ). For
example, 𝑦(𝑡)
¤ = 𝐴𝑦(𝑡) + 𝐵𝑢(𝑡) is linear and time-invariant; 𝑦(𝑡)
¤ = 2 𝑦(𝑡) −
sin(𝑦(𝑡))𝑢(𝑡) is nonlinear, yet time-invariant; 𝑦(𝑡)
¤ = 2 𝑦(𝑡) − 𝑡 sin(𝑦(𝑡))𝑢(𝑡)
is time-varying. Often, we abbreviate linear time-invariant systems as LTI
systems.

In mechanical systems, torque limits of motors, hardening spring, Coulomb

friction forces, etc. make systems nonlinear. However, even if physical
systems are nonlinear, they can often be linearized or be well represented
by linear systems under specific conditions, making linear analysis and
design tools powerful.

The system M is called

20 2 Modeling

▶ causal if 𝑦(𝑡) or 𝑦(𝑘) depends on 𝑢(𝜏) for 𝜏 ≤ 𝑡 or 𝑘 ,

▶ strictly causal if the inequality is strict.

For example, 𝑦(𝑡) = 𝑢(𝑡 − 10) is strictly causal.

A system that is not causal is said to be acausal.
For an LTI continuous-time system to be causal, the order of the denominator
must be greater than or equal to the order of the numerator in its transfer
function. For example, for the model in Equation 2.1 to be causal, it must
be that 𝑛 ≥ 𝑚 . (Check, e.g., the case with 𝑛 = 0 and 𝑚 = 1.)

2.7 Nonlinear Systems

Although we will be mostly focusing on linear systems, in practice, many

control systems are nonlinear. Several examples are presented next to show
how the nonlinearity plays a role in modeling.

2.7.1 Example: Magnetically Suspended Ball

A magnetically suspended ball consists of a ball of mass 𝑚 suspended by

an electromagnet as shown below.

Fup y

mg

Let 𝑦 be the position of the ball, measured down from the base of the
electromagnet. If a current 𝑢 is injected into the coils of the electromagnet,
it will produce a magnetic field which in turn exerts an upward force on
the ball given by 𝐹𝑢𝑝 = − 𝑐𝑢 . Note that the force decreases as 𝑦 2 increases
2

𝑦2
because the effect of the magnet weakens when the ball is further away,
and that the force is proportional to 𝑢 2 which is representative of the power
supplied to the magnet.
Let us assume that we can measure the position 𝑦 of the ball. This can be
arranged optically using light-emitting diodes with photocells.
We can then write a simple model for the motion of the ball from Newton’s
second law as
𝑐𝑢 2
𝑚 𝑦¥ = 𝑚 𝑔 − 2 . (2.6)
𝑦
Note that this is a single-input, single-output nonlinear model. The input
and output of this model are:
 2.7 Nonlinear Systems 21

𝑢 the current injected into the magnet coils

𝑦 the position of the ball

2.7.2 Example: Water Tank

Consider the water tank in the following figure:

𝑢(𝑡)
?
ℎ(𝑡)
6
𝐴
𝑎, 𝑞 𝑜𝑢𝑡 (𝑡)

Denote by 𝑞 𝑖𝑛 (𝑡) the water flow entering the tank and 𝑞 𝑜𝑢𝑡 (𝑡) the water
flow leaving the tank from a hole of area 𝑎 . Out goal is to obtain a model
that describes the evolution of the tank height ℎ(𝑡) as a function of the
external input 𝑢(𝑡) = 𝑞 𝑖𝑛 (𝑡). Let 𝐴 be the area of the tank. By using the
conservation law we can state that

¤ = 𝑞 𝑖𝑛 (𝑡) − 𝑞 𝑜𝑢𝑡 (𝑡).

𝐴 ℎ(𝑡)

Let 𝑣 𝑜𝑢𝑡 (𝑡) be the speed of the water at the outlet. Then, 𝑞 𝑜𝑢𝑡 (𝑡) = 𝑎𝑣 𝑜𝑢𝑡 (𝑡).

Torricelli’s law states that the speed of a fluid through a sharp-edged hole
under the force of gravity is the same as the speed p that a body would
acquire in falling freely from a height ℎ , i.e. 𝑣 𝑜𝑢𝑡 (𝑡) = 2 𝑔 ℎ(𝑡), where 𝑔 is
the acceleration due to gravity. Considering all the above, our nonlinear
model is q
1
ℎ¤ = (𝑢(𝑡) − 𝑎 2 𝑔 ℎ(𝑡)).
𝐴

2.7.3 Example: Pendulum

Consider the pendulum shown in the following figure.

We assume that the pendulum has mass 𝑚 = 0.333 kg which is concentrated

at the end point and length 𝑙 = 0.75 meters. The angle 𝜃 is measured, and
an actuator can supply a torque 𝑢(𝑡) = 𝑇𝑐 (𝑡) to the pivot point. The moment
of inertia about the pivot point is 𝐼 = 𝑚𝑙 2 . By analyzing the rigid body
 22 2 Modeling

dynamics and writing Euler’s equation for the pendulum, we can readily
arrive at a differential equation model for this system:

¥ = 𝑇𝑐 − 𝑚 𝑔𝑙 sin(𝜃(𝑡)),
𝐼 𝜃(𝑡)

or,
𝑔
¥ = 1 𝑇𝑐 − sin(𝜃(𝑡)),
𝜃(𝑡)
𝑚𝑙 2 𝑙
which is a second-order nonlinear differential equation.

2.7.4 Example: Vehicle Steering

Figure 2.10: Kinematic modelling of the

four-wheel vehicle steering system.

Figure 2.10 shows the kinematics of a four-wheeled vehicle. If we assume

the front and rear pairs of wheels act the same as a single wheel on each axle
at the centerline of the vehicle, we arrive at a two-wheel abstraction known
as the bicycle model. The bicycle model does not take into account tilting of
the vehicle. A four-wheeled vehicle would experience very different tilting
mechanics than those of a two-wheeled vehicle. However, when only the
steering of the vehicle is of concern with the assumption that it cannot
tilt, such as in vehicles with a low center of gravity and operating at slow
Read more about the bicycle model at, enough speeds, the bicycle model is appropriate.
e.g., https://thef1clan.com/2020/09/21/
When
vehicle-dynamics-the-kinematic-bicycle-model/. the vehicle turns, the inner and outer tires exhibit different steering
angles due to their varying distances from point 𝑂 in the bicycle model in
Figure 2.11. Here, 𝛿 is the steering angle of the front wheels, 𝑏 is the wheel
base, 𝑎 is the distance between the rear axle and the center of mass, 𝑥 and 𝑦
are the positions of the center of mass, 𝜃 is the heading, and 𝛼 is the angle
between the velocity 𝑣 and the centerline of the vehicle. The point 𝑂 is the
intersection of the centerlines of the of the front and rear wheels. By letting
the distance from the center of rotation 𝑂 to the rear wheel contact point
be 𝑟𝑟 , we can deduce that 𝑏 = 𝑟𝑟 tan 𝛿 and 𝑎 = 𝑟𝑟 tan 𝛼 . This leads to the
relationship between 𝛼 and the steering angle 𝛿 :

𝑎 tan 𝛿
 
𝛼 = arctan . (2.7)
𝑏
 2.7 Nonlinear Systems 23

O
<latexit sha1_base64="aX5OSAcyff457KE0qOMYCjDyXZI=">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</latexit>

↵
<latexit sha1_base64="AFdd/xiA8hcOU64JgcPkJJYRgVA=">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</latexit>

<latexit sha1_base64="Bm2V1vBvPkyplDlkMqKDJuhIxAY=">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</latexit>

rc
<latexit sha1_base64="2WKT/3GF2EkETQkRV91n+cE9nEE=">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</latexit>

rr
<latexit sha1_base64="S5/XCeA9IBcwcMPECZar3TDMbBk=">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</latexit>

a
b

y a
↵
<latexit sha1_base64="AFdd/xiA8hcOU64JgcPkJJYRgVA=">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</latexit>

✓
<latexit sha1_base64="jKWX8jQMkgy52Q4032scWgd1ILQ=">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</latexit>

x Figure 2.11: The bicycle model of the ve-

hicle steering system.

Given a vehicle speed 𝑣 at its center of mass, the motion of the center of
mass is expressed as:
𝑑𝑥
= 𝑣 cos(𝛼 + 𝜃),
𝑑𝑡
(2.8)
𝑑𝑦
= 𝑣 sin(𝛼 + 𝜃).
𝑑𝑡

To determine the influence of the steering angle on the heading angle 𝜃 , we

note that the distance from the center of mass to the center of rotation 𝑂 is
𝑟 𝑐 = 𝑎/sin 𝛼. As the vehicle rotates around point 𝑂 , the angular velocity is
given by 𝑣/𝑟 𝑐 = (𝑣/𝑎) sin 𝛼 . Therefore,

𝑣 𝑣 sin 𝛼 𝑣 𝑎 tan 𝛿
  
𝜃¤ = = = sin arctan . (2.9)
𝑟𝑐 𝑎 𝑎 𝑏

When the steering angle 𝛿 and the angle 𝛼 (known as the slip angle) is
small, the above equation becomes
𝑣
𝜃¤ ≈ 𝛿. (2.10)
𝑏

Let the input 𝑢 be the steering angle 𝛿 . The full set of nonlinear equations
of motion for the vehicle steering problem is now:

𝑥 𝑣 cos (𝛼 (𝑢) + 𝜃)

𝑑  𝑦  = 𝑓 (x, 𝑢) =
 
 𝑣 sin (𝛼 (𝑢) + 𝜃)  , 𝛼 (𝑢) = arctan 𝑎 tan 𝑢 . (2.11)
 
𝑑𝑡  
𝜃 

 𝑣 sin 𝛼(𝑢)

 𝑏
   𝑎 
|{z}
x
 24 2 Modeling

2.8 “All Models are Wrong, but Some are Useful”

A perfect model is very hard to be obtained in reality. In the Journal of

[3]: Box (1976), Science and statistics the American Statistical Association [3], British statistician George Box
famously wrote in 1976, that “all models are wrong, but some are useful.”
The aphorism acknowledges that statistical models always fall short of
the complexities of reality but can still be useful nonetheless. A dynamic
system may simply be too complex (consider the neural system of human
brains), or there are inevitable hardware uncertainties such as the fatigue
of gears or bearings in a car. Modeling thus involves varying degrees of
approximation. For example in modeling a car, we may chose to ignore
rolling resistance, aero-dynamic effects, road-tire interactions, etc.

However, feedback control can empower a system to tolerate model un-

certainties. In a 5-year-old car, the same engine control unit (ECU) can
maintain smooth performance of the internal combustion engine despite
the accumulated mileage and wear, whether they come from highways
Robust control is not tied to complex con- or mountains. In the hard disk drive example in Section 2.1, the high-
trol. A PID controller can create robust frequency vibration modes can change their shapes and locations in a batch
stability too.
of products and when working under different temperatures, leading to the
[4]: Atsumi (2022), Magnetic-head Position- frequency responses in Figure 2.12 [4]. However, thanks to the robustness
ing Control System in HDDs from feedback controls, when making thousands of hard disk drives per
day, the manufactures do not need to tune the controller in each individual
drive. Instead, each batch can robustly read and write data using the same
The subfield of robust control covers de- servo controller.
tails of why and how to handle the model
uncertainties. P cv

-20
Gain [dB]

-40

-60

-80

10 3 10 4
Frequency [Hz]

-90
Phase [deg.]

-180

-270

-360
Figure 2.12: Example frequency responses
10 3 10 4
of the voice coil motor stage in a batch of Frequency [Hz]
hard disk drives.

The following MATLAB codes generate the perturbed HDD model in

Figure 2.12.
 2.8 “All Models are Wrong, but Some are Useful” 25

% modeling/hdddsa.m
% Dual-stage HDD model
num_sector=420; % Number of sector
num_rpm=7200; % Number of RPM
Ts = 1/(num_rpm/60*num_sector); % Sampling time

% VCM
Kp_vcm=3.7976e+07;
omega_vcm=[0, 5300 ,6100 ,6500 ,8050 ,9600 ,14800 ,17400 ,21000 ,26000
↩→ ,26600 ,29000 ,32200 ,38300 ,43300 ,44800]*2*pi;
kappa_vcm=[1, -1.0 ,+0.1 ,-0.1 ,0.04 ,-0.7 ,-0.2 ,-1.0 ,+3.0 ,-3.2
↩→ ,2.1 ,-1.5 ,+2.0 ,-0.2 ,+0.3 ,-0.5 ];
zeta_vcm =[0, 0.02 ,0.04 ,0.02 ,0.01 ,0.03 ,0.01 ,0.02 ,0.02 ,0.012
↩→ ,0.007 ,0.01 ,0.03 ,0.01 ,0.01 ,0.01 ];

% PZT
omega_pzt=[14800 ,21500 ,28000 ,40200 ,42050,44400,46500
↩→ ,100000]*2*pi;
kappa_pzt=[-0.005,-0.01 ,-0.1 ,+0.8 ,0.3 ,-0.25 ,0.3 ,10.0 ];
zeta_pzt =[0.025 ,0.03 ,0.05 ,0.008 ,0.008 ,0.01 ,0.02 ,0.3 ];

%% LT(Low temp.) model: VCM: +4 % resonance shift from nominal

↩→ model, PZT actuator: +6 % resonance shift from nominal
↩→ model
% VCM
Sys_Pc_vcm_c1=0;
for i=1:length(omega_vcm)
Sys_Pc_vcm_c1=Sys_Pc_vcm_c1+ss(tf([0,0,kappa_vcm(i)]*Kp_-
↩→ vcm,[1, 2*zeta_vcm(i)*0.8*omega_vcm(i)*1.04,
↩→ (omega_vcm(i)*1.04)^2]));
Sys_Pc_vcm_c1=ssbal(Sys_Pc_vcm_c1);
end

% PZT
Sys_Pc_pzt_c1=0;
for i=1:length(omega_pzt)
Sys_Pc_pzt_c1=Sys_Pc_pzt_c1+ss(tf([0,0,kappa_pzt(i)],[1,
↩→ 2*zeta_pzt(i)*0.8*omega_pzt(i)*1.06,
↩→ (omega_pzt(i)*1.06)^2]));
Sys_Pc_pzt_c1=ssbal(Sys_Pc_pzt_c1);
end
Sys_Pc_pzt_c1=Sys_Pc_pzt_c1/abs(freqresp(Sys_Pc_pzt_c1,0));

%% RT(Room temp.) model: Same as nominal models

% VCM
Sys_Pc_vcm_c2=0;
for i=1:length(omega_vcm)
Sys_Pc_vcm_c2=Sys_Pc_vcm_c2+ss(tf([0,0,kappa_vcm(i)]*Kp_-
↩→ vcm,[1, 2*zeta_vcm(i)*omega_vcm(i),
↩→ omega_vcm(i)^2]));
Sys_Pc_vcm_c2=ssbal(Sys_Pc_vcm_c2);
end

% PZT
Sys_Pc_pzt_c2=0;
for i=1:length(omega_pzt)
Sys_Pc_pzt_c2=Sys_Pc_pzt_c2+ss(tf([0,0,kappa_pzt(i)],[1,
↩→ 2*zeta_pzt(i)*omega_pzt(i), omega_pzt(i)^2]));
Sys_Pc_pzt_c2=ssbal(Sys_Pc_pzt_c2);
end
Sys_Pc_pzt_c2=Sys_Pc_pzt_c2/abs(freqresp(Sys_Pc_pzt_c2,0));
26 2 Modeling

%% HT(High temp.) model: VCM: -4 % resonance shift from nominal

↩→ model, PZT actuator: -6 % resonance shift from nominal
↩→ model
% VCM
Sys_Pc_vcm_c3=0;
for i=1:length(omega_vcm)
Sys_Pc_vcm_c3=Sys_Pc_vcm_c3+ss(tf([0,0,kappa_vcm(i)]*Kp_-
↩→ vcm,[1, 2*zeta_vcm(i)*1.2*omega_vcm(i)*0.96,
↩→ (omega_vcm(i)*0.96)^2]));
Sys_Pc_vcm_c3=ssbal(Sys_Pc_vcm_c3);
end

% PZT
Sys_Pc_pzt_c3=0;
for i=1:length(omega_pzt)
Sys_Pc_pzt_c3=Sys_Pc_pzt_c3+ss(tf([0,0,kappa_pzt(i)],[1,
↩→ 2*zeta_pzt(i)*1.2*omega_pzt(i)*0.94,
↩→ (omega_pzt(i)*0.94)^2]));
Sys_Pc_pzt_c3=ssbal(Sys_Pc_pzt_c3);
end
Sys_Pc_pzt_c3=Sys_Pc_pzt_c3/abs(freqresp(Sys_Pc_pzt_c3,0));

%% LT / PZT gain +5% (Case 4)

Sys_Pc_vcm_c4=Sys_Pc_vcm_c1;
Sys_Pc_pzt_c4=Sys_Pc_pzt_c1*1.05;

%% RT / PZT gain +5% (Case 5)

Sys_Pc_vcm_c5=Sys_Pc_vcm_c2;
Sys_Pc_pzt_c5=Sys_Pc_pzt_c2*1.05;

%% HT / PZT gain +5% (Case 6)

Sys_Pc_vcm_c6=Sys_Pc_vcm_c3;
Sys_Pc_pzt_c6=Sys_Pc_pzt_c3*1.05;

%% LT / PZT gain -5% (Case 7)

Sys_Pc_vcm_c7=Sys_Pc_vcm_c1;
Sys_Pc_pzt_c7=Sys_Pc_pzt_c1*0.95;

%% RT / PZT gain -5% (Case 8)

Sys_Pc_vcm_c8=Sys_Pc_vcm_c2;
Sys_Pc_pzt_c8=Sys_Pc_pzt_c2*0.95;

%% HT / PZT gain -5% (Case 9)

Sys_Pc_vcm_c9=Sys_Pc_vcm_c3;
Sys_Pc_pzt_c9=Sys_Pc_pzt_c3*0.95;

%% All plant
Sys_Pc_vcm_all=[Sys_Pc_vcm_c1;Sys_Pc_vcm_c2;Sys_Pc_vcm_c3;Sys_Pc_vcm_-
↩→ c4;Sys_Pc_vcm_c5;Sys_Pc_vcm_c6;Sys_Pc_vcm_c7;Sys_Pc_vcm_c8;Sys_-
↩→ Pc_vcm_c9];
Sys_Pc_pzt_all=[Sys_Pc_pzt_c1;Sys_Pc_pzt_c2;Sys_Pc_pzt_c3;Sys_Pc_pzt_-
↩→ c4;Sys_Pc_pzt_c5;Sys_Pc_pzt_c6;Sys_Pc_pzt_c7;Sys_Pc_pzt_c8;Sys_-
↩→ Pc_pzt_c9];

%% Cotrolled object (Discrete-time system)

% Case 1
Sys_Pd_vcm_c1=c2d(Sys_Pc_vcm_c1,Ts,'ZOH');
Sys_Pd_pzt_c1=c2d(Sys_Pc_pzt_c1,Ts,'ZOH');

% Case 2
Sys_Pd_vcm_c2=c2d(Sys_Pc_vcm_c2,Ts,'ZOH');
Sys_Pd_pzt_c2=c2d(Sys_Pc_pzt_c2,Ts,'ZOH');
 2.8 “All Models are Wrong, but Some are Useful” 27

% Case 3
Sys_Pd_vcm_c3=c2d(Sys_Pc_vcm_c3,Ts,'ZOH');
Sys_Pd_pzt_c3=c2d(Sys_Pc_pzt_c3,Ts,'ZOH');

% Case4
Sys_Pd_vcm_c4=c2d(Sys_Pc_vcm_c4,Ts,'ZOH');
Sys_Pd_pzt_c4=c2d(Sys_Pc_pzt_c4,Ts,'ZOH');

% Case 5
Sys_Pd_vcm_c5=c2d(Sys_Pc_vcm_c5,Ts,'ZOH');
Sys_Pd_pzt_c5=c2d(Sys_Pc_pzt_c5,Ts,'ZOH');

% Case 6
Sys_Pd_vcm_c6=c2d(Sys_Pc_vcm_c6,Ts,'ZOH');
Sys_Pd_pzt_c6=c2d(Sys_Pc_pzt_c6,Ts,'ZOH');

% Case 7
Sys_Pd_vcm_c7=c2d(Sys_Pc_vcm_c7,Ts,'ZOH');
Sys_Pd_pzt_c7=c2d(Sys_Pc_pzt_c7,Ts,'ZOH');

% Case 8
Sys_Pd_vcm_c8=c2d(Sys_Pc_vcm_c8,Ts,'ZOH');
Sys_Pd_pzt_c8=c2d(Sys_Pc_pzt_c8,Ts,'ZOH');

% Case 9
Sys_Pd_vcm_c9=c2d(Sys_Pc_vcm_c9,Ts,'ZOH');
Sys_Pd_pzt_c9=c2d(Sys_Pc_pzt_c9,Ts,'ZOH');

% All
Sys_Pd_vcm_all=[Sys_Pd_vcm_c1;Sys_Pd_vcm_c2;Sys_Pd_vcm_c3;Sys_Pd_vcm_-
↩→ c4;Sys_Pd_vcm_c5;Sys_Pd_vcm_c6;Sys_Pd_vcm_c7;Sys_Pd_vcm_c8;Sys_-
↩→ Pd_vcm_c9];
Sys_Pd_pzt_all=[Sys_Pd_pzt_c1;Sys_Pd_pzt_c2;Sys_Pd_pzt_c3;Sys_Pd_pzt_-
↩→ c4;Sys_Pd_pzt_c5;Sys_Pd_pzt_c6;Sys_Pd_pzt_c7;Sys_Pd_pzt_c8;Sys_-
↩→ Pd_pzt_c9];

%% Frequency response
f=logspace(1,log10(60e3),3000);
Fr_Pc_vcm_all=squeeze(freqresp(Sys_Pc_vcm_all,f*2*pi)).';
Fr_Pc_pzt_all=squeeze(freqresp(Sys_Pc_pzt_all,f*2*pi)).';
Fr_Pd_vcm_all=squeeze(freqresp(Sys_Pd_vcm_all,f*2*pi)).';
Fr_Pd_pzt_all=squeeze(freqresp(Sys_Pd_pzt_all,f*2*pi)).';

figure
subplot(211)
semilogx(f,20*log10(abs(Fr_Pc_vcm_all(:,1:7))))
hold on
semilogx(f,20*log10(abs(Fr_Pc_vcm_all(:,8:9))),'--')
hold off
title('P_{cv}');xlabel('Frequency [Hz]');ylabel('Gain
↩→ [dB]');grid;axis([1e3 f(end) -90 10])
subplot(212)
semilogx(f,mod(angle(Fr_Pc_vcm_all(:,1:7))*180/pi+360,360)-360)
hold on
semilogx(f,mod(angle(Fr_Pc_vcm_all(:,8:9))*180/pi+360,360)-360,'--')
hold off
xlabel('Frequency [Hz]');ylabel('Phase [deg.]');grid;axis([1e3 f(end)
↩→ -360 0]);yticks(-360:90:0)
legend('Case 1','Case 2','Case 3','Case 4','Case 5','Case 6','Case
↩→ 7','Case 8','Case 9','Location','NorthWest')

figure
subplot(211)
28 2 Modeling

semilogx(f,20*log10(abs(Fr_Pc_pzt_all(:,1:7))))
hold on
semilogx(f,20*log10(abs(Fr_Pc_pzt_all(:,8:9))),'--')
hold off
title('P_{cp}');xlabel('Frequency [Hz]');ylabel('Gain
↩→ [dB]');grid;axis([1e3 f(end) -10 30])
subplot(212)
semilogx(f,angle(Fr_Pc_pzt_all(:,1:7))*180/pi)
hold on
semilogx(f,angle(Fr_Pc_pzt_all(:,8:9))*180/pi,'--')
hold off
xlabel('Frequency [Hz]');ylabel('Phase [deg.]');grid;axis([1e3 f(end)
↩→ -180 180]);yticks(-180:90:180)
legend('Case 1','Case 2','Case 3','Case 4','Case 5','Case 6','Case
↩→ 7','Case 8','Case 9','Location','NorthWest')

2.9 Exercise

1. What is the difference between a causal system and an acausal system?

2. Can a static system be acausal?
3. Can a linear system be time-varying?
4. Solve the following ODEs:
a) 𝑦¥ + 𝑦¤ − 2 𝑦 = 0, 𝑦(0) = 4, 𝑦(
¤ 0) = −5,
b) 𝑦¥ + 𝑦¤ + 0.25 𝑦 = 0, 𝑦(0) = 3, 𝑦(
¤ 0) = −3.5,
c) 𝑦¥ + 0.4 𝑦¤ + 9.04 𝑦 = 0.
5. A quadcopter is a type of unmanned aerial vehicle (UAV) that is
controlled by adjusting the speed of its four rotors. Read relevant
literature and develop a mathematical model of a quadcopter.
6. A robotic arm is used to move objects in a manufacturing facility.
The arm consists of several joints that can be controlled to move the
end-effector to a desired position. Read relevant literature. Develop a
dynamic model between the torque input to the motors and the joint
angles.
7. A power system consists of generators, transformers, transmission
lines, and loads. Read relevant literature and develop a simplified
model of a power system.
8. A cruise control system in a car maintains a constant speed by
adjusting the throttle. The system must be able to handle changes in
terrain and driving conditions. Read relevant literature and develop
a simplified model of a cruise control system.
9. A temperature control system in a chemical plant consists of a heater,
a temperature sensor, and a controller. The controller receives input
from the temperature sensor and sends a signal to the heater to adjust
the temperature. Develop a block diagram of the control system and
a mathematical model of the plant.