Smith and Fuzzy Control For Complex Systems: Huang Xiaogang

SMITH AND FUZZY CONTROL
FOR COMPLEX SYSTEMS
HUANG XIAOGANG
NATIONAL UNIVERSITY OF SINGAPORE

2003
Founded 1905
SMITH AND FUZZY CONTROL

FOR COMPLEX SYSTEMS
BY
HUANG XIAOGANG (B.ENG., M.ENG.)

DEPARTMENT OF ELECTRICAL & COMPUTER
ENGINEERING
A THESIS SUBMITTED
FOR THE DEGREE OF MASTER ENGINEERING
NATIONAL UNIVERSITY OF SINGAPORE

2003
Acknowledgments
I would like to express my sincere appreciation and gratitude to my supervisor,

Prof. Wang Qing-Guo, who has guided me through my Master’s study. Without his
gracious encouragement and generous guidance , I would not be able to finish the
work here. His uncompromising quest for excellence shapes me significantly. His
wealth of knowledge and accurate foresight have greatly impressed and benefited
me. I am indebted to him for his care and advice not only in my academic research
but also in my daily life. I would like to extend special thanks to Dr. Zheng Feng
and Dr. Zhang Yong. Their comments, advice and the inspiration given have
played a very important role in this piece of work.
Special thanks also go to Prof. Chen, Ben-Mei and Prof. Xu Jian-Xin who have
in one way or another given me their kind help. I would like to express my thanks
to Yang Yongsheng, Dr. Yang Xueping, Huang Bin, He Ru and many others in
the Advanced Control Technology Lab(Center for Intelligent Control) for their care
and kind help. I am also grateful to the National University of Singapore for the
research scholarship.
Finally, this thesis would not have been possible without the love, patience and
support from my family. Their encouragement from my parents and my wife has
been invaluable. I would like to dedicate this thesis to them and hope that they
will find joy in this humble achievement.
Huang Xiaogang
May, 2003
i
Contents
Acknowledgments i
Summary iv
List of Figures vii
1 Introduction 1
1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.3 Organization of the Thesis . . . . . . . . . . . . . . . . . . . . . . . 6
2 Modified Smith Predictor Control for General Unstable Processes 7

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.2 The Scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.3 Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.4 Internal Stability Analysis of Two Modified Smith Predictors . . . . 15
2.4.1 Astrom’s Modified Smith Predictor . . . . . . . . . . . . . . 15
2.4.2 Matausek’s Modified Smith Predictor . . . . . . . . . . . . . 16
2.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3 Robust PI controller design for nonlinear systems via fuzzy mod-

eling approach 20
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.2 Problem Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.3 Proportional Control . . . . . . . . . . . . . . . . . . . . . . . . . . 25
ii
3.4 PI Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.5 Numerical Example: Excitation Control of Power Systems . . . . . 31
3.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
4 Fuzzy Modelling and Control of F-16 Aircraft 47

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
4.2 F-16 Aircraft Model . . . . . . . . . . . . . . . . . . . . . . . . . . 51
4.2.1 Need for Modelling . . . . . . . . . . . . . . . . . . . . . . . 51
4.2.2 Modelling Method . . . . . . . . . . . . . . . . . . . . . . . 52
4.2.3 Workable Model . . . . . . . . . . . . . . . . . . . . . . . . . 54
4.3 TS Fuzzy Modelling . . . . . . . . . . . . . . . . . . . . . . . . . . 55
4.3.1 The Technique . . . . . . . . . . . . . . . . . . . . . . . . . 56
4.3.2 TS Model of F-16 . . . . . . . . . . . . . . . . . . . . . . . . 58
4.3.3 Model Validation . . . . . . . . . . . . . . . . . . . . . . . . 62
4.4 LQR Based Control . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
4.4.1 Stabilization . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
4.4.2 Tracking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
4.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
5 Conclusions 76
5.1 Main Findings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
5.2 Suggestions for Further Work . . . . . . . . . . . . . . . . . . . . . 77
Bibliography 79
Appendix A Original Model . . . . . . . . . . . . . . . . . . . . . . . . 86
Appendix B TS Fuzzy Basic Model . . . . . . . . . . . . . . . . . . . . 87
Appendix C TS Fuzzy Augmented Model . . . . . . . . . . . . . . . . 91
Author’s Publications 97
iii
Summary
In process control, many modified Smith predictor schemes have been applied to
stable process and integrator plus dead-time process, while few considered the
general unstable process and internal stability of proposed structures. Generally
speaking, many industrial processes can be approximated sufficiently well in con-
cerned operating region of state space by linear systems. However, many other
plants exist whose dynamics must be described by nonlinear systems. Compared
with the voluminous references on tuning PID controllers for linear systems, few
results have been reported in the literature on PID controller tuning for nonlinear
systems. Therefore, it is highly desirable to develop effective methods to determine
the parameters of PID controllers for nonlinear systems. Feedback linearization is
often used to control specific output variable sets in nonlinear flight control prob-
lems. However, it is difficult to identify accurately aerodynamic coefficients because
they are nonlinear functions of several physical variables. With these considera-
tions in mind, the thesis is mainly devoted the development of (i) modified Smith
Predictor control for general unstable processes; (ii) robust PI controller design
for nonlinear systems via fuzzy modeling approach; and (iii) fuzzy modelling and
control of nonlinear F-16 aircraft system.
In this thesis, the analysis and design of a modified Smith predictor control
scheme for general unstable processes is presented. We generalized Matausek and
Micic’s scheme from pure integral processes to general unstable delay processes,
analyzed internal stability, and discussed design issues. The proposed scheme can,
by both analysis and simulation, yield great improvement of both set-point and
load disturbance responses over conventional feedback systems.
iv
In the context of proportional-plus-integral (PI) controllers for nonlinear sys-
tems, first, the Takagi-Sugeno (T-S) fuzzy model with parameter uncertainties is
used to approximate the nonlinear systems. Then a numerically tractable algo-
rithm based on the technique of iterative linear matrix inequalities is developed to
design a proportional (static output feedback) controller to the robust stabilization
of the system in T-S fuzzy model. Third, we transform the problem of PI controller
design to that of proportional controller design for an augmented system and thus
bring the solution of the former problem into the configuration of the developed
algorithm. Finally, the proposed method is applied to the design of robust stabi-
lizing controllers for the excitation control of power systems. Simulation results
show that the transient stability can be improved by using a fuzzy PI controller
when large faults appear in the system, compared to the conventional PI controller
designed by using linearization method around the steady state.
A framework for control of F-16 aircraft with Takagi-Sugeno (TS) fuzzy systems
is developed. First, based on the best-available nonlinear dynamical model of F-
16 aircraft in the open domain, the TS fuzzy model of F-16 aircraft is presented
and validated with reasonable accuracy. Then, the LQR based control strategy is
proposed using the TS model. It is applied to synthesize a F-16 flight control system
for both stabilizing control and attitude tracking control. Extensive simulation
is carried out. It is concluded that the proposed control design is feasible and
outperformances the linear control design significantly.
The results presented in the thesis have very practical values as well as sound
theoretical contributions. The findings in the thesis can be applied to industrial
control systems, as show form several implementation tests.
v
List of Figures
2.1 Modified Smith predictor control scheme . . . . . . . . . . . . . . . 8

2.2 Conventional feedback system . . . . . . . . . . . . . . . . . . . . . 10
2.3 Step responses of modified Smith predictor control system . . . . . 14
2.4 Step responses of conventional feedback system . . . . . . . . . . . 14
2.5 The Smith predictor proposed by Astrom . . . . . . . . . . . . . . . 15
2.6 The Smith predictor proposed by Matausek . . . . . . . . . . . . . 17
3.1 Membership functions of the fuzzy models for power system (3.24) . 36
3.2 Response of system (3.24) with fuzzy proportional controller (3.26)
(tF = 0.13 seconds) . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.3 Response of system (3.24) with fuzzy PI controller (3.27) (tF = 0.13
seconds) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
seconds) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.5 Response of system (3.24) with linear PI controller (3.28) (tF = 0.08
seconds) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
seconds) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.7 Response of system (3.24) with linear PI controller (3.28) (tF =
0.163 seconds) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
3.8 Approximation errors of linear and fuzzy nonlinear models for power
system (3.24) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
4.1 Definition of aircraft axes and angles. . . . . . . . . . . . . . . . . . 53
vi
4.2 Fuzzy triangle membership functions . . . . . . . . . . . . . . . . . 62
4.3 Approximate error of TS-fuzzy and linear model (different α) . . . . 63
4.4 Approximate error of TS-fuzzy and linear model (different φ) . . . . 63
4.5 The LQR based stabilizing control I . . . . . . . . . . . . . . . . . . 68
4.6 The LQR based stabilizing control II . . . . . . . . . . . . . . . . . 69
4.7 The LQR based stabilizing control III . . . . . . . . . . . . . . . . . 70
4.8 The LQR based tracking control . . . . . . . . . . . . . . . . . . . . 74
vii
Chapter 1
Introduction
1.1 Motivation
Proportional-Integral-Derivative (PID) controllers are still widely used in the pro-
cess industries even though control theory and technology have been developed sig-
nificantly since they were first used decades ago. The popularity of PID controllers
is due to their functinoal simplicity, which allows process engineers to operate them
in a simple and straightforward manner. Also, they provide robust and reliable
performance for a wide range of operating conditions. PID controller exhibits good
performance for linear and small time delay systems. In process industries many
control loops show long deadtime characteristics. However, processes with long
time-delays can not be controlled effectively using a simple PID controller. This is
due to the additional phase lag contributed by the time delay which tends to desta-
bilize the closed loop system. The stability problem can be solved by decreasing
the controller gain. However, in this case the reponse obtained is very sluggish.
The Smith predictor (SP) developed by Otto Smith (1959) is the most popular
and effective dead-time compensator for a stable process with long time-delay.
However, this approach fails in a very significant way for an unstable process owing
to the problem of stabilization(Wang et al., 1999a). Many modifications to the
original Smith scheme have been proposed to overcome this obstacle. Astrom et
al. (1994) presented a modified Smith predictor which is applicable to integrator
1
Chapter 1. Introduction 2
plus dead-time process and can achieve faster setpoint response and better load
disturbance rejection. Matausek and Micic (1996) considered the same problem
with similar results but their scheme is easier to tune. The unstable process other
than the above class was considered only by Majhi and Atherton (1999) where
a rather complicated modified Smith predictor was developed and it has three
controllers to be turned. It is also noted that in all these works, internal stability
of the proposed structures is not analyzed. It will be thus useful in this respect to
generalize a scheme from pure integral processes to general unstable delay processes
and analyze internal stability.
Despite the developments of various kinds of modern or postmodern control
theories, such as LQG or LQR optimal control, H∞ control, and µ analysis and
synthesis, classical proportional-plus-integral (PI) or proportional-plus-integral-
derivative (PID) controllers (Astrom and Hagglund, 1995; Astrom et al., 1992; Tan
et al., 1999) are widely used in industry due to their relatively simple structure,
ready-in-hand implementation, and perhaps, being easily understood. Therefore
it is often the case (Astrom and Hagglund, 1995; Astrom et al., 1992; Wang et
al., 1999b) that in practical applications one first considers PID controllers un-
less evidence shows that they are insufficient to meet specification. Because of the
popularity of PID controllers in the real world, many approaches (Astrom and Hag-
glund, 1995; Astrom et al., 1993; Datta et al., 2000) and the references therein, have
been developed to determine the parameters of PID controllers. The first system-
atic tuning method for PID parameters was proposed by Ziegler and Nichols in the
early 1940s (Ziegler and Nichlos, 1942). Then came the well-known formulae such
as Cohen-Coon method (Cohen and Coon, 1953), integral absolute error (IAE) op-
timum method (Shinskey, 1988; Smith and Corripio, 1985), integral time-weighted
absolute error (ITAE) optimum method (Smith and Corripio, 1985), internal model
control (IMC) method (Chien and Fruehauf, 1990; Morari and Zafiriou, 1989) and
relay auto-tuning method (Astrom and Hagglund, 1984). Recently, many mod-
ified tuning methods (Bi et al., 2000; Ho et al., 1998; Ho et al., 1995; Wang et
al., 1999b; Zhuang and Atherton, 1993) have also been proposed associated with
different performance specifications or different methods used accordingly. Notice

that all the aforementioned tuning methods are only suitable to linear systems.
Compared with the voluminous references on tuning PID controllers for linear
systems, few results have been reported in the literature on PID controller tuning
for nonlinear systems. Indeed, many industrial processes can be approximated suf-
ficiently well in concerned operating region of state space by linear systems. But
there do exist many other plants whose dynamics must be described by nonlinear
systems. Therefore, it is highly desirable to develop effective methods to deter-
mine the parameters of PID controllers for nonlinear systems. To the best of the
author’s knowledge, the only reference discussing this problem is the monograph
by Freeman and Kokotovic (1996a), where backstepping and control Lyapunov
function methods are used to design PI control, but it is supposed that all state
variables are available.
As in the design methodologies of many other types of controllers, a crucial step
for the design of PID controllers is the modelling of the nonlinear plants. Since
the 1980s, fuzzy technique has been widely adopted to model complex nonlinear
plants. The so called T-S or T-S-K fuzzy model, first proposed by Takagi and
Sugeno (1985) and further developed by Sugeno and Kang (1988), is one of most
successful models in this direction. The basic idea in this approach is first to
decompose the model of a nonlinear system or other kinds of complex systems into
linear systems in accordance with the cases for which linear models are suitable to
describe and then to aggregate (fuzzy blend) each individual model (linear model)
into a single nonlinear model in terms of their membership function. Thus the
relatively complex consequence part allows the number of fuzzy rules (local models)
to be quite small in many applications. Consequently, the T-S fuzzy model is less
prone to the curse of dimensionality than other fuzzy models. As is well known,
the key problem in this approach is to what degree the nonlinear system can be
approximated by a convex (fuzzy) blending of several linear systems. Theoretical
justification of T-S fuzzy model as a universal approximator has been given by
Wang and Mendel (1992). Now T-S model has found wide applications in the
control of complex systems, for example, in the control of robot manipulators

(Palm et al., 1997) and time delay systems (Cao and Frank, 2000). There is thus a
high demand for a different approach to design PI controllers for nonlinear systems,
which will use output variables only or partial state variables.
The classical flight control technology can be used via local linearization method.
However, the aircraft model is inherently nonlinear, exact input-output lineariza-
tion is often used to control specific output variable sets in nonlinear flight control
problems (Lane and Stengel, 1988; Krstić and Tsiotras, 1999). Feedback lineariza-
tion is a theoretically established and widely used method in controlling nonlinear
systems. Unfortunately, this direct application of feedback linearization requires
the second or third derivatives of uncertain aerodynamic coefficients and does not
guarantee internal stability for non-minimum phase systems. Another difficulty re-
lated to the application of feedback linearization to a flight control system is that a
complete and accurate aircraft dynamic model including aerodynamic coefficients
is required. It is difficult to identify accurately aerodynamic coefficients because
they are nonlinear functions of several physical variables. The control problem of
a F-16 aircraft model has been addressed by many researchers for the purpose of
the control of spacecraft and aircraft (Debs and Athans, 1969; Vadali et al., 1984).
Also, there have been several works that consider performance indices such as
time and/or fuel in the formulation of the optimal control problems (Athans et
al., 1963; Scrivener and Thomson, 1994). These studies have mainly addressed the
regulation problem for the angular velocity subsystem and for some quadratic costs
(Athans et al., 1963; Windeknecht, 1963; Tsiotras et al., 1996). Recently, the atti-
tude control problem of the complete system that includes the dynamics as well as
the kinematics has been investigated by many researchers. There have been several
studies concerning the systematic design of stabilizing fuzzy controllers (Tanaka
and Sugeno, 1992; Tanaka et al., 1998), since Takagi and Sugeno (1985) opened a
new direction of research in the area of control by introducing the Takagi-Sugeno
(TS) fuzzy model. It is appealing to design both stabilizing and tracking control
with application to the complete attitude motion of a rigid spacecraft based on TS
fuzzy system model.
1.2 Contributions
In the present thesis, a modified Smith Predictor control scheme for general un-
stable processes is addressed. Robust PI controller design for nonlinear systems is
proposed via fuzzy modeling approach. A framework for control of F-16 aircraft
with Takagi-Sugeno (TS) fuzzy systems is developed.
A. Modified Smith Predictor Control for General Unstable Processes

The analysis and design of a modified Smith predictor control scheme for general
unstable processes is presented. We generalized Matausek and Micic’s scheme
from pure integral processes to general unstable delay processes, analyzed internal
stability, and discussed design issues. The proposed scheme can, by both analysis
and simulation, yield great improvement of both set-point and load disturbance
responses over conventional feedback systems.
B. Robust PI Controller Design for Nonlinear Systems via Fuzzy

Modeling Approach
The design problem of proportional-plus-integral (PI) controllers for nonlinear
systems is studied. First, the Takagi-Sugeno (T-S) fuzzy model with parameter
uncertainties is used to approximate the nonlinear systems. Then a numerically
tractable algorithm based on the technique of iterative linear matrix inequalities
is developed to design a proportional (static output feedback) controller to the
robust stabilization of the system in T-S fuzzy model. Third, we transform the
problem of PI controller design to that of proportional controller design for an
augmented system and thus bring the solution of the former problem into the
configuration of the developed algorithm. Finally, the proposed method is applied
to the design of robust stabilizing controllers for the excitation control of power
systems. Simulation results show that the transient stability can be improved by
using a fuzzy PI controller when large faults appear in the system, compared to the
conventional PI controller designed by using linearizion method around the steady
state.
C. Fuzzy Modelling and Control of F-16 Aircraft

First, based on the best-available nonlinear dynamical model of F-16 aircraft
in the open domain, the TS fuzzy model of F-16 aircraft is presented and validated
with reasonable accuracy. Then, the LQR based control strategy is proposed using
the TS model. It is applied to synthesize a F-16 flight control system for both sta-
bilizing control and attitude tracking control. Extensive simulation is carried out.
It is concluded that the proposed control design is feasible and outperformances
the linear control design significantly.
1.3 Organization of the Thesis

The thesis is organized as follows. Chapter 2 presents Smith predictor control
scheme for general unstable processes. Chapter 3 focuses on the design of robust
PI controller for nonlinear systems via fuzzy modeling approach. In Chapter 4,
a framework for control of F-16 aircraft with Takagi-Sugeno (TS) fuzzy systems
is presented. Finally, conclusions and suggestions for further works are drawn in
Chapter 5.
Chapter 2
Modified Smith Predictor Control

for General Unstable Processes
2.1 Introduction
In process control, the Smith predictor (SP) (Smith, 1959) is a well known dead-
time compensator. But the original Smith predictor control scheme will be unstable
when applied to unstable processes(Wang et al., 1999a). To overcome this obstacle,
many modifications to the original Smith scheme have been proposed. Astrom et
al. (1994) presented a modified Smith predictor which is applicable to integrator
plus dead-time process and can achieve faster setpoint response and better load
disturbance rejection. Matausek and Micic (1996) considered the same problem
with similar results but their scheme is easier to tune. To our best knowledge,
the unstable process other than the above class was considered only by Majhi
and Atherton (1999) where a rather complicated modified Smith predictor was
developed and it has three controllers to be turned. It is also noted that in all
these works, internal stability of the proposed structures is not analyzed.
The Chapter is organized as follows. Section 2.2 devives the proposed scheme.
Simulation examples are given in Section 2.3. In Section 2.4, the internal stability
of two typical modified Smith Predictors are analysized. Section 2.5 concludes the
chapter.
7
Chapter 2. Modified Smith Predictor Control for General Unstable Processes 8
d
r + ur u + y
K 1 (s )
+ G 0 (s)e - Ls
- -
d̂
ˆ (s)
G K 2 (s )
0
ym - +
e - Ls
+
+
Figure 2.1. Modified Smith predictor control scheme
2.2 The Scheme

Let the process be represented by
G(s) = G0 (s)e−Ls ,
where G0 is a rational function with at least one unstable pole. A modified Smith
predictor control scheme is depicted in Fig. 2.1. Note that it reduces to Matausek
and Micic’s scheme if
kp
G0 (s) =
s
and K2 (s) is a static gain. But in this chapter we consider general unstable G0
and general K2 (s).
Let G0 (s) = a(s)/b(s),K1 (s) = m(s)/n(s), and K2 (s) = α(s)/β(s) be coprime
polynomial fractions. Internal stability is a prerequisite for any control systems. It
is shown by Wang et al. (1999a) that an interconnected system consisting only of
single-input/single output(SISO) plants is internally stable if and only if p c (s) ,
Q
∆ i pi (s) has all its roots in the open left half of the complex plane, where pi (s)
are the denominators of the plant transfer functions and the ∆ is the system
determinant as defined in the Mason’s formula. To apply this theorem, one sees
that the system in Fig. 2.1 consists of five subsystems:
K1 (s), Ĝ0 , K2 (s), G0 e−Ls and e−Ls .
Their respective pi are
p1 = n(s), p2 = b̂(s), p3 = β(s), p4 = b(s), and p5 = 1.
The system has the following four loops:
L1 : −K1 (s)Ĝ0 , L2 : −K2 (s)G0 e−Ls , L3 : −K1 (s)G0 e−Ls and L4 : K1 (s)Ĝ0 e−Ls .
where L1 and L2 possess no common node, and thus they are non-touching. It
follows that the system determinant ∆ is
∆ = 1 − L 1 − L 2 − L 3 − L 4 + L 1 ∗ L2
= 1 + K1 (s)Ĝ0 + K2 (s)G0 e−Ls + K1 (s)G0 e−Ls − K1 (s)Ĝ0 e−Ls
+K1 (s)Ĝ0 K2 (s)G0 e−Ls

h i£
= 1 + K1 (s)Ĝ0 1 + K2 (s)G0 e−Ls + K1 (s)e−Ls (G0 − Ĝ0 ).
¤
In the case of perfect modeling, i.e., G0 = Ĝ0 , we have
∆ = [1 + K1 (s)G0 ] 1 + K2 (s)G0 e−Ls

£ ¤
(2.1)
Therefore, we have
5
Y
pc (s) = ∆(s) pi (s)
i=1
= [1 + K1 (s)G0 ][1 + K2 (s)G0 e−Ls ] · n(s) · b(s) · β(s) · b(s) · 1
= [n(s)b(s) + m(s)a(s)][β(s)b(s) + α(s)a(s)e−Ls ].
The polynomial, n(s)b(s) + m(s)a(s), reflects stabilization of delay-free G 0 by the

controller K1 , which is always possible, say, by pole placement. Thus, the overall
system is stabilizable if and only if the delay process G is stabilizable. For a general
unstable delay system, readers are referred to Bonnet and Partington (1999) for
design of a stabilizing controller K2 . For a low-order unstable process, the methods
in Walton and Marshall (1987) or root locus technique would be adequate.
d
r + u + y
K (s ) G 0 (s)e - Ls
-
Figure 2.2. Conventional feedback system
In view of the above analysis, it can be concluded that unlike the original Smith
system where the characteristic equation is delay-free, the modified Smith scheme
gets no simplification as far as stabilization is concerned. It is indeed the case
for all the existing modified Smith schemes as we found (Majhi and Atherton,
1999; Astrom et al., 1994; Matausek and Micic, 1996), that is, their characteristic
equations are all delay-dependent like (2.1). Then why do we use the modified
scheme? The merit lies in performance improvement but not stabilization, as will
be shown now.
For a conventional feedback system in Fig. 2.2, the input-output relation can
be derived as
Y (s) = H̃r (s)R(s) + H̃d (s)D(s),
where
K(s)G0 e−Ls
H̃r (s) = ,
1 + K(s)G0 e−Ls
G0 e−Ls
H̃d (s) = .
1 + K(s)G0 e−Ls
On the other hands, the modified Smith system in Fig. 2.1 gives
Y (s) = Hr (s)R(s) + Hd (s)D(s),

where
K1 (s)G0 e−Ls [1 + K2 (s)G0 e−Ls ] K1 (s)G0 e−Ls
Hr (s) = = ,
[1 + K1 (s)G0 ][1 + K2 (s)G0 e−Ls ] 1 + K1 (s)G0
G0 [e−Ls + K1 (s)G0 (e−Ls − e−2Ls )]

Hd (s) = .
[1 + K1 (s)G0 ][1 + K2 (s)G0 e−Ls ]
Note that the set-point response Hr (s) now has delay-free denominator and pro-
vides great potential improvement over conventional feedback systems with H̃r (s).
Though the load disturbance response Hd (s) is still delay-dependent, performance
enhancement can also be significant, as will be shown by simulation in the next
section, as well as a simplifying analysis as follows. We compare such responses
for two schemes:
Hd G0 [e−Ls + K1 (s)G0 (e−Ls − e−2Ls )] 1 + K(s)G0 e−Ls

= ·
H̃d [1 + K1 (s)G0 ][1 + K2 (s)G0 e−Ls ] G0 e−Ls
[1 + K1 (s)G0 (1 − e−Ls )][1 + K(s)G0 e−Ls ]

= (2.2)
[1 + K1 (s)G0 ][1 + K2 (s)G0 e−Ls ]
Notice that the major function of K(s) and K2 (s) is the same, i.e, to stabilize
the delay plant. Let us assume now K(s) = K2 (s). (2.2) then becomes
Hd K1 (s)G0 e−Ls
= 1− = 1 − Hr (s)
H̃d 1 + K1 (s)G0
For a reasonably designed controller K1 , Hr (s) should be close to unity in the
working frequency range for good set-point response. If this is the case, the above
ratio is close to zero, implying a much smaller response from Hd than H̃d .
2.3 Simulation
Consider the following unstable low-order plus dead time model:
G(s) = G0 (s)e−Ls
k(T3 s + 1)
= e−Ls
(T1 s − 1)(T2 s + 1)
where all coefficients are non-negative. Note that standard first-order or second-
order model can be obtained from the above G by setting the coefficient values
accordingly. Suppose that we use K2 to cancel the stable and minimum phase part
of the plant for easy stabilizing design, that is,
(T2 s + 1)kf
K2 (s) = .
T3 s + 1
Take, as usual, K1 (s) as a PID type:
K1 (s) = kp + Ti /s + kd s.
Then, pc (s) for this modified Smith system is

5
Y
pc (s) = ∆ pi
i=1
= [1 + K1 (s)G0 ][1 + K2 (s)G0 e−Ls ] · s · (T1 s − 1)(T2 s + 1) · (T3 s + 1)
·(T1 s − 1)(T2 s + 1) · 1
k p s + T i + k d s2
· ¸
k(T3 s + 1)
= 1+ ·
s (T1 s − 1)(T2 s + 1)
· ¸
(T2 s + 1)kf k(T3 s + 1)
· 1+ · · e−Ls
T3 s + 1 (T1 s − 1)(T2 s + 1)
·s · (T1 s − 1)(T2 s + 1) · (T3 s + 1) · (T1 s − 1)(T2 s + 1) · 1
(T1 T2 + kd kT3 )s3 + (T1 − T2 + kp kT3 + kd k)s2 + (kTi T3 + kp k − 1)s + kTi

£ ¤
=
· T1 s + kf ke−Ls − 1 (T3 s + 1)(T2 s + 1).

£ ¤
Obviously, pc1 = [(T1 T2 + kd kT3 )s3 + (T1 − T2 + kp kT3 + kd k)s2 + (kTi T3 +

kp k − 1)s + kTi ] can be assigned to an arbitrary stable polynomial by the PID
controller K1 (s). However, the quasi-polynomial pc2 = T1 s + kf ke−Ls − 1 has
£ ¤
infinite spectrum.
To be specific, let T1 = 10, k = 1 and T2 = T3 to get the same model as in

Majhi and Atherton (1999),
e−5s
G(s) = ,
10s − 1
K1 in PI form is adequate (let kd = 0) and pc (s) becomes
5
Y
pc (s) = ∆ pi
i=1
= [1 + K1 (s)G0 ][1 + K2 (s)G0 e−Ls ] · s · (10s − 1) · 1 · (10s − 1) · 1
· ¸· ¸
kp s + T i 1 1
= 1+ · 1 + kf · · e−Ls
s (10s − 1) (10s − 1)
·s · (10s − 1) · 1 · (10s − 1) · 1
10s2 + (kp − 1)s + Ti 10s + kf e−Ls − 1

£ ¤£ ¤
=
= pc1 · pc2
If the desired poles for pc1 are s1,2 = −0.95±0.31i, then we get K1 (s) = (20+ 10
s
).
To make all the roots of pc2 (s) = 0 lie in the left half of the complex s-plane, we
adopt the method introduced in Walton and Marshall (1987). The characteristic
equation of the system is
F (s, L) = 10s − 1 + kf e−Ls = A(s) + B(s)e−Ls = 0.
One sees that F (s, 0) = 10s − 1 + kf is stable when kf > 1. From
W (w2 ) = A(iw)A(−iw) − B(iw)B(−iw) = 100w 2 − kf2 + 1 = 0,

q
we get a positive solution,w = (kf2 − 1)/100. For such a w, the positive zeros of
F (iw, L) can be found from
© 10wi − 1 ª 1
cos wL = Re − = ,
kf kf
© 10wi − 1 ª 10w q
sin wL = Im = = 1 − 1/kf2 .
kf kf
1.25
0.75
0.5
0.25
0
0 10 20 30 40 50 60 70 80 90 100
Figure 2.3. Step responses of modified Smith predictor control system

2.5
1.5
0.5
0
0 100 200 300 400 500 600 700 800 900 1000
Figure 2.4. Step responses of conventional feedback system
√ √
2
√
2
Suppose kf = 2, then w = 0.1, cos wL = 2
and sin wL = 2
, so that
5π
L= + 20πq, q = 0, 1, 2, · · ·,
2
among which, the smallest L is 5π/2. It can be concluded that the only range of L
for stability is 0 ≤ L < 5π/2. In this example, L = 5 falls in this range. Therefore,
choosing kf = 1.414 can make all the roots of pc2 (s) = 0 lie in the left half of the
complex s-plane. The responses of the system to the unit step set-point change and
load disturbance of step size d = −0.1 are shown in Fig. 2.3, where the responses
to the same signals for conventional feedback system designed with the method of
Ho and Xu (1998) with controller K(s) = 1.506 + 0.012/s are shown in Fig. 2.4.
The performance enhancement from the proposed scheme is evident and big.
2.4 Internal Stability Analysis of Two Modified

Smith Predictors
In this section, we analyse the internal stability of two Modified Smith Predictors
proposed by Astrom et al. (1994) and Matausek and Micic (1996).
2.4.1 Astrom’s Modified Smith Predictor
A block diagram of Astrom’s modified Smith predictor is shown in Fig. 2.5.
d
r + u + + + 1 - Ls y
K e
- - s
dˆ
1
M(s)
s
yˆ - + y
e - Ls
Figure 2.5. The Smith predictor proposed by Astrom
The system consists of five subsystems:
1 −Ls 1
K, , e , M (s) and e−Ls .
s s
Let M (s) = a(s)/b(s) be comprime polynomial fraction. Their respective p i

are
p1 = 1, p2 = s, p3 = 1, p4 = b(s), and p5 = s.
The system has the following two loops:
1 1
L1 : − K and L2 : − e−Ls M (s).
s s
∆ = 1 − L 1 − L 2 + L 1 ∗ L2
1 1 1 1
= 1 + K + e−Ls M (s) + K e−Ls M (s)
· s ¸ ·s s ¸s
1 1 −Ls
= 1 + K 1 + e M (s) .
s s
Therefore, we have
5
Y
pc (s) = ∆(s) pi (s)
i=1
· ¸· ¸
1 1 −Ls
= 1 + K 1 + e M (s) · 1 · s · 1 · b(s) · s
s s
= (s + K)[b(s)s + a(s)e−Ls ].
where
a(s) K4 s2 + K 3 s
M (s) = =
b(s) (1 + K1 )s2 + (K2 − K4 )s − (K4 s + K3 )e−Ls
It can be inferred that
pc (s) = (s + K)[(1 + K1 )s3 + (K2 − K4 )s2 − (K4 s2 + K3 s)e−Ls
+(K4 s2 + K3 s)e−Ls ]
= s2 (s + K)[(1 + K1 )s + k2 − K4 ]
pc (s) = 0 at least have two roots not in the open left half of the complex plane,
no matter what controller K and M (s) were chosen. Therefore the plant is not
internally stable.
2.4.2 Matausek’s Modified Smith Predictor
A block diagram of Matausek’s modified Smith predictor is shown in Fig. 2.6.

The system consists of five subsystems:
Kp Kp −Ls
Kr , , K0 , e and e−Ls .
s s
d
r + ur u + Kp y
Kr e - Ls
- - s
dˆ
Kp s K0
ym - +
e - Ls
+
+
Figure 2.6. The Smith predictor proposed by Matausek
Their respective pi are
p1 = 1, p2 = s, p3 = 1, p4 = s, and p5 = 1.
The system has the following four loops:

1 1 1 1
L1 : −Kr Kp · , L2 : −K0 Kp e−Ls · , L3 : −Kr Kp e−Ls · and L4 : Kr Kp e−Ls · .
s s s s
∆ = 1 − L 1 − L 2 − L 3 − L 4 + L 1 ∗ L2
1 1 1 1
= 1 + Kr Kp · + K0 Kp e−Ls · + Kr Kp e−Ls · − Kr Kp e−Ls ·
s s s s
1 1
+Kr Kp · · K0 Kp e−Ls ·
s s
= (1 + Kr Kp /s)(1 + K0 Kp e−Ls /s).
Therefore, we have
5
Y
pc (s) = ∆(s) pi (s)
i=1
= (1 + Kr Kp /s)(1 + K0 Kp e−Ls /s) · 1 · s · 1 · s · 1
= (s + Kr Kp )(s + K0 Kp e−Ls ).
In the simulations(Matausek and Micic, 1996), Kp , K0 , Kr and L were given

by
Kp = 1, K0 = 0.1, Kr = 0.6, L = 5
So
pc (s) = (s + 0.6)(s + 0.1e−5s )
= pc1 · pc2
The pole for pc1 is s1 = −0.6, which lies in the left half of the complex s-plane.
To make all the roots of pc2 = 0 lie in the left half of the complex s-plane, we also
adopt the method introduced in Walton and Marshall (1987). The characteristic
equation of the system is
F (s, L) = s + 0.1e−Ls = A(s) + B(s)e−Ls = 0.
One sees that F (s, 0) = s + 0.1 is stable. From
W (w2 ) = A(iw)A(−iw) − B(iw)B(−iw) = w 2 − 0.01 = 0,
we get a positive solution,w = 0.1. For such a w, the positive zeros of F (iw, L)
can be found from
© iw ª
cos wL = Re − = 0,
0.1
© iw ª
sin wL = Im = 1.
0.1
so that
L = 5π + 20πq, q = 0, 1, 2, · · ·,
among which, the smallest L is 5π. It can be concluded that the only range of L
for stability is 0 ≤ L < 5π. In this example, L = 5 falls in this range. Therefore
all the roots of pc (s) = 0 lie in the left half of the complex s-plane. The plant is
internally stable.
2.5 Conclusion
The Smith predictor is an effective dead-time compensator for a processes with
long dead-time. Recently, there have been renewed interests in extension of the
original Smith predictor control from stable processes to unstable processes. But in
these works the stability of proposed structures was not explicitly analyzed and only
simulation examples were given. In this chapter, we have generalized Matausek and
Micic’s scheme from pure integral processes to general unstable delay processes,
analyzed internal stability, and discussed design issues. The proposed scheme can,
by both analysis and simulation, yield great improvement of both set-point and
load disturbance responses over conventional feedback systems.
Chapter 3
Robust PI controller design for

nonlinear systems via fuzzy
modeling approach
3.1 Introduction
Few results have been reported in the literature on PID controller tuning for nonlin-
ear systems, despite the voluminous references on tuning PID controllers for linear
systems. Many industrial processes can be approximated sufficiently well in con-
cerned operating region of state space by linear systems, but there do exist many
other plants whose dynamics must be described by nonlinear systems. Therefore,
it is highly desirable to develop effective methods to determine the parameters of
PID controllers for nonlinear systems. To the best of the authors’ knowledge, the
only reference discussing this problem is the monograph by Freeman and Kokotovic
(1996a), where backstepping and control Lyapunov function methods are used to
design PI control, but it is supposed that all state variables are available. In this
chapter, we will try to take a different approach to design PI controllers for non-
linear systems. As shown later, the proposed controllers will use output variables
only or partial state variables.
As in the design methodologies of many other types of controllers, a crucial step
20
Chapter 3. Robust PI controller design for nonlinear systems via fuzzy modeling
approach 21
for the design of PID controllers is the modelling of the nonlinear plants. Since
the 1980s, fuzzy technique has been widely adopted to model complex nonlinear
plants. The so called T-S or T-S-K fuzzy model, first proposed by Takagi and
Sugeno (1985) and further developed by Sugeno and Kang (1988), is one of most
successful models in this direction. Now T-S model has found wide applications in
the control of complex systems, for example, in the control of robot manipulators
(Palm et al., 1997) and time delay systems (Cao and Frank, 2000). In this chapter,
we will apply T-S fuzzy modelling approach to the design of robust PI controllers
for nonlinear systems.
This chapter is organized as follows. The problem formulation is introduced in
Section 3.2. In Section 3.3, proportional (static output feedback) controller design
for nonlinear systems using T-S fuzzy models is presented. PI controller design
for nonlinear systems is given in Section 3.4. A numerical example is provided
in Section 3.5 to show the design procedure and the effectiveness of the proposed
method. Finally concluding remarks are drawn in Section 3.6.
Notation: Rn denotes n−dimensional Euclidean space, Rn×m is the set of all
n × m real matrices, I is an identity matrix whose dimension is implied from
context, and k · k refers to either the Euclidean vector norm or the induced matrix
2-norm. The notation X > Y (or X ≥ Y, respectively) means that X − Y is
positive definite (or positive semidefinite, respectively). The symbol := means
“defined as.” In the augmented matrices, the dimensions of the added elements
will be implied from their associated variables and hence will not be mentioned
throughout this chapter.
3.2 Problem Formulation

Consider the following uncertain nonlinear system:

 ẋ(t) = f (x, t) + ∆f (x, t, ξ) + g(x, t)u(t),
(3.1)
 y(t) = Cx(t),
approach 22
where x(t) ∈ Rn is the vector of state variables, u(t) ∈ Rm is the vector of control
inputs, y ∈ Rl is the vector of outputs, ξ is a vector of uncertain parameters which
is restricted to a prescribed bounding set described later, f and g, nonlinear func-
tions of (x, t), representing the nominal model of the system under consideration,
are continuously differentiable with respect to (x, t), ∆f represents the model un-
certainties, and C ∈ Rl×n is a constant matrix. Ignoring the uncertainty term, we
obtain the nominal model of the system (3.1) as follows:
ẋ(t) = f (x, t) + g(x, t)u(t), (3.2)
y(t) = Cx(t). (3.3)
To deal with the problem for the system (3.2) to run at different operating
points or in different conditions, Takagi and Sugeno (1985) proposed an effective
method to represent the system (3.2). The main feature of a T-S fuzzy model is
to express the join dynamics of each fuzzy implication (rule) by a linear system.
The i-th rule of the T-S fuzzy model is of the following form:
Plant Rule i ( i = 1, . . . , r) :
IF θ1 is µi1 and · · · and θp is µip THEN
ẋ(t) = Ai x(t) + Bi u(t),
where θj (x) (j = 1, . . . , p) are the premise variables, which are functions of state
variables x, µij ( i = 1, . . . , r, j = 1, . . . , p) are fuzzy sets, r is the number of the IF-
THEN rules, p is the number of the premise variables, and Ai and Bi are constant
matrices of compatible dimensions. It is assumed that the premise variables are
independent of the input variables u(t). The overall fuzzy model is achieved by
fuzzy blending (aggregation) of each individual rule (model) as follows:
Pr
wi (θ)(Ai x + Bi u)
ẋ(t) = i=1 Pr , (3.4)
i=1 wi (θ)
where θ = [θ1 , . . . , θp ], and wi : Rp → [0, 1], i = 1, . . . , r, are the membership

function of the system’s belonging to plant rule i.
Define
wi (θ)
αi (θ) = Pr .
j=1 wj (θ)
approach 23
The system (3.4) admits the following form:

r
X
ẋ(t) = αi (θ)(Ai x + Bi u) (3.5)
i=1
with the constraints:

r
X
αi (θ) ≥ 0, αi (θ) = 1. (3.6)
i=1
Stripping off the fuzzy cover of system (3.5), one can see that the model (3.5) is
nothing but the convex combination of several linear systems. It has been shown
(Johansen and Foss, 1993; Wang and Mendel, 1992) that a nonlinear system can
be approximated by means of the above fuzzy basis functions to desired accuracy
inside an arbitrarily large compact subspace of Rn .
Correspondingly, we can model uncertain nonlinear system (3.1) as the follow-
ing uncertain fuzzy system:
Plant Rule i ( i = 1, . . . , r) :
IF θ1 is µi1 and · · · and θp is µip THEN
ẋ(t) = (Ai + ∆Ai (x, t, ξ))x(t) + Bi u(t), (3.7)
where ∆Ai represents the uncertainties in system matrix. We assume that ∆Ai
admits the following form:
∆Ai (x, t, ξ) = Di Υi (x, t, ξ)Ei , i = 1, 2, . . . , r, (3.8)
where Di and Ei are known real constant matrices of dimensions n×nDi and nEi ×n
respectively, and the uncertainty Υi (x, t, ξ), an unknown matrix-valued function of
(x, t, ξ), belongs to the following bounded set:
Γi := {ςi (x, t, ξ) ∈ RnDi ×nEi : kςi (x, t, ξ)k ≤ 1
the elements of ςi are Lebesque measurable},
i = 1, 2, . . . , r. (3.9)
Fuzzy blending of each individual model yields the overall fuzzy model as follows:

 ẋ(t) = Pr α (θ)[(A + ∆A (x, t, ξ))x(t) + B u(t)],
i=1 i i i i
(3.10)
 y(t) = Cx(t).
approach 24
Here the output equation is also added into the overall model for the convenience
of later citation. We assume that all the triples (Ai , Bi , C), i = 1, 2, . . . , r, are
controllable and observable.
It should be noted that the parameter uncertainty structure in (3.8) and (3.9)
describes how the uncertainties enter the system model and has been widely used
in the study of the problem of robust stability and stabilization of uncertain lin-
ear systems and it can represent parameter uncertainty in many physical cases
(Khargonekar et al., 1990). Actually, any norm-bounded parameter uncertainty
can be expressed in the form of (3.8) and (3.9).
Our objective in this chapter is to design a PI controller of the following form
Z t
u(t) = FP y(t) + FI y(τ )dτ, (3.11)
0
such that the closed loop system (3.10) and (3.11) is asymptotically stable, where
FP , FI ∈ Rm×l are constant matrices.
Controller (3.11) is an ideal PI controller. Readers are referred to (Astrom and
Hagglund, 1995) for how to change the ideal PI controllers into practical ones and
the relationships between the parameters of the two kinds of the controllers.
Remark 3.1. It is noticed that the output of the system is assumed to be a linear
function of the state in this chapter. If it is not the case, a coordinate change
(Isidori, 1995) can always been found such that the output equation is a linear
mapping. Actually, a pre-nonlinear-transformation between the sensed variables
and the actual input to the processor of the controller exists in many practical
plants. The rule of this transformation is just to make the equivalent output
equation of the whole plant become a linear mapping. Therefore the assumed
linear form in the output equation here has no harm on the generality of the
method developed in this chapter.
Remark 3.2. Due to some technical problem, no uncertainties are assumed to ap-
pear in the input matrices Bi and i = 1, 2, . . . , r.
approach 25
3.3 Proportional Control

In this section, we first study the stabilization of system (3.10) by a proportional
controller, i.e., by the following controller:
u(t) = F y(t), (3.12)
where F ∈ Rm×l is a constant matrix. To do it, we need the following proposition.
Proposition 3.3.1. System (3.10) is asymptotically stabilized via controller (3.12)

if there exists a matrix P > 0 such that the following matrix inequalities hold:
P (Ai + ∆Ai )T + (Ai + ∆Ai )P − P C T CP
+(Bi F + P C T )(Bi F + P C T )T < 0
∀Υi (x, t, ξ) ∈ Γi , i = 1, 2, . . . , r, (3.13)
Proof. First notice that system (3.10) is asymptotically stabilized via controller
(3.12) if there exists a matrix P > 0 such that
P (Ai + ∆Ai + Bi F C)T + (Ai + ∆Ai + Bi F C)P < 0
∀Υi (x, t, ξ) ∈ Γi , i = 1, 2, . . . , r. (3.14)
To show this claim, we choose V (x) := xT P −1 x(t) as a Lyapunov function candi-

date for system (3.10). Since Γi is a closed set and r is finite, there exists a positive
definite matrix Q0 such that
P (Ai + ∆Ai + Bi F C)T + (Ai + ∆Ai + Bi F C)P < −Q0
∀Υi (x, t, ξ) ∈ Γi , i = 1, 2, . . . , r.
if matrix inequality (3.14) holds. Then noticing the fact that αi (x) ≥ 0, i =
1, 2, . . . , r and ri=1 αi (x) = 1, we have
P
V̇ (x) = ẋT (t)P −1 x(t) + xT (t)P −1 ẋ(t)

( r
X
= xT P −1 αi (x)P (Ai + ∆Ai + Bi F C)T
i=1
r
)
X
+ αi (x)(Ai + ∆Ai + Bi F C)P P −1 x
i=1
T −1
< −x P Q0 P −1 x.
approach 26
Thus follows the claim.

Now, following the same procedure as the proof of Theorem 1 of Cao and Sun
(1998), we can prove that matrix inequality (3.14) holds if and only if matrix
inequality (3.13) holds. This completes the proof.
Theorem 3.1. Consider the uncertain nonlinear system (3.10). Suppose there ex-
ist positive definite matrices P, a matrix F, and positive numbers εi , i = 1, 2, . . . , r,
such that the following matrix inequalities hold:
P ATi + Ai P + εi Di DiT + ε−1 T T

i P Ei Ei P − P C CP
+(Bi F + P C T )(Bi F + P C T )T < 0
i = 1, 2, . . . , r, (3.15)
Then, system (3.10) can be robustly stabilized by controller u(t) = F y(t).
To prove this theorem, we need the following lemma.
Lemma 3.1. Let D, E, Υ be real matrices of appropriate dimensions with Υ

satisfying kΥk ≤ 1. Then for any real number ε > 0 we have
DΥE + E T ΥT DT ≤ εDDT + ε−1 E T E. (3.16)
Proof. The assumption kΥk ≤ 1 leads to I− Υ ΥT ≥ 0, which again yields the

following inequality:  
I Υ
 ≥0
ΥT I
by applying Schur complements (Boyd et al., 1994). Then for any two real numbers
ε1 and ε2 we will have
  
T
h i I Υ ε1 D
ε1 D −ε2 E    ≥ 0. (3.17)
T T
Υ I −ε2 E
If ε1 ε2 = 1, the inequality (3.17) will be of the following form:
ε21 DDT + ε22 E T E − DΥE − E T ΥT DT ≥ 0,
which is equivalent to the inequality (3.16).

approach 27
Proof of Theorem 3.1. From equations (3.8) and (3.9) and applying Lemma
3.1, we have
P (∆Ai )T + ∆Ai P = P EiT ΥTi (x, t, ξ)DiT + Di Υi (x, t, ξ)Ei P
≤ εi Di DiT + ε−1 T
i P Ei Ei P,
where εi , i = 1, 2, . . . , r, are real positive numbers. Thus it follows that matrix

inequalities (3.13) hold if so do matrix inequalities (3.15). Applying Proposition
3.3.1 we arrive at this theorem. ¤
One cannot directly use matrix inequalities (3.15) to calculate the required
feedback matrix F due to the existence of the term −P BB T P in (3.15), which
makes the solution of (3.15) very complicated. Neither is itself convex, nor can
it be transformed into a convex problem in the space of unknown parameters
(P, F, εi , i = 1, 2, . . . , r). In Cao and Sun (1998), an iterative linear matrix inequal-
ity (ILMI) algorithm was developed to solve the similar problem. Following the
same idea as in Cao and Sun (1998), we can also develop an ILMI algorithm to
solve matrix inequalities (3.15), which is summarized in the sequel.
Algorithm 3.1. (ILMI algorithm for proportional nonlinear controllers).
• Initial data: system’s state space parameters (Ai , Bi , Ci , Di , Ei , i = 1, 2, . . . , r).
• Step 1. Choose P0 > 0. Set j = 1 and X1 = P0 .
• Step 2. Solve the following optimization problem for Pj , F and αj .
OP1: Minimize αj subject to the following LMI constraints

 
T T
Σ1ij Pj Ei Bi F + P j C
 
Ei Pj −εi I 0  < 0 (3.18)
 

 
T T
(Bi F + Pj C ) 0 −I
Pj > 0 (3.19)
i = 1, 2, . . . , r
where Σ1ij = Pj ATi +Ai Pj +εi Di DiT −Xj C T CPj −Pj C T CXj +Xj C T CXj −
αj Pj . Denote by αj∗ the minimized value of αj .
approach 28
• Step 3. If αj∗ ≤ 0, the obtained matrices (Pj , F, εi , i = 1, 2, . . . , r) solve the

problem. Stop. Otherwise go to Step 4.
• Step 4. Solve the following optimization problem for Pj , F and εi , i =

1, 2, . . . , r.
OP2: Minimize tr(Pj ) subject to LMI constraints (3.18) and (3.19) with
αj = αj∗ , where tr stands for the trace of a square matrix. Denote by
Pj∗ the optimal Pj .
• Step 5. If k Xj − Pj∗ k< δ, where δ is a prescribed tolerance, go to Step 6;

otherwise set j := j + 1, Xj = Pj∗ , and go to Step 2.
• Step 6. It cannot be decided by this algorithm whether the problem is solvable

for the system. Stop.
The form of inequality (3.18) is different from the form of inequalities (3.15) in
that there is an additional term −αj Pj in (3.18). The introduction of this term
is to guarantee LMI (3.18) has solutions for a fixed Xj , which is evident for a
sufficiently large positive number αj .
The optimization problem OP1 in Step 2 is a generalized eigenvalue problem
(see pages 10-11 of (Boyd et al., 1994)) in LMI language, which can be solved
by the command gevp in LMI tool box of Matlab (Gahinet et al., 1995); while
the optimization problem OP2 in Step 4 is an eigenvalue problem (see page 10 of
(Boyd et al., 1994)) in LMI language, which can be solved by the command mincx
in LMI tool box of Matlab (Gahinet et al., 1995).
In Theorem 3.1, a common feedback matrix F is to be found for the r matrix
inequalities. This constraint is not necessary and sometimes it is difficult to find
such a common feedback matrix F. For some systems such as the one we will study
later, all the r matrices Bi , i = 1, 2, . . . , r, are identical, i.e., B1 = B2 = · · · =
Br := B. In this case, we can design a controller of the following form
r
X
u(t) = αi (θ)Fi y(t) (3.20)
i=1
to stabilize system (3.10). To this end, we first establish the following.

approach 29
Theorem 3.2. Consider the uncertain nonlinear system (3.10) with B1 = B2 =

· · · = Br := B. If there exist a positive definite matrix P, matrices Fi and positive
numbers εi , i = 1, 2, . . . , r, such that the following matrix inequalities hold:
P ATi + Ai P + εi Di DiT + ε−1 T T

i P Ei Ei P − P C CP
+(BFi + P C T )(BFi + P C T )T < 0,
i = 1, 2, . . . , r,
then system (3.10) can be robustly stabilized by controller (3.20).
The proof of this theorem is similar to that of Theorem 3.1, hence omitted.
A corresponding algorithm can also be developed accordingly to solve feedback
matrices Fi , i = 1, 2, . . . , r. This is only to change LMI (3.18) to
 
T T
Σ1ij Pj Ei BFi + Pj C
 
Ei Pj −εi I 0  < 0, (3.21)
 

 
T T
(BFi + Pj C ) 0 −I
i = 1, 2, . . . , r.
Then Algorithm 3.1 applies verbatim.
Remark 3.3. For the general case that Bi , i = 1, 2, . . . , r, are not identical, it is still
possible in theory to develop a method for the design of a stabilizing controller
of the form (3.20) using the technique developed in reference (Wang et al., 1996).
But the number of the involved matrix inequalities is still considerably large, which
would make it difficult to realize the corresponding algorithm to solve the feedback
matrices with numerical efficiency.
Notice that if we assume that all the uncertainties ∆Ai , i = 1, 2, . . . , r, are

bounded and satisfy matching conditions and that all the state variables are avail-
able (complete state information), a linear robust controller with sufficiently large
“gain” has also been developed in (Zak, 1999) to stabilize the fuzzy model (3.10).
The method in (Zak, 1999) only requires to solve a group of Lyapunov matrix
equation at the expense of the above assumption; while our method requires to
solve a group of ILMI without the assumptions of either matching conditions or
complete state information.
approach 30
3.4 PI Control
We again consider system (3.10), but now we use PI controller (3.11) instead of
proportional controller (3.12). Our objective here is to design the feedback matrices
FP and FI such that system (3.10) is robustly stabilized by the controller.
Let
z1 = x,
Z t
z2 = ydt.
0
Denote z = [z1T , z2T ]T . The variable z can be viewed as the state vector of a new
system, whose dynamics is governed by
r
X
ż1 = ẋ = αi (θ)[(Ai + ∆Ai (z1 , t, ξ))z1 (t) + Bi u(t)],
i=1
ż2 = y = Cz1 .
i.e.,
r
X
ż = αi (θ)[(Āi + ∆Āi (x, t, ξ))z(t) + B̄i u(t)],
i=1
where    
Ai 0 Bi
Āi =  , B̄i =  ,
C 0 0
∆Āi (x, t, ξ)) = D̄i Υ(x, t, ξ)Ēi ,
 
Di h i
D̄i =  , Ēi = Ei 0
0
approach 31
Define
ȳ1 := y = Cz1 = [C 0]z,
Z t
ȳ2 := ydt = z2 = [0 I]z,
0
 
ȳ1
ȳ :=  ,
ȳ2
and denote C̄1 := [C 0], C̄2 := [0 I], C̄ := [C̄1T C̄2T ]T , F̄ := [FP FI ]. Then the
problem of PI controller design for system (3.10) is reduced to that of proportional
controller design for the following system

ż = ri=1 αi (θ)[(Āi + ∆Āi (x, t, ξ))z(t) + B̄i u(t)],
 P



ȳ = C̄z, (3.22)


 u = F̄ ȳ.

Thus the feedback matrices FP and FI can be calculated by applying Algorithm

3.1 to system (3.22).
In the case where all Bi , i = 1, 2, . . . , r, are equal, the following kind of controller
can be also designed by applying Algorithm 3.1 and Theorem 3.2:
r
X
u = αi (θ)F̄i ȳ
i=1
Xr µ Z t ¶
= αi (θ) FP i y + FIi ydt
i=1 0
with F̄i := [FP i FIi ]
3.5 Numerical Example: Excitation Control of

Power Systems
Most existing excitation controllers of power systems are designed by application
of linear control theory to the approximate linearized models of power systems.
Thus they work well only when the disturbance caused by faults in the systems
approach 32
is relatively small. Under large disturbance, the systems might be out of synchro-
nization. In this section, we will apply the results obtained in the previous sections
to the design of excitation controllers for a thermal turbine synchronous generator.
The simplified dynamical model of the single-machine infinite-bus power system
with a silicon-controlled rectifier direct excitor is as follows (Bergen, 1986):
δ̇ = 2πf0 (ω − ω0 ),
D ω0
ω̇ = − (ω − ω0 ) + (Pm − Pe ), (3.23)
H H
0 1
Ėq = (Ef − Eq )
TD0
where
0
xdΣ 0 xd − xd
Eq = 0 Eq − 0 cos δ · Vs ,
xdΣ xdΣ
Vs E q
Pe = sin δ,
xdΣ
xad
Ef = kA (u + h(t)),
Rf
where δ is the angular position of the rotor of generator (G) with respect to a
synchronously rotating reference, which is selected here to be the infinite bus, ω is
the angular velocity of the rotor, Pe and Pm are the active power and mechanical
0
power of G, respectively, Eq is the EMF in the q-axis of G, Eq is transient EMF
in the q-axis of G, Ef the equivalent EMF in the excitation winding of G, xd
0
is the d-axis reactance of G, xd is the d -axis transient reactance of G, xT and
xL are the reactances of the transformer and the transmission line, respectively,
0 0
xdΣ = xd + xT + xL , xdΣ = xd + xT + xL , xad is the mutual reactance between the
excitation winding and the stator windings, Rf is the resistance of the excitation
winding, kA is the gain of the amplifier, Vs is the bus voltage, u is the control input,
and h(t) denotes the external disturbance.
approach 33
Suppose δ0 be the value of δ under steady operating condition. For this δ0 , let
Pm xdΣ
Eq0 = ,
Vs sin δ0
0 1 0 0
Eq0 = [xdΣ Eq0 + (xd − xd )Vs cos δ0 ],
xdΣ
Rf
u0 = Eq0 .
kA xad
Define new state variables and control input as
x1 = δ − δ 0 ,
x2 = ω − ω 0 ,
0 0
x3 = Eq − Eq0 ,
∆u = u − u0 .
The dynamical model of the power system (3.23) can be rewritten as




 ẋ1 = 2πf0 x2 =: f1 (x),

 D
ẋ2 = − H x2




 · 0
¸
(xd −xd )Vs2
 0
ω V


 + H Pm − x0 (x3 + Eq0 ) sin(x1 + δ0 ) + x x0 sin(x1 + δ0 ) cos(x1 + δ0 )
0 s
dΣ dΣ dΣ
=: f2 (x),




 0
0 xd −xd
 xdΣ kA xad
ẋ = − (x + E ) + V cos(x1 + δ0 ) + (u0 + ∆u + h(t))

3 3 TD0 xdΣ s
 0 0

 TD0 xdΣ q0 TD0 Rf


=: f3 (x) + TkD0
A xad


Rf
(u0 + ∆u + h(t)).
(3.24)
The parameters of the system are as follows:
◦
f0 = 50 Hz, ω0 = 1 p.u., δ0 = 60 , H = 8.0 seconds, D = 0.8, Pm = 0.79 p.u.,
0
Vs = 1.0 p.u., xd = 1.5 p.u., xd = 0.3 p.u., xad = 1.3 p.u., xL = (0.8 + ∆xL ) p.u.,
∆xL = 0.0008 p.u., xT = 0.01 p.u., kA = 10, TD0 = 3.0 seconds, Rf = 0.0045 p.u..
0
These parameters yield Eq0 = 1.2723 p.u. and u0 = 7.2942 × 10−4 p.u.. Notice that
all parameters except xL are supposed to be known here. The parameter xL is
supposed to have some perturbation. The reason why we choose the perturbation
in xL to illustrate the robustness of our controller is that xL is one of the system
parameters which are most uncertain and difficult to measure.
approach 34
Generally the state variable δ is difficult to measure. Hence the system output
equation is as follows:
 
0 1 0
y = Cx with C =  .
0 0 1
The physical limit of the excitation voltage leads to the constraint on the control
input ∆u as follows:
kA xad
| u0 + ∆u |≤ 3.0 p.u.,
Rf
or equivalently
−0.0018 p.u. ≤ ∆u ≤ 0.00031 p.u.. (3.25)
The local fuzzy models of the system are obtained through the linearization
of the system (3.24) around the points x1 = [−30, 0, 0]T , x2 = [0, 0, 0]T and
x3 = [+30, 0, 0]T , respectively. In terms of IF-THEN rules, the fuzzy models
admit the following form:
◦
Rule 1. IF x1 (or δ) is small (i.e. x1 is about −30 , or equivalently, δ is about
◦
30 ) THEN
ẋ(t) = (A1 + ∆A1 )x(t) + B1 (u0 + ∆u + h(t)),
◦
Rule 2. IF x1 (or δ) is middle (i.e. x1 is about 0 , or equivalently, δ is about
◦
60 ) THEN
ẋ(t) = (A2 + ∆A2 )x(t) + B2 (u0 + ∆u + h(t)),
◦
Rule 3. IF x1 (or δ) is large (i.e. x1 is about +30 , or equivalently, δ is about
◦
90 ) THEN
ẋ(t) = (A3 + ∆A3 )x(t) + B3 (u0 + ∆u + h(t)),
where ∆Ai ( i = 1, 2, 3 ) account for the parameter perturbation in xL . By using

approach 35
Teixeira-Zak’s formula (Teixeira and Zak, 1999):
Ai = [ai1 ai2 ai3 ]T ,
∂fj

∂fj fj (xi )−(xi )T (xi ) i

∂x
(xi ) + kxi k2
∂x
x if xi 6= 0,
aij =
∂fj

∂x
(xi ) if xi = 0,
for i = 1, 2, 3, j = 1, 2, 3,
matrices Ai and Bi ( i = 1, 2, 3 ) are obtained as follows (noticing that we set

∆xL = 0 while solving Ai and Bi ):
 
0 314.1600 0
 
A1 =  −0.1002 −0.1000 −0.0563 
 
 
−0.2519 0 −0.6937
 
0 314.1600 0
 
A2 =  −0.1009 −0.1000 −0.0975 
 
 
−0.3121 0 −0.6937
 
0 314.1600 0
 
A3 =  −0.0850 −0.1000 −0.1126 
 
 
−0.3441 0 −0.6937
 
0
 
B1 = B2 = B3 =  0 .
 
 
962.9630
There are many options to assign membership functions. For the sake of conve-
nience in computation, we select triangular function as our membership functions,
approach 36
α1
1.4
α2
α
3
1.2
membership functions 1
0.8
0.6
0.4
0.2
−0.2
0 20 40 60 80 100 120
δ in degrees
Figure 3.1. Membership functions of the fuzzy models for power system (3.24)
which are:



 1 if δ ≤ 0.5236

α1 = 1.0472−δ
1.0472−0.5236
if 0.5236 < δ < 1.0472


 0

if δ ≥ 1.0472



 0 if δ ≤ 1.0472

α3 = δ−1.0472
1.5708−1.0472
if 1.0472 < δ < 1.5708


 1

if δ ≥ 1.5708
α2 = 1 − α 1 − α 3 ,
as depicted in Fig. 3.1.

Since the right hand side of the first equation of the power system (3.24) includes
no uncertainties and it is also linear function of the state variables, we let the first
row of ∆Ai ( i = 1, 2, 3) be zero. To account for the parameter perturbation and
approximation error caused by linearization, we let the second and third rows be
approach 37
0.1% times the corresponding elements of Ai ( i = 1, 2, 3), i.e.

 
0 0 0
 
−3
∆A1 =  −0.1002 −0.1000 −0.0563  × 10
 
 
−0.2519 0 −0.6937
 
0 0 0
 
−3
∆A2 =  −0.1009 −0.1000 −0.0975  × 10
 
 
−0.3121 0 −0.6937
 
0 0 0
 
−3
∆A3 =  −0.0850 −0.1000 −0.1126  × 10 .
 
 
−0.3441 0 −0.6937
For ∆Ai to be in the form (3.8), we can let
DAi = ∆Ai , EAi = I.
Since B1 = B2 = B3 , we can use Theorem 3.2 combined with Algorithm 3.1

to calculate the feedback matrices of proportional and PI controllers, which are as
follows. For proportional controller
h i
F1 = 0.0398 −0.0071
h i
F2 = 0.0353 −0.0065
h i
F3 = 0.0212 −0.0027
u = α1 F1 y + α2 F2 y + α3 F3 y. (3.26)
For PI controller:
h i h i
FP 1 = 0.0163 −0.0021 , FI1 = −0.0037 −0.0125
h i h i
FP 2 = 0.0171 −0.0028 , FI2 = −0.0028 −0.0125
h i h i
FP 3 = 0.0241 −0.0033 , FI3 = +0.0050 −0.0125
Z t Z t
u = α1 (FP 1 y + FI1 ydt) + α2 (FP 2 y + FI2 ydt)
0 0
Z t
+α3 (FP 3 y + FI3 ydt). (3.27)
0
approach 38
The typical fault chosen for simulation is a balanced three phase shorted to
ground. The fault happens at the side of the transformer connected to high voltage
from the time t = 0 to t = tF seconds. At t = tF the fault is cleaned. The simulation
results are shown in Fig. 3.2 and Fig. 3.3 respectively, where tF = 0.13 seconds. As
can be seen, both controllers (3.26) and (3.27) can stabilize the system successfully.
To compare the results obtained by fuzzy nonlinear control, the response of the
system (3.24) under the controller designed on the basis of the linearization model
around the steady-state operating point is also investigated. Here we only study
PI controller. The linear PI controller is as follows
Z t
∆u = FP 2 y + FI2 ydt. (3.28)
0
Simulation results are depicted in Fig. 3.4 and Fig. 3.6 (for nonlinear fuzzy PI
controller) and Fig. 3.5 and Fig. 3.7 (for linear PI controller) respectively. Fig.
3.4 and Fig. 3.5 are for the case of small fault (tF = 0.08 seconds), while Fig.
3.6 and Fig. 3.7 are for the case of large fault (tF = 0.163 seconds). Notice that
in order to show the difference in the performances between the controllers (3.27)
and (3.28), it is necessary to remove the constraint (3.25) on the controllers in the
case of large faults, otherwise the controllers will lose their regulation rule since
both the controller outputs will take the same saturation value under large faults.
It is observed from Fig. 3.4 and Fig. 3.5 that the responses of the closed loop
system under both controllers show little difference in the case of small faults.
Fig. 3.6 shows that controller (3.27) can stabilize the system under large fault
(tF = 0.163 seconds), while Fig. 3.7 illustrates that controller (3.28) fails to do
so. This phenomenon can be explained clearly from the approximation errors of
linearization model and fuzzy nonlinear model, which is shown in Fig. 3.8, where
∆f2 = |f2 (x)− fˆ2 (x)| and fˆ2 (x) is the approximation value of f2 (x) by linearization
approach and fuzzy approach, respectively. It follows from Fig. 3.8 that when ∆δ
◦
varies within 20 , the errors in both approaches are very small and are almost
◦
equal, while when ∆δ > 20 the error in the linearization approach is much larger
than that in the fuzzy approach.
It is pointed out that α1 , α2 and α3 in controllers (3.26) and (3.27) are the
approach 39
∆δ in degrees 50
−50
0 5 10 Time 15
0.02
∆ω in p.u.
−0.02
0 5 10 Time 15
0.2
∆ Eq in p.u.
0
′
−0.2
0 −4 5 10 Time 15
x 10
4
2
∆ u in p.u.
0
−2
−4
0 5 10 Time 15
Figure 3.2. Response of system (3.24) with fuzzy proportional controller (3.26)
(tF = 0.13 seconds)
approach 40
∆δ in degrees 50
−50
0 5 10 Time 15
0.02
∆ω in p.u.
−0.02
0 5 10 Time 15
0.2
∆ Eq in p.u.
0
′
−0.2
0 −4 5 10 Time 15
x 10
4
2
∆ u in p.u.
0
−2
−4
0 5 10 Time 15
Figure 3.3. Response of system (3.24) with fuzzy PI controller (3.27) (tF = 0.13
seconds)
approach 41
∆δ in degrees 20
−20
0 5 10 Time 15
0.01
∆ω in p.u.
−0.01
0 5 10 Time 15
0.1
∆ Eq in p.u.
0
′
−0.1
0 −4 5 10 Time 15
x 10
4
2
∆ u in p.u.
0
−2
−4
0 5 10 Time 15
seconds)
approach 42
∆δ in degrees 20
−20
0 5 10 Time 15
0.01
∆ω in p.u.
−0.01
0 5 10 Time 15
0.1
∆ Eq in p.u.
0
′
−0.1
0 −4 5 10 Time 15
x 10
4
2
∆ u in p.u.
0
−2
−4
0 5 10 Time 15
Figure 3.5. Response of system (3.24) with linear PI controller (3.28) (t F = 0.08
seconds)
approach 43
100
∆δ in degrees
50
0
−50
0 5 10 Time 15
0.02
∆ω in p.u.
−0.02
0 5 10 Time 15
0.2
∆ Eq in p.u.
0
′
−0.2
0 −4 5 10 Time 15
x 10
10
∆ u in p.u.
−5
0 5 10 Time 15
seconds)
approach 44
4
x 10
15
∆δ in degrees
10
0
0 5 10 Time 15
0.8
∆ω in p.u.
0.6
0.4
0.2
0
−0.2
0 5 10 Time 15
0.2
∆ Eq in p.u.
0
′
−0.2
0 −4 5 10 Time 15
x 10
10
∆ u in p.u.
0
0 5 10 Time 15
Figure 3.7. Response of system (3.24) with linear PI controller (3.28) (t F = 0.163
seconds)
approach 45
fuzzy method
0.06 linearization method
0.05
0.04
∆ f2
0.03
0.02
0.01
0
0 20 40 60 80 100 120
δ in degrees
Figure 3.8. Approximation errors of linear and fuzzy nonlinear models for power
system (3.24)
functions of power angle δ, which would make it difficult to realize the controllers.
However, the response of the closed loop system is not sensitive to the form of
the functions α1 , α2 and α3 . Therefore some kind of rough estimates of δ can be
used to realize the controllers. If this approach fails, one can apply Theorem 3.1
to design the corresponding proportional or PI controllers, in which only output
variables are used.
3.6 Conclusion
The design problem of proportional and PI controllers for nonlinear systems has
been studied in this chapter. First, Takagi-Sugeno fuzzy model with parameter
uncertainties is used to model the nonlinear systems. The introduction of parame-
ter uncertainties in the T-S fuzzy model can account for both approximation error
of fuzzy model and the actual parameter uncertainties in the practical process.
Secondly an algorithm based on iterative linear matrix inequality is developed to
design a proportional controller to robustly stabilize the system which can be ex-
pressed by a Takagi-Sugeno fuzzy model. Thirdly, we transform the problem of
approach 46
PI controller design to that of proportional controller design for an augmented

system and thus bring the solution of the former problem into the configuration of
the developed algorithm. The characteristics of the developed method is that it is
numerically tractable and can be applied to general multi-variable nonlinear sys-
tems. Finally, the proposed method is applied to the robust stabilizing controller
design for the excitation control of power systems. Simulation results show that
the transient stability can be improved by using a fuzzy PI controller when large
faults appear in the system, compared to the conventional PI controller designed
by using linearization method around the steady state.
Notice that the controllers proposed in this chapter use only output variables
(in the case of common feedback matrix corresponding to all the fuzzy rules) or
output variables plus some other state variables needed in the fuzzy premise (in the
case of different feedback matrix corresponding to each fuzzy rule accordingly). In
the latter case, the state variables needed in the fuzzy premise should be available
normally.
Chapter 4
Fuzzy Modelling and Control of

F-16 Aircraft
4.1 Introduction
F-16 aircraft is designed for all electric control in 1970s. The classical flight con-
trol technology can be used via local linearization method. However, the aircraft
model is inherently nonlinear, exact input-output linearization is often used to
control specific output variable sets in nonlinear flight control problems (Lane and
Stengel, 1988; Krstić and Tsiotras, 1999). Feedback linearization is a theoretically
established and widely used method in controlling nonlinear systems. Unfortu-
nately, this direct application of feedback linearization requires the second or third
derivatives of uncertain aerodynamic coefficients and does not guarantee internal
stability for non-minimum phase systems.
Another approach to designing flight control laws with feedback linearization
is to separate the flight dynamics into fast and slow dynamics by using timescale
properties (Menon et al., 1987; Snell et al., 1992). This method allows the con-
troller design process to be performed without state transformation. Because each
separated subsystem is square, the number of control inputs is equal to the number
of states. The design process of this method can be divided into two steps. In the
outer loop, the controller for the slow states α, β, and φ is designed to facilitate
47
Chapter 4. Fuzzy Modelling and Control of F-16 Aircraft 48
tracking of the given commands by assuming that the fast states p, q, and r are
control inputs, which achieve their commanded values instantaneously. With the
slow states controller designed in the outer loop, a separated inner-loop controller
is designed to make the fast states p, q, and r follow the outer loop’s control in-
put trajectories using the real control inputs: aileron, rudder, and elevator. This
method can be justified only if there is sufficient timescale separation between the
inner- and outer-loop dynamics because the fast states p, q, and r are used as
control inputs in the outer-loop system. Hence p, q, and r in the inner loop should
be much faster than the states α, β, and φ in the outer loop. The stability of
this timescale separation approach may be analyzed by the singular perturbation
theory. However, in most nonlinear flight control research, the gain of the inner-
loop controller is set much larger than that of the outer-loop controller, and it is
assumed that the aircraft dynamics satisfies this property. This, therefore, does
not guarantee closed-loop stability.
Schumacher and Khargonekar (1998) analyzed theoretically the stability of the
flight control system with the two-timescale separation assumption. Using the Lya-
punov theory, they determined the minimal inner-loop gain guaranteeing closed-
loop stability. However, this approach is so complicated and conservative that the
calculated value of minimal inner-loop gain is too large to be applied in the flight
controller. It may excite un-modelled dynamics or saturate the control inputs and,
therefore, cause robustness problems. Another difficulty related to the application
of feedback linearization to a flight control system is that a complete and accu-
rate aircraft dynamic model including aerodynamic coefficients is required. It is
difficult to identify accurately aerodynamic coefficients because they are nonlinear
functions of several physical variables.
An inverse optimal design is developed by Sepulchre et al. (1997) to derive the
optimal stabilizing controller. This approach was first proposed by Kalman (1964)
to establish the gain and phase margins of linear quadratic regulators and was re-
cently revised by Freeman and Kokotović (1996b) to develop a design methodology
of robust nonlinear controllers.
The control problem of a F-16 aircraft model has been addressed by many
researchers for the purpose of the control of spacecraft and aircraft (Debs and
Athans, 1969; Vadali et al., 1984) . Also, there have been several works that con-
sider performance indices such as time and/or fuel in the formulation of the optimal
control problems (Athans et al., 1963; Scrivener and Thomson, 1994) . These stud-
ies have mainly addressed the regulation problem for the angular velocity subsys-
tem and for some quadratic costs (Athans et al., 1963; Windeknecht, 1963; Tsiotras
et al., 1996) . Recently, the attitude control problem of the complete system that
includes the dynamics as well as the kinematics has been investigated by many
researchers. Carrington and Junkins (1986) has used a polynomial expansion ap-
proach to approximate the solution to the HJB equation. Rotea et al. (1998) has
shown that, for some special cases of performance outputs, Lyapunov functions
that include a logarithmic term in the kinematic parameters result in linear con-
trollers with a finite quadratic cost. For the general quadratic cost, they have also
presented sufficient conditions that guarantee the existence of a linear, suboptimal,
stabilizing controller. Tsiotras (1996) has derived a new class of globally asymp-
totically stabilizing feedback control laws for the complete attitude motion of a
nonsymmetric rigid body and has also presented a family of exponentially stabi-
lizing optimal control laws for the complete system. Tsiotras (1997) , by using
the natural decomposition of the complete system into its kinematics and dynam-
ics subsystems and the inherent passivity properties of these two subsystems, has
presented a partial solution to the optimal regulation of the symmetry axis of a
spinning rigid body. Bharadwaj et al. (1998) have derived a couple of new globally
stabilizing attitude control laws using the inverse optimal approach of Freeman and
Kokotović (1996b) where minimal, exponential coordinates are used to represent
the kinematic equations. To design a stabilizing control law for F-16 aircraft, the
method of backstepping (Krstić et al., 1995) can be considered, which was used
by Sontag and Sussmann (1988) for the first time to design feedback contros laws
for an underachieved rigid body. Tsiotras and Longuski (1994) have employed this
method for the attitude stabilization of an axis symmetric spacecraft with two
control torques. In the their design, a control Lyapunov function along with a
stabilizing controller is derived and thus the stabilization of closed-loop system is
guaranteed.
In this chapter, we consider the complete attitude motion of a rigid spacecraft
(Shuster, 1963) . Moreover, both stabilizing and attitude tracking controller is
needed to be developed based on nonlinear F-16 model.
Since Takagi and Sugeno (1985) opened a new direction of research in the area
of control by introducing the Takagi-Sugeno (TS) fuzzy model, there have been
several studies concerning the systematic design of stabilizing fuzzy controllers
(Tanaka and Sugeno, 1992; Tanaka et al., 1998) . In this chapter, we propose a
design procedure yielding the stabilizing controller for the nonlinear system de-
scribed by a TS fuzzy model. In the TS fuzzy model (Takagi and Sugeno, 1985)
, the overall system is described by several fuzzy IF-THEN rules, each of which
represents a local linear state equation ẋ = Ai x + Bi u. To derive the stabilizing
controller, we employ the Lyapunov stabilizing theory and LMIs method. The
resulting controllers are time-invariant state feedback controllers. Also, we show
that the parameters of the stabilizing controller can be found by solving an linear
matrix inequality (LMI) problem. The LMI formulation of the controller synthesis
problems is of great practical value because it can be solved by using reliable and
efficient convex optimization techniques (Boyd et al., 1994) , for example, the LMI
Control Toolbox of MATLAB (Gahinet et al., 1995).
Our proposed method is based on the design of the control law for the TS fuzzy
model. To the authors’best knowledge, the proposed approach is the first attempt
to design both stabilizing and tracking control with application to the complete
attitude motion of a rigid spacecraft. A minor disadvantage of the proposed method
is that it needs a computation procedure based on an LMI solver.
This chapter is organized as follows. The F-16 aircraft model is introduced in
Section 4.2 . In Section 4.3, the modelling of TS fuzzy system of F-16 aircraft
is described. Section 4.4 studies the LQR based controller design. Finally, the
discussion and conclusions are given in Section 4.5.
4.2 F-16 Aircraft Model
4.2.1 Need for Modelling
Model building is a fundamental process. An aircraft designer has a mental model

of the type of aircraft that is needed, uses physical models to gather wind-tunnel
data, and designs with mathematical models that incorporate the experimental
data (Taeyoung and Youdan, 2001) . The modelling process is often iterative; a
mathematical model based on the laws of physics will suggest what experimental
data should be taken, and the model may then undergo considerable adjustment in
order to fit the data. In building the mathematical model we recognize the onset
of the law of diminishing returns and build a model that is good enough for our
purposes but has known limitations. Some of the limitations of the models involve
uncertainty in the values of their parameters. Later, we attempt to characterize
this uncertainty mathematically and allow for it in control system design.
Because of the high cost of building and flight testing a real aircraft, the impor-
tance of aircraft mathematical models goes far beyond control system design. The
mathematical model is used, in conjunction with computer simulation, to evalu-
ate the performance of the prototype aircraft and hence improve the design. It
can also be used to drive training simulators, to reconstruct the flight conditions
involved in accidents, and to study the effects of modifications to the design. Fur-
thermore, other mathematical models are used in all aspects of the aircraft design
(e.g., structural models for studying stress distribution and predicting fatigue life).
In the aerospace industry it is necessary for a wide range of specialists to work
together, thus flight control engineers must be able to work with the aerodynam-
icists as well as with structural and propulsion engineers. Each have some un-
derstanding of the terms and mathematical models used by the other. This is
becoming increasingly important as designers seek to widen aircraft performance
envelopes by integrating the many parts of the whole design process. Furthermore,
at the prototype stage the controls designer must work closely with the test pilots
to make the final adjustments to the control systems. This may take many hours
of simulator time and flight test, and the flight controls engineer involved in the
final stages of the design is also likely to be a qualified pilot.
4.2.2 Modelling Method
The aerodynamic forces and moments on an aircraft are produced by the rela-
tive motion with respect to the air and depend on the orientation of the aircraft
with respect to the airflow. In a uniform airflow these forces and moments are
unchanged after a rotation around the free-stream velocity vector. Therefore, only
two orientation angles (with respect to the relative wind) are needed to specify the
aerodynamic forces and moments. The angles that are used are the angle of attack
(alpha) and the sideslip angle (beta). They are known as the aerodynamic angles
and will now be defined by means of coordinate rotations in three dimensions. Note
that the aerodynamic force and moments are also dependent on angular rates, but
for the moment we are concerned only with their dependence on orientation.
The angle of attack usually specified in the aerodynamic data for an aircraft is
measured with respect to a fuselage reference line and denoted by αf lr . We shall
assume that our aircraft body-fixed axes are aligned with this fuselage reference
line; otherwise, the alpha used in the body-axes equations of motion would differ
by a constant from the alpha used in the aerodynamic data base.
Fig. 4.1 shows an aircraft with the relative wind on its side (i.e., sideslipping)
and with the conventional right-handed (forward, starboard, and down) set of
body-fixed axes illustrated. The relative wind vector is equal and opposite to
the cg relative velocity vector vB used in the equations of motion. The angles
of attack and sideslip are defined by performing a plane rotation about the body
y-axis, followed by a plane rotation about the new z-axis, such that the final x-axis
is aligned directly into the relative wind. The first rotation defines the stability
axes, and the angle of attack is the angle between the body-fixed x-axis and the
stability x-axis. Alpha is positive if the rotation about the body-fixed y-axis was
negative; thus a positive alpha is shown in the figure. The second rotation leads to
a set of wind axes, and the sideslip angle is the angle between the stability x-axis
Figure 4.1. Definition of aircraft axes and angles.
and the wind x-axis. Beta is positive if the rotation about the stability z-axis was
positive, and a positive beta is shown in the figure.
An aircraft of conventional shape must fly more or less directly into the appar-
ent wind in order to have low drag; therefore, beta is usually very small in steady
fight. Alpha must be large enough to generate the required lift but is also usually
quite small. Therefore, although the stability axes and wind axes have a variable
orientation depending on the flight condition, they essentially point forward, star-
board, and down, the same as the body-fixed axes. Note that technically all three
sets of axes are ”body axes” but only one is body-fixed; we shall drop the term
”fixed” and simply refer to them as body, stability,and wind axes.
A left-handed wind-axes system, aligned backwards, and ”up” relative to the
aircraft, has often been used in the past for wind tunnel data. Lift L, drag D, and
cross-wind force C were defined naturally in these axes as the aerodynamic force
components along the respective positive axes. In the right-handed wind axes that
we have defined, we use −L and −D for the z and x force components, and define
a component Y for the aerodynamic side-force measured along the positive y-axis.
The symbol Y is also commonly used for the aerodynamic side-force component
along the body y-axis, but it will be clear from the context which axes are being
used.
4.2.3 Workable Model
The body-fixed axes, nonlinear equations of motion for an aircraft over a flat Earth
are given by
cos α cos β sin β sin α cos β

V̇ = [T + Fx ] + [Fy ] + [Fz ]
m m m
+g[− cos α cos β sin θ + sin β sin φ cos θ + sin α cos β cos θ], (4.1a)
sin α
α̇ = −p cos α tan β + q − r sin α tan β − [T + Fx ]
mV cos β
cos α g
+ Fz + [sin α sin θ + cos α cos φ cos θ], (4.1b)
mV cos β V cos β
cos α sin β cos β sin α cos β
β̇ = p sin α − r cos α − [T + Fx ] + [Fy ] − [Fz ]
mV mV mV
g
+ [cos α sin β sin θ + cos β cos θ sin φ − sin α sin β cos φ cos θ], (4.1c)
V
ṗ = I2 pq + I1 qr + I3 L + I4 N, (4.2a)
q̇ = I5 pr − I6 (p2 − r2 ) + I7 M, (4.2b)
ṙ = −I2 qr + I8 pq + I4 L + I9 N, (4.2c)
φ̇ = p + tan θ(q sin φ + r cos φ), (4.3a)
θ̇ = q cos φ − r sin φ, (4.3b)

q sin φ + r cos φ
ψ̇ = , (4.3c)
cos θ
where the moments of inertia Ii , i = 1, 2, ..., 9, are defined as follows:

2
Iz (Iz − Iy ) + Ixz Ixz (Ix − Iy + Iz ) Iz
I1 = 2
, I2 = 2
, I3 = 2
,
Ix Iz − Ixz Ix Iz − Ixz Ix Iz − Ixz
Ixz Iz − I x Ixz 1
I4 = 2
, I5 = , I6 = , I7 = ,
Ix Iz − Ixz Iy Iy Iy
2
Ix (Ix − Iy ) + Ixz Ix
I8 = 2
, I 9 = 2
.
Ix Iz − Ixz Ix Iz − Ixz
It is assumed that the aerodynamic forces and moments are expressed as functions
of angle of attack, sideslip angle, angular rates, and control surface deflection. For
example, Fx , L and the other symbols are expressed as follows:
Fx = [Cx (α, δe ) + (c̄q/2V )Cxq (α)]q̄S, (4.4a)
Fy = [Cy (β, δα , δr ) + (bp/2V )Cyp (α) + (br/2V )Cyr ]q̄S, , (4.4b)
Fz = [Cz (α, β, δe ) + (c̄q/2V )Czq (α)]q̄S, (4.4c)
L = [Cl (α, β) + Clδα (α, β)δα + Clδr (α, β)δr + (p/2V )Clp (α) + (r/2V )Clr (α)]q̄Sb, (4.5a)
M = [Cm (α, δe ) + (q/2V )Cmq (α)]q̄Sc̄ (4.5b)
N = [Cn (α, β) + Cnδα (α, β)δα + Cnδr (α, β)δr + (p/2V )Cnp (α) + (r/2V )Cnr (α)]q̄Sb, (4.5c)
For the F-16 aircraft, the force coefficients Cx , Cxq , Cy , Cyp , Cyr , Cz , Czq and mo-
ment coefficients Cl , Clδα , Clδr , Clp , Clr , Cm , Cmq , Cn , Cnδα , Cnδr , Cnp , Cnr and other
necessary parameters can be obtained from Brian and Frank (1992) . The control
input are thrust T , elevator angle (el) δa , aileron angle (ail) δa and rudder angle
(rdr) δr . The nine states are speed V , angle of attack (alpha) α, sideslip angle
(beta) β, roll rate p, pitch rate q, yawing rate r, roll angle (phi) φ, pitch angle
(theta) θ and yaw angle (psi) ψ.
The complete F-16 aircraft nonlinear model with all the data is given in Ap-
pendix A for easy reference.
4.3 TS Fuzzy Modelling

From the development of F-16 aircraft nonlinear model in the last section, we
observe that the nonlinear model is rather complex and coefficients can be only
obtained from the look-up tables. Thus, the control design for such model is quite
difficult. We need a simpler model which is easier to be used for control design.
This section reviews the Takagi and Sugeno Fuzzy modelling approach and develop
TS model of F-16 nonlinear model.
4.3.1 The Technique
The fuzzy model proposed by Takagi and Sugeno consists of several fuzzy IF-THEN
rules, each of which represents the local linear state equation of a nonlinear system.
In this chapter, we consider the fuzzy rules described as follows.
Rule i:
IF x1 (t) is Mi1 and · · · and xn (t) is Min ,
THEN
ẋ(t) = Ai x(t) + Bi u(t), i = 1, 2, ..., N, (4.6)
Here, xi (t), i = 1, 2, ..., n, and Mij , i = 1, 2, ..., N , j = 1, 2, ..., n, are state variables
and fuzzy sets, respectively, and N is the number of IF-THEN rules; u ∈ Rm is the
input vector and Ai ∈ Rn×n and Bi ∈ Rn×m , i = 1, 2, ..., N . Following the usual
inference method of the TS fuzzy model, the state equation at time t is represented
in the form of weighted average along the trajectory x(t) ∈ Rn :
N
X N
.X
ẋ(t) = wi (x(t)){Ai x(t) + Bi u(t)} wi (x(t)), (4.7)
i=1 i=1
In (4.7), the weight functions are defined as

n
Y
wi (x(t)) = Mij (xj (t)),
j=1
where Mij (xj (t)) is the grade of xj (t) in the fuzzy set Mij . The weight functions
wi , which are nonnegative and measurable, usually satisfy
N
X
wi (x(t)) > 0, ∀ t > 0. (4.8)
i=1
Throughout this chapter, it is assumed that Eq.(4.8) always holds and that the
vector x(t) can be measured in real time. With the normalization of weight func-
tions
N
.X
hi (x(t)) = wi (x(t)) wi (x(t)), i = 1, 2, ..., N, (4.9)
i=1
the state equation (4.7) can be written in the polytopic form

N
X
ẋ(t) = hi (x(t))(Ai x(t) + Bi u(t)), (4.10)
i=1
where the normalized weights hi satisfy hi (x(t)) ≥ 0, i = 1, 2, ..., N , and

N
X
hi (x(t)) = 1, ∀ t ≥ 0.
i=1
Suppose that the truth model of a plant has the form
ẋ = F (x, u). (4.11)
Expanding F by means of a Taylor series around (x0 , u0 ) yields
∂F ∂F
ẋ = F (x0 , u0 ) + |x=x0 ,u=u0 (x − x0 ) + |x=x0 ,u=u0 (u − u0 ) (4.12)
∂x ∂u
The point (x0 , u0 ) is called an equilibrium point of (4.11) if at (x0 , u0 ) we have

ẋ = F (x0 , u0 ) = 0. Let δx = x − x0 and δu = u − u0 , the linearized model about
the equilibrium (x0 , u0 ) is obtained by neglecting higher order terms. Observing
that for F (x0 , u0 ) = 0, the linearized model has the form
d
δx = Aδx + Bδu, (4.13)
dt
∂F ∂F
where A = |
∂x x=x0 ,u=u0
and B = |
∂u x=x0 ,u=u0
. We start with generating linear local
models describing plant’s behavior about the equilibrium state (x0 , u0 ) = (0, 0), we
obtain the first linear model by using the above described linearization technique.
The resulting model is given by ẋ = A1 x + B1 u. Next, we need to construct
local linear models describing plant’s behavior at the remaining operating points.
Suppose that x = xj is the next operating state of interest. The result of Talor’s
linearization of a nonlinear model about an operating nonequilibrium point is an
affine rather than linear model. Even if the operating point is an equilibrium point,
Taylor series linearization, in general, will not yield a local model that is linear in
x and u. Indeed, suppose that the operating point (xj , uj ) is an equilibrium point,
the resulting linearized model is
d
(x − xj ) = Aj (x − xj ) + Bj (u − uj ). (4.14)
dt
We can represent model (4.14) in the form
ẋ = Aj x + Bj u − (Aj xj + Bj uj ). (4.15)
The term (Aj xj + Bj uj ) does not have to be equal to zero and, hence, model
(4.15) is not a linear model, but rather an affine model. However, we note that
Taylor’s linearization will yield a linear system in x and u if the equilibrium point
(x0 , u0 ) = (0, 0). Suppose that we are given an operating state x0 that does not
have to be an equilibrium state. Our goal is to construct a linear model in x and u
that approximates the behavior of (4.11) in the vicinity of the operating state x 0 ;
that is, we wish to find constant matrices A and B such that in a neighborhood of
x0
F (x, u) ≈ Ax + Bu (4.16)
and
F (x0 , u0 ) = Ax0 + Bu0 . (4.17)
Let A = [a1 a2 · · · an ] and B = [b1 b2 · · · bn ]. We wish to have
Fi (x0 , u0 ) = aTi x0 + bTi u0 (4.18)
and
∂Fi
bi = |x=x0 ,u=u0 . (4.19)
∂u
where Fi is the ith row of F . Moreover, we also wish that
∂Fi
|x=x0 ,u=u0 ≈ ai . (4.20)
∂x
Using the method in Marcelo and Stanislaw (1999)], we obtain that
∂Fi Fi (x0 , u0 ) − xT0 ∂F i

|
∂x x=x0 ,u=u0
ai = |x=x0 ,u=u0 + x0 , x0 6= 0. (4.21)
∂x kx0 k2
∂Fi
ai = |x=x0 ,u=u0 , x0 = 0.
∂x
4.3.2 TS Model of F-16
We now apply the proposed fuzzy modelling method to the F-16 aircraft modelling,
where the F-16 nonlinear model is described by (??) - (4.3c). The model can be
written as
ẋ = F (x, u),
y = Cx, (4.22)
where x = [V, α, β, p, q, r, φ, θ, ψ]0 and u = [T, δe , δa , δr ]0 . In the following, we will

present the details of constructing a F-16 TS model: 1) Selection of equilibrium
state of the nonlinear model. 2) Selection of the operating points around the
equilibrium state. 3) Computation of the linear models at the chosen operating
points. 4) Determination of the weighting function of the TS model.
It has been shown (Brian and Frank, 1992) that one of the equilibrium states
of the F-16 aircraft is
[x0 , u0 ] = [502 0.03936 0 0 0 0 0 0.03936 0.1485, 9.6427 − 1.931 0.0 0.0]0
For the convenience of analysis and simulation, we move this point to the original
point x0 = 0. Substituting [x, u] = [x̂, û] + [x0 , u0 ] into (4.22), we have a new
nonlinear model with the equilibrium point [x0 , u0 ] = [0, 0].
From our extensive simulation, we choose nine operating points to build the TS
fuzzy model. If more points are chosen, the error of the model does not decrease
rapidly and less points result in generally bad approximation. For the states of
the nonlinear F-16 system near the equilibrium point with α ∈ [−0.2, 0.2] and
φ ∈ [−0.2, 0.2], the TS model constructed by the proposed method is described by
9
X
ẋ = hi (α, φ)(Ai x + Bi u), (4.23)
i=1
where hi (α, φ) = Mαi Mφi . Mαi and Mφi are the fuzzy membership functions of α
and φ respectively. In our simulation, the nine operating points evenly distributed
in the state space. That is, we choose
(α1 , φ1 ) = (−0.2, −0.2),
(α2 , φ2 ) = (−0.2, 0),
(α3 , φ3 ) = (−0.2, 0.2),
(α4 , φ4 ) = (0, −0.2),
(α5 , φ5 ) = (0, 0),
(α6 , φ6 ) = (0, 0.2),
(α7 , φ7 ) = (0.2, −0.2),
(α8 , φ8 ) = (0.2, 0),
(α9 , φ9 ) = (0.2, 0.2)
as the nine operating points.

At each operating point, we can obtain the linear model at these operating
points from (4.19) and (4.21). For example, at the operating point (α, φ) =
(−0.2, −0.2), from (4.21), we get
 
-3.0489 0.0003 -0.9873 0 0 -3.7977 0 0
 
0 -1.3815 -0.0029 0.0009 -1.1356 0 -0.0335 0
 
 
 
0.0312 0.0025 -0.5181 0 0 9.7538 0 0
 
 
 
 0 2.5473 0 -0.0832 191.1245 -6.3862 -80.3479 -31.5328 
A1 =  ,
 

 0 0.9639 0 0.0003 -1.4637 0 0.1846 -0.0127 

 

 -0.1609 0 -0.9798 0.0000 0.0004 -0.2779 0.0632 0.0005 
 
 1.0000 -0.0078 0.0386 0 0 0 0 0 
 
0 0.9801 0.1987 0 0 0 0 0
and from (4.19), we obtain

 
0 0 -0.5983 0.1588
 
0 -0.1925 0 0
 
 
 
0 0 -0.0337 -0.0675 
 

 
 0 0.4584 0 0 
B1 =  .
 

 0 -0.0020 0 0 

 

 0 0 0.0003 0.0008 
 
 0 0 0 0 
 
0 0 0 0
Similarly, we can obtain the linear models (Ai , Bi ), i = 2, · · · , 9, at each operating

point. The details are given in Appendix B.
For weightings, we select triangle function as the fuzzy membership functions,
which are 


 1 if α ≤ −0.2

M α1 = 2.5 ∗ α − 0.5 if − 0.2 < α < 0.2


0 if α ≥ 0.2





 0 if α ≤ −0.2


 5∗α+1

if − 0.2 < α ≤ 0
M α2 =


 −5 ∗ α + 1 if 0 < α < 0.2


0 if α ≥ 0.2





 0 if α ≤ −0.2

M α3 = 2.5 ∗ α + 0.5 if − 0.2 < α < 0.2


1 if α ≥ 0.2





 1 if φ ≤ −0.2

Mφ 1 = 2.5 ∗ φ − 0.5 if − 0.2 < φ < 0.2


0 if φ ≥ 0.2





 0 if φ ≤ −0.2


 5∗φ+1

if − 0.2 < φ ≤ 0
Mφ 2 =


 −5 ∗ φ + 1 if 0 < φ < 0.2


0 if φ ≥ 0.2


0.9
0.8
0.7
membership function
0.6
0.5
0.4
0.3
0.2
0.1
0
−0.4 −0.3 −0.2 −0.1 0 0.1 0.2 0.3
alpha(theta)
Figure 4.2. Fuzzy triangle membership functions



 0 if φ ≤ −0.2

Mφ 3 = 2.5 ∗ φ + 0.5 if − 0.2 < φ < 0.2


1 if φ ≥ 0.2


The function is depicted in Fig. 4.2.

The complete model with all the data is given in Appendix B.
4.3.3 Model Validation
To show the effectiveness of the obtained TS fuzzy model to approximate the

original nonlinear model, we compare the error between the TS fuzzy model and
linear model. The results are shown in Fig. 4.3 (with different α) and Fig.
4.4(with different φ), where δFT S = kF (x, u) − F̂T s k2 and δFL = kF (x, u) −
P9
F̂L k2 . In Fig. 4.3, we obtain F̂T S = i hi (α, φ0 )(Ai x + Bi u) from TS fuzzy
model, F̂L = A5 x + B5 u is from the linear model at equilibrium point, where

x = [p0 , q0 , r0 , V0 , α, β0 , φ0 , θ0 , ψ0 ]0 and u = [T0 , δe0 , δa0 , δr0 ]0 , α ∈ [−0.2, 0.2]. In
P9
Fig. 4.4, we have F̂T S = i hi (α0 , φ)(Ai x + Bi u), F̂L = A5 x + B5 u, where
x = [p0 , q0 , r0 , V0 , α0 , β0 , φ, θ0 , ψ0 ]0 and u = [T0 , δe0 , δa0 ,δr0 ]0 , φ ∈ [−0.2, 0.2]. It

is shown that the TS model has smaller error than the linear model around the
equilibrium point. At the interval α ∈ [−0.2, 0.2] , the maximal model error is
δFTmax
S = 2.0214 for TS fuzzy model and the maximal error is δFLmax = 20.6833
for linear model in Fig. 4.3. At the interval φ ∈ [−0.2, 0.2], the maximal model
error is δFTmax
S = 1.0209e − 004 for TS fuzzy model and the maximal error is
δFLmax = 0.0252 for linear model in Fig. 4.4.
140
120
100
80
error
60
40
20
0
−0.5 −0.4 −0.3 −0.2 −0.1 0 0.1 0.2 0.3 0.4 0.5
alpha(rad)
( —– δFLS , - - - δFL )
Figure 4.3. Approximate error of TS-fuzzy and linear model (different α)
0.16
0.14
0.12
0.1
error
0.08
0.06
0.04
0.02
0
−0.5 −0.4 −0.3 −0.2 −0.1 0 0.1 0.2 0.3 0.4 0.5
phi(rad)
( —– δFLS , - - - δFL )
Figure 4.4. Approximate error of TS-fuzzy and linear model (different φ)

4.4 LQR Based Control

F-16 Aircraft is internally a nonlinear system. The nonlinear system control prob-
lem is generally difficult to solve. Moreover, even the nonlinear controller is found,
it is often very complicated and hard to be implemented. On the other hand, the
linear control is usually designed based on linear model. Because of the severely
nonlinear of the real F-16 aircraft, the linear model cannot represent the dynamics
of the system in large scale. Thus, we try to design controller based on TS fuzzy
model. From the simulation in last section, we observe that TS fuzzy model has
much less model error than linear model. In our approaches, the LQR based and
Lyaponov based controllers are with linear controller and the gain schedule control
is with fuzzy controller. It is obvious that both of the controllers are of simpler
structure than most nonlinear controller.
In control design,it is often of interest to synthesize a controller to satisfy, in
an optimal fashion, certain performance criteria and constraints in addition to sta-
bility. For linear system, the problem of designing optimal controllers reduces to
solving algebraic Riccati equations (AREs), which are usually easy to solve and
detailed discussion of their solutions can be found in many textbooks. However,
for a general nonlinear system, the optimization problem reduces to the so-called
Hamiltion-Jacobi(HJ) equations, which are nonlinear partial differential equations
(PDEs). Different from their counterpart for linear systems, HJ equations are usu-
ally hard to solve both numerically and analytically. In this section, an alternative
approach is developed to nonlinear optimal control based on TS fuzzy model.
4.4.1 Stabilization
Suppose that the TS model of the nonlinear systems is described by

N
X
ẋ = hi (Ai x + Bi u), (4.24)
i=1
y = Cx. (4.25)
For stabilizing control, we choose the state feedback control law as
u = Kx. (4.26)
The optimal control methodology presented here is based on a quadratic per-

formance function. The control objective of optimal control is to minimize certain
performance functions with optimal controller. For all method, the controller is
designed to minimize the upper bound of the following quadratic performance
function
Z ∞
J= {y T (t)W y(t) + uT (t)Ru(t)}dt, (4.27)
0
The fuzzy system can be stabilized by a controller K if there exist a common

positive definite matrix P satisfying
 
(Ai + Bi K)T P + P (Ai + Bi K) C T K T
 
−1
C W 0  < 0. (4.28)
 

 
K 0 R−1
From (4.28) and using Schur complement, we can obtain
  
h i W 0 C
(Ai + Bi K)T P + P (Ai + Bi K) + C T K T     < 0. (4.29)
0 R K
It is obvious that
  
h i W 0 C
CT KT     > 0,
0 R K
so we have
(Ai + Bi K)T P + P (Ai + Bi K) < 0,
thus the closed-loop system is stable. Moreover, consider a Lyapunov function

candidate xT (t)P x(t), then we have
d T
(x P x)
dt
= ẋT P x + xT P ẋ
Xr
= xT ( hi (ATi P + K T BiT P + P Ai + P Bi K))x
i=1
< −x (C W C + K T RK)x.
T T
Therefore, we obtain
d T
(x P x) < −xT (C T W C + K T RK)x. (4.30)
dt
Integrating both side from 0 to ∞, we get
Z ∞
J= (y T W y + uT Ru)dt < xT (0)P x(0) − xT (∞)P x(∞). (4.31)
0
Thus, if the closed-loop system can be stabilized and the input is zero, then all
states approach to zero, the performance function satisfies
J < xT (0)P x(0), (4.32)
where xT (0)P x(0) acts as an upper bound of J.

The optimal control design problem is to minimizes the upper bound of the
performance function based on the above result. xT (0)P x(0) gives an upper bound
of the performance function J. The optimal controller introduced here is in the
strict sense a suboptimal controller since the xT (0)P x(0) is minimized instead of
J in the control design procedure.
Let Σ = P −1 and Ω = KP −1 , the feedback gains to minimize the upper bound
of the performance function can be obtained by solving the following LMIs.
minimize λ (4.33)
Σ,Ω
subject to
Σ > 0, (4.34)
 
λ xT (0)
  > 0, (4.35)
x(0) Σ
 
ΣATi + ΩT BiT + Ai Σ + Bi Ω ΣC T ΩT
 
−1
CΣ W 0 <0 (4.36)
 

 
−1
Ω 0 R
To find a minimal λ and Σ, Ω satisfying (4.34), (4.35) and (4.36) called the LMI
minization problem. An LMI is any constraint of the form
A(x) = A0 + x1 A1 + · · · + xN AN < 0 (4.37)

where x=[x1 , ..., xN ]T is the variable and A0 , ..., AN are given symmetric matri-
ces. Since A(y) < 0 and A(z) < 0 implies A[(y + z)/2] < 0, the LMI (4.37) is a
convex constraint on the variable x. It is well known that LMI-based optimiza-
tion problems as well as LMI feasibility problems can be solved by interior-point
algorithms with polynomial time, and a toolbox of MATLAB that is dedicated to
convex problems involving LMIs is now available.
From the solution of the LMIs, the feedback gains are obtained as
K = ΩΣ−1 . (4.38)
Then the performance function satisfies J < xT (0)P x(0) < λ.

Now we apply the proposed method in the F-16 control. Suppose the TS fuzzy
model of F-16 nonlinear system is (4.23), which is developed in last section. Using
the proposed control design method and choose W = 0.001 ∗ I8∗8 , R = 0.01 ∗ I4∗4
and C = I8∗8 , where I is a unity matrix, we obtain the optimal stabilizing controller
 
0 0 0 0 0 0 0 0
 
−11.4157 255.4694 −198.6310 −5.1515 429.9190 −318.2809 −386.3673 915.4586 
K= .
 
 35.0530 −91.2672 142.1502 2.0526 −217.1499 298.5508 458.7839 −339.5245 
 
−1.8140 −77.8953 199.6748 1.6734 −145.2588 −45.1900 160.6066 −298.3513
The simulation results are shown in Fig. 4.5.

We note that the only tuning parameters in this design are R and W matrices.
The larger the W matrix, the response will be faster, but the control signal will be
lager. The larger the R matrix, the more expensive the control signal. That means
the control signal is smaller and the dynamic of the states will be more sluggish.
If we choose R = 0.1 ∗ I4∗4 and W = 0.001 ∗ I8∗8 , the controller obtained is
 
0 0 0 0 0 0 0 0
 
−23.2854 198.4983 −12.0994 −2.7259 247.4012 −742.9609 −250.5109 488.8682 
K= .
 
 24.1550 −84.0403 21.8790 1.2085 −126.0828 358.9566 191.6083 −207.8943 
 
−5.1680 18.0407 40.2367 −0.2114 18.4618 −108.6397 −31.3616 35.1284
The response curves shown is Fig. 4.6. We can obtain that maximal amplitude of
the control signal is less than the result is Fig. 4.5, but the response of α and φ
are a little sluggish than those in Fig. 4.6.
5 5
0
alpha(deg)
0
el(deg)
−5
−5
−10
−15 −10
0 2 4 6 8 0 2 4 6 8
time(s) time(s)
5 20
0 0
phi(deg)
ail(deg)
−5 −20
−10 −40
−15 −60
0 2 4 6 8 0 2 4 6 8
time(s) time(s)
2
0
rdr(deg)
−2
−4
0 2 4 6 8
time(s)
Figure 4.5. The LQR based stabilizing control I
In another case, we choose R = 0.0001 ∗ I4∗4 and W = 0.0001 ∗ I8∗8 and obtain
5 2
0
alpha(deg)
1
el(deg)
−5
0
−10
−15 −1
0 2 4 6 8 0 2 4 6 8
time(s) time(s)
5 5
0 0
phi(deg)
ail(deg)
−5 −5
−10 −10
−15 −15
0 2 4 6 8 0 2 4 6 8
time(s) time(s)
4
2
rdr(deg)
−2
0 2 4 6 8
time(s)
Figure 4.6. The LQR based stabilizing control II
 
0 0 0 0 0 0 0 0
 
−0.0127 0.4500 −0.2407 −0.0115 1.0703 −0.3410 −0.7514 2.0573 
K = 103 ∗  .
 
 0.1332 −0.8641 0.9011 0.0240 −2.4380 1.2719 3.3564 −4.1577 
 
0.0034 −0.7199 1.3607 0.0189 −1.9330 0.3611 2.2602 −3.3008
5 50
alpha(deg)
0
0
el(deg)
−5
−50
−10
−15 −100
0 2 4 6 8 0 2 4 6 8
time(s) time(s)
5 100
0
0
phi(deg)
ail(deg)
−5
−100
−10
−15 −200
0 2 4 6 8 0 2 4 6 8
time(s) time(s)
50
0
rdr(deg)
−50
−100
0 2 4 6 8
time(s)
Figure 4.7. The LQR based stabilizing control III

In this case, the control signal is the largest and the state response is fastest,
although the response does not have much difference with the last two cases. The
response is shown in Fig. 4.7.
4.4.2 Tracking
In control system design, we often need the process to track some reference signals.
In order to obtain zero steady tracking error, we often add a set of new states on
the system, which are the integrals of the error signals. If tracking error is not
zero, the differential of the augmented state, i.e, the error between the specified
signal and the reference signal, will no be zero, that is, the augmented system is
not in steady state or can not be stabilized. Thus, with the augmented states, the
specified state will track the reference signal at the steady state if the augmented
system is stabilized by a controller. The augmented system can be described by
   
ẋ F (x, u)
 = ,
x˙a x − xr
Rt
where xa (t) = 0
(xr (τ ) − x(τ )d)τ , xr are reference signals, and F (x, u) is the
original system dynamics.
In aircraft control, we are often interested not only in regulating the state near
zero, but also in following a nonzero reference command signal. For example, we
may be interested in designing a control system for optimal step-response shap-
ing. This reference-input tracking or servo design problem is important in the
design of command augmentation system, where the reference signal may be, for
instance, the roll and pitch angle. According to the proposed design, we augment
the F-16 aircraft model with corresponding states in order to obtain zero steady
tracking error. Then we can use similar stabilizing control algorithm to stabilize
the augmented system. If the augmented system is stabilized, the differential of
the augmented stats approach to zero when time approaches infinity, that is, the
output is tracking the reference-input. Suppose that we wish the F-16 nonlinear
system to track the step signal of θr and φr , while keeping other states around the
zero. The states are augmented for the nonlinear F-16 model. The two augmented
Rt Rt
states θa (t) = 0
(θr (τ ) − θ(τ ))dτ and φa (t) = 0
(φr (τ ) − φ(τ ))dτ are incorporated
into the nonlinear F-16 model, where θr and φr are reference signal of θ and φ
respectively. The nonlinear system can be written as
   
ẋ F (x, u)
   
˙ 
 θa  =  θ − θ r 

   
φ˙a φ − φr
The augmented nonlinear system is different from the original F-16 model. If we
want to design a controller for augmented system, we need to build a new TS
model. The TS fuzzy model of the augmented system is described by
9
X
ẋ = hi (φ, θ)(Ai x + Bi u), (4.39)
i=1
where hi (φ, θ) = Mφi Mθi . Mφi and Mθi are the fuzzy membership functions of φ
and θ respectively. In our simulation, the nine operating points evenly distributed
in the state space. That is, we choose
(φ1 , θ1 ) = (−0.15, −0.2),
(φ2 , θ2 ) = (−0.15, 0),
(φ3 , θ3 ) = (−0.15, 0.2),
(φ4 , θ4 ) = (0, −0.2),
(φ5 , θ5 ) = (0, 0),
(φ6 , θ6 ) = (0, 0.2),
(φ7 , θ7 ) = (0.15, −0.2),
(φ8 , θ8 ) = (0.15, 0),
(φ9 , θ9 ) = (0.15, 0.2)
as the nine operating points. At each operating point, we can obtain the lin-
ear model at these operating points from (4.19) and (4.21). For example, at the
operating point (φ, θ) = (−0.2, −0.2), from (4.21), we get
 
-3.6722 0.0003 0.6741 0 0 -30.9113 0 0 0 0 0 0
 0 -1.3859 -0.0029 0.0000 -2.4977 0 -0.0003 -0.0002 0 0 0 0 
 -0.0263 0.0025 -0.4986 0 0 9.4747 0 0 0 0 0 0 
 
 0 -0.6501 0 -0.0202 6.8833 -6.3521 0.0559 -31.9571 0 0 0 0 
 0 0.9048 0 -0.0002 -1.0285 0 0.0064 0.0047 0 0 0 0 
 
A1 = 

0.0389
1.000
0
0.0221
-0.9916
-0.1089
0.0000
0
-0.0000
0
-0.3315
0
0.0636
0
-0.0005
0
0
0
0
0
0
0
0
0
,

 0 0.9801 0.1987 0 0 0 0 0 0 0 0 0

 
 0 -0.1999 0.9861 0 0 0 0 0 0 0 0 0 
 0 0 0 0 0 0 1.0000 -0.0000 0 0 0 0

 
0 0 0 0 0 0 -0.0000 1.0000 0 0 0 0
0 0 0 0 0 0 0 0 1.0000 0 0 0
and from (4.19), we obtain

 
0 0 -0.7336 0.1310
 0 -0.1824 0 0 
 0 0 -0.0327 -0.0643 
 
 0 0.1704 0 0 
 0 -0.0022 0 0 
 
B1 = 

0
0
0
0
0.0003
0
0.0008
0
.

 0 0 0 0

 
 0 0 0 0 
 0 0 0 0

 
0 0 0 0
0 0 0 0
The resulting (Ai , Bi ), i = 1, 2, · · · , 9, are given in Appendix B. Moreover, we select

triangle function in last section as the fuzzy membership functions, which are the
function is depicted in Fig. 4.2.
The complete augmented model with all the data is given in Appendix C.
The reference signals are step with amplitude 0.1 rad ( 5.73 degree). With
the proposed method and choose R = 0.0001 ∗ I4∗4 and W = 0.001 ∗ I12∗12 , the
controller gives

K = 1.0e + 004 * A 
0 0 0 0 0 0 0 0 0 0 0 0
 
 -0.0000 0.2323 0.0000 -0.0063 -0.0057 -0.0002 0.0000 0.8482 0.0001 -0.0000 0.0814 0.0000 
A= .
 

 0.0019 0.0000 0.0426 -0.0000 -0.0000 -0.1846 0.0193 0.0001 0.2651 -0.0047 0.0000 0.0089 

-0.0013 0.0000 0.2077 -0.0000 -0.0000 -1.3254 0.0320 0.0004 0.8747 -0.0329 0.0000 0.0167
The simulation results are shown in Fig. 4.8. The simulations show that the
LQR based control can track the given altitude step reference signals, however the
response is very sluggish. The reason is the method is a suboptimal design, we fail
to increase the speed of the response by changing the weighting W and R.
6 0.5
4 0
phi(deg)
el(deg)
2 −0.5
0 −1
−2 −1.5
0 200 400 600 0 200 400 600
time(s) time(s)
6 1
theta(deg)
4 0
ail(deg)
2 −1
0 −2
0 200 400 600 0 200 400 600
time(s) time(s)
15
10
rdr(deg)
−5
0 200 400 600
time(s)
Figure 4.8. The LQR based tracking control

4.5 Conclusion
In this chapter, we have carried out and achieved the following:
(i) The problem to design both stabilizing and tracking controller for F-16
aircraft systems has been addressed via the TS fuzzy modelling approach.
(ii) Both basic and augmented TS fuzzy models of F-16 have been obtained
using the best-available F-16 nonlinear model the literature and validated to be
reasonable accurate in the operating range of interests.
(iii) The LQR based control method has been proposed and shown to be feasible
and better than the normal linear control.
(iv) The LQR based control method is developed for stabilization of a TS
fuzzy model and extended to handle attitude tracking problem when the model
is augmented with integrator at each output. The most important feature of this
method is guaranteed stability and step-signal tracking of the closed-loop system
if the TS fuzzy model error is small enough, while most other nonlinear controllers
cannot. Besides, The solution is simple to get since it involves a LMI feasibility
problem only.
(v) These results could be useful to understand and design a flight control
system for rapid maneuver of F-16 aircraft.
Chapter 5
Conclusions
5.1 Main Findings

Most work of the author are focused on the modified Smith predictor for unstable
processes, PI controller design for nonlinear system and fuzzy modelling and control
of F-16 aircraft. In detail, the thesis has investigated and contributed to the
following areas:
A. Modified Smith Predictor Control for General Unstable Processes

The analysis and design of a modified Smith predictor control scheme for general
unstable processes is presented. We generalized Matausek and Micic’s scheme
from pure integral processes to general unstable delay processes, analyzed internal
stability, and discussed design issues. The proposed scheme can, by both analysis
and simulation, yield great improvement of both set-point and load disturbance
responses over conventional feedback systems.
B. Robust PI Controller Design for Nonlinear Systems via Fuzzy

Modeling Approach
The design problem of proportional-plus-integral (PI) controllers for nonlinear
systems is studied. First, the Takagi-Sugeno (T-S) fuzzy model with parameter
uncertainties is used to approximate the nonlinear systems. Then a numerically
tractable algorithm based on the technique of iterative linear matrix inequalities
76
Chapter 5. Conclusions 77
is developed to design a proportional (static output feedback) controller to the

robust stabilization of the system in T-S fuzzy model. Third, we transform the
problem of PI controller design to that of proportional controller design for an
augmented system and thus bring the solution of the former problem into the
configuration of the developed algorithm. Finally, the proposed method is applied
to the design of robust stabilizing controllers for the excitation control of power
systems. Simulation results show that the transient stability can be improved by
using a fuzzy PI controller when large faults appear in the system, compared to the
conventional PI controller designed by using linearizion method around the steady
state.
C. Fuzzy Modelling and Control of F-16 Aircraft

A framework for control of F-16 aircraft with Takagi-Sugeno (TS) fuzzy systems
is developed. First, based on the best-available nonlinear dynamical model of F-
16 aircraft in the open domain, the TS fuzzy model of F-16 aircraft is presented
and validated with reasonable accuracy. Then, the LQR based control strategy is
proposed using the TS model. It is applied to synthesize a F-16 flight control system
for both stabilizing control and attitude tracking control. Extensive simulation
is carried out. It is concluded that the proposed control design is feasible and
outperformances the linear control design significantly.
5.2 Suggestions for Further Work
Automatic Tuning of The Modified Smith Predictor Controllers of Un-

stable Processes
The autotuning approach is very important in industrial practice as no math-
ematical model of the process is required and also the Smith predictor strategy
is commonly used for processes with large time delays. Hang et al. (1995) have
presented methods to autotune and self-tune a modified Smith predictor of stable
precesses by relay feedback using an analysis based on the describing function ap-
proximation to a relay. The modified Smith predictor scheme in Chapter 2 can be
Chapter 5. Conclusions 78
extended to have a relay feedback automatic tuning method to provide a controller

for unstable processes with long dead time.
Bibliography
Astrom, K. J. and T. Hagglund (1984). Automatic tuning of simple regulators with

specifications on phase and amplitude margin. Automatica 20(5), 645–651.
Astrom, K. J. and T. Hagglund (1995). PID Controllers: Theory, Design, and

Tuning, 2nd Edition. Instrument Society of America. NC, USA.
Astrom, K. J., C. C. Hang and B. C. Lim (1994). A new smith predictor for
controlling a process with an integrator and long dead time. IEEE Trans.
Automatic Control 39(2), 343–345.
Astrom, K. J., C. C. Hang, P. Persson and W. K. Ho (1992). Towards intelligent

PID control. Automatica 28, 1–9.
Astrom, K. J., T. Hagglund, C. C. Hang and W. K. Ho (1993). Automatic tuning

and adaptation for PID controllers – a survey. Control Eng. Practice 1(4), 699–
714.
Athans, M., P. L. Falb and R. T. Lacoss (1963). Time-, fuel-, and energy-optimal
control of nonlinear norm-invariant systems. IRE Trans. on Automatic Control
8, 196–202.
Bergen, A. R. (1986). Power System Analysis. Prentice Hall. Englewood Cliffs, NJ.
Bharadwaj, S., M. Osipchuk, K. D. Mease and F. C. Park (1998). Geometry and

inverse optimality in global attitude stabilization. Journal of Guidance, Con-
trol, and Dynamics 21, 930–939.
79
Bibliography 80
Bi, Q., W. J. Cai and Q. G. Wang (2000). Advanced controller auto-tuning and
its application in hvac systems. Control Eng. Practice 8, 633–644.
Bonnet, C. and J. R. Partington (1999). Bezout factors and l 1 -optimal controllers

for delay systems using a two-parameter compensator scheme. IEEE Trans.
Automatic Control 44(8), 1512–1521.
Boyd, S., EL. L. Ghaoui, E. Feron and V. Balakrishnan (1994). Linear Matrix
Inequalities in System and Control Theory. SIAM. Philadelphia.
Brian, L. S. and L. L. Frank (1992). Aircraft Control and Simulation. John Wiley
& Sons. New York.
Cao, Y. Y. and P. M. Frank (2000). Analysis and synthesis of nonlinear time-delay

systems via fuzzy control approac. IEEE Trans. Fuzzy Systems 8, 200–211.
Cao, Y. Y. and Y. X. Sun (1998). Static output feedback stabilization: An ILMI

approach. Automatica 34, 1641–1645.
Carrington, C. K. and J. L. Junkins (1986). Optimal nonlinear feedback control for

spacecraft attitude maneuvers. Journal of Guidance, Control, and Dynamics
9, 99–107.
Chien, I. L. and P. S. Fruehauf (1990). Consider IMC tuning to improve controller

performance. Chemical Eng. Progress 86(10), 33–41.
Cohen, G. H. and G. A. Coon (1953). Theoretical consideration of retarded control.

Trans. Amer. Soc. Mech. Eng. 75, 827–834.
Datta, A., M. T. Ho and S. P. Bhattacharyya (2000). Structure and Synthesis of

PID Controllers. Springer. London.
Debs, A. S. and M. Athans (1969). On the optimal angular velocity control of

asymmetrical space vehicles. IEEE Trans. on Automatic Control 14, 80–83.
Freeman, R. A. and P. K. Kokotovic (1996a). Robust Nonlinear Control Design.

Birkhauser. Boston.
Bibliography 81
Freeman, R. A. and P. V. Kokotović (1996b). Inverse optimality in robust stabi-

lization. SIAM Journal on Control and Optimization 34, 1365–1391.
Gahinet, P., A. Nemirovski, A. J. Laub and M. Chilali (1995). LMI Control Toolbox.
The Math Works Inc.
Hang, C. C., Q. G. Wang and L. S. Cao (1995). Selftuning smith predictors for pro-
cesses with long dead time. Int. J. of Adaptive Control and Signal Processing
9(3), 255–270.
Ho, W. K. and W. Xu (1998). Pid tuning for unstable processes based on gain and
phase-margin specifications. IEE proc. Control Theory Appl. 145(5), 392–396.
Ho, W. K., C. C. Hang and L. S. Cao (1995). Tuning of PID controllers based on
gain and phase margin specifications. Automatica 31, 497–502.
Ho, W. K., Q. G. Wang and T. H. Lee (1998). PI tuning in terms of gain and phase
margins. Automatica 34, 1145–1149.
Isidori, A. (1995). Nonliear Control Systems: An Introduction. 3rd ed.. Berlin:

Springer-Verlag.
Johansen, T. A. and B. A. Foss (1993). Constructing NARMAX models using

ARMAX models. Int. J. Control 58, 1125–1153.
Kalman, R. E. (1964). When is a linear control system optimal?. Journal of Basic

Engineering 86, 51–60.
Khargonekar, P. P., I. R. Petersen and K. Zhou (1990). Robust stabilization of un-

certain linear systems: quadratic stabilizability and H∞ control. IEEE Trans.
Automatic Control 35, 356–361.
Krstić, M. and P. Tsiotras (1999). Inverse optimal stabilization of a rigid spacecraft.

IEEE Trans. on Automatic Control 44, 1042–1049.
Krstić, M., I. Kanellakopoulos and P. V. Kokotović (1995). Nonlinear and Adaptive

Control Design. Wiley. New York.
Bibliography 82
Lane, S. H. and R. F. Stengel (1988). Flight control design using non-linear inverse
dynamics. Automatica 24(4), 471–483.
Majhi, S and D. P. Atherton (1999). Modified smith predictor and controller for
processes with time delay. IEE Proc. -Control Theory Appl. 146(5), 359–366.
Marcelo, C. M. and H. Zak Stanislaw (1999). Stabilzing controller design for un-
certain nonlinear system using fuzzy models. IEEE Transactions on Fuzzy
Systems 7(2), 133–142.
Matausek, M. R. and A. D. Micic (1996). A modified smith predictor for controlling

a process with an integrator and long dead-time. IEEE Trans. Automatic
Control 41(8), 1199–1203.
Menon, P., P. Badgett and R. Walker (1987). Nonlinear flight test trajectory con-
trollers for aircraft. Journal of Guidance, Control, and Dynamics 10(1), 67–72.
Morari, M. and E. Zafiriou (1989). Robust Process Control. Prentice-Hall. Engle-

wood Cliffs, NJ.
Palm, R., D. Driankov and H. Hellendoorn (1997). Model based fuzzy control.
Springer-Verlag. Berlin, Germany.
Rotea, M., P. Tsiotras and M. Corless (1998). Suboptimal control of rigid body
motion with a quadratic cost. Dynamics and Control 8, 55–81.
Schumacher, C. and P. P. Khargonekar (1998). Stability analysis of a missile control

system with a dynamic inversion controller. Journal of Guidance, Control, and
Dynamics 21(3), 508–515.
Scrivener, S. L. and R. C. Thomson (1994). Survey of time-optimal attitude ma-

neuvers. Journal of Guidance, Control, and Dynamics 17, 225–233.
Sepulchre, R., M. Janković and P. V. Kokotović (1997). Constructive Nonlinear

Control. Springer-Verlag.
Bibliography 83
Shinskey, F. G. (1988). Process Control Systems: Application, Design and Tuning.

3rd ed.. McGraw-Hill. New York.
Shuster, M. D. (1963). A survey of attitude representations. SIAM Journal of

Astronautical Sciences 41, 439–517.
Smith, C. A. and A. B. Corripio (1985). Principles and Practice of Automatic

Process Control. Wiley.
Smith, O. J. (1959). A controller to overcome dead time. ISA J. 6(2), 28–33.
Snell, S. A., D. F. Enns and W. L. Garrard (1992). Nonlinear inversion flight

control for a supermaneuverable aircraft. Journal of Guidance, Control, and
Dynamics 15(4), 976–984.
Sontag, E. and H. Sussmann (1988). Further comments of the stabilizability of the

angular velocity of a rigid body. Systems and Control Letters 12, 213–217.
Sugeno, M. and G. T. Kang (1988). Structure identification of fuzzy model. Fuzzy

Sets and Systems 28, 15–33.
Taeyoung, L. and K. Youdan (2001). Nonlinear adaptive flight control using back-
stepping and neural networks controller. Journal of Guidance, Control, and
Dynamics 24, 675–682.
Takagi, T. and M. Sugeno (1985). Fuzzy identification of systems and its appli-
cations to modeling and control. IEEE Trans. Systems, Man, Cybernetics
15, 116–132.
Tan, K. K., Q. G. Wang and C. C. Hang (1999). Advances in PID Control. Springer.
London.
Tanaka, K. and M. Sugeno (1992). Stability analysis and design of fuzzy control
systems. Fuzzy Sets and Systems 45(2), 135–156.
Bibliography 84
Tanaka, K., T. Ikeda and H. Wang (1998). Fuzzy regulators and fuzzy observers:
Relaxed stability conditions and lmi-based designs. IEEE Trans. on Fuzzy
Systems 6(2), 250–265.
Teixeira, M. C. M. and S. H. Zak (1999). Stability analysis and design of fuzzy

control systems. IEEE Trans. Fuzzy Systems 7, 133–142.
Tsiotras, P. (1996). Stabilization and optimality results for the attitude control
problemin robust stabilization. Journal of Guidance, Control, and Dynamics
19, 772–779.
Tsiotras, P. (1997). Optimal regulation and passivity results for axisymmetric

rigid bodies using two controls. Journal of Guidance, Control, and Dynamics
20, 457–463.
Tsiotras, P. and J. M. Longuski (1994). Spin-axis stabilization of symmetric space-

craft with two control torques. Systems and Control Letters 23, 395–402.
Tsiotras, P., M. Corless and M. Rotea (1996). Optimal control of rigid body angular
velocity with quadratic cost. Proceedings of the 35th Conference on Decision
and Control pp. 1630–1635.
Vadali, S. R., L. G. Kraige and J. L. Junkins (1984). New results on the optimal
spacecraft attitude maneuver problem. SIAM Journal of Guidance, Control,
and Dynamics 7, 378–380.
Walton, K. and J. E. Marshall (1987). Direct method for tds stability analysis.
Proc. IEE 134 pt(D) pp. 101–107.
Wang, H. O., K. Tanaka and M. F. Griffin (1996). An approach to fuzzy control

of nonlinear systems: stability and design issues. IEEE Trans. Fuzzy Systems
4, 14–23.
Wang, L. X. and J. M. Mendel (1992). Fuzzy basis functions, universal approxima-

tiors and orthogonal least-square learning. IEEE Trans. on Neural Networks
3, 807–814.
Bibliography 85
Wang, Q. G., T. H. Lee and J. B. He (1999a). Internal stability of interconnected

systems. IEEE Trans. Automatic Control 44(3), 593–596.
Wang, Q. G., T. H. Lee, H. W. Fung, Q. Bi and Y. Zhang (1999b). PID tuning for
improved performance. IEEE Trans. Control Systems Technology 7, 457–465.
Windeknecht, T. G. (1963). Optimal stabilization of rigid body attitude. Journal

of Mathematical Analysis and Applications 6, 325–335.
Zak, S. H. (1999). Stabilizing fuzzy system models using linear controllers. IEEE
Trans. Fuzzy Systems 7, 236–240.
Zhuang, M. and D. P. Atherton (1993). Automatic tuning of optimum PID con-

trollers. IEE Proc. D: Contr. Theory and Appl. 140, 216–224.
Ziegler, J. G. and N. B. Nichlos (1942). Optimum settings for automat. Trans.

Amer. Soc. Mech. Eng. 64, 759–768.
Bibliography 86
Appendix A
Original Model
The body-fixed axes, nonlinear equations of motion for an aircraft over a flat Earth
are given by
cos α cos β sin β sin α cos β

V̇ = [T + Fx ] + [Fy ] + [Fz ]
m m m
+g[− cos α cos β sin θ + sin β sin φ cos θ + sin α cos β cos θ],
sin α
α̇ = −p cos α tan β + q − r sin α tan β − [T + Fx ]
mV cos β
cos α g
+ Fz + [sin α sin θ + cos α cos φ cos θ],
mV cos β V cos β
cos α sin β cos β sin α cos β
β̇ = p sin α − r cos α − [T + Fx ] + [Fy ] − [Fz ]
mV mV mV
g
+ [cos α sin β sin θ + cos β cos θ sin φ − sin α sin β cos φ cos θ],
V
ṗ = I2 pq + I1 qr + I3 L + I4 N,
q̇ = I5 pr − I6 (p2 − r2 ) + I7 M,
ṙ = −I2 qr + I8 pq + I4 L + I9 N,
φ̇ = p + tan θ(q sin φ + r cos φ),
θ̇ = q cos φ − r sin φ,
q sin φ + r cos φ
ψ̇ = ,
cos θ
where the moments of inertia Ii , i = 1, 2, ..., 9, are defined as follows:

2
Iz (Iz − Iy ) + Ixz Ixz (Ix − Iy + Iz ) Iz Ixz
I1 = 2
, I2 = 2
, I3 = 2
, I4 = 2
,
Ix Iz − Ixz Ix Iz − Ixz Ix Iz − Ixz Ix Iz − Ixz
2
Iz − I x Ixz 1 Ix (Ix − Iy ) + Ixz Ix
I5 = , I6 = , I7 = , I8 = 2
, I 9 = 2
.
Iy Iy Iy Ix Iz − Ixz Ix Iz − Ixz
It is assumed that the aerodynamic forces and moments are expressed as functions
of angle of attack, sideslip angle, angular rates, and control surface deflection. For
Bibliography 87
example, Fx , L and the other symbols are expressed as follows:
Fx = [Cx (α, δe ) + (c̄q/2V )Cxq (α)]q̄S,
Fy = [Cy (β, δα , δr ) + (bp/2V )Cyp (α) + (br/2V )Cyr ]q̄S, ,
Fz = [Cz (α, β, δe ) + (c̄q/2V )Czq (α)]q̄S,
L = [Cl (α, β) + Clδα (α, β)δα + Clδr (α, β)δr + (p/2V )Clp (α) + (r/2V )Clr (α)]q̄Sb,
M = [Cm (α, δe ) + (q/2V )Cmq (α)]q̄Sc̄
N = [Cn (α, β) + Cnδα (α, β)δα + Cnδr (α, β)δr + (p/2V )Cnp (α) + (r/2V )Cnr (α)]q̄Sb,
For the F-16 aircraft, the force coefficients Cx , Cxq , Cy , Cyp , Cyr , Cz , Czq and mo-
ment coefficients Cl , Clδα , Clδr , Clp , Clr , Cm , Cmq , Cn , Cnδα , Cnδr , Cnp , Cnr and other
necessary parameters can be obtained from the model programs. The control in-
puts are thrust T , elevator angle (el) δa , aileron angle (ail) δa and rudder angle
(rdr) δr . The nine states are speed V , angle of attack (alpha) α, sideslip angle
(beta) β, roll rate p, pitch rate q, yawing rate r, roll angle (phi) φ, pitch angle
(theta) θ and yaw angle (psi) ψ.
Appendix B
TS Fuzzy Basic Model
The model assumes the form:
9
X
ẋ = hi (α, φ)(Ai x + Bi u).
i=1
Selection of the operating points and the fuzzy rules
Rule 1: If α = −0.2 and φ = −0.2, then the linearized model is ẋ = A1 x + B1 u.

Rule 2: If α = 0 and φ = −0.2, then the linearized model is ẋ = A2 x + B2 u.
Bibliography 88
Rule 3: If α = 0.2 and φ = −0.2, then the linearized model is ẋ = A3 x + B3 u.

Rule 4: If α = −0.2 and φ = 0, then the linearized model is ẋ = A4 x + B4 u.
Rule 5: If α = 0 and φ = 0, then the linearized model is ẋ = A5 x + B5 u.
Rule 6: If α = 0.2 and φ = 0, then the linearized model is ẋ = A6 x + B6 u.
Rule 7: If α = −0.2 and φ = 0.2, then the linearized model is ẋ = A7 x + B7 u.
Rule 8: If α = 0 and φ = 0.2, then the linearized model is ẋ = A8 x + B8 u.
Rule 9: If α = 0.2 and φ = 0.2, then the linearized model is ẋ = A9 x + B9 u.
Resulting linear models:
 
-3.0489 0.0003 -0.9873 0 0 -3.7977 0 0
 

 0 -1.3815 -0.0029 0.0009 -1.1356 0 -0.0335 0 

 

 0.0312 0.0025 -0.5181 0 0 9.7538 0 0 


0 2.5473 0 -0.0832 191.1245 -6.3862 -80.3479 -31.5328

 
A1 =  ,
0 0.9639 0 0.0003 -1.4637 0 0.1846 -0.0127
 
 
 
-0.1609 0 -0.9798 0.0000 0.0004 -0.2779 0.0632 0.0005
 
 
 
1.0000 -0.0078 0.0386 0 0 0 0 0
 
 
0 0.9801 0.1987 0 0 0 0 0
 
0 0 -0.5983 0.1588
 

 0 -0.1925 0 0 

 

 0 0 -0.0337 -0.0675 


0 0.4584 0 0

 
B1 =  
0 -0.0020 0 0
 
 
 
0 0 0.0003 0.0008
 
 
 
0 0 0 0
 
 
0 0 0 0
 
-3.6722 0.0003 0.6741 0 0 -30.9113 0 0
 

 0 -1.3859 -0.0029 0.0000 -2.4977 0 -0.0005 0 

 

 -0.0263 0.0025 -0.4986 0 0 9.4747 0 0 


0 -0.6501 0 -0.0202 7.2411 -6.3862 0.1259 -32.169

 
A2 =  ,
0 0.9048 0 -0.0003 -1.0189 0 0.0064 0.0001
 
 
 
0.0389 0 -0.9916 0.0000 0 -0.3220 0.0636 0.0005
 
 
 
1.0000 -0.0078 0.0386 0 0 0 0 0
 
 
0 0.9801 0.1987 0 0 0 0 0
 
0 0 -0.7336 0.1310
 

 0 -0.1824 0 0 

 

 0 0 -0.0327 -0.0643 


0 0.1704 0 0

 
B2 =  
0 -0.0022 0 0
 
 
 
0 0 0.0003 0.0008
 
 
 
0 0 0 0
 
 
0 0 0 0
 
-3.2067 0.0003 1.8464 0 0 -65.6296 0 0
 

 0 -1.6530 -0.0029 -0.0016 -0.7348 0 1.2180 0 

 

 -0.0826 0.0025 -0.5456 0 0 8.8395 0 0 


0 -7.4582 0 -0.0917 -92.9348 -6.3862 -33.758 -31.5228

 
A3 =  ,
0 0.9031 0 -0.0006 -1.0248 0 0.0054 0.0128
 
 
 
0.2390 0 -0.9633 0.0000 -0.0004 -0.2984 0.0632 0.0005
 
 
 
1.0000 -0.0078 0.0386 0 0 0 0 0
 
 
0 0.9801 0.1987 0 0 0 0 0
Bibliography 89
 
0 0 -0.6839 0.1258
 

 0 -0.2000 0 0 

 

 0 0 -0.0254 -0.0641 


0 -0.1139 0 0

 
B3 =  
0 -0.0021 0 0
 
 
 
0 0 0.0003 0.0008
 
 
 
0 0 0 0
 
 
0 0 0 0
 
-3.0489 0.0003 -0.9873 0 0 -3.7977 0 0
 

 0 -1.3815 -0.0029 0.0009 -1.1690 0 0 0 

 

 0.0312 0.0025 -0.5181 0 0 9.7538 0 0 


0 2.5473 0 -0.0832 111.2891 0 0 -31.5328

 
A4 =  ,
0 0.9639 0 0.0003 -1.2854 0 0 -0.0127
 
 
 
-0.1609 0 -0.9798 0.0000 0.000 -0.2779 0.0640 0.00
 
 
 
1.0000 0 0.0394 0 0 0 0 0
 
 
0 1.0000 0 0 0 0 0 0
 
0 0 -0.5983 0.1588
 

 0 -0.1925 0 0 

 

 0 0 -0.0337 -0.0675 


0 0.4584 0 0

 
B4 =  
0 -0.0020 0 0
 
 
 
0 0 0.0003 0.0008
 
 
 
0 0 0 0
 
 
0 0 0 0
 
-3.6722 0.0003 0.6741 0 0 -30.9113 0 0
 

 0 -1.3859 -0.0029 0.0000 -2.4977 0 0 0 

 

 -0.0263 0.0025 -0.4986 0 0 9.4747 0 0 


0 -0.6501 0 -0.0202 7.8814 0 0 -32.169

 
A5 =  ,
0 0.9048 0 -0.0003 -1.0190 0 0 0
 
 
 
0.0389 0 -0.9916 0.0000 0 -0.3220 0.064 0
 
 
 
1.0000 0 0.0394 0 0 0 0 0
 
 
0 1.0000 0 0 0 0 0 0
 
0 0 -0.7336 0.1310
 

 0 -0.1824 0 0 

 

 0 0 -0.0327 -0.0643 


0 0.1704 0 0

 
B5 =  
0 -0.0022 0 0
 
 
 
0 0 0.0003 0.0008
 
 
 
0 0 0 0
 
 
0 0 0 0
 
-3.2067 0.0003 1.8464 0 0 -65.6296 0 0
 

 0 -1.6530 -0.0029 -0.0016 -1.9527 0 0 0 

 

 -0.0826 0.0025 -0.5456 0 0 8.8395 0 0 


0 -7.4582 0 -0.0917 -58.4168 0 0 -31.5287

 
A6 =  ,
0 0.9031 0 -0.0006 -1.0240 0 0 0.0127
 
 
 
0.2390 0 -0.9633 0.0000 -0.0000 -0.2987 0.0640 0
 
 
 
1.0000 0.00 0.0394 0 0 0 0 0
 
 
0 1.0000 0 0 0 0 0 0
 
0 0 -0.6839 0.1258
 

 0 -0.2000 0 0 

 

 0 0 -0.0254 -0.0641 


0 -0.1139 0 0

 
B6 =  
0 -0.0021 0 0
 
 
 
0 0 0.0003 0.0008
 
 
 
0 0 0 0
 
 
0 0 0 0
Bibliography 90
 
-3.0489 0.0003 -0.9873 0 0 -3.7977 0 0
 

 0 -1.3815 -0.0029 0.0009 -1.1356 0 0.0335 0 

 

 0.0312 0.0025 -0.5181 0 0 9.7538 0 0 


0 2.5473 0 -0.0832 191.1245 6.3862 80.3479 -31.5328

 
A7 =  ,
0 0.9639 0 0.0003 -1.4637 0 0.1846 -0.0127
 
 
 
-0.1609 0 -0.9798 -0.0000 -0.0004 -0.2779 0.0632 -0.0005
 
 
 
1.0000 0.0078 0.0386 0 0 0 0 0
 
 
0 0.9801 -0.1987 0 0 0 0 0
 
0 0 -0.5983 0.1588
 

 0 -0.1925 0 0 

 

 0 0 -0.0337 -0.0675 

 
 0 0.4584 0 0 
B7 =  
0 -0.0020 0 0
 
 
 
0 0 0.0003 0.0008
 
 
 
0 0 0 0
 
 
0 0 0 0
 
3.6722 0.0003 0.6741 0 0 -30.9113 0 0
 

 0 -1.3859 -0.0029 0.0000 -2.4977 0 0.0005 0 

 

 -0.0263 0.0025 -0.4986 0 0 9.4747 0 0 


0 -0.6501 0 -0.0202 7.2411 6.3862 -0.1259 -32.169

 
A8 =  ,
0 0.9048 0 -0.0003 -1.0189 0 -0.0064 0.0001
 
 
 
0.0389 0 -0.9916 -0.0000 0 -0.3220 0.0636 -0.0005
 
 
 
1.0000 0.0078 0.0386 0 0 0 0 0
 
 
0 0.9801 -0.1987 0 0 0 0 0
 
0 0 -0.7336 0.1310
 

 0 -0.1824 0 0 

 

 0 0 -0.0327 -0.0643 

 
 0 0.1704 0 0 
B8 =  
0 -0.0022 0 0
 
 
 
0 0 0.0003 0.0008
 
 
 
0 0 0 0
 
 
0 0 0 0
 
-3.2067 0.0003 1.8464 0 0 -65.6296 0 0
 

 0 -1.6530 -0.0029 -0.0016 -0.7348 0 -1.2180 0 

 

 -0.0826 0.0025 -0.5456 0 0 8.8395 0 0 


0 -7.4582 0 -0.0917 -92.9348 6.3862 33.7585 -31.5228

 
A9 =  ,
0 0.9031 0 -0.0006 -1.0248 0 -0.0054 0.0128
 
 
 
0.2390 0 -0.9633 -0.0000 0.0004 -0.2984 0.0632 -0.0005
 
 
 
1.0000 0.0078 0.0386 0 0 0 0 0
 
 
0 0.9801 -0.1987 0 0 0 0 0
 
0 0 -0.6839 0.1258
 

 0 -0.2000 0 0 

 

 0 0 -0.0254 -0.0641 


0 -0.1139 0 0

 
B9 =  
0 -0.0021 0 0
 
 
 
0 0 0.0003 0.0008
 
 
 
0 0 0 0
 
 
0 0 0 0
Resulting Weighting Functions:
hi (α, φ) = Mαi Mφi , i = 1, 2, 3,

Bibliography 91
where 


 1 if α ≤ −0.2

M α1 = 2.5 ∗ α − 0.5 if − 0.2 < α < 0.2


0 if α ≥ 0.2





 0 if α ≤ −0.2


 5∗α+1

if − 0.2 < α ≤ 0
M α2 =


 −5 ∗ α + 1 if 0 < α < 0.2


0 if α ≥ 0.2





 0 if α ≤ −0.2

M α3 = 2.5 ∗ α + 0.5 if − 0.2 < α < 0.2


1 if α ≥ 0.2





 1 if φ ≤ −0.2

Mφ 1 = 2.5 ∗ φ − 0.5 if − 0.2 < φ < 0.2


0 if φ ≥ 0.2





 0 if φ ≤ −0.2


 5∗φ+1

if − 0.2 < φ ≤ 0
Mφ 2 =


 −5 ∗ φ + 1 if 0 < φ < 0.2


0 if φ ≥ 0.2





 0 if φ ≤ −0.2

Mφ 3 = 2.5 ∗ φ + 0.5 if − 0.2 < φ < 0.2


1 if φ ≥ 0.2


Appendix C
TS Fuzzy Augmented Model
The model assumes the form:
9
X
x̄˙ = h̄i (φ, θ)(Āi x̄ + B̄i ū)
i=1
Bibliography 92
Selection of the operating points and the fuzzy rules
Rule 1: If φ = −0.2 and θ = −0.15, then the linearized model is x̄˙ = Ā1 x̄ + B̄1 ū.
Rule 2: If φ = 0 and θ = −0.15, then the linearized model is x̄˙ = Ā2 x̄ + B̄2 ū.
Rule 3: If φ = 0.2 and θ = −0.15, then the linearized model is x̄˙ = Ā3 x̄ + B̄3 ū.
Rule 4: If φ = −0.2 and θ = 0, then the linearized model is x̄˙ = Ā4 x̄ + B̄4 ū.
Rule 5: If φ = 0 and θ = 0, then the linearized model is x̄˙ = Ā5 x̄ + B̄5 ū.
Rule 6: If φ = 0.2 and θ = 0, then the linearized model is x̄˙ = Ā6 x̄ + B̄6 ū.
Rule 7: If φ = −0.2 and θ = 0.15, then the linearized model is x̄˙ = Ā7 x̄ + B̄7 ū.
Rule 8: If φ = 0 and θ = 0.15, then the linearized model is x̄˙ = Ā8 x̄ + B̄8 ū.
Rule 9: If φ = 0.2 and θ = 0.15, then the linearized model is x̄˙ = Ā9 x̄ + B̄9 ū.
Resulting linear models:
 
-3.6722 0.0003 0.6741 0 0 -30.9113 0 0 0 0 0 0
 

 0 -1.3859 -0.0029 0.0000 -2.4977 0 -0.0003 -0.0002 0 0 0 0 

 

 -0.0263 0.0025 -0.4986 0 0 9.4747 0 0 0 0 0 0 

 

 0 -0.6501 0 -0.0202 6.8833 -6.3521 0.0559 -31.9571 0 0 0 0 

 

 0 0.9048 0 -0.0002 -1.0285 0 0.0064 0.0047 0 0 0 0 

 
 0.0389 0 -0.9916 0.0000 -0.0000 -0.3315 0.0636 -0.0005 0 0 0 0 
Ā1 =  
1.000 0.0221 -0.1089 0 0 0 0 0 0 0 0 0
 
 
 
0 0.9801 0.1987 0 0 0 0 0 0 0 0 0
 
 
 
0 -0.1999 0.9861 0 0 0 0 0 0 0 0 0
 
 
 

 0 0 0 0 0 0 1.0000 -0.0000 0 0 0 0 

 

 0 0 0 0 0 0 -0.0000 1.0000 0 0 0 0 

0 0 0 0 0 0 0 0 1.0000 0 0 0
 
0 0 -0.7336 0.1310
 

 0 -0.1824 0 0 

 

 0 0 -0.0327 -0.0643 

 

 0 0.1704 0 0 

 

 0 -0.0022 0 0 

 
 0 0 0.0003 0.0008 
B̄1 =  
0 0 0 0
 
 
 
0 0 0 0
 
 
 
0 0 0 0
 
 
 

 0 0 0 0 

 

 0 0 0 0 

0 0 0 0
 
-3.6722 0.0003 0.6741 0.0000 0.0000 -30.9113 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
 

 0.0000 -1.3859 -0.0029 0.0000 -2.4977 0.0000 0.0000 -0.0007 0.0000 0.0000 0.0000 0.0000 

 

 -0.0263 0.0025 -0.4986 0.0000 0.0000 9.4747 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 

 

 0.0000 -0.6501 0.0000 -0.0202 7.5202 0.0000 0.0000 -32.0497 0.0000 0.0000 0.0000 0.0000 

 

 0.0000 0.9048 0.0000 -0.0003 -1.0286 0.0000 0.0000 0.0048 0.0000 0.0000 0.0000 0.0000 

 
 0.0389 0.0000 -0.9916 0.0000 0.0000 -0.3316 0.0637 0.0000 0.0000 0.0000 0.0000 0.0000 
Ā2 =  
1.0000 0.0000 -0.1111 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
 
 
 
0.0000 1.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
 
 
 
0.0000 0.0000 1.0062 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
 
 
 

 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 1.0000 0.0000 0.0000 0.0000 0.0000 0.0000 

 

 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 1.0000 0.0000 0.0000 0.0000 0.0000 

0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 1.0000 0.0000 0.0000 0.0000
Bibliography 93
 
0.0000 0.0000 -0.7336 0.1310
 

 0.0000 -0.1824 0.0000 0.0000 

 

 0.0000 0.0000 -0.0327 -0.0643 

 

 0.0000 0.1704 0.0000 0.0000 

 

 0.0000 -0.0022 0.0000 0.0000 


0.0000 0.0000 0.0003 0.0008

 
B̄2 =  
0.0000 0.0000 0.0000 0.0000
 
 
 
0.0000 0.0000 0.0000 0.0000
 
 
 
0.0000 0.0000 0.0000 0.0000
 
 
 

 0.0000 0.0000 0.0000 0.0000 

 

 0.0000 0.0000 0.0000 0.0000 

0.0000 0.0000 0.0000 0.0000
 
-3.6722 0.0003 0.6741 0.0000 0.0000 -30.9113 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
 

 0.0000 -1.3859 -0.0029 0.0000 -2.4977 0.0000 0.0003 -0.0002 0.0000 0.0000 0.0000 0.0000 

 

 -0.0263 0.0025 -0.4986 0.0000 0.0000 9.4747 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 

 

 0.0000 -0.6501 0.0000 -0.0202 6.8833 6.3521 -0.0559 -31.9571 0.0000 0.0000 0.0000 0.0000 

 

 0.0000 0.9048 0.0000 -0.0002 -1.0285 0.0000 -0.0064 0.0047 0.0000 0.0000 0.0000 0.0000 


0.0389 0.0000 -0.9916 -0.0000 0.0000 -0.3315 0.0636 0.0005 0.0000 0.0000 0.0000 0.0000

 
Ā3 =  
1.0000 -0.0221 -0.1089 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
 
 
 
0.0000 0.9801 -0.1987 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
 
 
 
0.0000 0.1999 0.9861 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
 
 
 

 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 1.0000 0.0000 0.0000 0.0000 0.0000 0.0000 

 

 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 1.0000 0.0000 0.0000 0.0000 0.0000 

0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 1.0000 0.0000 0.0000 0.0000
 
0.0000 0.0000 -0.7336 0.1310
 

 0.0000 -0.1824 0.0000 0.0000 

 

 0.0000 0.0000 -0.0327 -0.0643 

 

 0.0000 0.1704 0.0000 0.0000 

 

 0.0000 -0.0022 0.0000 0.0000 


0.0000 0.0000 0.0003 0.0008

 
B̄3 =  
0.0000 0.0000 0.0000 0.0000
 
 
 
0.0000 0.0000 0.0000 0.0000
 
 
 
0.0000 0.0000 0.0000 0.0000
 
 
 

 0.0000 0.0000 0.0000 0.0000 

 

 0.0000 0.0000 0.0000 0.0000 

0.0000 0.0000 0.0000 0.0000
 
-3.6722 0.0003 0.6741 0.0000 0.0000 -30.9113 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
 

 0.0000 -1.3859 -0.0029 0.0000 -2.4977 0.0000 -0.0005 0.0000 0.0000 0.0000 0.0000 0.0000 

 

 -0.0263 0.0025 -0.4986 0.0000 0.0000 9.4747 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 

 

 0.0000 -0.6501 0.0000 -0.0202 7.2411 -6.3862 0.1259 -32.1690 0.0000 0.0000 0.0000 0.0000 

 

 0.0000 0.9048 0.0000 -0.0003 -1.0189 0.0000 0.0064 0.0001 0.0000 0.0000 0.0000 0.0000 


0.0389 0.0000 -0.9916 0.0000 0.0000 -0.3220 0.0636 0.0005 0.0000 0.0000 0.0000 0.0000

 
Ā4 =  
1.0000 -0.0078 0.0386 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
 
 
 
0.0000 0.9801 0.1987 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
 
 
 
0.0000 -0.1988 0.9808 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
 
 
 

 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 1.0000 0.0000 0.0000 0.0000 0.0000 0.0000 

 

 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 1.0000 0.0000 0.0000 0.0000 0.0000 

0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 1.0000 0.0000 0.0000 0.0000
 
0.0000 0.0000 -0.7336 0.1310
 

 0.0000 -0.1824 0.0000 0.0000 

 

 0.0000 0.0000 -0.0327 -0.0643 

 

 0.0000 0.1704 0.0000 0.0000 

 

 0.0000 -0.0022 0.0000 0.0000 


0.0000 0.0000 0.0003 0.0008

 
B̄4 =  
0.0000 0.0000 0.0000 0.0000
 
 
 
0.0000 0.0000 0.0000 0.0000
 
 
 
0.0000 0.0000 0.0000 0.0000
 
 
 

 0.0000 0.0000 0.0000 0.0000 

 

 0.0000 0.0000 0.0000 0.0000 

0.0000 0.0000 0.0000 0.0000
Bibliography 94
 
-3.6722 0.0003 0.6741 0.0000 0.0000 -30.9113 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
 

 0.0000 -1.3859 -0.0029 0.0000 -2.4977 0.0000 -0.0005 0.0000 0.0000 0.0000 0.0000 0.0000 

 

 -0.0263 0.0025 -0.4986 0.0000 0.0000 9.4747 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 

 

 0.0000 -0.6501 0.0000 -0.0202 7.2411 -6.3862 0.1259 -32.1690 0.0000 0.0000 0.0000 0.0000 

 

 0.0000 0.9048 0.0000 -0.0003 -1.0189 0.0000 0.0064 0.0001 0.0000 0.0000 0.0000 0.0000 


0.0389 0.0000 -0.9916 0.0000 0.0000 -0.3220 0.0636 0.0005 0.0000 0.0000 0.0000 0.0000

 
Ā5 =  
1.0000 -0.0078 0.0386 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
 
 
 
0.0000 0.9801 0.1987 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
 
 
 
0.0000 -0.1988 0.9808 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
 
 
 

 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 1.0000 0.0000 0.0000 0.0000 0.0000 0.0000 

 

 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 1.0000 0.0000 0.0000 0.0000 0.0000 

0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 1.0000 0.0000 0.0000 0.0000
 
0.0000 0.0000 -0.7336 0.1310
 

 0.0000 -0.1824 0.0000 0.0000 

 

 0.0000 0.0000 -0.0327 -0.0643 

 

 0.0000 0.1704 0.0000 0.0000 

 

 0.0000 -0.0022 0.0000 0.0000 


0.0000 0.0000 0.0003 0.0008

 
B̄5 =  
0.0000 0.0000 0.0000 0.0000
 
 
 
0.0000 0.0000 0.0000 0.0000
 
 
 
0.0000 0.0000 0.0000 0.0000
 
 
 

 0.0000 0.0000 0.0000 0.0000 

 

 0.0000 0.0000 0.0000 0.0000 

0.0000 0.0000 0.0000 0.0000
 
-3.6722 0.0003 0.6741 0.0000 0.0000 -30.9113 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
 

 0.0000 -1.3859 -0.0029 0.0000 -2.4977 0.0000 0.0005 0.0000 0.0000 0.0000 0.0000 0.0000 

 

 -0.0263 0.0025 -0.4986 0.0000 0.0000 9.4747 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 

 

 0.0000 -0.6501 0.0000 -0.0202 7.2411 6.3862 -0.1259 -32.1690 0.0000 0.0000 0.0000 0.0000 

 

 0.0000 0.9048 0.0000 -0.0003 -1.0189 0.0000 -0.0064 0.0001 0.0000 0.0000 0.0000 0.0000 


0.0389 0.0000 -0.9916 -0.0000 0.0000 -0.3220 0.0636 -0.0005 0.0000 0.0000 0.0000 0.0000

 
Ā6 =  
1.0000 0.0078 0.0386 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
 
 
 
0.0000 0.9801 -0.1987 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
 
 
 
0.0000 0.1988 0.9808 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
 
 
 

 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 1.0000 0.0000 0.0000 0.0000 0.0000 0.0000 

 

 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 1.0000 0.0000 0.0000 0.0000 0.0000 

0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 1.0000 0.0000 0.0000 0.0000
 
0.0000 0.0000 -0.7336 0.1310
 

 0.0000 -0.1824 0.0000 0.0000 

 

 0.0000 0.0000 -0.0327 -0.0643 

 

 0.0000 0.1704 0.0000 0.0000 

 

 0.0000 -0.0022 0.0000 0.0000 


0.0000 0.0000 0.0003 0.0008

 
B̄6 =  
0.0000 0.0000 0.0000 0.0000
 
 
 
0.0000 0.0000 0.0000 0.0000
 
 
 
0.0000 0.0000 0.0000 0.0000
 
 
 

 0.0000 0.0000 0.0000 0.0000 

 

 0.0000 0.0000 0.0000 0.0000 

0.0000 0.0000 0.0000 0.0000
 
-3.6722 0.0003 0.6741 0.0000 0.0000 -30.9113 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
 

 0.0000 -1.3859 -0.0029 0.0000 -2.4977 0.0000 -0.0003 0.0002 0.0000 0.0000 0.0000 0.0000 

 

 -0.0263 0.0025 -0.4986 0.0000 0.0000 9.4747 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 

 

 0.0000 -0.6501 0.0000 -0.0202 6.8909 -6.2769 0.2860 -31.8332 0.0000 0.0000 0.0000 0.0000 

 

 0.0000 0.9048 0.0000 -0.0002 -1.0093 0.0000 0.0063 -0.0047 0.0000 0.0000 0.0000 0.0000 

 
 0.0389 0.0000 -0.9916 0.0000 -0.0000 -0.3124 0.0634 0.0011 0.0000 0.0000 0.0000 0.0000 
Ā7 =  
1.0000 -0.0381 0.1878 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
 
 
 
0.0000 0.9801 0.1987 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
 
 
 
0.0000 -0.2023 0.9979 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
 
 
 

 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 1.0000 0.0000 0.0000 0.0000 0.0000 0.0000 

 

 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 1.0000 0.0000 0.0000 0.0000 0.0000 

0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 1.0000 0.0000 0.0000 0.0000
Bibliography 95
 
0.0000 0.0000 -0.7336 0.1310
 

 0.0000 -0.1824 0.0000 0.0000 

 

 0.0000 0.0000 -0.0327 -0.0643 

 

 0.0000 0.1704 0.0000 0.0000 

 

 0.0000 -0.0022 0.0000 0.0000 


0.0000 0.0000 0.0003 0.0008

 
B̄7 =  
0.0000 0.0000 0.0000 0.0000
 
 
 
0.0000 0.0000 0.0000 0.0000
 
 
 
0.0000 0.0000 0.0000 0.0000
 
 
 

 0.0000 0.0000 0.0000 0.0000 

 

 0.0000 0.0000 0.0000 0.0000 

0.0000 0.0000 0.0000 0.0000
 
-3.6722 0.0003 0.6741 0.0000 0.0000 -30.9113 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
 

 0.0000 -1.3859 -0.0029 0.0000 -2.4977 0.0000 0.0000 0.0007 0.0000 0.0000 0.0000 0.0000 

 

 -0.0263 0.0025 -0.4986 0.0000 0.0000 9.4747 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 

 

 0.0000 -0.6501 0.0000 -0.0202 7.5202 0.0000 0.0000 -32.0493 0.0000 0.0000 0.0000 0.0000 

 

 0.0000 0.9048 0.0000 -0.0003 -1.0094 0.0000 0.0000 -0.0048 0.0000 0.0000 0.0000 0.0000 


0.0389 0.0000 -0.9916 0.0000 0.0000 -0.3124 0.0629 0.0000 0.0000 0.0000 0.0000 0.0000

 
Ā8 =  
1.0000 0.0000 0.1917 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
 
 
 
0.0000 1.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
 
 
 
0.0000 0.0000 1.0182 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
 
 
 

 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 1.0000 0.0000 0.0000 0.0000 0.0000 0.0000 

 

 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 1.0000 0.0000 0.0000 0.0000 0.0000 

0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 1.0000 0.0000 0.0000 0.0000
 
0.0000 0.0000 -0.7336 0.1310
 

 0.0000 -0.1824 0.0000 0.0000 

 

 0.0000 0.0000 -0.0327 -0.0643 

 

 0.0000 0.1704 0.0000 0.0000 

 

 0.0000 -0.0022 0.0000 0.0000 


0.0000 0.0000 0.0003 0.0008

 
B̄8 =  
0.0000 0.0000 0.0000 0.0000
 
 
 
0.0000 0.0000 0.0000 0.0000
 
 
 
0.0000 0.0000 0.0000 0.0000
 
 
 

 0.0000 0.0000 0.0000 0.0000 

 

 0.0000 0.0000 0.0000 0.0000 

0.0000 0.0000 0.0000 0.0000
 
-3.6722 0.0003 0.6741 0.0000 0.0000 -30.9113 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
 

 0.0000 -1.3859 -0.0029 0.0000 -2.4977 0.0000 0.0003 0.0002 0.0000 0.0000 0.0000 0.0000 

 

 -0.0263 0.0025 -0.4986 0.0000 0.0000 9.4747 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 

 

 0.0000 -0.6501 0.0000 -0.0202 6.8909 6.2769 -0.2860 -31.8332 0.0000 0.0000 0.0000 0.0000 

 

 0.0000 0.9048 0.0000 -0.0002 -1.0093 0.0000 -0.0063 -0.0047 0.0000 0.0000 0.0000 0.0000 


0.0389 0.0000 -0.9916 -0.0000 0.0000 -0.3124 0.0634 -0.0011 0.0000 0.0000 0.0000 0.0000

 
Ā9 =  
1.0000 0.0381 0.1878 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
 
 
 
0.0000 0.9801 -0.1987 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
 
 
 
0.0000 0.2023 0.9979 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
 
 
 

 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 1.0000 -0.0000 0.0000 0.0000 0.0000 0.0000 

 

 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 -0.0000 1.0000 0.0000 0.0000 0.0000 0.0000 

0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 1.0000 0.0000 0.0000 0.0000
 
0.0000 0.0000 -0.7336 0.1310
 

 0.0000 -0.1824 0.0000 0.0000 

 

 0.0000 0.0000 -0.0327 -0.0643 

 

 0.0000 0.1704 0.0000 0.0000 

 

 0.0000 -0.0022 0.0000 0.0000 


0.0000 0.0000 0.0003 0.0008

 
B̄9 =  
0.0000 0.0000 0.0000 0.0000
 
 
 
0.0000 0.0000 0.0000 0.0000
 
 
 
0.0000 0.0000 0.0000 0.0000
 
 
 

 0.0000 0.0000 0.0000 0.0000 

 

 0.0000 0.0000 0.0000 0.0000 

0.0000 0.0000 0.0000 0.0000
Resulting Weighting Functions:

Bibliography 96
h̄i (φ, θ) = Mφi Mθi , i = 1, 2, 3,
where 


 1 if φ ≤ −0.2

Mφ 1 = 2.5 ∗ φ − 0.5 if − 0.2 < φ < 0.2


0 if φ ≥ 0.2





 0 if φ ≤ −0.2


5∗φ+1 if − 0.2 < φ ≤ 0


Mφ 2 =


 −5 ∗ φ + 1 if 0 < φ < 0.2


0 if φ ≥ 0.2





 0 if φ ≤ −0.2

Mφ 3 = 2.5 ∗ φ + 0.5 if − 0.2 < φ < 0.2


1 if φ ≥ 0.2





 1 if θ ≤ −0.15

Mθ 1 = 2.5 ∗ θ − 0.5 if − 0.15 < θ < 0.15


0 if θ ≥ 0.15





 0 if θ ≤ −0.15


5∗θ+1 if − 0.15 < θ ≤ 0


Mθ 2 =


 −5 ∗ θ + 1 if 0 < θ < 0.15


0 if θ ≥ 0.15





 0 if θ ≤ −0.15

Mθ 3 = 2.5 ∗ θ + 0.5 if − 0.15 < θ < 0.15


1 if θ ≥ 0.15


List of Publications
[1] Feng Zheng, Qing-Guo Wang, Tong Heng Lee, Xiaogang Huang (2001). Robust
PI controller design for nonlinear systems via fuzzy modeling approach. IEEE
Trans. on Systems, Man, and Cybernetics–Part A: Systems and Humans Vol.31,
No.6, 666–675.
[2] Feng Zheng, Qing-Guo Wang, Tong Heng Lee, Xiaogang Huang (2001). Robust
PI controller design for nonlinear systems via fuzzy modeling approach. Proceeding
of the 40th IEEE Conference on Decision and Control Vol.5, 4346-4351.
[3] Qing-Guo Wang, Yong Zhang, Xiao-Gang Huang (2002). Virtual feedforward
control for asymptotic rejection of periodic disturbance. IEEE Trans. on Industrial
Electronics Vol.49, No.3, 566-573.
[4] Qing-Guo Wang, Ru He, Xiao-Gang Huang (2002). An Effective frequency

domain approach to tuning non-PID controllers for high performance. ISA Trans-
actions Vol.41, No.1, 37-49.
97

Smith and Fuzzy Control For Complex Systems: Huang Xiaogang

Загружено:

Сведения о документе

Исходное описание:

Оригинальное название

Авторское право

Доступные форматы

Поделиться этим документом

Поделиться или встроить документ

Параметры публикации

Этот документ был вам полезен?

Это неприемлемый материал?

Авторское право:

Доступные форматы

Smith and Fuzzy Control For Complex Systems: Huang Xiaogang

Загружено:

Авторское право:

Доступные форматы

SMITH AND FUZZY CONTROL

FOR COMPLEX SYSTEMS

NATIONAL UNIVERSITY OF SINGAPORE

SMITH AND FUZZY CONTROL

HUANG XIAOGANG (B.ENG., M.ENG.)

NATIONAL UNIVERSITY OF SINGAPORE

I would like to express my sincere appreciation and gratitude to my supervisor,

List of Figures vii

2 Modified Smith Predictor Control for General Unstable Processes 7

3 Robust PI controller design for nonlinear systems via fuzzy mod-

4 Fuzzy Modelling and Control of F-16 Aircraft 47

2.1 Modified Smith predictor control scheme . . . . . . . . . . . . . . . 8

4.1 Definition of aircraft axes and angles. . . . . . . . . . . . . . . . . . 53

different performance specifications or different methods used accordingly. Notice

control of complex systems, for example, in the control of robot manipulators

fuzzy system model.

A. Modified Smith Predictor Control for General Unstable Processes

B. Robust PI Controller Design for Nonlinear Systems via Fuzzy

C. Fuzzy Modelling and Control of F-16 Aircraft

1.3 Organization of the Thesis

Modified Smith Predictor Control

Figure 2.1. Modified Smith predictor control scheme

2.2 The Scheme

that the system in Fig. 2.1 consists of five subsystems:

K1 (s), Ĝ0 , K2 (s), G0 e−Ls and e−Ls .

Their respective pi are

p1 = n(s), p2 = b̂(s), p3 = β(s), p4 = b(s), and p5 = 1.

The system has the following four loops:

= 1 + K1 (s)Ĝ0 + K2 (s)G0 e−Ls + K1 (s)G0 e−Ls − K1 (s)Ĝ0 e−Ls

+K1 (s)Ĝ0 K2 (s)G0 e−Ls

In the case of perfect modeling, i.e., G0 = Ĝ0 , we have

∆ = [1 + K1 (s)G0 ] 1 + K2 (s)G0 e−Ls

= [n(s)b(s) + m(s)a(s)][β(s)b(s) + α(s)a(s)e−Ls ].

The polynomial, n(s)b(s) + m(s)a(s), reflects stabilization of delay-free G 0 by the

Figure 2.2. Conventional feedback system

Y (s) = H̃r (s)R(s) + H̃d (s)D(s),

Y (s) = Hr (s)R(s) + Hd (s)D(s),

G0 [e−Ls + K1 (s)G0 (e−Ls − e−2Ls )]

Hd G0 [e−Ls + K1 (s)G0 (e−Ls − e−2Ls )] 1 + K(s)G0 e−Ls

[1 + K1 (s)G0 (1 − e−Ls )][1 + K(s)G0 e−Ls ]

Take, as usual, K1 (s) as a PID type:

Then, pc (s) for this modified Smith system is

= [1 + K1 (s)G0 ][1 + K2 (s)G0 e−Ls ] · s · (T1 s − 1)(T2 s + 1) · (T3 s + 1)

·s · (T1 s − 1)(T2 s + 1) · (T3 s + 1) · (T1 s − 1)(T2 s + 1) · 1

(T1 T2 + kd kT3 )s3 + (T1 − T2 + kp kT3 + kd k)s2 + (kTi T3 + kp k − 1)s + kTi

· T1 s + kf ke−Ls − 1 (T3 s + 1)(T2 s + 1).

Obviously, pc1 = [(T1 T2 + kd kT3 )s3 + (T1 − T2 + kp kT3 + kd k)s2 + (kTi T3 +

To be specific, let T1 = 10, k = 1 and T2 = T3 to get the same model as in

= [1 + K1 (s)G0 ][1 + K2 (s)G0 e−Ls ] · s · (10s − 1) · 1 · (10s − 1) · 1

10s2 + (kp − 1)s + Ti 10s + kf e−Ls − 1

F (s, L) = 10s − 1 + kf e−Ls = A(s) + B(s)e−Ls = 0.

One sees that F (s, 0) = 10s − 1 + kf is stable when kf > 1. From

W (w2 ) = A(iw)A(−iw) − B(iw)B(−iw) = 100w 2 − kf2 + 1 = 0,

Figure 2.3. Step responses of modified Smith predictor control system

Figure 2.4. Step responses of conventional feedback system

2.4 Internal Stability Analysis of Two Modified

2.4.1 Astrom’s Modified Smith Predictor

A block diagram of Astrom’s modified Smith predictor is shown in Fig. 2.5.

Figure 2.5. The Smith predictor proposed by Astrom

The system consists of five subsystems:

Let M (s) = a(s)/b(s) be comprime polynomial fraction. Their respective p i

The system has the following two loops: