Вы находитесь на странице: 1из 47

具有非線性及截止區特性之輸入之非線性時間延遲

系統之模式參考適應模糊滑動模式控制

MODEL REFERENCE ADAPTIVE FUZZY SLIDING MODE


CONTROL OF TIME-DELAY UNCERTAIN NONLINEAR
SYSTEMS WITH INPUT CONTAINING SECTOR
NONLINEARITIES AND DEAD-ZONE

研究生:張碩傑(Shuo-Chieh Chang)

指導教授:龔宗鈞 教授(Prof. Chung-Chun Kung)

大同大學
電機工程研究所
碩士論文

Thesis for Master of Science


Department of Electrical Engineering
Tatung University

中華民國九十五年七月
July 2006
摘要

本論文是探討具有非線性輸入及截止區特性的非線性時間延遲系統,設計

一個模式參考之適應模糊滑動模式控制器。其中,系統的輸入訊號是一種具有扇

形特性的非線性函數。由於物理現象的關係,截止區的特性在是相當常見的,因

此在本論文中,非線性輸入項不僅僅包含了具有扇形特性的非線性函數且也同時

具有截止區的特性。而對於系統整體的擾動,則假設具有未知的上界。

根據 Lyapunov 穩定度定理分析的觀念,我們設計一個模式參考之適應模糊

滑動模式控制器,使系統達到追踪的效能。藉由滑動模式控制優越的強健性,克

服了系統延遲及不確定性的問題。此外,我們利用適應模糊技術來估測系統整體

的擾動。最後,為了展現本文所提出控制器的性能,我們提出一個例子來驗證其

成效。

i
ABSTRACT
In this thesis, a model reference adaptive fuzzy sliding mode control

(MRAFSMC) design for a class of time-delay uncertain nonlinear systems with input

containing sector nonlinearities and dead-zone is investigated. First, the fuzzy logic

systems are applied to estimate the boundary of the lumped uncertainties. Next, based

on the Lyapunov stability theorem, a MRAFSMC law to solve the control problem is

proposed. Moreover, the chattering around the sliding hyperplane in the sliding mode

control (SMC) can be reduced by the proposed design approach. The proposed

strategy not only possesses the advantages of SMC and adaptive fuzzy control, but

also eliminates the disadvantage of traditional SMC. A numerical example is given to

illustrate the feasibility of the proposed control strategy.

ii
誌謝

記得自己剛接觸自動控制這門學問已經過了兩年了,而自動控制這門學問對

我而言就好像是在品嘗一杯香濃的美式咖啡一樣,剛喝下去只覺得苦澀而已,但

是到了喉嚨一股清香湧入我的腦中,從此就欲罷不能。

在大同大學我們所學習的不再只是課本中的知識,而是研究的精神,生活的

態度及提高同學們之間的向心力。所以我首先要感謝大同大學提供我這麼好的學

習環境,而更要感謝我的指導教授,龔宗鈞老師,在我學習的過程中不斷的給我

鼓勵和意見,從老師身上學到了負責,細心且追根究底的學習態度。同時也非常

感謝呂虹慶教授,江江盛教授,及游文雄教授的指導。而這兩年來,在 800 實驗

室陪我度過的學長、同學、學弟們,我感到很榮幸認識了你們,你們的歡笑與快

樂永遠會留在我的心中。

最後,我要致上最誠摯的心給我的家人和怡鳳,感謝他們能夠體諒我的任性

及我的壞脾氣,沒有他們的支持,我今天就無法順利獲得碩士學位。在此,謹以

此論文獻給所有關愛我的人。

張碩傑

民國九十五年七月

大同大學

iii
TABLE OF CONTENTS

CHINESE ABSTRACT ……….……………….……………………………i

ENGLISH ABSTRACT ……….……………….…………………………...ii

ACKNOWLEDGEMENTS………………………………………………...iii

TABLE OF CONTENTS……….……………….………………………….iv

LIST OF FIGURES ……….……………….………………………………v

CHAPTER

I INTRODUCTION ……….……………….………………………1

II SYSTEM DESCRIPTION..…………….…………………………4

III DESIGN OF THE MODEL REFERENCE ADAPTIVE FUZZY

SLIDING MODE CONTROLLER ……….…….…………………8

3.1 Model Reference Sliding Mode Controller Design ……...……….8

3.2 Fuzzy Logic Systems …………………………………….….12

3.3 Model Reference Adaptive Fuzzy Sliding Mode Controller

(MRAFSMC) Design ………………………………………..14

IV SIMULATION EXAMPLE ………..…………………………….23

V CONCLUSIONS AND FUTURE STUDY …………….…………..36

REFERENCES ……...…………………………………………………....37

iv
LIST OF FIGURES

Fig. 1.1 The nonlinear continuous function Φ (u ) for control input ....……......7

Fig. 3.1 The overall scheme of the proposed controller …………………...….22

Fig. 4.1 The membership functions of x1 (t ) and x2 (t ) .……..……………..25

Fig. 4.2 The trajectories of x1 (t ) and xm1 (t ) for Case 1 ……………………29

Fig. 4.3 The trajectories of x2 (t ) and xm 2 (t ) for Case 1 ………………...…29

Fig. 4.4 The trajectory of the tracking error e1 (t ) for Case 1 ………...……...30

Fig. 4.5 The trajectory of the tracking error e 2 (t) for Case 1 ………………..30

Fig. 4.6 The phase plane between e1 (t) and e 2 (t) for Case 1 ………………31

Fig. 4.7 The nonlinear input Φ (u ) of the system for Case 1 …………..……31

Fig. 4.8 The control law u (t ) of the system for Case 1 ………………….....32

Fig. 4.9 The trajectories of x1 (t ) and xm1 (t ) for Case 2 ……………………32

Fig. 4.10 The trajectories of x2 (t ) and xm 2 (t ) for Case 2 ……………….…..33

Fig. 4.11 The trajectory of the tracking error e1 (t) for Case 2………………...33

Fig. 4.12 The trajectory of the tracking error e 2 (t) for Case 2 …………….....34

Fig. 4.13 The phase plane between e1 (t) and e 2 (t) for Case 2 .……..………34

Fig. 4.14 The nonlinear input Φ (u ) of the system for Case 2 .…………..……35

Fig. 4.15 The control law u (t ) of the system for Case 2 ……………………..35

v
CHAPTER 1

INTRODUCTION
Most studies of the robust control are carried out to stabilize the uncertain

nonlinear systems without nonlinear inputs and dead-zone [1-4]. In practice, due to

physical limitation, there do existence nonlinearities in the control input, such as

saturation, quantization, backlash, dead-zone and their effect cannot be ignored in

analysis of realization. The existence of input nonlinearities is a source of degradation or,

even worse, instability in system performance. Consequently, the stabilization of

uncertain nonlinear systems with input containing sector nonlinearities and dead-zone has

become of great interest in the recent years [5-9]. The problem of controlling nonlinear

systems with sector nonlinearities using a variable structure control approach has been

addressed in [5]. The use of fuzzy sliding mode control (FSMC) theory in dealing with

uncertain time-delayed systems with nonlinear input was proposed in [6]. In [7], a model

reference adaptive controller is applied to treat a continuous-time system with an

unknown input dead-zone. In [8], a robust stability criterion for uncertain large-scare

systems containing a dead-zone is given and the variable structure control is proved to be

applicable. Recently, it has been shown that SMC could make the nonlinear systems with

multiple inputs containing sector nonlinearities and dead-zones not only asymptotically

stable but also exponentially stable [9].

Time-delay commonly exists in various engineering systems, such as chemical

processes, electrical network, nuclear reactor, manual control, long transmission lines in

pneumatic, rolling mills and hydraulic system, etc. Its existence is frequently a source of

poor system performance, or instability. In order to obtain a more practical system model,

the information of , the existence of time-delay in systems should not be discarded.

1
Hence, the control problem of time-delay systems has received considerable attention

over the past years, and different design approaches have been proposed [1, 3, 4, 6,

10-12].

In addition to the input nonlinearities and the time-delay problem, we may be

facing the parameter uncertainty and the external disturbance of the system. When

designing a traditional SMC, the designers usually need the information of the upper

bound of lumped uncertainties consisted of parameter uncertainty and external

disturbance in order to guarantee the stability of the control systems [5, 6, 8, 9]. But in

practice, the information of such as upper bound is not easy to be obtained. Therefore, in

this thesis we will employ an adaptive fuzzy system [13] to estimate this upper bound of

lumped uncertainties.

Adaptive fuzzy control system designs have been extensively discussed in the

literature [13-15, 17, 18]. The fundamental idea of adaptive fuzzy control is as follows:

based on the universal approximation theorem [13], one firstly constructs a fuzzy model

to describe the input/output behavior of the controlled system. After that a controller is

designed based on the fuzzy model, and then design the adaptive laws to adjust the

parameters of the fuzzy models. In an effort to improve the robustness of the adaptive

fuzzy control system, many works have been published on the design of adaptive fuzzy

sliding mode controller [17, 18], which integrates the sliding mode controller [2, 4, 5, 6,

8-11] design technique into the adaptive fuzzy control to improve the stability and the

robustness of the control system.

In this thesis, we will investigate a MRAFSMC design for a class of time-delay

uncertain nonlinear systems with input containing sector nonlinearities and dead-zone.

First, we will employ adaptive fuzzy technique to estimate the upper bound of the lumped

uncertainties. Next, based on the Lyapunov stability theorem [20], we will propose a

2
systematic and effective strategy to solve the tracking problem for a class of time-delay

uncertain nonlinear systems with input containing sector nonlinearities and dead-zone.

Then, the adaptive fuzzy technique is applied in our propose control strategy so that the

tracking objective can be achieved. Finally, a numerical example is given to illustrate the

feasibility of the proposed control strategy.

This thesis is organized as follows: In chapter 2, the system description is presented.

In Chapter 3, we propose the model reference adaptive fuzzy sliding mode controller

design. In Chapter 4, a simulation example is given to illustrate the feasibility of the

propose control method. The conclusions and future study are given in Chapter 5.

3
CHAPTER 2

SYSTEM DESCRIPTION
Consider the following time-delay uncertain nonlinear system with input

containing sector nonlinearities and dead-zone:

x& (t ) = ( A + ΔA(x, t ))x(t ) + ( Aτ + ΔAτ (x, t ))x(t − h) + BΦ (u ) + d(t ) (2.1)

x(t ) = θ (t ) , t ∈ [− h ,0]

where x(t ) ∈ R n is the state vector of the system which is assumed to be available for

measurement, A ∈ R n×n , Aτ ∈ R n×n , and B ∈ R n are known constant matrices,

ΔA(x, t ) ∈ R n×n and ΔAτ (x, t ) ∈ R n×n are matrices representing time-varying parameter

uncertainties of the matrix A and Aτ , respectively, θ (t ) is a known continuous state

vector for specifying initial function, u (t ) ∈ R is a control input signal, Φ (u ) is a

continuous nonlinear input function, and satisfying Φ (0) = 0 , d(t ) ∈ R n is the external

disturbance, and h ∈ R + is the time-delay, and is bounded as

0<h≤h <∞

in which h is a known positive constant.

Now, let a reference model be given by

x& m (t ) = A m x m (t ) + B m r (t ) (2.2)

where x m (t ) ∈ R n is the state vector of the reference model, r (t ) ∈ R is the known

bounded input to the reference model, A m and B m are known and real constant

matrices with appropriate dimensions. In addition, A m is assumed to be a strictly

Hurwitz matrix [11], i.e., all the eigenvalues of A m lie in the left half-plane of the

complex plane.

4
Control objective: The control objective is to design a control law u (t ) for (2.1)

such that the state vector x(t ) can track the reference state vector x m (t ) .

Define the tracking error vector as

e(t ) = x(t ) − x m (t ) . (2.3)

Hence, the control objective is to design a control law for (2.1) so that lim e(t ) = 0 .
t →∞

Differentiating (2.3) with respect to time and using (2.1) and (2.2), we obtain the dynamic

equation of tracking error as

e& (t ) = ( A + ΔA(x, t ))x(t ) + ( Aτ + ΔAτ (x, t ))x(t − h) + BΦ (u ) + d(t ) − A m x m (t ) − B m r (t )


= ( A + ΔA(x, t ))x(t ) + ( Aτ + ΔAτ (x, t ))x(t − h) + BΦ (u ) + d(t )
− A m x m (t ) + A m x(t ) − A m x(t ) − B m r (t )
= A me(t ) + ( A − A m + ΔA(x, t ))x(t ) + ( Aτ + ΔAτ (x, t ))x(t − h)
+ BΦ (u ) + d(t ) − B m r (t ) . (2.4)

To achieve the control objective, the following assumptions are required.

Assumption 1 (Chou [10]). For the time-delay uncertain nonlinear system in (2.1),

the matrix pair ( A, B) and ( Aτ , B) are completely controllable.

Assumption 2 (Khalil [20]). The uncertain matrices ΔA(x, t ) and ΔAτ (x, t ) are

continuously differentiable in x , and are piecewise continuous in t.

Assumption 3 (Drazenovic [21]). There exist continuous vectors ΔA (x, t ) ,

ΔAτ (x, t ) , H m , the function d (t ) and the scalar k m such that the following matching

conditions are satisfied:

ΔA(x, t ) = BΔA (x, t )


ΔAτ (x, t ) = B ΔAτ (x, t )
A − A m = BH m
d(t ) = B d (t )
B m = B km .

Note that ΔA (x, t ) , ΔAτ (x, t ) , and d (t ) are unknown. However, H m and k m can be

5
obtained.

Remark 1. The existence of vectors ΔA (x, t ) , ΔAτ (x, t ) , H m , function d (t ) and

scalar k m are guaranteed if and only if

rank(B) = rank(B | ΔA(x, t ))


= rank(B | ΔAτ (x, t ))
= rank (B | d(t ))
= rank (B | A − A m )
= rank (B | B m ).

According to the above assumptions, the tracking error given by (2.4) can be

rewritten as

e& (t )= A me(t ) + BH m x(t ) + BΔA (x, t )x(t ) + Aτ x(t − h) + BΔAτ (x, t )x(t − h)
+ BΦ (u ) + B d (t ) − B k m r (t )
= A me(t ) + Aτ x(t − h) + BΦ(u ) + B( ΔA (x, t )x(t ) + ΔAτ (x, t )x(t − h) + d (t ) )
+ BH m x(t ) − B k m r (t )
= A me(t ) + Aτ x(t − h) + B Φ (u ) + BW (x, t ) + BH m x(t ) − B k m r (t ) (2.5)

where W (x, t ) = ΔA (x, t )x(t ) + ΔAτ (t )x(t − h) + d (t ) is the lumped uncertainties. For

the lumped uncertainties, we make the following assumption:

Assumption 4 (Shyu et al. [8]). There exist two positive constants g 0 and g1

such that

W (x, t ) ≤ g 0 + g1 x(t ) . (2.6)

Assumption 5 (Hsu et al. [9]). The nonlinear continuous input function Φ (u )

shown in Fig. 2.1 is assumed to satisfy the following property:

Φ (u ) = 0 for − u 0− ≤ u ≤ u 0+ , (2.7)

α 1 (u − u 0+ ) 2 ≤ (u − u 0+ )Φ(u ) ≤ α 2 (u − u 0+ ) 2 , for u > u 0+ , (2.8)

and

α 1 (u + u 0− ) 2 ≤ (u + u 0− )Φ (u ) ≤ α 2 (u + u 0− ) 2 , for u < −u 0− (2.9)

6
where 0 ≤ α1 ≤ α 2 .

Remark 2. In order to simplify the control problem, we assume that α 2 → ∞ ,

α 1 = α , and u 0+ = u 0− = u 0 . Then (2.7), (2.8) and (2.9) can be rewritten as follows:

Φ (u ) = 0 for − u0 ≤ u ≤ u0 , (2.10)

(u − u 0 )Φ (u ) ≥ α (u − u 0 ) 2 , for u > u 0 , (2.11)

and

(u + u 0 )Φ (u ) ≥ α (u + u 0 ) 2 , for u < −u 0 . (2.12)

If the nonlinear input function Φ (u ) satisfies (2.10), (2.11) and (2.12), then such

nonlinear input function not only contains a dead-zone but also allows the nonlinear input

outside of the dead-zone with gain reduction tolerance α .

Φ (u )

slope α 2

slope α 1
− u0−
u
slope α 1 u 0+

slope α 2

Fig. 2.1. The nonlinear continuous function Φ (u ) for control input.

7
CHAPTER 3

DESIGN OF THE MODEL REFERENCE

ADAPTIVE FUZZY SLIDING MODE

CONTROLLER

3.1 Model Reference Sliding Mode Controller Design

Let a function s (t ) as

s (t ) = CT e(t ) (3.1)

where C ∈ R n will be appropriately selected so that det(CT B) ≠ 0 . Define a sliding

hyperplane in the space of the error state as

s (t ) = CT e(t ) = 0 . (3.2)

Differentiating (3.2) with respect to time, we get

C T e& (t ) = C T [ A m e(t ) + A τ x(t − h) + BΦ (u ) + BW (x, t ) + BH m x(t ) − B k m r (t ) ] . (3.3)

Once the above the sliding hyperplane s (t ) = 0 is reached, it is always accompanied

s&(t ) = 0 , i.e.,

s&(t ) = CT [ A me(t ) + Aτ x(t − h) + BΦ (u ) + B W (x, t ) + BH m x(t ) − B k m r (t ) ] = 0 . (3.4)

If W (x, t ) is known, an equivalent control Φ (ueq ) can be generated by (3.4).

Thus the equivalent control is

Φ (ueq ) = −[ (CT B) −1 CT A me(t ) + (CT B) −1 CT Aτ x(t − h) + W (x, t ) + H m x(t ) − k m r (t ) ] .(3.5)

Remark 3 (Hsu [5]). It is noted that the equivalent control Φ (ueq ) is only a

mathematically derived tool for the analysis of a sliding motion rather than a real control

8
law being generated in practical systems. In fact, Φ (ueq ) is not realizable through a

nonlinear controller even if the system is nominal, or the system is in the absence of

lumped uncertainties. Therefore, the equivalent control generates an “ideal” sliding

motion on the sliding hyperplane while the real sliding mode controller generates a

trajectory close to the ideal sliding motion around the sliding hyperplane.

Introducing (3.5) into error dynamic system (2.5), we can obtain the equivalent

dynamic equation for the error system as

e& (t ) = A me(t ) + Aτ x(t − h) + BΦ (ueq ) + B W (x, t ) + BH m x(t ) − B k m r (t )


= A me(t ) + Aτ x(t − h) − B[ (CT B) −1 CT A me(t ) + (CT B) −1 CT Aτ x(t − h)
+ W (x, t ) + H m x(t ) − k m r (t ) ] + B W (x, t ) + BH m x(t ) − B k m r (t )
= [ I − B(CT B) −1 CT ]A me(t ) (3.6)

where I is an n × n identity matrix. Equation (3.6) shows that the invariance condition

[21].

Lemma. The motion of the sliding hyperplane (3.2) is asymptotically stable, if the

following condition is held:

s (t ) s&(t ) < 0, ∀t ≥ 0 . (3.7)

Proof. Let

1 2
V (t ) = s (t ) (3.8)
2

be the Lyapunov function for the system described in (2.5). According to Lyapunov

stability theorem, condition of (3.7) ensures that

V& (t ) = s (t ) s&(t ) < 0 . (3.9)

Then, s (t ) will decay to zero, and the motion of the sliding hyperplane (3.2) is

asymptotically stable. This completes the proof.

Now, consider the control problem of (2.1), if W (x, t ) is bounded such that

W (x, t ) ≤ g 0 + g1 x(t ) (3.10)

9
where g 0 and g1 are known positive constants. According to (3.10), we will propose

the control strategy for (2.1). To treat the dead-zone of the input, we define the controller

as follows:

⎧ sC T B
⎪− φ (x, t ) + u0 , for sCT B < 0
⎪ sC B
T


u (t ) = ⎨ 0 , for sCT B = 0 (3.11)
⎪ sC T B
⎪− φ (x, t ) − u0 , for sCT B > 0
⎪⎩ sC B
T

where

β
φ ( x, t ) = [ (CT B) −1 CT A m e(t ) + (CT B) −1 CT Aτ x sup (t ) + ( H m + g1 ) x(t )
α
− k m r (t ) + g 0 ] , β > 1, α > 0.

Substituting (3.11) into (2.10) and (2.11), we have

sC T B
(u − u0 )Φ (u ) = − φ (x, t )Φ(u ) ≥ α (u − u0 ) 2 = αφ (x, t ) 2 , for u > u 0 , (3.12)
sC B
T

and

sC T B
(u + u0 )Φ (u ) = − φ (x, t )Φ (u ) ≥ α (u + u0 ) 2 = αφ (x, t ) 2 , for u < −u0 . (3.13)
sC B
T

Therefore, we obtain

− sCT B α φ (x, t ) ≥ sCT B Φ (u ) . (3.14)

Theorem 1. Consider the time-delay uncertain nonlinear system (2.1) and suppose

(3.10) is satisfied. If the SMC laws are chosen as (3.11), then the tracking error

trajectories globally asymptotically converge to the sliding hyperplane (3.2).

Proof. Consider the Lyapunov function given as

1 2
V (t ) = s (t ) . (3.15)
2

Differentiating (3.14) with respect to time, we get

10
V& (t ) = s (t ) s&(t ) . (3.16)
By noting that

x(t − h) ≤ xsup (t ) Δ sup x(τ ) ,


t − h ≤τ ≤ t

and substituting (3.3) into (3.16), then we get

V& (t ) = sCT [ A me(t ) + Aτ x(t − h) + BΦ (u ) + B W (x, t ) + BH m x(t ) − B k m r (t ) ]


= s (CT B)(CT B) −1 CT [ A me(t ) + Aτ x(t − h) + BΦ (u ) + B W (x, t )
+ BH m x(t ) − B k m r (t ) ]
= s (CT B)(CT B) −1 CT [ A me(t ) + Aτ x(t − h) + BW (x, t ) + BH m x(t )
− B k m r (t ) ] + sCT BΦ (u )
≤ sCT B [ (CT B) −1 CT A me(t ) + (CT B) −1 CT Aτ x(t − h) + ( g 0 + g1 x(t ) )
+ H m x(t ) − k m r (t ) ] + sCT BΦ (u )
≤ sCT B [ (CT B) −1 CT A m e(t ) + (CT B) −1 CT Aτ x sup (t )
+ ( H m + g1 ) x(t ) − k m r (t ) + g 0 ] + sC BΦ (u ) . T
(3.17)

Form (3.14), we obtain

− sCT B α φ (x, t ) ≥ sCT B Φ (u ) .

So, (3.17) can be rewritten as

V& (t ) ≤ sC T B [ (C T B) −1 C T A m e(t ) + (C T B) −1 C T Aτ x sup (t ) + ( H m + g1 ) x(t )


− k m r (t ) + g 0 ] − sC T B α φ (x, t )
= sC T B [ (C T B) −1 C T A m e(t ) + (C T B) −1 C T Aτ x sup (t ) − α φ (x, t )
+ ( H m + g1 ) x(t ) − k m r (t ) + g 0 ] . (3.18)

According to (3.11), we have

β
φ (x, t ) = [ (CT B) −1 CT A m e(t ) + (CT B) −1 CT Aτ x sup (t ) + ( H m + g1 ) x(t )
α
− km r (t ) + g 0 ] . (3.19)

Substituting (3.19) into (3.18), we have

11
V& (t ) ≤ sC T B [ (C T B) −1 C T A m e(t ) + (C T B) −1 C T Aτ x sup (t )
β
− α ( ( (C T B) −1 C T A m e(t ) + (C T B) −1 C T Aτ x sup (t )
α
+ ( H m + g1 ) x(t ) − k m r (t ) + g 0 ))
+ ( H m + g1 ) x(t ) − k m r (t ) + g 0 ]
= (1 − β ) sC T B [ (C T B) −1 C T A m e(t ) + (C T B) −1 C T Aτ x sup (t )
+ ( H m + g1 ) x(t ) − k m r (t ) + g 0 ] < 0 . (3.20)

Because β > 1 , V& (t ) < 0 . Form Lemma, we confirm the error trajectories asymptotically

converge to the sliding hyperplane (3.2) using the SMC law (3.11). This completes the

proof.

In traditional SMC design, it is necessary to know the information of the upper

bound of (2.6) for controller design. But in practice, the information of the upper bound is

not easy to be obtained and hence g 0 and g1 are unknown. So, we employ adaptive

fuzzy technique to estimate g 0 and g1 , respectively.

3.2 Fuzzy Logic Systems

Since the g 0 and g1 are unknown, we apply the following two fuzzy systems

gˆ 0 (x θ gˆ 0 ) and gˆ1 (x θ gˆ 1 ) to estimate g 0 and g1 , respectively. The l th fuzzy rule of

gˆ 0 (x θ gˆ 0 ) is written as

R lgˆ 0 : IF x1 is Fxl1 and L and xn is Fxnl , THEN gˆ 0 (x θ gˆ 0 ) is Gglˆ 0 (3.21)

and the l th fuzzy rule of gˆ1 (x θ gˆ1 ) is written as

R lgˆ1 : IF x1 is Fxl1 and L and xn is Fxnl , THEN gˆ1 (x θ gˆ 1 ) is Gglˆ 1 (3.22)

where l = 1,2, L q , q is the total number of fuzzy rules for each of the fuzzy model,

Fxil ( i = 1, L , n ) are the fuzzy sets associated with xi ( i = 1, L , n ), and Gglˆ 0 and Gglˆ 1

12
are fuzzy singletons for gˆ 0 (x θ gˆ 0 ) and gˆ1 (x θ gˆ 1 ) , respectively. By using the singleton

fuzzifier, the product inference and the center average defuzzifier [13], the output of

gˆ 0 (x θ gˆ 0 ) and gˆ1 (x θ gˆ1 ) can be respectively obtained by

gˆ 0 (x θ gˆ 0 ) = θ Tgˆ 0 ξ x (x) , (3.23)

and

gˆ1 (x θ gˆ1 ) = θ Tgˆ1 ξ x (x) , (3.24)

where θ gˆ 0 = [G1gˆ 0 , Gg2ˆ0 ,L, Ggqˆ 0 ]T and θ gˆ 1 = [G1gˆ 1 , Gg2ˆ1 ,L, Ggqˆ 1 ]T are the adjustable

parameter vectors, and ξ x (x) = [ξ x1 (x), ξ x2 (x),L, ξ xq (x)]T is the vector of fuzzy basis

function [13] defined as

ξ (x) =
l ∏ F (x ) n
i =1
l
xi i
, l = 1,2, K q , (3.25)
∑ [∏ F ( x )]
x q n l
l =1 i =1 xi i

with Fxl i ( xi ) represents the membership function value of x i in Fxl i . Without loss of

generality, we make the following assumption:

Assumption 6 (Wang [13], Li and Tong [15], Kung et al [17], Kung and Chen [18]).

It is known a prior that the optimal vectors θ∗ĝ 0 and θ∗ĝ1 lie in some convex regions

Ω gˆ 0 = { θ gˆ 0 ∈ R q θ gˆ 0 ≤ r0 } , (3.26)

Ω gˆ1 = { θ gˆ1 ∈ R q θ gˆ1 ≤ r1} , (3.27)

in which the radiuses r0 and r1 are designed positive constants and θ∗ĝ0 and θ∗ĝ1 are

defined as

θ∗gˆ 0 = arg min [ sup gˆ 0 (x θ gˆ 0 ) − g 0 ] (3.28)


θ gˆ 0 ∈Ω gˆ 0 x∈U x

and

13
θ∗gˆ 1 = arg min [ sup gˆ1 (x θ gˆ 1 ) − g1 ] . (3.29)
θ gˆ 1 ∈Ω gˆ 1 x∈U x

Define the minimum approximation error as [13]:

w = [ g 0 − θ*gˆT0ξ x (x) ] + [ g1 − θ*gˆT1ξ x (x) ] x(t ) . (3.30)

3.3 Model Reference Adaptive Fuzzy Sliding Mode

Controller (MRAFSMC) Design

The result in (3.11) is realizable only while g 0 and g1 are known. However, g 0

and g1 are unknown an ideal controller (3.11) cannot be implemented. So we replace

the g 0 and g1 in (3.11) by ĝ 0 and ĝ1 , and hence the resulting controller is given as

⎧ sC T B
⎪− φ (x, t ) + u0 , for sCT B < 0
⎪ sC B
T


u (t ) = ⎨ 0 , for sCT B = 0 (3.31)
⎪ sC BT

⎪− φ (x, t ) − u0 , for sCT B > 0


⎪⎩ sC B
T

where

β
φ (x, t ) = [ (C T B) −1 C T A m e(t ) + (C T B) −1 C T A τ x sup (t ) + H m x(t )
α
1
− km r (t ) ] + [ gˆ 0 + gˆ 1 x(t ) + sC T B ], β > 1, α > 0.
α

Then, we can obtain the following main result:

Theorem 2. Consider the nonlinear system given as (2.1). Suppose that Assumptions

1-6 are satisfied and w ∈ L2 [0, T ] , ∀T ∈ [0, ∞) . The MRAFSMC laws are chosen as

(3.31). The two adaptive laws are given by

14
⎧ rgˆ sC T B ξ x (x) if { θ g) 0 < r0 }
⎪ 0
⎪ or{ θ g)0 = r0 and sC T B θ Tg)0 ξ x ≤ 0 }, (3.32a)

θ& gˆ 0 = ⎨ rgˆ 0 sC T B ξ x (x)

⎪ θ g) 0 θ Tg) 0
⎪ − rgˆ 0 sC B T) ) ξ x (x) if { θ g)0 = r0 and sC T B θ Tg)0 ξ x > 0 },
T
(3.32b)
⎩ θ θ
g0 g0

⎧ rgˆ sC T B ξ x (x) x(t ) if { θ g)1 < r1 }


⎪ 1
⎪ or{ θ g)1 = r1

⎪ and sC T B θ Tg)1 ξ x x(t ) ≤ 0 }, (3.33a)

θ& gˆ1 = ⎨ rgˆ1 sC T B ξ x (x) x(t )

⎪ θ g)1 θ Tg)1
⎪− rgˆ1 sC B T) ) ξ x (x) x(t ) if { θ g)1 = r1
T

⎪ θ g1 θ g1
⎪ and sC T B θ Tg)1 ξ x x(t ) > 0 }, (3.33b)

where rĝ 0 and rĝ1 are positive constants specified by designer. Then, the error

trajectories globally asymptotically converge to the sliding hyperplane (3.2).

Proof. Consider the Lyapunov function given as

1 2 1 ~T ~ 1 ~T ~
V (t ) = s (t ) + θgˆ 0 θgˆ 0 + θgˆ θgˆ (3.34)
2 2rgˆ 0 2rgˆ 1 1 1

~ ~
where θgˆ 0 = θ gˆ 0 − θ*gˆ 0 and θgˆ 1 = θ gˆ 1 − θ*gˆ 1 .

Differentiating (3.34) with respect to time, we get

1 ~ T ~& 1 ~ ~&
V& (t ) = s (t ) s&(t ) + θgˆ 0 θgˆ 0 + θgTˆ1 θgˆ 1 . (3.35)
rgˆ 0 rgˆ 1
By noting that

~&
θgˆ 0 = θ& gˆ 0

~&
θgˆ 1 = θ& gˆ 1

x(t − h) ≤ xsup (t ) Δ sup x(τ ) ,


t − h ≤τ ≤ t

and using (3.3) in (3.35), then we get

15
V& (t ) = sC T [ A m e(t ) + A τ x(t − h) + BΦ (u ) + BW (x, t ) + BH m x(t ) − B k m r (t ) ]
1 ~T & 1 ~T &
+ θ gˆ 0 θ gˆ 0 + θ gˆ θ gˆ
rgˆ 0 rgˆ1 1 1
= s (C T B)(C T B) −1 C T [ A m e(t ) + A τ x(t − h) + BΦ (u ) + BW (x, t ) + BH m x(t )
1 ~T & 1 ~T &
− B k m r (t ) ] + θ gˆ 0 θ gˆ 0 + θ gˆ θ gˆ
rgˆ 0 rgˆ1 1 1
= s (C T B)(C T B) −1 C T [ A m e(t ) + A τ x(t − h) + BW (x, t ) + BH m x(t ) − B k m r (t ) ]
1 ~T & 1 ~T &
+ sC T BΦ (u ) + θ gˆ 0 θ gˆ 0 + θ gˆ θ gˆ
rgˆ 0 rgˆ1 1 1
≤ sC T B [ (C T B) −1 C T A m e(t ) + (C T B) −1 C T A τ x(t − h) + ( g 0 + g1 x(t ) ) + H m x(t )
1 ~T & 1 ~T &
− k m r (t ) ] + sC T BΦ (u ) + θ gˆ 0 θ gˆ 0 + θ gˆ θ gˆ
rgˆ 0 rgˆ1 1 1
≤ sC T B [ (C T B) −1 C T A m e(t ) + (C T B) −1 C T A τ x sup (t ) + ( H m + g1 ) x(t )
1 ~T & 1 ~T &
− km r (t ) + g 0 ] + sC T BΦ (u ) + θ gˆ 0 θ gˆ 0 + θ gˆ θ gˆ . (3.36)
rgˆ 0 rgˆ1 1 1
Form (3.14), we obtain

− sCT B α φ (x, t ) ≥ sCT BΦ(u ) .

So, (3.36) can be rewritten as

V& (t ) ≤ sC T B [ (C T B) −1 C T A m e(t ) + (C T B) −1 C T A τ x sup (t ) + ( H m + g1 ) x(t )


1 ~T & 1 ~T &
− km r (t ) + g 0 ] − sC T B α φ (x, t ) + θ gˆ 0 θ gˆ 0 + θ gˆ θ gˆ
rgˆ 0 rgˆ1 1 1
= sC T B [ (C T B) −1 C T A m e(t ) + (C T B) −1 C T A τ x sup (t ) − α φ (x, t ) + H m x(t )
1 ~T & 1 ~T &
− km r (t ) ] + g 0 sC T B + g1 sC T B x(t ) + θ gˆ 0 θ gˆ 0 + θ gˆ θ gˆ . (3.37)
rgˆ 0 rgˆ1 1 1
According to (3.31), we have

β
φ (x, t ) = [ (CT B) −1 CT A m e(t ) + (CT B) −1 CT Aτ x sup (t ) + H m x(t )
α
1
− km r (t ) ] + [ gˆ 0 + gˆ 1 x(t ) + sC T B ] . (3.38)
α

Substituting (3.38) into (3.37), we have

16
V& (t ) ≤ sCT B [ (CT B) −1 CT A m e(t ) + (CT B) −1 CT Aτ x sup (t )
β
−α( ( (CT B) −1 CT A m e(t ) + (CT B) −1 CT Aτ x sup (t )
α
+ Hm x(t ) − k m r (t ) ) ) + H m x(t ) − k m r (t ) ]
1
− sC T B α [ ( gˆ 0 + gˆ1 x(t ) + sCT B ) ] + g 0 sCT B
α
1 ~T & 1 ~
+ g1 sCT B x(t ) + θ gˆ 0 θ gˆ 0 + θ gTˆ1 θ& gˆ1
rgˆ 0 rgˆ1
= (1 − β ) sCT B [ (CT B) −1 CT A m e(t ) + (CT B) −1 CT Aτ x sup (t ) + H m x(t )
− km r (t ) ] + sCT B [ − θ Tgˆ 0 ξ x ( x) − θ Tgˆ1 ξ x (x) x(t ) − sCT B
+ θ Tgˆ *0 ξ x (x) − θ Tgˆ *0 ξ x ( x) + θ Tgˆ1*ξ x ( x) x(t ) − θ Tgˆ1*ξ x (x) x(t )
1 ~T & 1 ~
+ g 0 + g1 x(t ) ] + θ gˆ 0 θ gˆ 0 + θ gTˆ1 θ& gˆ1
rgˆ 0 rgˆ1
= (1 − β ) sCT B [ (CT B) −1 CT A m e(t ) + (CT B) −1 CT Aτ x sup (t ) + H m x(t )
− km r (t ) ] + sCT B [ − θ Tgˆ 0 ξ x ( x) − θ Tgˆ1 ξ x (x) x(t ) − sCT B
1 ~T & 1 ~
+ θ Tgˆ *0 ξ x (x) + θ Tgˆ1*ξ x ( x) x(t ) + w ] + θ gˆ 0 θ gˆ 0 + θ gTˆ1 θ& gˆ1
rgˆ 0 rgˆ1
= (1 − β ) sCT B [ (CT B) −1 CT A m e(t ) + (CT B) −1 CT Aτ x sup (t ) + H m x(t )
~ ~
− k m r (t ) ] + sCT B [ − θ gTˆ0 ξ x (x) − θ gTˆ1 ξ x (x) x(t ) − sCT B + w]
1 ~T & 1 ~
+ θ gˆ 0 θ gˆ 0 + θ gTˆ1 θ& gˆ1
rgˆ 0 rgˆ1
= (1 − β ) sCT B [ (CT B) −1 CT A m e(t ) + (CT B) −1 CT Aτ x sup (t ) + H m x(t )

~ θ& gˆ
− km r (t ) ] + sCT B [− sCT B + w] + θ gTˆ0 [− sCT B ξ x (x) + 0 ]
rgˆ 0
~ θ& gˆ
+ θ gTˆ1 [− sCT B ξ x (x) x(t ) + 1 ] (3.39)
rgˆ1
where
~ θ& gˆ ~ θ& gˆ
Vθ gˆ 0 = θ gTˆ0 [− sCT B ξ x (x) + 0 ] , Vθ gˆ1 = θ gTˆ1 [− sCT B ξ x (x) x(t ) + 1 ] .
rgˆ 0 rgˆ1

According to (3.32a), Vθ gˆ 0 = 0 holds. Substituting (3.32b) into Vθ ĝ0 gives

~T
θ gˆ 0 θ g)0
Vθ gˆ 0 = − sC B T
θ Tg)0 ξ x (x) .
θ Tg)0 θ g)0

Because θ gˆ 0 = r0 > θ*gˆ 0

17
~T 1~ 1~
θ gˆ 0 θ g)0 = θ gTˆ0 θ g)0 + θ gTˆ0 θ g)0
2 2
1 1 ~
= (θ g)0 − θ*gˆ 0 ) T θ g)0 + (θ g)0 − θ*gˆ 0 ) T ( θ g)0 + θ*g)0 )
2 2
1 1 1 ~ 1 1 ~ 1
= θ Tg)0 θ g)0 − θ Tg)0*θ g)0 + θ Tg)0 θ g)0 + θ Tg)0 θ*g)0 − θ Tg)0* θ g)0 − θ Tg)0*θ*g)0
2 2 2 2 2 2
1 2 1 2 1 ~ 1 ~
= θ g)0 − θ*g)0 + θ Tg)0 θ g)0 − θ Tgˆ 0*θ g)0
2 2 2 2
1 2 1 2 1
= θ g)0 − θ*g)0 + (θ g)0 − θ*g)0 ) T (θ g)0 − θ*g)0 )
2 2 2
1 2 1 2 1 2
= θ g)0 − θ*g)0 + θ g)0 − θ*g)0 > 0
2 2 2

it can be conclude that Vθ gˆ 0 ≤ 0 . Then, according to (3.33a), Vθ gˆ1 = 0 can be obtained.

Substituting (3.33b) into Vθ ĝ1 , it is possible to obtain

~T
θ g)1 θ g)1
Vθ gˆ 1 = − sC B T
θ Tg)1 ξ x (x) x(t ) .
θ θ
T
)
g1
)
g1

Because θ gˆ1 = r1 > θ *gˆ1


~T 1~ 1~
θ gˆ1 θ gˆ1 = θ gTˆ1 θ gˆ1 + θ gTˆ1 θ gˆ1
2 2
1 1 ~
= (θ gˆ1 − θ *gˆ1 ) T θ gˆ1 + (θ gˆ1 − θ *gˆ1 ) T ( θ gˆ1 + θ *gˆ1 )
2 2
1 1 1 T~ 1 1 ~ 1
= θ Tgˆ1 θ gˆ1 − θ T* gˆ1 θ gˆ1 + θ gˆ1 θ gˆ1 + θ Tgˆ1 θ *gˆ1 − θ Tgˆ1* θ gˆ1 − θ Tgˆ1*θ *gˆ1
2 2 2 2 2 2
1 2 1 2 1 ~ 1 ~
= θ gˆ1 − θ *gˆ1 + θ Tgˆ1 θ gˆ1 − θ Tgˆ 0* θ g) 0
2 2 2 2
1 2 1 2 1
= θ gˆ1 − θ *gˆ1 + (θ gˆ1 − θ *gˆ1 ) T (θ gˆ1 − θ *gˆ1 )
2 2 2
1 2 1 2 1 2
= θ gˆ1 − θ *gˆ1 + θ gˆ1 − θ *gˆ1 > 0
2 2 2

it can be conclude that Vθ gˆ 0 ≤ 0 . Therefore, (3.39) can be rewritten as

V& (t ) ≤ (1 − β ) sC T B [ (C T B) −1 C T A m e(t ) + (C T B) −1 C T A τ x sup (t ) + H m x(t )


− km r (t ) ] + sC T B [− sC T B + w]

18
2
≤ − sC T B + w sC T B
21 1
= − sC T B +
w sC T B + w sC T B
2 2
2 1 2 1
≤ − sC T B + sC T B + w 2
2 2
1 2 1
= − sC T B + w 2
2 2
1 2 1
= − (C T B) 2 s (t ) + w 2 . (3.40)
2 2

Integrating both sides of (3.40) and after some manipulations yields

∞ 1 ∞ 1 ∞
∫ V& (t ) dt ≤ − (CT B) 2 ∫ s (t ) dt + ∫ w2 dt .
2
(3.41)
0 2 0 2 0

Thus

∞ 2 1 ∞
∫ s (t ) dt ≤ [V (0) − V (∞) ] + T 2 ∫
2
w2 dt . (3.42)
0 (C B)
T 2
(C B) 0

If w ∈ L2 [0, T ] , ∀T ∈ [0, ∞) , then from (3.42) we have s (t ) ∈ L2 . It was seen all the

right-hand side of (3.3) are bounded, so s&(t ) ∈ L∞ . Using the Barbalat’s Lemma [19], we

have lim s (t ) = 0 , thus lim e(t ) = 0 . The MRAFSMC system is shown in Fig. 3.1.
t →∞ t →∞

Remark 4. The sector bounded function in (2.7), (2.9) and (2.10) is only a special

case of series nonlinear function defined in (2.10), (2.11) and (2.12). Therefore, the

proposed control laws (3.31) ensure that the systems with sector bounded nonlinearities

can be applied.

sC T B
Remark 5. Because (3.31) contains the sign function sgn( sCT B) = . The
sC T B

direct application of such controller may give rise to undesirable chattering problem [10],

owning to it is discontinuous on the sliding hyperplane. In order to treat the problem, we

sC T B
will replace the sgn( sCT B) , (3.31) by sat ( ) where the saturation function instead
ε

of the sign function is defined as follows [22]:

19
⎧ 1, for s > ε
s ⎪ s
sat ( ) = ⎨ , for − ε ≤ s ≤ ε
ε ⎪ ε
for s < −ε
⎩ − 1,

sC T B
where ε is a small positive constant. Once the switching term is used sat ( )
ε

instead of sgn( sCT B) , the introduction of boundary layer is a continuous approximation

to the discontinuous sliding mode with the bounder layer. As ε approaches to zero, the

performance of the modified switching control law can be made arbitrarily close to that

control law (3.31). Then, we can modify the two adaptive gains (3.32) and (3.33) by

following equations:



⎪ 0 if s ≤ ε,


⎪ r sC T B ξ ( x ) if s > ε and { θ g) 0 < r0 }
⎪ gˆ 0 x
θ& gˆ 0 = ⎨
⎪ or{ θ g) 0 = r0 and sC T B θ Tg) 0 ξ x ≤ 0 }, (3.43)

⎪rgˆ 0 sC B ξ x (x)
T


⎪− r sC T B θ g 0 θ g 0 ξ (x) if
T
) )
s > ε and { θ g) 0 = r0 and sC T B θ Tg)0 ξ x > 0 }.
⎪ gˆ 0
θ g) 0 θ g) 0
T x



⎪ 0 if s ≤ ε ,


⎪ r sC T B ξ (x) x(t ) if s > ε and { θ g)1 < r1 }
&θ = ⎪⎨ gˆ 1 x

or{ θ g)1 = r1 and sC T B θ Tg)1 ξ x ≤ 0 },
1
⎪ (3.44)

⎪rgˆ sC B ξ x (x) x(t )
T
1

⎪ θ ) θ T)
⎪− r sC T B g g ξ (x) x(t )
1 1
if s > ε and { θ g)1 = r1 and sC T B θ Tg)1 ξ x > 0 }.
⎪ gˆ 1
θ Tg) θ g)
x
⎩ 1 1

From the previous discussion, one can conclude that designer has the trade-off among

chattering, tracking accuracy, and adaptive performance.

To summarize the above analysis, the design procedures for the MRAFSMC

20
algorithm are proposed as follow.

Design procedure:

Step 1. Specify the desired constant vector C in (3.2).

Step 2. Define qi fuzzy sets Fxi for linguistic variable xi and the membership

functions Fxl i ( xi ) is uniformly cover the universe of discourse, for i = 1,2, L, n.

Then, construct the fuzzy systems gˆ 0 (x θ gˆ 0 ) = θ Tgˆ 0 ξ x (x) and

gˆ1 (x θ gˆ1 ) = θ Tgˆ1 ξ x (x) .

Step 3. Construct the control law (3.31).

Step 4. Construct the adaptive law in (3.43) and (3.44).

Step 5. Select the learning positive constants rĝ 0 and rĝ1 .

Step 6. Obtain the control law (3.31) and apply to the plant (2.1), then compute the

adaptive law (3.43-44) to adjust the parameter vector θ ĝ0 and θ ĝ1 .

21
Φ(u ) System x(t )
x& (t ) = ( A + ΔA(x, t ))x(t ) + ( Aτ + ΔAτ (x, t ))x(t − h) + BΦ (u ) + d(t )
(2.1)
+
Reference Input Reference Model x n (t ) -
Σ
r (t ) x& m (t ) = A m x m (t ) + B m r (t ) (2.2)

s (t ) SMC e(t )
s (t ) = CT e(t ) (3.2)

Adaptive Laws
&θ , θ& given as (3.43-44) x(t )
gˆ 0 gˆ 1

θ ĝ 0 θ ĝ1
Fuzzy Logic Systems
gˆ 0 (x θ gˆ 0 ) = θ Tgˆ 0 ξ x (x) (3.23)
gˆ1 (x θ gˆ1 ) = θ Tgˆ1 ξ x (x) (3.24)

gˆ 0 (x θ gˆ 0 ) gˆ1 (x θ gˆ1 )

β
φ (x, t ) = [ (CT B) −1 CT A m e(t ) + (CT B) −1 CT Aτ x sup (t )
α
1
+ Hm x(t ) − k m r (t ) ] + [ gˆ 0 + gˆ1 x(t ) + sCT B ]
α

φ (x, t )

Controller

⎧ sC T B
⎪− φ (x, t ) + u0 , for sCT B < 0
⎪ sC B
T


u (t ) = ⎨ 0 , for sCT B = 0 (3.31)
⎪ sCT B
⎪− φ (x, t ) − u0 , for sCT B > 0
⎪⎩ sC B
T

u (t )
Nonlinear Function Φ(⋅)

Fig. 3.1. The overall scheme of the proposed controller.

22
CHAPTER 4

SIMULATION EXAMPLE
In this chapter, we use a numerical example to illustrate the proposed control

strategy.

Example:

Consider a time-delay uncertain nonlinear system with input containing sector

nonlinearities and dead-zone which is described by (2.1)

x& (t ) = ( A + ΔA(x, t ))x(t ) + ( Aτ + ΔAτ (x, t ))x(t − h) + BΦ (u ) + d(t )

where

⎡ 0 1⎤ ⎡ 0 0.1⎤ ⎡0⎤
A=⎢ ⎥ , Aτ = ⎢ ⎥ , B=⎢ ⎥,
⎣− 4 5⎦ ⎣− 0.1 1 ⎦ ⎣1 ⎦

⎡ 0 0 ⎤
Δ A ( x, t ) = ⎢ ⎥,
⎣0.5 sin( x1 (t )) + 0.3 cos( x2 (t )) 0.02 sin( x2 (t )) + 0.2 cos( x1 (t ))⎦

⎡ 0 0 ⎤
ΔAτ (x, t ) = ⎢ ⎥ and d(t )
⎣0.15 sin( x1 (t )) + 0.5 cos( x2 (t )) 0.02 sin( x2 (t )) + 0.01 cos( x1 (t ))⎦

is the external disturbance. Assume that the nonlinear input containing sector

nonlinearities and dead-zone are represented as [9]:

⎧(l + γ sin(u ) + pe( h1 cos(u ) ) )(u − u0 ), if u > u0,



Φ (u ) = ⎨ 0, if − u0 ≤ u ≤ u0 ,
⎪(l + γ sin(u ) + pe ( h1 cos(u ) ) )(u + u ), if u < −u0 ,
⎩ 0

in which l , γ , p , h1 , u0 , and ε are designed parameters.

Let the desired reference model as

x& m (t ) = A m x m (t ) + B m r (t )

23
where

⎡0 1⎤ ⎡0 ⎤
Am = ⎢ ⎥ , B = ⎢1 ⎥ ,
⎣ − 6 − 9⎦
m
⎣ ⎦

and r (t ) is the reference input.

In this thesis, the following two cases are simulated:

Case 1: x(t ) = θ (t ) = 0 for t ∈ [− h 0] , h = 1 , l = −0.35 , γ = 0.3 , p = 1 , h1 = 0.02 ,

u0 = 1 , ε = 0.02 , r (t ) = 5 sgn(sin(t )) , and external disturbance

d(t ) = [0 d (t )] , d (t ) = 0.25 sin(2t ) . The initial values are given as

x1 (0) = 0.5 , x 2 (0) = −0.2 , x m1 (0) = −0.3 , and x m 2 (0) = −0.6 .

Case 2: x(t ) = θ (t ) = 0 for t ∈ [ − h 0] , h =1 , l = − 0 .4 , γ = 0.2 , p =1 ,

h1 = 0.9 , u 0 = 1.5 , ε = 0.02 , r (t ) = 4.5 sin(t ) + cos(4t ) , and external

disturbance d(t ) = [0 d (t )] , d (t ) = 0.2 cos(2t ) . The initial values are given as

x1 (0) = 0.25 , x 2 (0) = 0.2 , x m1 (0) = −0.55 , and x m 2 (0) = −0.15 .

The main objective is to use the proposed control strategy to design a controller

u (t ) so that the trajectory of the system x(t ) can track the state of the reference model

x m (t ) .

According to the design procedure, the controller can be designed in the following

steps:

Step 1. The sliding hyperplane is selected as s (t ) = CT e(t ) , where e(t ) = x(t ) − x m (t )

and we select the constant vector C = [2 1] .

Step 2. There exist two unknown positive constant parameters g 0 and g1 such that

W (x, t ) ≤ g 0 + g1 x(t ) ,

we employ adaptive fuzzy technique to estimate g 0 and g1 , respectively. To construct

24
two fuzzy logic systems gˆ 0 (x θ gˆ 0 ) , gˆ1 (x θ gˆ1 ) as given in (3.22) and (3.23),

respectively. We select the membership functions for xi (i = 1,2) from the following

fuzzy sets:

1/{1 + exp[10( xi + 0.4)]} , exp{−2[( xi + 0.2)]2 } , exp{−2( xi2 )} ,

exp{−2[( xi − 0.2)]2 } , and 1/{1 + exp[10( xi − 0.4)]} .

Twenty-five fuzzy rules for fuzzy logic system gˆ 0 (x θ gˆ 0 ) are included in the fuzzy rule

bases:

R lgˆ 0 : IF x1 is Fxl1 and x2 is Fxl2 , THEN gˆ 0 (x θ gˆ 0 ) is Gglˆ 0 , for l = 1,2, K ,25.

Hence, the unknown positive constant g 0 is estimated by gˆ 0 (x θ gˆ 0 ) = θ Tgˆ 0 ξ x (x) where

θ gˆ 0 = [G1gˆ 0 , Gg2ˆ0 ,L, Gg25


ˆ0 ] ∈ R
T 25

and

0.9

0.8
Membership Function

0.7

0.6

0.5

0.4

0.3

0.2

0.1

0
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2
xi

Fig. 4.1. The membership functions of x1 (t ) and x2 (t ) .

25
ξ x (x) = [ξ x1 (x), ξ x2 (x),L, ξ x25 (x)]T ∈ R 25

with components as

ξ ( x) =
l ∏ F (x ) n
i =1
l
xi i
, l = 1,2, K 25 , i = 1,2 .
∑ [∏ F ( x )]
x q n l
l =1 i =1 xi i

Moreover, twenty-five fuzzy rules for fuzzy logic system gˆ1 (x θ gˆ1 ) are included in the

fuzzy rule bases:

R lgˆ1 : IF x1 is Fxl1 and x2 is Fxl2 , THEN gˆ1 (x θ gˆ1 ) is Gglˆ 1 , for l = 1,2, K ,25.

Hence, the unknown positive constant g1 is estimated by gˆ1 (x θ gˆ1 ) = θ Tgˆ1 ξ x (x) where

θ gˆ1 = [G1gˆ1 , Gg2ˆ1 ,L, Gg25


ˆ1 ] ∈ R
T 25

and

ξ x (x) = [ξ x1 (x), ξ x2 (x),L, ξ x25 (x)]T ∈ R 25

with components as

ξ xl (x) =
∏ F (x ) n
i =1
l
xi i
, l = 1,2, K 25 , i = 1,2 .
∑ [∏ F ( x )]
q n l
l =1 i =1 xi i

Step 3. Construct the control law (3.31)

⎧ sC T B
⎪− φ (x, t ) + u0 , if sCT B < 0
⎪ sC B
T


u (t ) = ⎨ 0 , if sCT B = 0 ,
⎪ sC T B
⎪− φ (x, t ) − u0 , if sCT B > 0
⎪⎩ sC B
T

where
β
φ (x, t ) = [ (CT B) −1 CT A m e(t ) + (CT B) −1 CT Aτ x sup (t ) + H m x(t )
α
1
− k m r (t ) ] + [ gˆ 0 + gˆ1 x(t ) + sCT B ].
α

Set α = 0.35 and β = 1.5 .

26
Step 4. Design the adaptive laws for adjusting θ ĝ0 and θ ĝ1 are redesigned as



⎪ 0 if s ≤ ε ,


⎪ r sC T B ξ ( x ) if s > ε and { θ g) 0 < r0 }
⎪ gˆ 0 x
θ& gˆ 0 =⎨
⎪ or{ θ g) 0 = r0 and sC T B θ Tg) 0 ξ x ≤ 0 },

⎪rgˆ 0 sC B ξ x (x)
T


⎪− r sC T B θ g 0 θ g 0 ξ (x) if s > ε and { θ ) = r and sC T B θ T) ξ > 0 },
T
) )

⎪ gˆ 0 θ Tg) 0 θ g) 0
x g0 0 g0 x



⎪ 0 if s ≤ ε ,


⎪ r sC T B ξ (x) x(t ) if s > ε and { θ g)1 < r1 }
⎪ gˆ1 x
θ& gˆ1 =⎨
⎪ or{ θ g)1 = r1 and sC T B θ Tg)1 ξ x ≤ 0 },

⎪rgˆ1 sC B ξ x (x) x(t )
T

⎪ θ ) θ T)
⎪− r sC T B g1 g1 ξ (x) x(t ) if s > ε and { θ g)1 = r1 and sC T B θ Tg)1 ξ x > 0 }.
⎪ gˆ1 θ g 1 θ g1
T x
⎩ ) )

Step 5. We select the learning positive constants rgˆ 0 = 1 and rgˆ1 = 1 .

Step 6. Obtain the control law and apply to the plant, then compute the adaptive law

(3.33) and (3.34) to adjust the parameter vector θ ĝ0 and θ ĝ1 .

The simulation results are shown in Figs. 4.2-15. Fig. 4.2 shows the trajectories of

x1 (t ) and xm1 (t ) for Case 1. Fig. 4.3 shows the trajectories of x2 (t ) and xm 2 (t ) for

Case 1. Fig. 4.4 shows the trajectory of the tracking error e1 (t ) for Case 1. Fig. 4.5

shows the trajectory of the tracking error e 2 (t ) for Case 1. Fig. 4.6 shows the phase

plane between e1 (t ) and e 2 (t ) for Case 1. Fig. 4.7 shows the nonlinear input Φ (u )

for Case 1. Fig. 4.8 shows the control law u (t ) of the system for Case 1. Fig. 4.9 shows

the trajectories of x1 (t ) and xm1 (t ) for Case 2. Fig. 4.10 shows the trajectories of

27
x2 (t ) and xm 2 (t ) for Case 2. Fig. 4.11 shows the trajectory of the tracking error e1 (t )

for Case 2. Fig. 4.12 shows the trajectory of the tracking error e 2 (t ) for Case 2. Fig.

4.13 shows the phase plane between e1 (t ) and e 2 (t ) for Case 2. Fig. 4.14 shows the

nonlinear input Φ (u ) for Case 2. Fig. 4.15 shows the control law u (t ) of the system

for Case 2. From these simulation results, we see the state x(t ) can asymptotically track

the reference mode x m (t ) .

28
0.8
x1
0.6 xm1

0.4
x1(t) , xm1(t), rad

0.2

-0.2

-0.4

-0.6

-0.8
0 5 10 15 20
time, sec

Fig. 4.2. The trajectories of x1 (t ) and xm1 (t ) for Case 1.

1
x2
xm2
0.5

0
x2(t) , xm2(t), rad

-0.5

-1

-1.5

-2
0 5 10 15 20
time, sec

Fig. 4.3. The trajectories of x2 (t ) and xm 2 (t ) for Case 1.

29
0.9

0.8

0.7

0.6

0.5
e1(t), rad

0.4

0.3

0.2

0.1

-0.1
0 5 10 15 20
time, sec

Fig. 4.4. The trajectory of the tracking error e1 (t ) for Case 1.

0.5

-0.5
e2(t), rad

-1

-1.5

-2
0 5 10 15 20
time, sec

Fig. 4.5. The trajectory of the tracking error e 2 (t ) for Case 1.

30
0.5

-0.5
e2(t), rad

-1

-1.5

-2
-0.2 0 0.2 0.4 0.6 0.8 1 1.2
e1 (t), rad

Fig. 4.6. The phase plane between e1 (t ) and e 2 (t ) for Case 1.

200

150

100

50
Phi(u), N

-50

-100

-150

-200
-150 -100 -50 0 50 100 150
u(t), N

Fig. 4.7. The nonlinear input Φ (u ) of the system for Case 1.

31
200

150

100

50
u(t), N

-50

-100

-150

-200
0 5 10 15 20
time, sec

Fig. 4.8. The control law u (t ) of the system for Case 1.

0.6
x1
0.4 xm1

0.2
x1 (t) , xm 1 (t), rad

-0.2

-0.4

-0.6

-0.8
0 5 10 15 20
time, sec

Fig. 4.9. The trajectories of x1 (t ) and xm1 (t ) for Case 2.

32
1
x2
xm2
0.5

0
x2(t) , xm2(t), rad

-0.5

-1

-1.5

-2
0 5 10 15 20
time, sec

Fig. 4.10. The trajectories of x2 (t ) and xm 2 (t ) for Case 2.

0.9

0.8

0.7

0.6

0.5
e (t), rad

0.4
1

0.3

0.2

0.1

-0.1
0 5 10 15 20
time, sec

Fig. 4.11. The trajectory of the tracking error e1 (t ) for Case 2.

33
0.5

-0.5
e2(t), rad

-1

-1.5

-2
0 5 10 15 20
time, sec

Fig. 4.12. The trajectory of the tracking error e 2 (t ) for Case 2.

0.5

-0.5
e2(t), rad

-1

-1.5

-2
-0.2 0 0.2 0.4 0.6 0.8 1 1.2
e1 (t), rad

Fig. 4.13. The phase plane between e1 (t ) and e 2 (t ) for Case 2.

34
400

300

200

100
Phi(u), N

-100

-200

-300

-400
-150 -100 -50 0 50 100 150
u(t), N

Fig. 4.14. The nonlinear input Φ (u ) of the system for Case 2.

200

150

100

50
u(t), N

-50

-100

-150

-200
0 5 10 15 20
time, sec

Fig. 4.15. The control law u (t ) of the system for Case 2.

35
CHAPTER 5

CONCLUSIONS AND FUTURE STUDY


In this thesis, a MRAFSMC for a class of time-delay uncertain nonlinear systems

with input containing sector nonlinearities and dead-zone has been proposed. First, we

employed adaptive fuzzy technique to estimate the upper bound of the lumped

uncertainties. Next, based on the Lyapunov stability theorem, we proposed a systematic

and effective strategy to achieve the control objective. By using the proposed control

strategy, the tracking performance can be guaranteed. Finally, the simulation results

demonstrate the effectiveness of the proposed control strategy.

In this thesis, we assume that all the state variables are available for measurement.

However, in practice, not all the state variables are available for measurement. It implies

that as one could not obtain all elements of the tracking error vector, the conventional

adaptive laws would be difficult to realize. In order to achieve the control objective, we

assume that the lumped uncertainties must be satisfied for matching conditions. This

restriction would limit the application of the MRAFSMC. Therefore, in the future we

will try to overcome these problems.

36
REFERENCES
[1] S. Phoojaruenchanachai, K. Uahchinkul, and Y. Prempraneerach, “Robust

stabilisation of a state delayed system,” IEE Proc. Control Theory Appl., vol. 145,

no. 1, pp. 87-91, 1998.

[2] S. K. Park and H. K. Ahn, “Robust controller design with novel sliding surface,”

IEE Proc. Control Theory Appl., vol. 146, no. 3, pp. 242-246, 1999.

[3] D. Yue and J. Lam, “Stabilising controller design for uncertain systems with

time-varying input delay,” IEE Proc. Control Theory Appl., vol. 151, no. 6, pp.

699-705, 2004.

[4] Y. Q. Xia and Y. M. Jia, “Robust sliding-mode control for uncertain time-delay

systems: An LMI approach,” IEEE Trans. Automat. Contr., vol. 48, no. 6, pp.

1086-1092, 2003.

[5] K. C. Hsu, “Variable structure control design for uncertain dynamic systems with

sector nonlinearities,” Automatica, vol. 34, no. 4, pp. 505-508, 1998.

[6] F. M. Yu, H. Y. Chung, and S. Y. Chen, “Fuzzy sliding mode controller design for

uncertain time-delayed systems with nonlinear input,” Fuzzy Sets and Systems, vol.

140, pp. 359-374, 2003.

[7] X. S. Wang, H. Hong, and C. Y. Su, “Model reference adaptive control of

continuous-time systems with an unknown input dead-zone,” IEE Proc. Control

Theory Appl., vol. 150, no. 3, pp. 261-266, 2003.

[8] K. K. Shyu, W. J. Liu, and K. C. Hsu, “Decentralized variable structure control of

uncertain large-scare systems containing a dead-zone,” IEE Proc. Control Theory

Appl., vol. 150, no. 5, pp. 467-475, 2003.

[9] K. C. Hsu, W. Y. Wang, and P. Z. Lin, “Sliding mode control for uncertain nonlinear

systems with multiple inputs containing sector nonlinearities and dead-zones,” IEEE

37
Trans. Syst., Man. Cybern., B, vol. 34, no. 1, pp. 374-380, 2004.

[10] C. H. Chou and C. C. Cheng, “Design of adaptive variable structure controllers for

perturbed time-varying state delay systems,” J. Franklin Inst., vol. 338, pp. 35-46,

2001.

[11] C. H. Chou and C. C. Cheng, “A decentralized model reference adaptive variable

structure controller for large-scale time-varying delay systems,” IEEE Trans.

Automat. Contr., vol. 48, no. 7, pp. 1213-1217, 2003.

[12] R. J. Wang, W. W. Lin, and W. J. Wang, “Stabilizability of linear quadratic state

feedback for uncertain fuzzy time-delay systems,” IEEE Trans. Syst., Man. Cybern.,

B, vol. 34, no. 2, pp. 1288-1292, 2004.

[13] L. X. Wang, Adaptive Fuzzy Systems and Control: Design and Stability Analysis.

Englewood Cliffs, NJ: Prentice Hall, 1994.

[14] Y. C. Chang, “Adaptive fuzzy-based tracking control for nonlinear SISO systems via

VSS and H ∞ approaches,” IEEE Trans. Fuzzy Syst., vol. 9, no. 2, pp. 278-292,

2001.

[15] H. X. Li and S. Tong, “A hybrid adaptive fuzzy control for a class of nonlinear

MIMO systems,” IEEE Trans. Fuzzy Syst., vol. 11, no. 1, pp. 24-34, 2003.

[16] Y. Liu and X. Y. Li, “Robust adaptive control of nonlinear systems with unmodelled

dynamics,” IEE Proc. Control Theory Appl., vol. 151, no. 1, pp. 83-88, 2004.

[17] C. C. Kung, T. H. Chen, and L. H. Kung, “Modified adaptive fuzzy sliding mode

controller for uncertain nonlinear systems,” IEICE Trans. Fundamentals, vol. E88-A,

pp. 1328-1334, 2005.

[18] C. C. Kung and T. H. Chen, “Observer-based indirect adaptive fuzzy sliding mode

control with state variable filters for unknown nonlinear dynamical systems,” Fuzzy

Sets and Systems, vol. 155, pp. 292-308, 2005.

38
[19] J. J. E. Slotine and W. Li, Applied Nonlinear Control. Englewood Cliffs, NJ:

Prentice Hall, 1991.

[20] H. K. Khalil, Nonlinear Systems. Englewood Cliffs, NJ: Prentice Hall, 1992.

[21] B. Drazenovic, “The invariance condition in variable structure system,” Automatica,

vol. 5, pp. 287-295, 1969.

[22] P. Kachroo and M. Tomizuka, “Chattering reduction and error convergence in the

sliding-mode control of a class of nonlinear systems,” IEEE Trans. Automat. Contr.,

vol. 41 no. 7, pp. 1063-1068, 1996.

39

Вам также может понравиться