IAENG International Journal of Computer Science, 36:3, IJCS_36_3_02
______________________________________________________________________________________
Synthesis of Fuzzy Control for Inverter Pendulum
Robot with H ∞ Performance Constraint
Wen-Jer Chang, Wei-Han Huang and Wei Chang
Abstract—The purpose of this paper is to give one set of
stability and stabilization conditions for an inverted pendulum
robot which simulating human stance under the framework of
Discrete Perturbed Time-Delay Affine (DPTDA) Takagi–
Sugeno (T–S) fuzzy models. In the beginning, the mathematical
model of inverted pendulum robot system and the
corresponding DPTDA T–S fuzzy model are introduced. Next,
some sufficient conditions are derived on robust H ∞
disturbance attenuation, in which the robust stability and
prescribed performance are achieved. In order to find suitable
fuzzy controllers, the Iterative Linear Matrix Inequality (ILMI)
algorithm is employed to solve these sufficient conditions.
Finally, a numerical simulation for the nonlinear inverted
pendulum robot system is given to show the applications of the
presented controller design approach.
Index Terms—Inverted Pendulum Robot, Takagi-Sugeno
Fuzzy Model, Time-delay, S-procedure, Iterative Linear Matrix
Inequality.
I. INTRODUCTION
Human being stance has been investigated in detail for
a long time [1]. In recent years, the researchers wish to
simulate human stance on the machine. In this paper, the
model is constructed based on purely inverted pendulum
dynamics and on a movable supportive base. This work was
based on the assumption that the act of maintaining an erect
posture could be viewed. However, the problems often are a
complicated nonlinear system. In general, the methods of
linear control and those of local linearization and moving
linearization are not well suited for the control problem of
inverted pendulums. This is due to the fact that inverted
pendulums constantly move among widely separated regions
of their workspace such that no linearization valid for all
regions can be found. In fact, in many practical systems, the
system plants contain severe nonlinear properties. Therefore,
many researchers have studied to solve the difficulties of
nonlinear control methods. One of them is the fuzzy logic
control [2-4]. It is a successful control approach to many
complex nonlinear systems or even non-analytic systems
[5-6]. Some remarkable studies on the stabilizing controller
design for fuzzy systems can be referred to [7-10], in which
Manuscript received on June 25, 2009. This work was supported by the
National Science Council of Taiwan of ROC under Contract
NSC97-2221-E-019-023.
W. J. Chang is with the National Taiwan Ocean University, Taiwan
(corresponding author tel: +886-224622192; fax: +886-224633765; e-mail:
wjchnag@mail.ntou.edu.tw).
W. H. Huang is with the National Taiwan Ocean University, Taiwan
(e-mail: M95660001@mail.ntou.edu.tw).
W. Chang is with the National Taiwan Ocean University, Taiwan (e-mail:
wolf57.tw@yahoo.com.tw).
(Advance online publication: 1 August 2009)
the so-called T–S fuzzy model [11] has mainly been used to
represent fuzzy systems.
Generally, the T–S fuzzy models can be separated into
the homogeneous fuzzy models [7-11] and the affine fuzzy
models [12-15]. The homogeneous one can be referred to the
T–S fuzzy model of which consequent part is linear without
constant bias term. In contrast, the affine one means the T–S
fuzzy model of which consequent part is affined by a
constant bias term. In general, the affine T–S fuzzy model
can preserve diverse nonlinear systems more than the
homogeneous one. However, the analysis and synthesis of
affine T–S fuzzy model are more difficult than the
homogeneous one.
Due to there are few studies dealing with the H ∞
constraints and time delay effects for the DPTDA T–S fuzzy
models. Hence, one major target presented in this paper is the
robust stability for the affine T-S fuzzy models. In this paper,
the issue of robust stability is proposed in the presence of
norm-bounded uncertainty. The class of perturbed affine T–S
fuzzy models is defined by a state-space model and
time-varying norm-bounded parameter uncertainties.
Moreover, the H ∞ control scheme [16] is used in this paper
to attack the problem of robust performance design problems
for the perturbed affine T–S fuzzy models. Besides, the
presence of time delays in control loops usually degrades
system performance and is even a source of instability [17].
The other target presented in this paper which we have
to pay attention is the fuzzy controller design problem under
the framework of LMI method [18]. The fuzzy controller
design of the DPTDA T–S fuzzy models is a challenging
problem for the designers because the closed-loop stability
conditions are not LMI formulations but Bilinear Matrix
Inequalities (BMI) ones. The BMI conditions can not be
easily solved via a convex optimization algorithm. For this
reason, an ILMI algorithm [12, 14-15] has been presented to
solve the BMI problem. In this paper, an ILMI algorithm is
developed to find feasible solutions for the synthesis problem
of fuzzy controller design for the DPTDA T–S fuzzy models.
Finally, in order to illustrate the applications of proposed
fuzzy controller design approach for the inverted pendulum
robot system, a simulation is provided in this paper.
II. SYSTEM DESCRIPTIONS AND PROBLEM FORMULATIONS
A. Inverted pendulum robot system
In this section, the mathematical model of the simple
inverted pendulum robot system is introduced. Referring to
Fig. 1, a simplified dynamic model for describing inverted
pendulum robot system to simulate human stance is proposed
as follows [4].
 IAENG International Journal of Computer Science, 36:3, IJCS_36_3_02
______________________________________________________________________________________
x3 ( k + 1) = x3 ( k ) + 0.1 x4 ( k ) (2c)
x1 ( k +1) = x1 ( k ) + T x2 ( k ) +e v ( k ) (1a) 0.1
x4 ( k + 1) = x4 ( k ) +
x2 ( k + 1) = x2 ( k ) +
T
M + m ( sin x2 ( k ) )
2 (
0.4478 0.0166 ( sin x3 ( k ) )
2
)
× (13.4162 sin x3 ( k ) − u ( k ) cos x3 ( k )
× ( u ( k ) + ml x42 ( k ) sin x3 ( k ) − b x2
+12.98 x2 ( k ) cos x3 ( k )
− mg cos x3 ( k ) sin x3 ( k ) ) (1b)
−0.0166 x4 2 ( k ) sin x3 ( k ) cos x3 ( k ) ) (2d)
x3 ( k + 1) = x3 ( k ) + T x4 ( k ) (1c)

x4 ( k + 1) = x4 ( k ) +
T where σ ( k ) = ρ x1 ( k ) + (1 − ρ ) x1 ( k − τ ) and σ ( k ) is a
(
l M + m ( sin x3 ( k ) )
2
) time-delay function.
× ( ( M + m ) g sin x3 ( k ) − u ( k ) cos x3 ( k ) B. DPTDA T–S fuzzy model
Based on the nonlinear inverted pendulum robot system
+ b x2 ( k ) cos x3 ( k )
(2) presented above, the stability analysis and fuzzy
− ml x4 2 ( k ) sin x3 ( k ) cos x3 ( k ) ) (1d) controller design problems for the nonlinear system (2) via
T-S fuzzy model are introduced. Consider the DPTDA T–S
where
fuzzy model described by the following IF-THEN rules.
m is the mass of the black on the pendulum.
l is length of the pendulum.
Rule i: IF z1 ( k ) is M i1 and z2 ( k ) is M i2 and ··· and
g is acceleration due to gravity.
b is coefficient of viscous friction for motion of the cart. zp ( k ) is M ip THEN
u is applied force. x ( k + 1) = ( A i + ΔA i ) x ( k ) + ( A id + ΔA id ) x ( k − τ )
v ( t ) is the denotes the disturbances. + ( B i + ΔB i ) u ( k ) + ( a i + Δa i ) + E v ( k ) ,
The four state variables stand for x1 = x , x2 = x ,
i = 1, 2,… , r , x ( k ) ∈ Xi , i ∈ ˆI
x2 = θ , x4 = θ with the position of the cart, and θ the angle
x ( k ) = ψ ( k ) , for k ∈ [ −τ , 0] (3)
the pendulum makes with vertical. This model is obtained by
discretizing the continuous time model via Euler’s method
where A i ∈ ℜn× n , A id ∈ ℜn× n , Bi ∈ ℜn× m , ai ∈ ℜn and
with Τ is 0.1s, b = 12.98 kg , Μ = 1.378 kg, l = 0.325 m,
s E ∈ ℜn are constant matrices, z1 ( k ) , ... , zp ( k ) are known
m
g = 9.8 2 , m = 0.051 kg. premise variables that may be functions of the state variables,
s
p is the premise variable number and r is the number of

m
fuzzy model rules. x ( k ) ∈ ℜn is the state vector, u ( k ) ∈ ℜm
is the input vector, v ( k ) ∈ ℜ denotes the disturbance which

θ
belongs to L2 [ 0, k f ] , where L2 [ 0, k f ] denotes the
Lebsegue space consists of square-integrable functions on
l the interval [ 0 , k f ] and k f is the terminal time of the
control. M ip is the fuzzy set, τ is the constant time delay in
M the state and τ > 0 . ψ ( k ) is the initial condition of the state
u
defined on −τ ≤ k ≤ 0 . Besides, the region Xi ⊆ ℜn is
assumed to be a fuzzy subspace and Xi is called as a cell.
Fig. 1 Inverted pendulum robot system to simulate human
stance The set of cell indices is denoted as ˆI and the union of all
cells x ( k ) = conv ( ∪ i∈ˆI Xi ) is referred to as the whole fuzzy
Considering premise nominal parameter uncertainties, the space, where conv ( i ) refers to the convex combination. Let
modified dynamic model for the inverted pendulum robot
ˆI ⊆ ˆI be the set of indices for the fuzzy rules that contain the
system can be described as follows: 0

origin and ˆI1 ⊆ ˆI be the set of indices for the fuzzy rules that
x1 ( k + 1) = ( 0.1cos ( t ) + 1) σ ( k ) + 0.1 x2 ( k ) +0.2 v ( k ) (2a) does not contain the origin. The origin is an equilibrium point
of the DPTDA T–S fuzzy models and it is assumed that
0.1 ai = 0 for i ∈ ˆI0 . Besides, ΔA i , ΔA id , ΔBi and Δai are
x2 ( k + 1) = x2 ( k ) +
1.378 + 0.051( sin x3 ( k ) )
2
time-varying matrices with appropriate dimensions and they
are structured in the following norm-bounded form:
× ( u ( k ) −12.98 x2 +0.0166 x4 2 ( k ) sin x3 ( k )
−0.4998 cos x3 ( k ) sin x3 ( k ) ) (2b) [ ΔAi ΔA id ΔB i Δa i ]

(Advance online publication: 1 August 2009)

 IAENG International Journal of Computer Science, 36:3, IJCS_36_3_02
______________________________________________________________________________________
= DΔ ( t ) [Q1i Q 2i Q3i Q 4i ] (4) r
u ( k ) = −∑ h i ( z ( k ) ) {Fi x ( k )} (9)
i =1

where D , Q1i , Q 2i , Q3i and Q 4i are known real constant

matrices of appropriate dimensions, and Δi ( t ) is an Substituting (9) into (6), one can obtain corresponding
closed-loop system as follows:
unknown matrix function with Lebesgue-measurable
elements and satisfies ΔiT ( k ) Δi ( k ) ≤ I . r r
x ( k + 1) = ∑∑ h i ( z ( k ) ) h j ( z ( k ) )
i =1 j =1
Lemma 1 [19]:
Let Γ , D , Q and Δi ( k ) be real matrices of {( ) (
× H1ij + DΔ ( k ) H1ij x ( k ) + H 2ij + DΔ ( k ) H 2ij x ( t − τ ( k ) ) )
appropriate dimensions with Δi ( k ) Δi ( k ) ≤ I . Then for
T
(
+ H 3ij + DΔ ( k ) H 3ij )} +E v ( k ) (10)
T
Ξ > 0 any scalar ε > 0 satisfying εI − D ΞD > 0 , one has
where
G ij + G ji A id + A jd ai + a j
( Γ + DΔ ( k ) Q ) Ξ ( Γ + DΔ ( k ) Q )
T
i i H ij = , H 2ij = ,, H 3ij =
2 2 2
( )
−1
≤ Γ T ΞΓ + Γ T ΞD εI − D ΞD
T T
D ΞΓ + εQ Q
T
(5) G ij + G ji Q 2i + Q 2j Q 4i + Q 4j
H1ij = , H 2ij = , H 3ij = ,
2 2 2
□
G ij = A i − Bi Fj and G ij = Q1i − Q3i Fj (11)
Given a pair of ( x ( t ) , u ( t ) ) , the final outputs of the
DPTDA T–S fuzzy model (3) are inferred as follows: For the more, in order to deal with the robust performance
design problems, we have
x ( k + 1) =
r Definition 1 ( H ∞ Performance Constraint)
∑ ω ( z ( k ) ) {( A
i =1
i i + ΔA i ) x ( k ) + ( A id + ΔA id ) x ( k − τ )
Given a positive real number γ , the model of the form
r
(10) is said to have L2 [ 0, k f ] gain less than γ if
+ ( Bi + ΔB i ) u ( k ) + ( ai + Δai )} ∑ ω ( z ( k )) + E v ( k )
i
i =1
r kf kf
= ∑ h i ( z ( k ) ) {( A i + ΔA i ) x ( k ) + ( A id + ΔA id ) x ( k − τ ) ∑ x (k)S x (k) < γ ∑ v (k) v (k) , ∀v (k) ≠ 0
T 2 T
(12)
i =1 k =0 k =0

+ ( Bi + ΔBi ) u ( k ) + ( ai + Δai )} +E v ( k ) (6)

with zero initial condition for all v ( k ) ∈ L2 [ 0 , k f ] , where
where k f is the terminal time of the control, γ is a prescribed value
x ( k ) = ⎡⎣ x1 ( k ) , x2 ( k ) , ... , xn ( k ) ⎤⎦ , which denotes the worst case effect of v ( k ) on x ( k ) .
T

n Besides, S = S T > 0 is a positive definite weighting matrix

ωi ( x ( k ) ) = ∏ M ij x j ( k ) , ( ) and S ∈ ℜn× n .
j=1
□
ωi ( x ( k ) )
hi ( x ( k )) = r
, hi ( x ( k )) ≥ 0
The purpose of this paper is to find a fuzzy controller
∑ ωi ( x ( k ) )
i =1
(9) such that the closed-loop system (10) is quadratically
r stable which satisfy the H ∞ constraint (12). In next section,
and ∑ h ( x ( k )) = 1
i =1
i (7) we analyze the quadratically stable conditions for DPTDA
T–S fuzzy model (10) firstly. According to these stability
conditions, a fuzzy controller is developed via ILMI
The PDC [8] offers a scheme to design a fuzzy controller
algorithm in section IV.
from the given T–S fuzzy model (6). The PDC type fuzzy
controller has the following form:
III. STABILITY ANALYSIS FOR DPTDA T-S FUZZY MODEL
Rule i: IF z1 ( k ) is M i1 and z2 ( k ) is M i2 and ··· and
Stability analysis for a closed-loop DPTDA T–S fuzzy
zp ( k ) is M ip THEN model (10) is discussed in this section. It is shown that the
u ( k ) = −Fi x ( k ) , i = 1, 2,… , r for x ( k ) ∈ Xi , i ∈ ˆI (8) stability analysis issue to closed-loop DPTDA T–S fuzzy
models is considered based on Lyapunov stability criterion
and Razumikhin theorem. The sufficient condition for
where Fi ∈ ℜm× n are constant matrices. The output of the guaranteeing the closed-loop stability is introduced in the
PDC type fuzzy controller is determined by the following following theorem.
summation:
Theorem 1

(Advance online publication: 1 August 2009)

 IAENG International Journal of Computer Science, 36:3, IJCS_36_3_02
______________________________________________________________________________________
Given a H ∞ attenuation parameter γ > 0 . The DPTDA Proof:
T–S fuzzy model described in (10) is quadratically stable in Select a discrete-type Lyapunov function as
the large and the H ∞ control performance (12) is guaranteed
k −1
for an attenuation γ , if there exist positive definite matrices V ( x ( k ) ) = xT ( k ) P x ( k ) + ∑ x T ( ϖ ) Pd x ( ϖ ) (18)
ϖ= k − τ
P > 0 , S > 0 , Pd > 0 , control gains Fi and scalars ξijq ≥ 0
such that By evaluating the first-forward difference of the Lyapunov
function V ( x ( k ) ) along the trajectories of DPTDA T–S
ϒij < 0 for i ∈ ˆI0 (13)
fuzzy model (10), one has
and
n
ϒij − ∑ ξijq Ωijq ( s ) < 0 for i ∈ Î1 (14) ΔV ( x ( k ))
q =1
= V ( x ( k + 1) ) − V ( x ( k ) )
where = x T ( k + 1) P x ( k + 1) − x T ( k ) P x ( k ) + x T ( k ) Pd x ( k )
⎧ ⎡ H ijT ⎤ ⎫
− x ( k − τ ) Pd x ( k − τ )
T
⎪ ⎢ ⎥ ⎪⎪
( )
−1
ϒij ⎨φ + ⎢ H 2ijT ⎥ PD εI − DT PD D P ⎡⎣ H ij
T
H 2ij E ⎤⎦ ⎬ r r r r
⎪ ⎢ ET ⎥ ⎪ = ∑∑∑∑ h i ( z ( k ) ) h j ( z ( k ) ) h k ( z ( k ) ) h l ( z ( k ) )
⎩ ⎣ ⎦ ⎭⎪ i =1 j=1 k =1 l =1

⎡ H T PH − P + P + S + H ε H1ij T
1ij
(15a)
×x ( k )
T
{(O + DΔ ( k ) J ) P (O + DΔ ( k ) J )} x ( k )
ij ij
T
ij ij

⎢ 1ij 1ij d

⎢ T −x T
( k ) P x ( k ) + x ( k ) Pd x ( k ) − x ( k − τ ) Pd x ( k − τ )
T T

φ= H 2ij PH1ij + H 2ijε H1ij

T
⎢ (19)
⎢ ET PH1ij where
⎢⎣
* * ⎤ Oij = ⎡⎣ H ij H 2ij E H 3ij ⎤⎦ , J ij = ⎡ H ij H 2ij 0 H 3ij ⎤ ,
T ⎥ ⎣ ⎦
H PH 2ij − Pd + H ε H 2ij
T
* ⎥ (15b)
and x ( k ) = ⎡⎣ x T ( k ) x T ( k − τ ) v T ( k ) 1⎤⎦
2ij 2ij T
⎥ (20)
ET PH 2ij ET PE − γ 2 I ⎥
⎦
ϒij From Lemma 1, one can obtain
⎧ ⎡ H ijT ⎤ ⎫
⎪ ⎢ ⎪ ΔV ( x ( k ))
T⎥
⎪ ⎢ H 2ij ⎥ PD εI − DT PD ⎪
( )
−1
⎨φ + D P ⎡⎣ H ij
T
H 2ij E H 3ij ⎤⎦ ⎬ r r r r

⎪
⎢ ET ⎥
⎪ ≤ ∑∑∑∑ h i ( z ( k ) ) h j ( z ( k ) ) h k ( z ( k ) ) h l ( z ( k ) )
⎢ ⎥ i =1 j=1 k =1 l =1
⎪ T ⎪
⎣⎢ H 3ij ⎦⎥
⎩
(15c)
⎭
{
×x T ( k ) OijT POij + OijT PD εI − DT PD ( )
−1
DT POij

φij
⎡
⎢ T
φij
T
}
+εJ ijT J ij x ( k ) − x T ( k ) P x ( k ) + x T ( k ) Pd x ( k )
⎢ H 3ij P ⎡⎣ H1ij H 2ij E ⎤⎦ + H ε ⎡ H1ij H 2ij 0⎤ −x T
( k − τ ) Pd x ( k − τ ) (21)
⎣ ⎣ 3ij
⎦
* ⎤
T ⎥ (15d) Next, let us define the following performance index
T
H 3ij PH 3ij + H ε H 3ij ⎥⎦
3ij
kf
⎡ Tijq 0
⎢ 0 0
0 n ijq ⎤ J∞ ∑{x ( k ) S x ( k ) − γ
T 2
vT ( k ) v ( k ) } (22)
⎢ 0 0 ⎥⎥ k =0

Ωijq ( s ) = ⎢ ⎥ (16)
⎢ ⎥ with zero initial condition for all v ( k ) ∈ L2 [ 0 , k f ] . Hence,
⎢ 0 0 0 0 ⎥
⎢n T 0 0 vijq ⎥⎦ for any nonzero v ( k ) one has
⎣ ijq s× s

kf
Besides, the S-procedure [12, 18] weighting parameters {
J ∞ ≤ ∑ xT ( k ) S x ( k ) − γ 2 vT ( k ) v ( k ) + V ( x ( k ) ) }
Tijq ∈ ℜn× n , nijq ∈ ℜn×1 , and vijq ∈ ℜ are defined such that k =0
kf kf

{
= ∑ xT ( k ) S x ( k ) − γ 2 vT ( k ) v ( k ) + ∑ Δ V ( x ( k ) ) }
σijq ( x ( t ) ) x T ( t ) Tijq x ( t ) + 2n ijq
T
x ( t ) + vijq ≤ 0 , k =0 k =0

{ }
kf
q =1 p and i = 1 r (17) = ∑ xT ( k ) S x ( k ) − γ 2 vT ( k ) v ( k ) + Δ V ( x ( k ) )
k =0

for all x ( t ) which activates rule i (i.e., h i ( x ( t ) ) > 0 ).

(Advance online publication: 1 August 2009)

 IAENG International Journal of Computer Science, 36:3, IJCS_36_3_02
______________________________________________________________________________________
kf
by setting the state bias term ai = 0 and ignoring the
∑ {H ( x,v,k )}
k =0
(23)
S-procedure from the similar proof procedure.
□
where H ( x,v,k ) is defined as follows according to (21).
From Theorem 1, it can be noted that the matrix
inequalities in P , S , Pd , Fi and ξijq belong to the class of
H ( x,v,k )
BMIs and the controller synthesis can not be solved by the
= Δ V ( x ( k ) ) + xT ( k ) S x ( k ) − γ 2 vT ( k ) v ( k ) MATLAB LMI-toolbox. In the next section, Theorem 2 is
provided to introduce modified stability conditions which
{ ( k ) ϒij x ( k )}
r r
= ∑∑ h i ( x ( k ) ) h j ( x ( k ) ) x ( k )
T
(24) can be solved by MATLAB LMI-toolbox through an ILMI
i =1 j =1 algorithm [12, 14-15].

where ϒij is defined in (15c). Converting (24) to an LMI by

applying the S-procedure described in [12, 18], one has IV. FUZZY CONTROLLER DESIGN FOR DPTDA T-S FUZZY
MODELS VIA ILMI ALGORITHM

∑∑ h ( x ( k ) ) h ( x ( k ) ) {x ( k ) ( k ) ϒ x ( k )}
r r In this section, the ILMI algorithm [12, 14-15] is used
H ( x,v,k ) ≤
T
i j ij to develop a fuzzy controller design procedure for the
i =1 j=1
n
DPTDA T–S fuzzy model (10). The idea of the ILMI
−∑ ξijq σijq ( x ( k ) ) (25) algorithm used in solving BMI problems is based on holding
q =1 some matrix variables as constant values and then converting
it into a LMI problem. One can thus use the MATLAB
where σijq ( x ( k ) ) ∈ ℜ is defined in (17). Since ξijq ≥ 0 and LMI-toolbox to solve the proposed fuzzy controller design
problem.
σijq ( x ( k ) ) ≤ 0 , then (25) can be represented as
A. Stabilization condition of DPTDA T–S fuzzy model
r r Theorem 2
H ( x,v,k ) ≤ ∑∑ h i ( x ( k ) ) h j ( x ( k ) ) Given a H ∞ attenuation parameter γ > 0 and the
i =1 j=1
auxiliary constant matrix R > 0 . The conditions of Theorem
⎧⎪ T ⎛
n ⎞ ⎫⎪ 1 are satisfied if there exist α < 1 , positive definite matrices
× ⎨x ( k ) ⎜ ϒij − ∑ ξijq Ωijq ( s ) ⎟ x ( k ) ⎬ (26)
⎩⎪ ⎝ q =1 ⎠ ⎭⎪ P > 0 , S > 0 , Pd > 0 , control gains Fi and scalars ξijq ≥ 0
such that
Obviously, if (14) hold for all x ( k ) ∈ Xi , i ∈ Î1 , then
⎪⎧Θij < 0
for i ∈ ˆI0
kf

∑ {H ( x,v,k )} < 0 . It means that

k =0
⎨ T
⎪⎩R PR − R ≤ 0
(30)

and
kf kf
⎧⎪Θij < 0
J ∞ < 0 or ∑ { x ( k ) S x ( k )} < γ ∑ {v ( k ) v ( k )}
T 2 T
(27) ⎨ T for i ∈ ˆI1 (31)
k =0 k =0 ⎪⎩R PR − R ≤ 0

Since (27) is equivalent to (12), it is easy to find that H ∞ where

performance constraint (12) is achieved with a prescribed γ . Θij
In the next step, we have to show that the DPTDA T–S fuzzy
⎡ −αP + Pd + S *
model in (10) is quadratically stable. From (26), if ⎢ T −1 T T
n
⎢ H 2ij R H1ij + H 2ijε H1ij H T2ij R −1H 2ij − Pd + H 2ijε H 2ij
ϒij − ∑ ξijq Ωijq < 0 hold, it implies that H ( x,v,k ) < 0 . ⎢
q =1
⎢ ET R −1H1ij ET R −1H 2ij
Assume that the disturbance v ( t ) is zero, and then one has ⎢ T −1
D R H ij DT R −1H 2ij
⎢
⎢ H1ij 0
H ( x,v,k ) = Δ V ( x ( k ) ) + x T ( k ) S x ( k ) − γ 2 v T ( k ) v ( k ) < 0 ⎢ H1ij 0
⎣
(28) * * * * ⎤
or
* * * * ⎥⎥
Δ V ( x ( k )) < − xT ( k ) S x ( k ) (29)
T −1
E R E−γ I 2
* * * ⎥
−1 −1 ⎥ (32)
T
D R E −εI + D R D *
T
* ⎥
Therefore, the DPTDA T–S fuzzy model described in (10) is 0 0 −ε −1 * ⎥
⎥
quadratically stable in the large. Besides, for the case of 0 0 0 −R ⎥⎦
x ( t ) ∈ Xi , i ∈ ˆI0 , the stability condition (13) can be obtained

(Advance online publication: 1 August 2009)

 IAENG International Journal of Computer Science, 36:3, IJCS_36_3_02
______________________________________________________________________________________
Θij ⎡( α − 1) P * * * * * *⎤
⎢ ⎥
⎡ −αP + Pd + S − ξijq Tijq * ⎢ 0 0 * * * * *⎥
⎢ T T ⎢ 0 0 0 * * * *⎥
⎢ H T2ij R −1H1ij + H 2ijε H1ij H T2ij R −1H 2ij − Pd + H 2ijε H 2ij ⎢ ⎥
⎢ <⎢ 0 0 0 0 * * *⎥ (34)
⎢ ET R −1H1ij ET R −1H 2ij ⎢ 0 0 0 0 0 * *⎥
⎢ T −1 T T ⎢ ⎥
⎢ H 3ij R H1ij + H 3ij ε H1ij − ξijq n
T T
H 3ij R −1H 2ij +H 3ijε H 2ij ⎢ 0 0 0 0 0 0 *⎥
⎢ ⎢ 0 0 0 0 0 0 0 ⎥⎦
−1
T
D R H ij DT R −1H 2ij
⎢ ⎣
⎢ H1ij 0
⎢ By using the Schur-complement [18], the inequality (34)
⎣ H1ij 0
becomes
* *
* * ⎡ H T R −1H − P + P + S + H1ij T
ε H1ij − ξijq Tij
E T R −1 E − γ 2 I * ⎢ 1ij 1ij d

T ⎢ −
T
H 2ij R H1ij + H 2ijε H1ij
T 1
T
H 3ij R −1 E T
H 3ij R −1H 3ij + H 3ijε H 3ij − ξijq v ij ⎢
⎢ ET R −1H1ij
D T R −1 E DT R −1H 3ij ⎢ T
0 0 ⎢⎣ T
H 3ij R −1H1ij +H 3ijε H1ij − ξijq n T
0 0 *
* * * ⎤ T
H T2ij R −1H 2ij − Pd + H 2ijε H 2ij
* * * ⎥⎥
ET R −1H 2ij
* * * ⎥ T
⎥ T
H 3ij R −1H 2ij + H 3ijε H 2ij
* * * ⎥ (33)
−εI+DT R −1D * * ⎥
⎥ * * ⎤
0 −ε −1 * ⎥ ⎥
* *
0 0 − R ⎥⎦ ⎥
−1
E R E − γ2I
T
* ⎥+Z
T
⎥
Proof: T
H 3ij R −1E T
H 3ij R −1H 3ij + H 3ijεH 3ij − ξijq v ij ⎥⎦
Rewriting (31), one has
⎡( α − 1) P * * *⎤
− P + Pd + S − ξijq Tijq ⎢ ⎥
⎡ * 0 0 * *⎥
⎢ <⎢ (35)
−1
T
−1
T ⎢ 0 0 0 *⎥
⎢ H 2ij R H1ij + H 2ijε H1ij
T
H 2ij R H 2ij − Pd + H 2ijε H 2ij
T
⎢ ⎥
⎢ ⎣⎢ 0 0 0 0 ⎦⎥
⎢ ET R −1H1ij ET R −1H 2ij
⎢ T −1 T T

⎢ H 3ij R H1ij + H 3ijε H1ij − ξijq n

T T
H 3ij R −1H 2ij + H 3ijε H 2ij where
⎢ DT R −1H ij DT R −1H 2ij Z=
⎢
⎢ H1ij 0 ⎡ H1ijT ⎤
⎢ ⎢ T⎥
H1ij 0 ⎢ H 2ij ⎥ R −1D εI − DT PD
( )
−1
⎣ DT R −1 ⎡⎣ H ij H 2ij E H 3ij ⎤⎦
⎢ ET ⎥
* * ⎢ ⎥
T
* * ⎣⎢ H 3ij ⎦⎥
−1
E R E−γ I
T 2
* (36)
T If the matrix P > 0 exist such that R PR − R ≤ 0 is held, T
T
H 3ij R −1E T
H 3ij R −1H 3ij + H ε H 3ij − ξijq v ij
3ij
then the following inequality is obvious.
DT R −1E DT R −1H 3ij
0 0 ⎡ H T PH − P + P + S + H1ij T
ε H1ij − ξijq Tij
0 0 ⎢ 1ij 1ij d

⎢ T
H T2ij PH1ij + H 2ijε H1ij
* * * ⎤ ⎢
* * * ⎥⎥ ⎢ ET PH1ij
⎢
* * * ⎥ ⎢⎣ T
H 3ij
T
PH1ij + H 3ijε H1ij − ξijq n T
⎥
* * * ⎥
−εI+DT R −1D * * ⎥
⎥
0 −ε −1 * ⎥
0 0 − R ⎥⎦

(Advance online publication: 1 August 2009)

 IAENG International Journal of Computer Science, 36:3, IJCS_36_3_02
______________________________________________________________________________________
* as follows:
T
H T2ij PH 2ij − Pd + H ε H 2ij
R (1) = P −1 ( 0 )
2ij
(39)
ET PH 2ij
For the given initial P ( 0 ) , one can solve it from the
T
T
H 3ij PH 2ij + H 3ijε H 2ij
* * ⎤ following discrete Riccati equation.
* * ⎥
⎥ ˆ T P (0) A
A ˆ − P (0)
ET PE − γ 2 I * ⎥+Z
⎥
( )( ) ( Bˆ P ( 0) Aˆ ) + Q = 0
−1
T ˆ T P ( 0) B
− A ˆ T P (0) B
ˆ 1+ B ˆ T
(40)
T
H 3ij PE T
H 3ij PH 3ij + H 3ijε H3ij − ξijq v ij ⎥⎦
⎡( α − 1) P * * * ⎤ r r
ˆ =1 A , B ˆ = 1 B and Q > 0 . The matrix Q
⎢
0 0 * *⎥
⎥ where A ∑ i ∑ i
<⎢ (37) r i =1 r i =1
⎢ 0 0 0 *⎥ is assigned by the designers.
⎢ ⎥
⎢⎣ 0 0 0 0 ⎥⎦ Step 2 Set κ = 1 and start the algorithm.
Step 3 Given the auxiliary variables R ( κ ) to solve the
where α < 1 . Thus, one has optimization problem for

⎡ H T PH − P + P + S + H1ij T Minimize α ( κ)
ε H1ij − ξijq Tij
⎢ 1ij 1ij d

⎢ T Subject to P ( κ ) > 0 , S ( κ ) > 0 , Pd ( κ ) > 0 ,

H 2ij PH1ij + H 2ijε H1ij
T
⎢ Fi ( κ ) and ξijq ( κ ) ≥ 0 ,
⎢ ET PH1ij
⎢ T (30) for i ∈ ˆI0 , and (31) for i ∈ Î1 (41)
⎢⎣ T
H 3ij PH1ij + H 3ijε H1ij − ξijq n T
* If α ( κ ) < 1 , then P ( κ ) , S ( κ ) , Pd ( κ ) , Fi ( κ ) , and ξijq ( κ )
T
H T2ij PH 2ij − Pd + H ε H 2ij
2ij obtained in (41) are feasible solutions for the Theorem 2 and
ET PH 2ij stop the iterative manner. Otherwise, if α ( κ ) ≥ 1 then go to
T Step 4.
T
H 3ij PH 2ij + H 3ijε H 2ij
Step 4 Given α ( κ ) obtained in Step 3 and the same
* * ⎤ auxiliary variables R ( κ ) used in Step 3. Resolve the
* * ⎥
⎥ optimization problem for P ( κ ) , S ( κ ) , Pd ( κ ) , Fi ( κ ) and
ET PE − γ 2 I * ⎥+Z<0 (38)
T
⎥ ξijq ( κ ) such that
T
H 3ij PE T
H 3ij PH 3ij + H 3ijεH 3ij − ξijq vij ⎥⎦
Minimize trace ( P ( κ ) )
The inequality (38) is equivalent to (14). Thus, the proof is
Subject to P ( κ ) > 0 , S ( κ ) > 0 , Pd ( κ ) > 0 , Fi ( κ )
completed. Besides, for the case of x ( t ) ∈ Xi , i ∈ ˆI0 , the
and ξijq ( κ ) ≥ 0 ,
stability condition (30) can be obtained by setting the state
bias term ai = 0 and ignoring the S-procedure from the (30) for i ∈ ˆI0 , and (31) for i ∈ Î1 (42)
similar proof procedure.
□ If the condition R −1 ( κ ) − P ( κ ) < υ is satisfied for a
predetermined small value υ . Then the Theorem 2 may not
B. ILMI Algorithm be feasible and stop the iterative manner. Otherwise, go to
According to Theorem 2, an ILMI algorithm [12, 14-15] Step 5.
is developed to find the feasible solutions for the stability Step 5 Update the auxiliary variables R ( κ + 1) using
conditions (30-31). The purpose of this algorithm is to
P ( κ ) , where P ( κ ) is determined in (42). Set κ = κ + 1 and
iteratively search for P , S , Pd , Fi , ξijq , α and to update
go back to Step 3.
the auxiliary constant matrix R until α < 1 . The detail of □
the proposed fuzzy controller design procedure is concluded In next section, a numerical example is presented to show
as follows. the usefulness of the above fuzzy controller design procedure
for the inverted pendulum robot system under the framework
<ILMI Algorithm> of DPTDA T–S fuzzy model.
Step 1 Define the iterative auxiliary variables as
R ( κ ) = P −1 ( κ − 1) , where κ denotes an iteration index.
When κ = 1 , the initial conditions of R (1) can be obtained

(Advance online publication: 1 August 2009)

 IAENG International Journal of Computer Science, 36:3, IJCS_36_3_02
______________________________________________________________________________________
V. A NUMERICAL EXAMPLE ⎡ 0 ⎤ ⎡0⎤ ⎡ 0 ⎤
⎢ 0.0549 ⎥ ⎢0⎥ ⎢ ⎥
⎥ , a = ⎢ ⎥ , a = ⎢ −0.0549 ⎥ ,
According to the results developing in previous
sections, this section provides a numerical simulation for the a1 = ⎢
⎢ 0 ⎥ 2 ⎢0⎥ 3 ⎢ 0 ⎥
inverted pendulum robot system presented in section Ⅱ. ⎢ ⎥ ⎢ ⎥ ⎢ ⎥
Considering the inverted pendulum robot system (2), one can ⎣ −3.0154 ⎦ ⎣0⎦ ⎣ 3.0154 ⎦
choose three operating points to obtain the linearized models ⎡ 0 ⎤ ⎡ 0 ⎤
for the system (2). Let us choose three operating points as ⎢ 0.0726 ⎥ ⎢ ⎥
B1 = B 2 = B3 = ⎢ ⎥ , and E = ⎢ 0 ⎥ . (45)
follows: ⎢ 0 ⎥ ⎢0.2 ⎥
⎢ ⎥ ⎢ ⎥
⎣ −0.2233⎦ ⎣ 0 ⎦
(x +
, xd+ , u + ) oper1
(
= 0 0 88 0 0 0 0 0 0 , )
(x, xd , u )oper 2 = 0 0 0 ( 0 0 0 0 0 0 , ) and the corresponding matrices of S-procedure [12, 18] are
(x −
, xd− , u − ) oper3
(
= 0 0 −88 0 0 0 0 0 0 ) presented as follows:
(43)
For Rules 11, i.e., 90 ≤ x3 ( t ) ≤ 80
Then, three linear subsystems can be constructed by these ⎡ 0 ⎤
⎡0 0 0 0⎤ ⎢ ⎥
operating points. In which, ( x , xd , u )oper 2 is the maintain ⎢0 0 0 0 ⎥⎥ ⎢ 0 ⎥
equilibrium point and the others are the off-equilibrium T111 =⎢ , n111 = ⎢ 1 ⎥
points. Through the above three linear subsystems and
⎢0 0 1 0⎥ ⎢ − ⎣⎡( 80 + 90 ) π 180 ⎦⎤ ⎥
⎢ ⎥ ⎢ 2 ⎥
membership functions defining in Fig. 2, one can obtain the ⎣0 0 0 0⎦
⎣⎢ 0 ⎥⎦
DPTDA T–S fuzzy model for the inverted pendulum robot
system (2), which is composed by three fuzzy rules as and v111 = ( 90π / 180 ) × ( 80π / 180 ) (46)
follows:
For Rules 33, i.e., −90 ≤ x3 ( t ) ≤ −80 , the matrices of
Rule i: IF x3 ( k ) is about M i1 THEN S-procedure are given as follows:
x ( k+1) = ( A i + ΔA i ) x ( k ) + ( A id + ΔA id ) x ( k − τ )
⎡ 0 ⎤
⎡0 0 0 0⎤
+B i u ( k ) + ai + E v ( k ) , i = 1 3 (44) ⎢0
⎢ ⎥
0 ⎥⎥ ⎢ 0 ⎥
0 0
T331 =⎢ , n331 = ⎢ 1 ⎥
where ⎢0 0 1 0⎥ ⎢ − ⎣⎡( −80 − 90 ) π 180 ⎦⎤ ⎥
⎢ ⎥ ⎢ 2 ⎥
⎡0.998 0.1 0 0⎤ ⎣0 0 0 0⎦
⎢ 0 ⎣⎢ 0 ⎥⎦
0.0917 0.035 0 ⎥⎥
A1 = ⎢ ,
⎢ 0 0 1 0.1⎥ and v331 = ( −90π / 180 ) × ( −80π / 180 ) (47)
⎢ ⎥
⎣ 0 0 0 1⎦
⎡ 0.998 0.1 0 0⎤
⎢ 0 M 31 M 21
0 ⎥⎥
1 M11
0 . 0581 − 0 . 0363
A2 = ⎢ ,
⎢ 0 0 1 0.1⎥
⎢ ⎥
⎣ 0 2.8983 3.127 1⎦
⎡ 0.998 0.1 0 0 ⎤
⎢ 0 ⎥ x3 ( k )
0. 0917 0 .035 0. 0464
A3 = ⎢ ⎥, −90° −80° 0 80° 90°
⎢ 0 0 1 0.1 ⎥
⎢ ⎥ Fig. 2 Membership functions of x3 ( k )
⎣ 0 0 0 1 ⎦
⎡0.001 sin ( t ) 0 0 0⎤
⎢ ⎥ For the DPTDA T–S fuzzy model (44), the fuzzy
0 0 0 0⎥
ΔA1 = ΔA 2 = ΔA 3 = ⎢ , controller can be designed by applying Theorem 2 and the
⎢ 0 0 0.003 cos ( t ) 0⎥ ILMI algorithm [12, 14-15]. In this example, it is assumed
⎢ ⎥
⎣⎢ 0 0 0 0 ⎦⎥ that the H ∞ control performance is guaranteed for an
⎡ 0.002 0 0 0⎤ attenuation γ 2 = 0.32 . Then, we can get a feasible solution
⎢ 0 0 0 0 ⎥⎥ after four iterations of the ILMI algorithm. The final decay
A1d = A 2d = A 3d = ⎢ , rate α is 0.9999 and the feasible solutions are obtained as
⎢ 0 0 0 0⎥
⎢ ⎥ follows:
⎣ 0 0 0 0⎦

(Advance online publication: 1 August 2009)

 IAENG International Journal of Computer Science, 36:3, IJCS_36_3_02
______________________________________________________________________________________
⎡ 4.4530 2.0824 7.3099 1.4670 ⎤
⎢ 2.0824 3.3681 9.5516 1.8320 ⎥⎥
P=⎢ ,
⎢ 7.3099 9.5516 41.5488 7.8826 ⎥
⎢ ⎥
⎣1.4670 1.8320 7.8826 1.5450 ⎦
⎡0.0327 0.0186 0.0430 0.0118 ⎤
⎢ 0.0186 0.0238 0.0366 0.0134 ⎥
Pd = ⎢ ⎥,
⎢ 0.0430 0.0366 0.1136 0.0225 ⎥
⎢ ⎥
⎣ 0.0118 0.0314 0.0225 0.0081⎦
⎡ 0.0076 0.0046 0.0107 0.0030 ⎤
⎢ 0.0046 0.0059 0.0092 0.0034 ⎥
S=⎢ ⎥,
⎢ 0.0107 0.0092 0.0284 0.0056 ⎥
⎢ ⎥
⎣ 0.0030 0.0034 0.0056 0.0020 ⎦
⎡ 0.3419 −0.1082 0.0601 −0.5031⎤
⎢ −0.1082 0.8916 −0.1470 −0.2054 ⎥
R=⎢ ⎥,
⎢ 0.0601 −0.1470 0.7670 −3.7935 ⎥
⎢ ⎥
⎣ −0.5031 −0.2054 −3.7935 20.7108 ⎦
ξ111 = 186.4773 , ξ331 = 186.4773
and the fuzzy controller has the following form:
Rule 1: IF x3 ( k ) is about M11 THEN
u ( k ) = − [ −5.2530 −1.0054 −43.6819 −9.2813] x ( k )
Rule 2: IF x3 ( k ) is about M 21 THEN
u ( k ) = − [ −5.1953 −18.5910 −49.8976 −9.2234] x ( k )
Rule 3: IF x3 ( k ) is about M 31 THEN
u ( k ) = − [ −5.2530 −1.0054 −43.6819 −9.2813] x ( k )
The output of the PDC type fuzzy controller (49) is
determined by the following summation
3 [12]
u ( k ) = −∑ h i ( x1 ( k ) ) {Fi x ( k )} (50)
i =1
The disturbance input noise v ( k ) is given with
variance one. The simulation results are shown in Fig. 3 to
Fig. 6. From the simulated results, one can find that the
controlled nonlinear perturbed time-delay inverted pendulum
robot system (2) is quadratically stable under the fuzzy
controller (50).
VI. CONCLUSION
A robust fuzzy controller design procedure has been
developed for the nonlinear inverted pendulum robot system
which can achieve the H ∞ performance constraints and cope
with the worst case effect of disturbances. Firstly, the
Lyapunov criterion was applied to analyze the stability
conditions for the nonlinear inverted pendulum robot system.
Secondly, an ILMI algorithm was developed to solve the
stabilization conditions of synthesis problems for the
nonlinear inverted pendulum robot system. Finally, in order
to illustrate the applicability of the present fuzzy controller
(Advance online publication: 1 August 2009)
design procedure, a numerical simulation for the inverted
pendulum robot system has been shown.
ACKNOWLEDGMENT
This work was supported by the National Science Council
of the Republic of China under contract
NSC97-2221-E-019-023.
REFERENCES
[1] I. D. Loram, P. J. Gawthrop and M. Lakie, “The Frequency of Human,
Manual Adjustments in Balancing an Inverted Pendulum is
Constrained by Intrinsic Physiological Factors”, Physiology, Vol. 577,
No. 1, pp. 417-432, 2006.
[2] H. J. Zimmermann, Fuzzy Set Theory and Its Application, MA: Kluwer,
Boston, 1991.
[3] S. Dutta, “Fuzzy Logic Applications: Technological and Strategic
Issues,” IEEE Transactions on Engineering Management, Vol. 40, No.
3, pp. 237–254, 1993.
[4] R. Gurumoorthy and S. R. Sanders, “Controlling Nonminimum Phase
Nonlinear Systems- the Inverted Pendulum on a Cart Example”,
Control Conference on American, pp. 680-685, 1994.
[5] B. Chen and X. Liu, “Fuzzy Approximate Disturbance Decoupling of
MIMO Nonlinear Systems by Back Stepping and Application to
(48) Chemical Processes,” IEEE Transactions on Fuzzy Systems, Vol. 13,
No. 6, pp. 832–847, 2005.
[6] S. R. Munasinghe, M. S. Kim, and J. J. Lee, “Adaptive Neuro Fuzzy
Controller to Regulate UTSG Water Level in Nuclear Power Plants,”
IEEE Transactions on Nucleus Science, Vol. 52, No. 1, pp. 421–429,
2005.
[7] K. Tanaka and M. Sugeno, “Stability Analysis and Design of Fuzzy
Control System,” Fuzzy Sets and Systems, Vol. 45, No. 2, pp. 135–156,
1992.
[8] K. Tanaka and H. O. Wang, Fuzzy Control Systems Design and
Analysis-A Linear Matrix Inequality Approach, John Wiley & Sons,
Inc., New York, 2001.
[9] W. J. Chang, “Fuzzy Controller Design via the Inverse Solution of
Lyapunov Equations”, ASME, J. Dynamic Systems, Measurement and
Control, Vol. 125, No. 1, pp. 42-47, 2003.
[10] W. J. Chang, C. C. Sun and H. Y. Chung, “Fuzzy Controller Design for
(49) Discrete Controllability Canonical Takagi-Sugeno Fuzzy Systems”,
IEE Proceeding, Part D, Control Theory and Applications, Vol. 151,
No. 3, pp. 319-328, 2004.
[11] T. Takagi and M. Sugeno, “Fuzzy Identification of Systems and Its
Applications to Modeling and Control,” IEEE Transactions on
Systems, Man, Cybernetics, Vol. 15, No. 1, pp. 116–132, 1985.
E. Kim and D. Kim, “Stability Analysis and Synthesis for an Affine
Fuzzy System via LMI and ILMI: Discrete Case,” IEEE Trans.
Systems, Man, and Cybernetics, Part B, Vol. 31, No. 1, pp. 132-140,
2001.
[13] C. C. Hsiao, S. F. Su, T. T. Lee and C. C. Chuang, “Hybrid
Compensation Control for Affine TSK Fuzzy Control Systems,” IEEE
Transactions on Systems Man and Cybernetics, Part B -Cybernetics,
Vol. 34, No. 4, pp. 1865-1873, 2004.
[14] W. J. Chang and W. Chang, “Synthesis of Nonlinear Discrete Control
Systems via Time-Delay Affine Takagi-Sugeno Fuzzy Models”, ISA
Transactions, Vol. 44, No. 2, pp. 243-257, 2005.
[15] W. J. Chang and W. Chang, “Discrete Fuzzy Control of Time-Delay
Affine Takagi-Sugeno Fuzzy Models with H ∞ Constraint”, IEE
Proceeding, Part D, Control Theory and Applications, Vol. 153, No. 6,
pp.745-752, 2006.
[16] I. R. Petersen, V. A. Ugrinovskii and A. V. Savkin, Robust Control
Design Using H ∞ Methods, Springer, 2000
[17] Y. Y. Cao and P. M. Frank, “Analysis and Synthesis of Nonlinear
Time-Delay Systems via Fuzzy Control Approach,” IEEE Transactions
on Fuzzy Systems, Vol. 8, No. 2, pp. 200-211, 2000.
[18] S. P. Boyd, Linear Matrix Inequalities in System and Control Theory,
Society for Industrial and Applied Mathematics, 1994.
[19] Y. Wang, L. Xie and C. E. de Souna, “Robust Control of A Class of
Uncertain Nonlinear Systems,” Systems and Control Letters, Vol. 19,
pp. 139-149, 1992.
Wen-Jer Chang received the the MS degree in the Institute of Computer
Science and Electronic Engineering from the National Central University in
 IAENG International Journal of Computer Science, 36:3, IJCS_36_3_02
______________________________________________________________________________________
1990, and the PhD degree from the Institute of Electrical Engineering of the
National Central University in 1995.
Since 1995, he has been with National Taiwan Ocean University,
Keelung, Taiwan, R.O.C. He is currently the Vice Dean of Academic Affairs,
Director of Center for Teaching and Learning and a full Professor of the
Department of Marine Engineering of National Taiwan Ocean University.
Dr. Chang is now a life member of the CIEE, CACS, CSFAT and
SNAME. Since 2003, Dr Chang was listed in the Marquis Who’s Who in
Science and Engineering. In 2003, he also won the outstanding young
control engineers award granted by the Chinese Automation Control Society
(CACS). In 2004, he won the universal award of accomplishment granted by
ABI of USA. In 2005, he was selected as an excellent teacher of the National
Taiwan Ocean University. His recent research interests are fuzzy control,
robust control, performance constrained control.

Wei-Han Huang received the BS and MS degree from the Department of

Marine Engineering of the National Taiwan Ocean University, Taiwan,
R.O.C. in 2006 and 2008, respectively. His research interests focus on fuzzy Fig. 5 Responses of x3 ( k )
control and linear system control.

Wei Chang received the BS and MS degree from the Department of Marine
Engineering of the National Taiwan Ocean University, Taiwan, R.O.C. in
2003 and 2005, respectively. His research interests focus on fuzzy control
and dynamic system analysis.

Fig. 6 Responses of x4 ( k )

Fig. 3 Responses of x1 ( k )

Fig. 4 Responses of x2 ( k )

(Advance online publication: 1 August 2009)