You are on page 1of 23

Fuzzy Sets and Systems 161 (2010) 2081 – 2103

www.elsevier.com/locate/fss

Design of a parallel distributed fuzzy LQR controller for the twin


rotor multi-input multi-output system夡
C.W. Taoa , J.S. Taurb,∗ , Y.C. Chena
a Department of Electrical Engineering, National ILan University, Taiwan
b Department of Electrical Engineering, National Chung-Hsing University, Taiwan

Received 27 May 2008; received in revised form 7 April 2009; accepted 9 December 2009
Available online 23 December 2009

Abstract
In this paper, a helicopter-like twin rotor multi-input multi-output system (TRMS) is decoupled and is fuzzy Takagi–Sugeno
modeled with the complex nonlinear functions simplified as the propositional combination of linear functions. The design procedures
of the fuzzy Takagi–Sugeno model of TRMS are detailed. Based on the derived fuzzy Takagi–Sugeno model, parallel distributed
fuzzy LQR controller are designed to control the positions of the pitch and yaw angles in TRMS. The stability of the TRMS system
with the proposed fuzzy controllers is discussed. Moreover, simulation results are included to indicate the effectiveness of the
presented parallel distributed fuzzy LQR controllers for the TRMS.
© 2009 Elsevier B.V. All rights reserved.

Keywords: Nonlinear twin rotor multi-input multi-output system (TRMS); Fuzzy Takagi–Sugeno model; T–S parallel distributed fuzzy LQR
controller

1. Introduction

It is known that to design effective controllers for the highly nonlinear and complex systems is not straightforward
[1–3]. One way to alleviate this difficulty is to linearize the nonlinear complex systems and then to apply the classical
linear control laws to stabilize the systems. However, the performance of the control system might be degraded due to
the differences between the linearized and the original nonlinear systems.
Recently, the Takagi–Sugeno fuzzy model with a parallel distributed fuzzy controller seems to be a good alternative to
simplify the control of nonlinear complex systems with desired performance. With the Takagi–Sugeno fuzzy modeling
[23], the nonlinear systems are approximated by the combination of several linear subsystems in the corresponding
fuzzy state space regions. Since the linear subsystems are fuzzily combined, the Takagi–Sugeno fuzzy model is a
nonlinear model which can conserve the nonlinear characteristics of the original complex nonlinear system. Moreover,
a parallel distributed fuzzy controller can be easily designed by constructing a linear subcontroller according to each
linear subsystem. Further, the fuzzy control technique [20–22] has been shown to the robust to the possible model
uncertainties and the external disturbances. However, in many of the applications of the fuzzy T–S model techniques,
the approach to obtain the fuzzy T–S model is not mentioned. Although Tanaka and Wang [31] present an approach to

夡 This work is supported by the R.O.C. National Science Council through Grant NSC 95-2221-197-007.
∗ Corresponding author.
E-mail address: jstaur@dragon.nchu.edu.tw (J.S. Taur).

0165-0114/$ - see front matter © 2009 Elsevier B.V. All rights reserved.
doi:10.1016/j.fss.2009.12.007
2082 C.W. Tao et al. / Fuzzy Sets and Systems 161 (2010) 2081 – 2103

obtain the T–S model, the model might become complicated when the number of the nonlinear terms is large. Therefore,
an approach to obtain the fuzzy T–S model based on the system behavior combined with the similar method in [31] to
simplify the complicated nonlinear functions might alleviate the difficulties.
Because of the complicated nonlinearity and the high coupling effect between two propellers, the control problem of
the twin rotor multi-input multi-output system (TRMS) has been considered as a challenging research topic [4,5]. The
limited maximum of control actions makes the control of the TRMS even more difficult. Moreover, with the dynamic
of the TRMS similar to the dynamic of a helicopter in certain aspects [6], the control of the TRMS has gained a lot of
attention [11–13].
In the literature, two-dimensional fuzzy rules are used in [7,8] to design fuzzy controllers for the pitch and yaw angle
tracking. However, it is not easy to figure out the organization of the fuzzy rule base, since the fuzzy rules are more
complicated than the rules based on the common sense. Moreover, the number of rules is large and the complexity of
the fuzzy controllers would be high. The linear quadratic regulator (LQR) is used to control a two-degree-of-freedom
helicopter plant when the external disturbances do not exist [9]. Although the pitch and yaw angles of the helicopter
plant in [10] can be stabilized at the desired positions, a sliding mode controller is needed to improve the system
performances and to relieve the difficulties from disturbances.
With the discussion above, the Takagi–Sugeno fuzzy model for the nonlinear complex TRMS is first presented
in detail in this paper. To simplify the design of a Takagi–Sugeno (T–S) fuzzy model for the TRMS, the complex
nonlinear functions in terms of state variables with very high degree are equivalently represented as the proportional
combinations of linear functions. The state space of state variables which dominate the nonlinear behavior of TRMS
is fuzzily partitioned into suitable fuzzy regions. According to each fuzzy region, the TRMS is quasi-linearized as a
linear subsystem and the Takagi–Sugeno fuzzy model of the TRMS is obtained. Then based on the T–S fuzzy model
of the TRMS, a parallel distributed fuzzy LQR controller is designed. Applying the proposed parallel distributed fuzzy
LQR for the original nonlinear complex TRMS, the simulation results are included to indicate the effectiveness of the
parallel distributed fuzzy LQR. Moreover, the stability of the TRMS with the parallel distributed fuzzy LQR controller
is also guaranteed according to Lyapunov’s Theorem [27–32].
The remainder of this paper is organized as follows. The dynamics and mathematical model of the TRMS are
described in Section 2. In Section 3, the related mathematical background for this paper is prepared. Also, a Takagi–
Sugeno fuzzy model is created to approximate the twin rotor multi-input multi-output system in Section 4. Further
in Section 4, a parallel distributed fuzzy LQR controller is designed. The stability condition for the TRMS with the
proposed controller is discussed in Section 5. The simulation results are included in Section 6. Section 7 provides the
conclusion for this paper.

2. Model descriptions of the TRMS

2.1. Model description of the twin rotor multi-input multi-output system (TRMS)

The TRMS (shown in Fig. 1) consists of a beam pivoted on its base in such a way that it can rotate freely both in the
horizontal and vertical planes. Also, a counterbalance arm with a weight at its end is fixed to the beam at the pivot to
have the TRMS stabilizable. At both ends of the beam, there are two propellers driven by two independent DC motors.
For the control of the TRMS, the output voltages of a controller are applied to the DC motors. A change in the voltage
value results in a change of the rotation speed of the propeller to adjust the corresponding position of the beam. The
TRMS is a high order non-linear cross-coupled system.

2.2. System mathematic model

For the derivation of the approximate mathematic model of the twin rotor multi-input multi-output helicopter system,
the parameters in the TRMS are first presented as follows, lm (lt ) is the main (tail) length of the beam, Jmr (Jtr ) is the
moment of inertia for the main (tail) propeller subsystem, Tmr (Ttr ) is the time constant of the main (tail) motor-
propeller system, wm (wt ) is the rotational speeds of the main (tail) DC-motor, kv (kh ) is the friction constant of the
main (tail) propeller subsystem, S f is balanced scale, D, E and G are constants, v (h ) is the pitch (yaw) angle, v (h )
is the angular velocity around the vertical (horizontal) axis, i v (i h ) is the armature current of the main (tail) propeller
subsystem, u v (u h ) is the control input for the main (tail) DC-motor, Fv (Fh ) is the propulsive force to move the joined
C.W. Tao et al. / Fuzzy Sets and Systems 161 (2010) 2081 – 2103 2083

Fig. 1. The twin rotor multi-input multi-output system (TRMS).

beam in the vertical (horizontal) direction. Fv (Fh ) is approximately described by a nonlinear function of the angular
velocity wm (wt ), and g is gravitational acceleration. From Newton’s second law of motion the dynamic equations
and the angular velocity functions of TRMS can be obtained. With the dynamic equations modified, the mathematical
model of the twin rotor multi-input multi-output helicopter system is shown as
ẋ 1 = x2
lt S f Fh (t ) cos x4 − kh x2 − x2 x4 (D − G) sin 2x4
ẋ2 =
D sin2 x4 + E cos2 x4 + G
Jmr m x5 sin x4 + Jmr  ˙ m (u v − x6 )/Ttr
+
D sin x4 + E cos2 x4 + G
2
ẋ3 = (u h − x3 )/Ttr
ẋ4 = x5
ẋ5 = 9.1[lm S f Fv (m ) − g(0.0099 cos x 4 + 0.0168 sin x4 )
−kv x5 − 0.0252x22 sin 2x4 + Jtr 
˙ t (u h − x3 )/Ttr ]
ẋ6 = (u v − x6 )/Tmr (2.1)
where
x = [x1 x2 x3 x4 x5 x6 ] = [h h i h v v i v ]


⎪ t = 2020x35 − 194.69x 34 − 4283.15x33 + 262.27x 32

+3769.83x3
(2.2)

⎪  = 90.99x 6 + 599.73x 6 − 129.26x 6 − 1238.64x 6
6 5 4 3
⎩ m
+63.45x62 + 1283.41x6


⎪ Fh (t ) = −3 × 10−14 5t − 1.595 × 10−11 4t + 2.511 × 10−7 3t

−1.808 × 10−4 2t + 0.0801t
(2.3)
⎪ Fv (m ) = −3.48 × 10−12 5m + 1.09 × 10−9 4m + 4.123 × 10−6 3m


−1.632 × 10−4 2m + 9.544 × 10−2 t
According to the above given information, the block diagrams which indicate the relationship between the input voltages
and the propulsive forces are presented in Fig. 2.
2084 C.W. Tao et al. / Fuzzy Sets and Systems 161 (2010) 2081 – 2103

uh 1 ih t Fh
t (ih) Fh (t)
Ttrs+1

uv 1 iv m Fv
m (iv) Fv (m)
Tmrs+1

Fig. 2. The block diagrams for the propulsive forces: (a) Fh ; (b) Fv .

3. Mathematical background

3.1. Optimal LQR control

Let the state equation of a linear system be


ẋ = Ax + Bu
y = Cx
where x is the state vector, A, B are coefficient matrix, u is the control action, any y is the output vector. According to
the optimal LQR control technique [14,15], a quadratic cost function J is defined as

1 1 T T
J = xT Px + (x Qx + u T u) dt (3.1)
2 2 0
where P,  are positive definite weight matrices, and Q is a positive semi-definite weight matrix. To minimize the
quadratic cost function for the control system with these matrices Q and  specified, an optimal LQR controller can
be designed. With the Lagrangian multiplier  applied, the cost function J becomes

1 T 1 T
J = x Px + (x Qx + u T u) + T ( Ax + Bu − ẋ) dt (3.2)
2 2
Let the Hamilton–Jacobi H be

H = 21 (x T Qx + u T u) + T ( Ax + Bu) (3.3)

Then the cost function J would be minimized if both the first derivatives of J with respect to x and u vanish. That is,
*H
= u + B T  = 0 (3.4)
*u
*H
ẋ = = Ax + Bu (3.5)
*
*H
= −˙ = Qx + A T  (3.6)
*x
and  could be simply assumed as

 = Px (3.7)

From, Eq. (3.4), u can be derived as

u = −−1 B T  (3.8)

Substituting Eq. (3.8) into Eq. (3.5), the state equation can be reformulated to the

ẋ = Ax − B−1 B T  (3.9)
C.W. Tao et al. / Fuzzy Sets and Systems 161 (2010) 2081 – 2103 2085

Differentiating the Lagrange multiplier, yields

˙ = Ṗx + P ẋ (3.10)

With Eq. (3.6) and (3.9) substituted into Eq. (3.10), the Riccati equation can be expressed as

− Ṗ = A T P + P A − P B−1 B T P + Q (3.11)

Thus, using the matrix P in Eq. (3.11), the optimal control input u ∗ can be obtained as

u ∗ = −−1 B T Px = −Kx (3.12)

where K = −1 B T P is a Kalman gain.

3.2. Takagi–Sugeno fuzzy modeling

Recently, Takagi–Sugeno fuzzy modeling technique has been successfully applied to approximate complex nonlinear
systems. According to the input state variables (other input variables are possible), the state space of the nonlinear system
is fuzzily partitioned into fuzzy regions. To build a Takagi–Sugeno fuzzy model, the nonlinear system is approximated
by a linear subsystem in each fuzzy regions. That is, for a nonlinear system

ẋ = f (x, u) (3.13)

with a nonlinear function f (x, u), the Takagi–Sugeno fuzzy model can be formulated by fuzzy rules like
Model Rule j :
IF x1 is M j1 and · · · x p is M j p
THEN L j : ẋ = A j x + B j u, j = 1, 2, . . . , r (3.14)
where A j and B j are constant coefficient matrices. If the sum–product inference and centroid defuzzification are used,
the state equation of the system with a Takagi–Sugeno fuzzy model can be globally expressed as
r
j=1 w j (x){A j x + B j u}
ẋ = r
j=1 w j (x)

r
= h j (x){A j x + B j u} (3.15)
j=1
p r r
where w j (x) = k=1 M jk (xk ), h j (x) = w j (x)/ j=1 w j (x), and 0 ≤ h j (x) ≤ 1, j=1 h j (x) = 1.

3.3. Parallel distributed fuzzy LQR controller

To design a parallel distributed [16–19] fuzzy LQR controller is simply to develop an optimal LQR controller for
each linear subsystem of the Takagi–Sugeno fuzzy model. With the same antecedent as in the fuzzy rules for the T–S
model, the parallel distributed fuzzy LQR controller can be presented as
Control Rule j :
IF x1 is M j1 and · · · x p is M j p
THEN u = −K j x, j = 1, 2, . . . , r (3.16)
with K j obtained by the optimal LQR technique. Also, the global control action would be


r
u=− h j (x)K j x, j = 1, 2, . . . , r (3.17)
j=1
2086 C.W. Tao et al. / Fuzzy Sets and Systems 161 (2010) 2081 – 2103

3.4. Quasi-linearization approach

In order to construct a Takagi–Sugeno (T–S) fuzzy model, the nonlinear TRMS system with state equation in Eq.
(2.1) is quasi-linearized into several linear time invariant (LTI) subsystems with respect to different operating points.
Let the nonlinear state equation of the TRMS system be represented as

ẋ = f (x, u) (3.18)

where
⎡ ⎤
f1
⎢ f2 ⎥
⎢ ⎥
⎢ f3 ⎥
f (x, u) = ⎢


⎢ f4 ⎥

⎣ f5 ⎦
f6
⎡ ⎤
x2
⎢ lt S f Fh (t ) cos x4 −kh x2 −x2 x4 (D−G) sin 2x4 +Jmr m x5 sin x4 +Jmr 
˙ m (u v −x6 )/Ttr ⎥
⎢ ⎥
⎢ Jh ⎥
⎢ (u h −x3 )/Ttr ⎥
=⎢ ⎥ (3.19)
⎢ ⎥
⎢ x5 ⎥
⎣9.1[l S F ( ) − g(0.0099 cos x + 0.0168 sin x ) − k x − 0.0252x 2 sin 2x + J  ⎦
m f v m 4 4 v 5 2 4 tr ˙ t (u h − x 3 )/Ttr ]
(u v − x6 )/Tmr
The Taylor series expansion of Eq. (3.18) using the appropriate operating points (x ∗ , u ∗ ) can be obtained as

ẋ = Ax + Bu + Err (x ∗ , u ∗ , t) (3.20)

where
 
* f (x, u)  * f (x, u) 
A= and B = (3.21)
*x x=x ∗ *u x=x ∗

Note that * f (x, u)/*x is a matrix of x and * f (x, u)/*u is a constant matrix for the TRMS considered in this paper
(refer to Eq. (3.19)).
With the compensation of an effective controller, the system approximation error term Err (x ∗ , u ∗ , t) can be neglected.
Thus, the quasi-linearization approach is to have the nonlinear systems simply linearized as

ẋ = Ax + Bu (3.22)

4. Design of a T–S parallel distributed fuzzy LQR controller

4.1. Decoupled models of the TRMS

Since the characteristic of TRMS is very complex in the nature, it would be convenient to design a controller for
TRMS with the TRMS decoupled into horizontal and vertical subsystems. From Eq. (2.1), it is easy to see that state
equations with the state vector x h for the horizontal subsystem of the TRMS could be defined as
x h = [x1 x2 x3 ]
ẋ1 = x2
lt S f Fh (t ) cos x4 − kh x2 − x2 x4 (D − G) sin 2x4
ẋ2 =
Jh
Jmr m x5 sin x4 + [Jmr  ˙ m (u v − x6 )/Ttr ]
+
Jh
ẋ3 = (u h − x3 )/Ttr (4.1)
C.W. Tao et al. / Fuzzy Sets and Systems 161 (2010) 2081 – 2103 2087

where u h is a control action of the horizontal subsystem. Likewise, we have


xv = [x4 x5 x6 ] = [v v i v ]
ẋ4 = x5
ẋ5 = 9.1[lm S f Fv (m ) − g(0.0099 cos x4 + 0.0168 sin x4 )
−kv x5 − 0.0252x22 sin 2x4 + Jtr 
˙ t (u h − x3 )/Ttr ]
ẋ6 = (u v − x6 )/Tmr (4.2)
with u v being a control action for the vertical subsystem.
It is known that the gravity force will affect the dynamics of the vertical subsystem. In order to eliminate the steady
state error for the vertical subsystem, the integral of the angle is included to be one part of the controller output.
Accordingly, the state equations of the vertical subsystem of TRMS are modified as
  
xv = [x4 x5 x6 x7 ] = v v v i v
ẋ4 = x5
ẋ5 = 9.1[lm S f Fv (m ) − g(0.0099 cos x4 + 0.0168 sin x4 )
−kv x5 − 0.0252x22 sin 2x4 + Jtr 
˙ t (u h − x3 )/Ttr ]
ẋ6 = kiv (rv − x4 )
ẋ7 = (u v − x6 )/Tmr (4.3)
Note that with the state equations modified in Eq. (4.3), the wm becomes a function of x7 .

4.2. Simplification of nonlinear functions

To obtain a suitable Takagi–Sugeno fuzzy model with more accuracy, highly nonlinear functions in the state equations
are simplified as the propositional combination of linear functions. It can be seen from Eqs. (2.2) and (2.3) that
Fh (wt ) = F̃h (x3 ) and Fv (wm ) = F̃v (x7 ) are complex nonlinear functions in terms of state variables with very high
degree. Therefore, Fh (wt ) and Fv (wm ) would be represented as the proportional combination of linear functions.
As in Fig. 3, two linear functions
L h1 (x3 ) = 575x3 + 300
L h2 (x3 ) = 850x3 − 800 (4.4)
are arbitrarily selected to approach the tendency of the variation of F̃h (x3 ) so that the nonlinear function F̃h (x3 )
lies between them for the possible region of x3 . Let m h1 (x3 )(m h2 (x3 )) be the distance between F̃h (x3 ) and L h1 (x3 )
(L h2 (x3 )). Then the nonlinear function Fh (wt ) = F̃h (x3 ) can be represented as the proportional combination of L h1 (x3 )
and L h2 (x3 ) with the weight factor h h1 and h h2 , respectively. That is


2
F̃h (x3 ) = h h1 L h1 (x3 ) + h h2 L h2 (x3 )= h hi L hi (x3 ) (4.5)
i=1

where
m h2 (x3 )
h h1 =
m h1 (x3 ) + m h2 (x3 )
m h1 (x3 )
h h2 = (4.6)
m h1 (x3 ) + m h2 (x3 )
It can be easily seen that h h1 + h h2 = 1. By substituting the Fh (wt ) = F̃h (x3 ) in Eq. (4.5) into Eq. (4.1), the state
equations of the horizontal subsystem become


2
ẋ h = h hi f hi (x h , u h ) (4.7)
i=1
2088 C.W. Tao et al. / Fuzzy Sets and Systems 161 (2010) 2081 – 2103

where
⎡ ⎤
x2
⎢ lt S f L h1 cos x4 − kh x2 − x2 x4 (D − G) sin 2x4 + Jmr m x5 sin x4 + Jmr 
˙ m (u v − x7 )/Ttr ⎥
f h1 (x h , u h )= ⎣ ⎦
Jh
(u h − x3 )/Ttr
(4.8)
and
⎡ ⎤
x2
⎢ lt S f L h2 cos x4 − kh x2 − x2 x4 (D − G) sin 2x4 + Jmr m x5 sin x4 + Jmr 
˙ m (u v − x7 )/Ttr ⎥
f h2 (x h , u h ) = ⎣ ⎦
Jh
(u h − x3 )/Ttr
(4.9)
Likewise, Fv (wm ) = F̃v (x7 ) can be represented as


2
F̃v (x7 ) = h v1 L v1 (x7 ) + h v2 L v2 (x7 )= h vi L vi (x7 ) (4.10)
i=1

with
L v1 (x7 ) = 810x7 + 500
L v2 (x7 ) = 720x7 − 500 (4.11)
shown in Fig. 4, and
m v2 (x7 )
h v1 =
m v1 (x7 ) + m v2 (x7 )
m v1 (x7 )
h v2 = (4.12)
m v1 (x7 ) + m v2 (x7 )
By substituting the Fv (wm ) = F̃v (x7 ) in Eq. (4.10) into Eq. (4.3), the state equations of the vertically subsystem become


2
ẋv = h vi f vi (xv , u v ) (4.13)
i=1

where
⎡ ⎤
x5
˙ t (u h − x3 )/Ttr ] ⎥
⎢ 9.1[lm S f L v1 − g(0.0099 cos x 4 + 0.0168 sin x4 ) − kv x5 − 0.0252x 2 sin 2x4 + Jtr 
f v1 (xv , u v )= ⎢

2 ⎥
⎦ (4.14)
kiv (rv − x4 )
(u v − x7 )/Tmr

and
⎡ ⎤
x5
⎢ 9.1[lm S f L v2 − g(0.0099 cos x4 + 0.0168 sin x4 ) − kv x5 − 0.0252x 2 sin 2x4 + Jtr 
˙ t (u h − x3 )/Ttr ] ⎥
f v2 (xv , u v ) = ⎢

2 ⎥
⎦ (4.15)
kiv (rv − x4 )
(u v − x7 )/Tmr

4.3. Takagi–Sugeno fuzzy model of TRMS

With the quasi-linearization approach in Section 3.4, the nonlinear horizontal subsystem in Eq. (4.7) can be simply
linearized as

2 
2
ẋ h = h hi f hi (x h , u h ) = h hi ( Ahi x h (t) + Bhi u h (t)) (4.16)
i=1 i=1
C.W. Tao et al. / Fuzzy Sets and Systems 161 (2010) 2081 – 2103 2089

2000

1500
Lh1 = 575x3+300
1000

500 Fh
0

Fh
mh1
-500
mh2 Lh2 = 850x3-800
-1000

-1500

-2000
-1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1
x3

Fig. 3. Two lines selected for equivalent representation of F̃h (x3 ).

2000

1500
Lv1 = 810x7+500
1000 Fv
500
Fv

0
mv1
-500 mv2 Lv2 = 720x7-500
-1000

-1500

-2000
-1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1
x7

Fig. 4. Two lines selected for equivalent representation of F̃v (x7 ).

where
 
* f hi (x h , u h )  * f hi (x h , u h ) 
Ahi =  ∗ and Bhi =  ∗ (4.17)
*x h x=x *u h x=x
hi hi

Also, the nonlinear vertically subsystem in Eq. (4.13) is linearized to be


2 
2
ẋv = h vi f vi (xv , u v ) = h vi ( Avi xv (t) + Bvi u v (t)) (4.18)
i=1 i=1

with
 
* f vi (xv , u v )  * f vi (xv , u v ) 
Avi =  ∗ and Bvi =  ∗ (4.19)
*xvi x=x v *u v x=x v
i i
2090 C.W. Tao et al. / Fuzzy Sets and Systems 161 (2010) 2081 – 2103

0.3

0.2

Angle Velocity (rad/s)


0.1

-0.1

-0.2

-0.3
-0.6 -0.5 -0.3 -0.1 0 0.1 0.3 0.5 0.6
Angle (rad)

Fig. 5. State route graph on (x4 , x5 ) state space.

Let the horizontal and vertical subsystems of the decoupled TRMS be linearized with respect to the equilibrium
∗ = x ∗ = (0, 0, 0) and x ∗ = x ∗ = (0, 0, 0, 0). The decoupled linearized TRMS is first controlled with
point x h1 h2 v1 v2
parallel distributed LQR controllers u h and u v for the horizontal and vertical subsystems, respectively, to provide the
necessary knowledge for the design of a Takagi–Sugeno fuzzy model of TRMS. For the horizontal subsystem


2
ẋ h = h hi (Ahi x h + Bhi u h ) (4.20)
i=1

the output of a parallel distributed LQR for the horizontal subsystem would be
u h = −(h h1 K h1 + h h2 K h2 )x h (4.21)
where optimal LQR controllers −K hi x h can be obtained according to ( Ahi , Bhi ), i = 1, 2 with the Matlab function
and
Q h = diag{[1, 1, 1]}
Rh = [1] (4.22)
Likewise, the output of a parallel distributed LQR for the vertical subsystem would be
u v = −(h v1 K v1 + h v2 K v2 )xv (4.23)
for the vertical subsystem


2
ẋv = h vi (Avi xv + Bvi u v ) (4.24)
i=1

with optimal LQR controllers −K vi xv , i = 1, 2, obtained using Matlab and


Q v = diag{[1, 1, 15, 45]}
Rv = [1] (4.25)
Since the state variables x 4 and x5 dominate the behavior of the decoupled linearized system, the state route of (x4 , x5 )
are taken into consideration for linearly approximation of the decoupled TRMS. From Fig. 5, the state route of (x4 , x5 )
with the initial conditions x h0 = (0, 0, 0) and xv0 = (−0.5, 0, 0, 0) can be roughly approximated and idealized to be a
piecewise linear route consisting of three linear parts (A, B, C) (see Fig. 6).
Based on the piecewise linear routes in Fig. 6, the state space of (x4 , x5 ) is fuzzily partitioned into four fuzzy regions
(Rk , k = 1.4). To simplify the definitions of fuzzy regions Rk (k = 1.4), the variables x4 , x5 are partitioned into
fuzzy sets NB, NM, NS, ZO, PS, PM, PB with membership functions shown in Fig. 7 and the state space of (x4 , x5 ) is
C.W. Tao et al. / Fuzzy Sets and Systems 161 (2010) 2081 – 2103 2091

0.3
R1
0.2

Angle Velocity (rad/s)


B R2
0.1
A R3
C
0
R4
-0.1

-0.2

-0.3
-0.6 -0.5 -0.3 -0.1 0 0.1 0.3 0.5 0.6
Angle (rad)

Fig. 6. Approximated and idealized state route graph.

NB NM NS ZO PS PM PB

X4O1 X4O2 X4O3 X4O4 X4O5 X4O6 X4O7


x4

NB NM NS ZO PS PM PB

X5O1 X5O2 X5O3 X5O4 X5O5 X5O6 X5O7


x5

Fig. 7. Input membership functions.

partitioned into 49 fuzzy regions. The T–S fuzzy model of the decoupled TRMS is then formulated with fuzzy rules in
the following form.
Model rule r j

⎪ 
2

⎨ ẋ h = h hi ( Ah ji x h + Bh ji u h )
j
If x4 is M j4 , x5 is M j5 then STRMS : i=1 (4.26)

⎪ 2
⎩ ẋv = h vi ( Av ji xv + Bv ji u v )
i=1

and
⎧  
⎪ * f hi (x h , u h )  * f hi (x h , u h ) 

⎨ Ah ji =

*x h  ∗ , Bh ji *u h  ∗
 x=xh j x=x h j
 (4.27)
⎪ * f (x , u )  * f vi v v ) 
(x , u


vi v v 
⎩ Av ji = *xv  ∗ , Bv ji = *u v  ∗
x=x vj x=x vj

where x h∗j and xv∗j represent the equilibrium point or appropriate operating points in the corresponding fuzzy regions.
The complete rules are represented in Table 1. It can be seen that the 49 fuzzy regions can be combined into four
2092 C.W. Tao et al. / Fuzzy Sets and Systems 161 (2010) 2081 – 2103

Table 1
Fuzzy partition table.

Fuzzy regions x4

NB NM NS ZO PS PM PB

x5
NB 1
STRMS 2
STRMS 3
STRMS 4
STRMS 5
STRMS 6
STRMS 7
STRMS
NM 24
STRMS 25
STRMS 26
STRMS 27
STRMS 28
STRMS 29
STRMS 8
STRMS
NS 23
STRMS 40
STRMS 41
STRMS 42
STRMS 43
STRMS 30
STRMS 9
STRMS
ZO S 22
TRMS
39
STRMS 48
STRMS 49
STRMS 44
STRMS 31
STRMS 10
STRMS
PS 21
STRMS 38
STRMS 47
STRMS 46
STRMS 45
STRMS 32
STRMS 11
STRMS
PM 20
STRMS 37
STRMS 36
STRMS 35
STRMS 34
STRMS 33
STRMS 12
STRMS
PB 19
STRMS 18
STRMS 17
STRMS 16
STRMS 15
STRMS 14
STRMS 13
STRMS

Table 2
Fuzzy rule table.

Subsystem x4

NB NM NS ZO PS PM PB

x5

NB S1 S1 S1 S1 S1 S1 S1

NM S1 S2 S2 S2 S2 S2 S1

NS S1 S2 S3 S3 S3 S2 S1

ZO S1 S2 S3 S4 S3 S2 S1

PS S1 S2 S3 S3 S3 S2 S1

PM S1 S2 S2 S2 S2 S2 S1

PB S1 S1 S1 S1 S1 S1 S1

corresponding fuzzy regions with the consequent parts of the fuzzy rules for the T–S model specially designed (as
shown in Table 2). That is
  1
j ( Ah ji , Bh ji ) ( Ahi , Bhi
1 ),
STRMS with = S with
1
i = 1, 2 when j = 1.24,
( Av ji , Bv ji ) ( A , Bvi ),
1 1
  vi
j ( Ah ji , Bh ji ) ( A2hi , Bhi
2 ),
STRMS with = S 2 with i = 1, 2 when j = 25.40,
( Av ji , Bv ji ) ( A , B ),
2 2
  3vi vi
j ( Ah ji , Bh ji ) ( Ahi , Bhi
3 ),
STRMS with = S 3 with i = 1, 2 when j = 41.48,
( Av ji , Bv ji ) ( A , Bvi ),
3 3
  vi
j ( Ah ji , Bh ji ) ( A4hi , Bhi
4 ),
STRMS with = S 4 with i = 1, 2 when j = 49.
( Av ji , Bv ji ) ( Avi , Bvi ),
4 4
C.W. Tao et al. / Fuzzy Sets and Systems 161 (2010) 2081 – 2103 2093

Angular velocity (rad/s)

Angle (rad)

Fig. 8. Grided space.

Let x ∗j = [x h∗j xv∗j ]. Since the linearized subsystems S 4 for the region R4 are obtained with respect to the equilibrium
∗ = x 4∗ = (0, 0, 0, 0, 0, 0, 0), three appropriate operating points
point x49
⎧ 1∗
⎨x , j = 1.24
x ∗j = x 2∗ , j = 25.40
⎩ 3∗
x , j = 41.48

are necessary to be found and to be substituted into Eq. (4.27) to have appropriate linear subsystems S 1 , S 2 , S 3 for
fuzzy regions R1 , R2 , R3 , respectively.
The approach used to find the candidate operating points for fuzzy region Rk (1 ≤ k ≤ 4) is as follows.
Step 1. Grid the state space of (x4 , x5 ) to generate grid points (see Fig. 8).
Step 2. Substitute grid points in Rk into Eq. (4.19) to create possible linear models, Mik (1 ≤ i ≤ n), n is the number
of the grid points in fuzzy region Rk .
Step 3. Define a quadratic cost function J = xQx + uu for each Rk , where Q and  are selected as in Eq. (4.25).
Since the control action is constrained, the most appropriate linear model obtained by the grid point substitution
is considered to be the one which is stable with the minimum initial quadratic cost.
Step 4. According to the optimal LQR control theorem, we can obtain the control action u ik (0) for each model Mik at
the initial state (xik (0)). The initial states are assumed as

xi1 (0) = (0, 0, 0, −0.5, 0, 0, 0)


xi2 (0) = (0, 0, 0, −0.3, 0, 0, 0)
xi3 (0) = (0, 0, 0, −0.1, 0, 0, 0)
xi4 (0) = (0, 0, 0, 0, 0, 0, 0)

for four fuzzy region, respectively.



Step 5. Find minimum initial quadratic cost value J k = min1≤i≤n Jik , where Jik = xik (0)Qxik (0) + u ik (0)u ik (0).

Step 6. Define the most appropriate linear model to be the model M k which has minimum initial quadratic cost value.
Also, the most appropriate operating point x k∗ (as in Fig. 9) is defined to be the grided operating point which
generates the most appropriate linear model.

Example 1. Given two grided operating points 1 = (x14 , x15 ) and 2 = (x24 , x25 ) in Region 1. With the two grided
operating points substituted into Eq. (3.21), two linearized models, M11 and M21 are developed. Let the initial quadratic
cost functions of M11 and M21 be J11 and J21 . Assume that J11 and J21 satisfy the following relations:

J11 < J21


2094 C.W. Tao et al. / Fuzzy Sets and Systems 161 (2010) 2081 – 2103

0.3

0.2

Angle Velocity (rad/s)


0.1

-0.1

-0.2

-0.3
-0.6 -0.5 -0.3 -0.1 0 0.1 0.3 0.5 0.6
Angle (rad)

Fig. 9. The most appropriate operating points.

After sorting in step 5, the M11 is the most appropriate model for Region 1, and 1 is the most appropriate operating
point.

Up to now, the most appropriate operating points and the most appropriate linear models for all the fuzzy regions
can be developed for Eqs. (4.26) and (4.27). The T–S fuzzy model for the TRMS is then constructed as


⎪ 49 2
⎨ ẋ h =
⎪ j h hi ( Ah ji x h + Bh ji u h )
j=1 i=1
(4.28)

⎪ 49 2

⎩ ẋ v =  j h vi ( A v ji x v + Bv ji u v )
j=1 i=1

where

w j (x) 49
w j (x) = M j4 (x4 )M j5 (x5 ),  j = 49 and j = 1 (4.29)
j=1 w j (x) j=1

4.4. Design of parallel distributed fuzzy LQR controllers (PFLQRs)

With the T–S fuzzy model of the twin rotor multi-input multi-output system being built in Section 4.3, the parallel
distributed compensation concept is utilized to design fuzzy LQR controllers. In order to design an optimal LQR
controller for each linear subsystem of the T–S fuzzy model, weight matrices Q hi (Q vi ) and hi (vi ) are first be
determined for the subsystems in fuzzy region Ri , 1 ≤ i ≤ 4. Then, the optimal Kalman gains K hl j (K vl j ), j =
1.49, l = 1.2 can be generated by substituting the corresponding Q hi (Q vi ) and hi (vi ), i = 1.4 into Riccati
equation in Eq. (3.11). With the optimal Kalman gains K hl j (K vl j ), the fuzzy parallel distributed fuzzy LQR controllers
(PFLQRs) are developed to have rules as
Rule j : IF x4 is M j4 and x5 is M j5

u h j = −(h h1 K h1j + h h2 K h2j )x h , j = 1.49
T HEN (4.30)
u v j = −(h v1 K v1j + h v2 K v2j )xv , j = 1.49

and the outputs of PFLORs are



49 
2
uh = − j h hi K hi j x h
j=1 i=1

49 
2
uv = − j h vi K vi j xv (4.31)
j=1 i=1
C.W. Tao et al. / Fuzzy Sets and Systems 161 (2010) 2081 – 2103 2095

5. Stability of the TRMS with PFLQRs

In this section, the stability discussion of the TRMS with PFLQRs is provided. For a proposed T–S fuzzy model
with parallel distributed fuzzy LQR controllers, the state equation of the horizontal subsystem is

49  2 
49 2
ẋ h = j h hi k h hl ( Ah ji − Bh ji K hk
l
)x h
j=1 i=1 k=1 l=1

49 
49 
2 
2
= j k h hi h hl ( Ah ji − Bh ji K hk
l
)x h
j=1 k=1 i=1 l=1

49 
49
= j k [h h1 h h1 ( Ah j1 − Bh j1 K hk
1
) + h h1 h h2 ( Ah j1 − Bh j1 K hk
2
)
j=1 k=1
+h h2 h h1 (Ah j2 − Bh j2 K hk
1
) + h h2 h h2 ( Ah j2 − Bh j2 K hk
2
)]x h

49 
49
= j h j − 
k ( A B K h jk )x h
j=1 k=1
 

49 
49  h jk + G
G hk j
= h j j x h + 2
jjG  j k xh (5.1)
2
j=1 j=1 j<k
where
h j − 
A h jk = [h h1 h h1 ( Ah j1 − Bh j1 K hk
B K h jk = G 1
) + h h1 h h2 ( Ah j1 − Bh j1 K hk
2
)
+h h2 h h1 ( Ah j2 − Bh j2 K hk
1
) + h h2 h h2 ( Ah j2 − Bh j2 K hk
2
)]
h j j = A
G h j − 
B K h j j = [h h1 h h1 ( Ah j1 − Bh j1 K h1j ) + h h1 h h2 ( Ah j1 − Bh j1 K h2j )
+h h2 h h1 ( Ah j2 − Bh j2 K h1j ) + h h2 h h2 ( Ah j2 − Bh j2 K h2j )]
Likewise the state equation of the vertically subsystem is

49  2 
49 2
ẋv = j h vi k h vl ( Av ji − Bv ji K vk
l
)xv
j=1 i=1 k=1 l=1

49 
49 
2 
2
= j k h vi h vl ( Av ji − Bv ji K vk
l
)xv
j=1 k=1 i=1 l=1

49 
49
= j k [h v1 h v1 ( Av j1 − Bv j1 K vk
1
) + h v1 h v2 ( Av j1 − Bv j1 K vk
2
)
j=1 k=1
+h v2 h v1 (Av j2 − Bv j2 K vk
1
) + h v2 h v2 ( Av j2 − Bv j2 K vk
2
)]xv

49 
49
= j v j − 
k ( A B K v jk )xv
j=1 k=1
 

49 
49  v jk + G
G vk j
= v j j xv + 2
jjG  j k xv (5.2)
2
j=1 j=1 j<k
where
v j − 
A v jk = [h v1 h v1 ( Av j1 − Bv j1 K vk
B K v jk = G 1
) + h v1 h v2 ( Av j1 − Bv j1 K vk
2
)
+h v2 h v1 ( Av j2 − Bv j2 K vk
1
) + h v2 h v2 ( Av j2 − Bv j2 K vk
2
)]
v j j = A
G v j − 
B K v j j = [h v1 h v1 ( Av j1 − Bv j1 K v1j ) + h v1 h v2 ( Av j1 − Bv j1 K v2j )
+h v2 h v1 ( Av j2 − Bv j2 K v1j ) + h v2 h v2 ( Av j2 − Bv j2 K v2j )]
2096 C.W. Tao et al. / Fuzzy Sets and Systems 161 (2010) 2081 – 2103

With the Lyapunov theorem [24–30], the stability condition of the presented TRMS is stated in Theorem 1.

Theorem 1. The TRMS approximated by a T–S fuzzy model with a parallel distributed fuzzy LQR controller is asymp-
totically stable [31,32] if there exists common symmetric and positive definite matrices Ph , Pv and common positive
semi-definite matrices Qh , Qv such that
hT j j Ph + Ph G
G h j j + (s − 1)Qh < 0, j = 1.49
 T  
h jk + G
G hk j h jk + G
G hk j
Ph + Ph − Qh ≤ 0 (5.3)
2 2

and
vTj j Pv + Pv G
G v j j + (s − 1)Qv < 0 j = 1.49
 T  
v jk + G
G vk j v jk + G
G vk j
Pv + Pv − Qv ≤ 0 (5.4)
2 2

where j = 1.49, j < k ≤ 49, s.t. U j ∩ Uk , and s is the maximal number of rules fired at each moment. Also, U j
and Uk are the fuzzy set with membership functions  j and k , respectively.

The T–S modeled TRMS using the proposed controller is indicated to be stable with the common positive definite
matrices Ph , Pv and common positive semi-definite matrices Qh , Qv which would be found in the next section.

6. Simulation results

Although the parallel distributed fuzzy LQR controllers (PFLQRs) are designed based on the proposed T–S fuzzy
model of the TRMS, the simulation results are provided with the parameters given in Table 3 to indicate the effectiveness
and robustness of PFLQRs for the original nonlinear twin rotor multi-input multi-output system.
The appropriate operating points x k∗ , 1 ≤ k ≤ 4, obtained through the process in Section 4.3 are

x 1∗ = (0 0 0 −0.3889 0.1418 0 0)
x 2∗ = (0 0 0 −0.2950 −0.0813 0 0)
x 3∗ = (0 0 0 0.0778 − 0.0420 0 0)
x 4∗ = (0 0 0 0 0 0 0)

Table 3
The physical parameters of the TRMS.

Parameter Values Unit

lt 0.25 m
kh 0.0095 N/A
Jtr 0.000016543 kg m2
Ttr 0.3842 s
D 0.0016065 kg m2
G 0.0063306 kg m2
g 9.81 m/s2
lm 0.236 m
kv 0.00545371 N/A
Jmr 0.0000265 kg m2
Tmr 1.432 s
E 0.0490092 kg m2
Sf 0.000843318 N/A
C.W. Tao et al. / Fuzzy Sets and Systems 161 (2010) 2081 – 2103 2097

Substituting x k∗ into Eq. (4.27), the matrices ( Akhi , Bhi


k ) and (Ak , B k ), i = 1, 2 for the linear subsystems S k in region
vi vi
Rk , 1 ≤ k ≤ 4 are represented by matrices presented as follows:

⎡ ⎤ ⎡ ⎤
0 1 0 0
A1h1 = ⎣ 0 −0.2930 2.4982 ⎦ , Bh1
1
=⎣ 0 ⎦
0 0 −2.6028 2.6028
⎡ ⎤ ⎡ ⎤
0 1 0 0
A1h2 = ⎣ 0 −0.2930 3.6930 ⎦ , Bh2
1
=⎣ 0 ⎦
0 0 −2.6028 2.6028
⎡ ⎤ ⎡ ⎤
0 1 0 0 0
⎢ −1.7206 −0.0496 0 1.4670 ⎥ ⎢ ⎥
A1v1 =⎢ ⎥ , B1 = ⎢ 0 ⎥
⎣ −0.6000 0 0 0 ⎦ v1 ⎣ 0 ⎦
0 0 0 −0.6983 0.6983
⎡ ⎤ ⎡ ⎤
0 1 0 0 0
⎢ −1.7206 −0.0496 0 1.3040 ⎥ ⎢ ⎥
A1v2 =⎢ ⎥ , B1 = ⎢ 0 ⎥
⎣ −0.6000 0 0 0 ⎦ v2 ⎣ 0 ⎦
0 0 0 −0.6983 0.6983
⎡ ⎤ ⎡ ⎤
0 1 0 0
A2h1 = ⎣ 0 −0.1430 2.3677 ⎦ , Bh1
2
=⎣ 0 ⎦
0 0 −2.6028 2.6028
⎡ ⎤ ⎡ ⎤
0 1 0 0
A2h2 = ⎣ 0 −0.1430 3.5001 ⎦ , Bh2
2
=⎣ 0 ⎦
0 0 −2.6028 2.6028
⎡ ⎤ ⎡ ⎤
0 1 0 0 0
⎢ −1.6918 −0.0496 0 1.4670 ⎥ ⎢ ⎥
A2v1 =⎢ ⎥ , B2 = ⎢ 0 ⎥
⎣ −0.6000 0 0 0 ⎦ v1 ⎣ 0 ⎦
0 0 0 −0.6983 0.6983
⎡ ⎤ ⎡ ⎤
0 1 0 0 0
⎢ −1.6918 −0.0496 0 1.3040 ⎥ ⎢ ⎥
A2v2 =⎢ ⎥ , B2 = ⎢ 0 ⎥
⎣ −0.6000 0 0 0 ⎦ v2 ⎣ 0 ⎦
0 0 0 −0.6983 0.6983
⎡ ⎤ ⎡ ⎤
0 1 0 0
A3h1 = ⎣ 0 −0.1782 2.2020 ⎦ , Bh1
3
=⎣ 0 ⎦
0 0 −2.6028 2.6028
⎡ ⎤ ⎡ ⎤
0 1 0 0
A3h2 = ⎣ 0 −0.1782 3.2551 ⎦ , Bh2
3
=⎣ 0 ⎦
0 0 −2.6028 2.6028
⎡ ⎤ ⎡ ⎤
0 1 0 0 0
⎢ −1.4251 −0.0496 0 1.4670 ⎥ ⎢ ⎥
A3v1 =⎢ ⎥ , B3 = ⎢ 0 ⎥
⎣ −0.6000 0 0 0 ⎦ v2 ⎣ 0 ⎦
0 0 0 −0.6983 0.6983
⎡ ⎤ ⎡ ⎤
0 1 0 0 0
⎢ −1.4251 −0.0496 0 1.3040 ⎥ ⎢ ⎥
A3v2 =⎢ ⎥ , B3 = ⎢ 0 ⎥
⎣ −0.6000 0 0 0 ⎦ v2 ⎣ 0 ⎦
0 0 0 −0.6983 0.6983
⎡ ⎤ ⎡ ⎤
0 1 0 0
A4h1 = ⎣ 0 −0.1717 2.1906 ⎦ , Bh1
4
=⎣ 0 ⎦
0 0 −2.6028 2.6028
2098 C.W. Tao et al. / Fuzzy Sets and Systems 161 (2010) 2081 – 2103

Angle of Horizontal Angle of Vertical


1.5 0.5

0.2

Angle (rad)
1 0

Angle (rad)
0.5 -0.5

0 -1
0 5 10 15 20 0 5 10 15 20
Time (sec) Time (sec)

Angle Velocity of Horizontal Angle Velocity of Vertical


0.3 0.6

Angular Velocity (rad/s)


0.2 0.4
Velocity (rad/s)

0.1 0.2

0 0

-0.1 -0.2
0 5 10 15 20 0 5 10 15 20
Time (sec) Time (sec)

Fig. 10. Performance of the TRMS with LQR controller.

⎡ ⎤ ⎡ ⎤
0 1 0 0
A4h2 = ⎣ 0 −0.1717 3.2382 ⎦ , Bh2
4
=⎣ 0 ⎦
0 0 −2.6028 2.6028
⎡ ⎤ ⎡ ⎤
0 1 0 0 0
⎢ −1.4981 −0.0496 0 1.4670 ⎥ ⎢ ⎥
A4v1 =⎢ ⎥ , B4 = ⎢ 0 ⎥
⎣ −0.6000 0 0 0 ⎦ v1 ⎣ 0 ⎦
0 0 0 −0.6983 0.6983
⎡ ⎤ ⎡ ⎤
0 1 0 0 0
⎢ −1.4981 −0.0496 0 1.3040 ⎥ ⎢ ⎥
A4v2 =⎢ ⎥ , B4 = ⎢ 0 ⎥
⎣ −0.6000 0 0 0 ⎦ v2 ⎣ 0 ⎦
0 0 0 −0.6983 0.6983
To design a parallel distributed optimal LQR controller to satisfy the constraint on the control action, different suitable
matrices
Q h1 = diag{[1, 2, 13]}, Q v1 = diag{[100, 1, 40, 60]}
Rh1 = [1], Rv1 = [10]
Q h2 = diag{[1, 3, 1]}, Q v2 = diag{[240, 1, 40, 60]}
Rh2 = [1], Rv2 = [8]
Q h3 = diag{[1, 5, 1]}, Q v3 = diag{[210, 1, 40, 60]}
Rh3 = [1], Rv3 = [8]
Q h4 = diag{[1, 6, 1]}, Q v4 = diag{[210, 1, 40, 60]}
Rh4 = [1], Rv4 = [8.2]
C.W. Tao et al. / Fuzzy Sets and Systems 161 (2010) 2081 – 2103 2099

Angle of Horizontal Angle of Vertical


1.5 0.5

0.2

Angle (rad)
1 0

Angle (rad)
0.5 -0.5

0 -1
0 5 10 15 20 0 5 10 15 20
Time (sec) Time (sec)

Angle Velocity of Horizontal Angle Velocity of Vertical


0.3 0.6

Angular Velocity (rad/s)


0.2 0.4
Velocity (rad/s)

0.1 0.2

0 0

-0.1 -0.2
0 5 10 15 20 0 5 10 15 20
Time (sec) Time (sec)

Fig. 11. Performance of the TRMS with PFLQR.

0.5
Uh

-0.5
0 2 4 6 8 10 12 14 16 18 20
Time (sec)
0.8
0.6
Uv

0.4
0.2

0
0 2 4 6 8 10 12 14 16 18 20
Time (sec)

Fig. 12. Control action of the PFLQR for the TRMS system.

are selected for the corresponding fuzzy regions. Substituting Q hi (Q vi ) and hi (vi ), i = 1.4, into the Riccati
equation, the optimal Kalman gains are obtained as
1
K h1 = [1 1.8230 3.1832], 1
K v1 = [0.3712 3.0689 −2.0000 3.4603]
1
K h2 = [1 1.7330 3.3495], 1
K v2 = [0.2040 3.0651 −2.0000 3.2950]
2
K h1 = [1 2.0874 1.4079], 2
K v1 = [1.5218 4.7686 −2.2361 4.3418]
2100 C.W. Tao et al. / Fuzzy Sets and Systems 161 (2010) 2081 – 2103

Disturbance
0.5

0.4

Control Force (V)


0.3

0.2

0.1

0
15 20 25 30
Time (sec)

Fig. 13. Disturbance d(t) = 0.2V .

2
K h2 = [1 2.0137 1.7232], 2
K v2 = [1.2468 4.8063 −2.2361 4.1908]
3
K h1 = [1 2.5080 1.4987], 3
K v1 = [1.9513 4.8773 −2.2361 4.3844]
3
K h2 = [1 2.4516 1.8517], 3
K v2 = [1.7017 4.9388 −2.2361 4.1908]
4
K h1 = [1 2.6992 1.5580], 4
K v1 = [1.7320 4.7266 −2.2086 4.3081]
4
K h2 = [1 2.6476 1.9305], 4
K v2 = [1.4809 4.7775 −2.2086 4.1146]
The parameters used for the membership functions in Fig. 7 are simply determined by almost equal spacing the universes
of x4 and x5 ,
[X 4O1 X 4O2 X 4O3 X 4O4 X 4O5 X 4O6 X 4O7 ] = [−0.6 −0.3 −0.1 0 0.1 0.3 0.6]
[X 5O1 X 5O2 X 5O3 X 5O4 X 5O5 X 5O6 X 5O7 ] = [−0.3 −0.2 −0.1 0 0.1 0.2 0.3]
From Figs. 10 and 11, it can be seen that the TRMS with parallel distributed fuzzy LQR controllers has better perfor-
mance then the TRMS with simple LQR controllers in the sense of less overshoot. In Fig. 12, the control actions for
PFLQR are provided to illustrate that the constraints of the control actions, |u h | ≤ 1V, |u v | ≤ 1V are satisfied.
In order to indicate the robustness of the TRMS with PFLQR, a disturbance

0.2V, 20 ≤ t ≤ 25
d(t) =
0, otherwise
shown in Fig. 13 is applied. Fig. 14 indicates that the TRMS with PFLQR is robust to the disturbance and stays at the
desired positions. The control action of PFLQR when the disturbance is applied is also shown in Fig. 15.
For the angle tracking, square wave functions are used as the reference commands for the pitch and yaw angles of
the TRMS. The tracking performance in Fig. 16 is depicted to have less overshoot when compared to the tracking
performance in [12].
Moreover, semi-definite matrices Qh , Qv
⎡ ⎤
⎡ ⎤ 0.5495 −0.0900 0.0299 −0.0678
0.6482 −0.1838 0.2245 ⎢ −0.0900 0.4679 −0.0602 −0.0575 ⎥
Q h = ⎣ −0.1838 1.1802 0.0065 ⎦ , Q v = ⎢ ⎣ 0.0299 −0.0602 0.3808 −0.0093 ⎦

0.2245 0.0065 1.3399
−0.0678 −0.0575 −0.0093 0.7549
C.W. Tao et al. / Fuzzy Sets and Systems 161 (2010) 2081 – 2103 2101

Angle of Horizontal Angle of Vertical


1.5 0.5

0.2

Angle (rad)

Angle (rad)
1 0

0.5 -0.5

0 -1
0 10 20 30 40 0 10 20 30 40
Time (sec) Time (sec)

Angle Velocity of Horizontal Angle Velocity of Vertical


0.3 0.6

Angular Velocity (rad/s)


0.2 0.4
Velocity (rad/s)

0.1 0.2

0 0

-0.1 -0.2
0 10 20 30 40 0 10 20 30 40
Time (sec) Time (sec)

Fig. 14. Performance of the TRMS with PFLQR when d(t) is included.

0.5
Uh

-0.5
0 5 10 15 20 25 30 35 40
Time (sec)
1
Uv

0.5

0
0 5 10 15 20 25 30 35 40
Time (sec)

Fig. 15. Control action of the PFLQR for the TRMS with disturbance d(t).

and symmetric positive definite matrices Ph , Pv


⎡ ⎤
⎡ ⎤ 19.7186 6.3878 −10.2099 0.9178
21.4139 8.6918 2.4568 ⎢ 6.3878 13.0224 −7.6392 3.3547 ⎥
Ph = ⎣ 8.6918 12.3647 2.7109 ⎦ , Pv = ⎢
⎣ −10.2099

−7.6392 23.5358 −2.0772 ⎦
2.4568 2.7109 2.4539
0.9178 3.3547 −2.0772 2.6800
2102 C.W. Tao et al. / Fuzzy Sets and Systems 161 (2010) 2081 – 2103

Angle of Horizontal Angle of Vertical


1
0.2
0.5

Angle (rad)
Angle (rad)
0

-0.5
-0.5
-1
0 20 40 60 0 20 40 60
Time (sec) Time (sec)
Angle Velocity of Horizontal Angle Velocity of Vertical
0.4 0.5

Angular Velocity (rad/s)


Velocity (rad/s)

0.2

0 0

-0.2

-0.4 -0.5
0 20 40 60 0 20 40 60
Time (sec) Time (sec)

Fig. 16. Tracking performance of the TRMS with the PFLQR and square wave reference.

can be obtained using Matlab software to prove that the T–S fuzzy modeled TRMS with the parallel distributed fuzzy
LQR controller is asymptotically stable.

7. Conclusion

In this paper, the nonlinear complex TRMS is decoupled and Takagi–Sugeno fuzzy modeled in detail. Based on the
provided T–S fuzzy model of TRMS, parallel distributed fuzzy LQR controllers (PFLQRs) are presented. With the
proposed parallel distributed fuzzy LQR controllers being applied to the original nonlinear complex TRMS, the yaw
and pitch angle are positioned at the desired values. The simulation results are included to indicate the effectiveness of
the designed parallel distributed fuzzy LQR controllers. Moreover, the stability of the T–S fuzzy modeled TRMS with
the PFLQRs is discussed.

References

[1] J.Q. Huang, F.L. Lewis, Neural-network predictive control for nonlinear dynamic-systems with time-delay, IEEE Trans. Neural Networks 14
(2) (2003) 377–389.
[2] H.S. Binning, D.P. Goodall, Constrained feedback-control of imperfectly known singularly perturbed nonlinear-systems, Int. J. Control 73 (1)
(2000) 49–62.
[3] W.M. Haddad, T. Hayakawa, V. Chellaboina, Robust adaptive-control for nonlinear uncertain system, Automatica 39 (3) (2003) 551–556.
[4] W.-Y. Wang, T.-T. Lee, H.-C. Huang, Evolutionary design of PID controller for twin rotor multi-input multi-output system, in: IEEE Congress
on Intelligent Control and Automation, Vol. 2, June 2002, pp. 913–917.
[5] S.M. Ahmad, A.J. Chipperfield, O. Tokhi, Dynamic modeling and optimal control of a twin rotor MIMO system, in: IEEE Conf. on National
Aerospace and Electronics, October 2000, pp. 391–398.
[6] B.U. Islam, N. Ahmed, D.L. Bhatti, S. Khan, Controller design using fuzzy logic for a twin rotor MIMO system, in: IEEE Conf. on multi topic,
December 2003, pp. 264–268.
[7] A. Rahideh, M.H. Shaheed, Hybrid fuzzy-PID-based control of a twin rotor MIMO system, in: IEEE Conf. Industrial Electronics, November
2006, pp. 48–53.
C.W. Tao et al. / Fuzzy Sets and Systems 161 (2010) 2081 – 2103 2103

[8] C.-S. Liu, L.-R. Chen, B.-Z. Li, S.-K. Chen, Z.-S. Zeng, Improvement of the twin rotor MIMO system tracking and transient response using
fuzzy control technology, in: IEEE Conf. on Industrial Electronics and Applications, May 2006, pp. 1–6.
[9] Z. Jiang, J. Han, Y. Wang, Q. Song, Enhanced LQR control for unmanned helicopter in hover, in: IEEE Conf. on Systems and Control in
Aerospace and Astronautics, January 2006, pp. 1438–1443.
[10] G.-R. Yu, H.-T. Liu, Sliding mode control of a two degree of freedom helicopter via linear quadratic regulator, in: IEEE Conf. on Systems, Man
and Cybernetics, Vol. 4, October 2005, pp. 3299–3304.
[11] I.Z. Mat Darus, F.M. Aldebrez, M.O. Tokhi, Parametric modeling of a twin rotor system using genetic algorithms, in: IEEE Sympos. on
Communications and Signal Processing, 2004, pp. 115–118.
[12] J.-G. Juang, K.-T. Tu, W.-K. Liu, Optimal fuzzy switching grey prediction with RGA for TRMS control, in: IEEE Conf. on Systems, Man, and
Cybernetics, Vol. 1, October 2006, pp. 681–686.
[13] T.-S. Kim, J.-H. Yan, Y.-S. Lee, O.-K. Kwo, Twin rotors system modeling and bumpless transfer implementation algorithm for LQ control, in:
IEEE Conf. on International Joint, October 2006, pp. 114–119.
[14] H. Zhang, L. Xie, Linear quadratic regulation for discrete-time systems with multiplicative noises and input delays, in: Proc. of the 2006
American Control Conf., June, 2006, pp. 1718–1723.
[15] J. Lavaei, A.G. Aghdam, Optimal generalized sampled-data hold functions with a constrained structure, in: Proc. of the 2006 American Control
Conf., June 2006, pp. 200–206.
[16] C.-S. Chiu, Mixed feedforward feedback based adaptive fuzzy control for a class of MIMO nonlinear systems, IEEE Transactions on Fuzzy
Systems January (2005) (NSC-93-2213-E-270-001).
[17] B.M. Al-Hadithi, F. Matia, A. Jimenez, Global stability of fuzzy systems based variable structure control, in: IEEE Fuzzy Inf. Process. Soc.,
2002, pp. 210–215.
[18] G. Xijin, Z. Xianlin, The optimization of T–S fuzzy model feedback control matrix based on genetic arithmetic, in: IEEE Region 10 Conf. on
Computers, Communications, Control and Power Engineering, 2002, pp. 1467–1470.
[19] S. Wang, P. Hu, L. Lin, Study on state feedback fuzzy-predictive control system of T–S fuzzy model, in: Proc. of the First Internat. Conf. on
Innovative Computing, Information and Control, 2006.
[20] X.-J. Ma, Z.-Q. Sun, Y.-Y. He, Analysis and design of fuzzy controller and fuzzy observer, IEEE Transactions on Fuzzy Systems 6 (1) (1998).
[21] J. Zhang, M. Fei, Analysis and design of robust fuzzy controllers and robust fuzzy observers of nonlinear systems, in: Proc. of the 6th World
Congress on Intelligent Control and Automation, 2006, pp. 3767–3771.
[22] G.J. Klir, B. Yuan, Fuzzy Sets and Fuzzy Logic Theory and Application, Prentice-Hall, Englewood Cliffs, NJ, 1995.
[23] T. Takagi, M. Sugeno, Fuzzy identification of systems and its applications to modeling and control, IEEE Trans. Syst. Man. Cybern. 15 (1985)
116–132.
[24] H.O. Wang, K. Tanaka, M.F. Griffin, An approach to fuzzy control of nonlinear systems: stability and design issues, IEEE Trans. Fuzzy Syst.
4 (1) (1996) 14–23.
[25] M. Vidyasagar, Nonlinear Systems Analysis, Prentice-Hall, Englewood Cliffs, NJ, 1993.
[26] K. Tanaka, T. Taniguchi, H.O. Wang, Model-based fuzzy control of TORA system: fuzzy regulator and fuzzy observer design via LMIs that
represent decay rate, disturbance rejection, robustness, optimality, in: Seventh IEEE Internat. Conf. on Fuzzy Systems, Alaska, 1998, pp.
313–318.
[27] K. Tanaka, T. Ikeda, H.O. Wang, Design of fuzzy control systems based on relaxed LMI stability conditions, in: 35th IEEE Conf. on Decision
and Control, Kobe, Vol. 1, 1996, pp. 598–603.
[28] K. Tanaka, T. Ikeda, H.O. Wang, Fuzzy regulators and fuzzy observers, IEEE Trans. Fuzzy Syst. 6 (2) (1998) 250–265.
[29] K. Tanaka, M. Sugeno, Stability analysis of fuzzy systems using Lyapunov’s direct method, in: Proc. of NAFIPS’90, 1990, pp. 133–136.
[30] K. Tanaka, M. Sugeno, Stability analysis and design of fuzzy control systems, Fuzzy Sets Syst. 45 (2) (1992) 135–156.
[31] K. Tanaka, H.O. Wang, Fuzzy Control Systems Design and Analysis, Prentice-Hall International Editions, 2001.
[32] L.-X. Wang, A Course in Fuzzy Systems and Control, Prentice-Hall International Editions, 2005.

You might also like