You are on page 1of 9

Soft Comput (2005) 9: 430–438

DOI 10.1007/s00500-004-0358-7

ORIGINAL PAPER

C. W. Tao  J. S. Taur

An approach for the robustness comparison between piecewise


linear PID -like fuzzy and classical PID controllers

Published online: 20 April 2004


Ó Springer-Verlag 2004

Abstract The comparison of the stability robustness clarify the reason why the fuzzy control techniques are
between the classical PID controller and two piecewise preferable to the classical control techniques. One of the
linear PID-like fuzzy controllers to the variations of the most significant advantages fuzzy control techniques
parameters in the second order plant is provided in this have is the stability robustness to the structured (para-
paper. The definition of a stability robust controller (to metric) and the unstructured (unmodelled) uncertainties.
the parameter variations of the plant model) is pre- In the literature, many robust fuzzy controllers are
sented. Then Kharitonov’s theorem is applied to find the successfully designed and the robustness of the fuzzy
regions of robustness to the parameter variations for the controller is described. For example, Kawaji et al. [5]
control systems with different controllers. Based on the discuss the robustness of a fuzzy control system based
size of regions of robustness, the relative robustness on the theory of the variable structure system. Also, in
factor is defined, and the robustness comparison is the work of Yi [12], the robustness of the fuzzy con-
provided. For every classical PID controller with gain troller with the sliding mode structure is analyzed in the
coefficients determined and fixed, it is shown that we can sense of Lyapunov. The fuzzy control system is claimed
always design the piecewise linear PID-like fuzzy con- to be robust by Tanaka [8] if the modified H 1 stability
trollers to be more robust than the specific classical PID conditions are satisfied. Other techniques like quadratic
controller. The results of robustness comparison is fur- stabilization can be used to design robust fuzzy con-
ther confirmed in the simulation included for the second trollers if the stability conditions are matched. Although
order uncertain plant. there exist works which illustrate that fuzzy control
system is more robust than the classical control system
Keywords Robustness  PID-like fuzzy controller  with the simulation or experimental results [2], the the-
Classical PID controller oretical-like comparison of the robustness between the
fuzzy control system and the system with classical con-
troller has not been provided. Without the robustness
1 Introduction comparison, it is not easy to convince people that robust
fuzzy control is necessary since the classical control
Fuzzy logic technique [13] has been successfully applied techniques might also be robust to the specific systems.
on the control system without exact mathematical plant For the sake of complexity reduction, the PID-like
model in the past several years [3]. The fuzzy controllers fuzzy controller has been designed to be a piecewise
are also shown to have satisfactory system performance linear controller with one dimensional fuzzy rule
for the complex systems [4, 9]. As the fuzzy control base [11, 9]. The piecewise linear PID-like fuzzy con-
techniques get more popular, it is more important to troller is considered to be a generalized PID controller
and is expected to have better performance than the
classical PID controller. In this paper, the robustness of
C. W. Tao the piecewise linear PID-like fuzzy and classical PID
Department of Electrical Engineering, controllers is compared for a plant with second order
National Ilan University,
I-Lan, Taiwan
uncertain model. The controller which is stability robust
to the parameter variations of the plant model is defined.
J. S. Taur (&) With the application of the Kharitonov’s theorem, the
Department of Electrical Engineering,
National Chung Shing University, piecewise linear robust PID-like fuzzy and robust clas-
Taichung, Taiwan sical PID controllers can be developed. Also, the regions
e-mail: jstaur@dragon.nchu.edu.tw of robustness to the parameter variations for the control
431

systems with different controllers can be determined 2.1 Control system with classical PID controller
with the Kharitonov’s theorem. Based on the size of
regions of robustness, the relative robustness factor is Without loss of generality, the transfer function GPID ðsÞ
defined, and the robustness comparison is provided. The of a PID controller is defined to have the gain coefficient
controller is said to be more robust to the parameter of the system error be equal to one, i.e.,
variations if its region of robustness is larger. For every kI
classical PID controller with gain coefficients determined GPID ðsÞ ¼ 1 þ kd s þ ;
and fixed, it is indicated that we can always design s
piecewise linear PID-like fuzzy controllers to be more then the output of the PID controller is
robust than the specific classical PID controller. More- Zt
over, the robustness comparison between two piecewise u ¼ e þ kd e_ þ kI e dt
linear PID-like fuzzy controllers indicate the better
choice of design of piecewise linear PID-like fuzzy con- 0
trollers. The results of robustness comparison is empir- and the Laplace transform of the output of the control
ically confirmed in the simulation included for the system is
second order uncertain plant. It is easy to find that the
GPID ðsÞP ðsÞ<ðsÞ
proposed approach for the robustness comparison can YPID ðsÞ ¼
be extended to the linear system with higher order. 1 þ GPID ðsÞP ðsÞ
The remainder of this paper is organized as follows.   g 
1 þ kd s þ ksI s2 þdsþm <ðsÞ
In Sect. 2, the second order plant model with uncertain ¼  
 
parameters is described. The control systems with clas- g
1 þ 1 þ kd s þ ksI s2 þdsþm
sical PID and two different piecewise linear PID-like
fuzzy controllers are reviewed in Sect. 2. In Sect. 3, the gðkd s2 þ s þ kI Þ<ðsÞ
¼ ; ð2Þ
robust controller and relative robustness factor are de- s3 þ ðd þ kd gÞs2 þ ðm þ gÞs þ kI g
fined. Also, Kharitonov’s theorem is introduced to
analyze the stability of the system with interval uncer- where kd , kI are the positive gain factors, and the <ðsÞ is
tainties in Sect. 3. In Sect. 4, the robustness of the the transfer function of the reference input.
piecewise linear PID-like fuzzy and classical PID con-
troller is compared. Simulation results are given in
Sect. 5. Finally, a conclusion is provided in Sect. 6. 2.2 Control system with piecewise linear PID-like fuzzy
controller FC1

The schematic diagram for the robust complexity


2 Control systems with classical and fuzzy controllers
reduced PID-like fuzzy controller FC1 is shown in Fig. 2.
In Fig. 2, the error e ¼ r  y, where r is a constant ref-
In this section, the transfer functions of the outputs of
erence input and y is the output of the system. The
the control systems with classical PID and two different
inputs for the PID-like fuzzy controller FC1 are error (e),
piecewise linear PID-like fuzzy controllers are derived.
derivative
Rt of the error (e),
_ and the integral of the error
The control system in Fig. 1 is considered to have a
( 0 edt). The fuzzy mechanism of the fuzzy controller
second order plant without exact values of the parame-
FC1 is designed to have the fuzzy if-then rules with the
ters in this paper. Let the uncertain second order plant
input [3]:
be
Zt
y€ þ d y_ þ my ¼ gu
x1 ¼ e þ kd e_ þ kI e dt ð3Þ
with zero initial conditions, where g > 0 is assumed to be
0
a known constant for simplicity and d, m are the possible
uncertain parameters with the known nominal values. where kd and kI are positive gain factors. As for the
Thus, the plant model can be defined with the transfer classical PID controller, the combination coefficient for
function as the error in Eq. (3) is adopted to be one. The output
g
P ðsÞ ¼ 2 : ð1Þ
s þ ds þ m

Fig. 1 Block diagram of the closed loop control system with a


uncertain plant Fig. 2 Structure of the PID-like fuzzy controller FC1
432

variable of the fuzzy controller is u1 . The universes of For each region R1i , i ¼ 3; 2; . . . ; 2; in the input
discourse of x1 and u1 are fuzzily partitioned into seven space (see Fig. 3), two of the fuzzy rules are activated.
fuzzy sets as shown in Fig. 3. Then the jth fuzzy if-then The output of the fuzzy controller u1 is
rule can be expressed as the following: 1
x1  b1i 1 biþ1  x1
u1 ¼c1iþ1 þ c i
Rule1j : ‘‘If x1 is A1j , then u1 is Cj1 .’’ b1iþ1  b1i b1iþ1  b1i
where the A1j and Cj1 are one of the seven fuzzy sets of the c1iþ1  c1i b1iþ1 c1i  b1i c1iþ1
¼ x 1 þ
input variable and output variable of the fuzzy mecha- b1iþ1  b1i b1iþ1  b1i
nism in FC1 , respectively. Based on the knowledge of
conventional PID controllers, the fuzzy if-then rules are ¼b1i x1 þ c1i ; ð4Þ
developed in Table 1. To calculate the output u1 , the where
centroid method is used for defuzzification. Then, as in
Kosko’s work [6], the crisp output u1 of the fuzzy con- c1iþ1  c1i b1iþ1 c1i  b1i c1iþ1
b1i ¼ ; and c1i ¼ :
troller FC1 can be derived: b1iþ1  b1i b1iþ1  b1i
P3
c1i fi1 ðxÞ with the output of the fuzzy controller u1 in Eq. (4), the
u1 ¼ Pi¼3
3 1 Laplace transform of the output of the control system in
i¼3 fi ðxÞ
region R1i is obtained as
where fi1 is one of the input membership functions of the
fuzzy controller FC1 . Since the inputs of the fuzzy con- YiFC1 ðsÞ
troller FC1 are linearly combined, the number of fuzzy
sP ðsÞC1i ðsÞ þ b1i P ðsÞðkd s2 þ s þ kI Þ<ðsÞ
if-then rules of the fuzzy mechanism used in FC1 depends ¼
only on the partition of the single input variable x1 . s þ b1i P ðsÞðkd s2 þ s þ kI Þ
Thus, the complexity of fuzzy controller FC1 is signifi- sgC1i ðsÞ þ b1i gðkd s2 þ s þ kI Þ<ðsÞ
cantly reduced. ¼ ; ð5Þ
s3 þ ðd þ b1i kd gÞs2 þ ðm þ b1i gÞs þ b1i kI g
c1
where C1i ðsÞ ¼ si is the Laplace transform of c1i , and c1i is
considered as a constant input in region R1i . Also,
<ðsÞ ¼ rs is the Laplace transform of the reference input r.

2.3 Control system with piecewise linear PID-like fuzzy


controller FC2

The diagram for the structure of the robust complexity


reduced PID-like fuzzy controller FC2 is shown in Fig. 4.
Since FC2 is also a PID-like fuzzy controller, the inputs
for the FC2 are error (e), derivative
Rt of the output (y),
_ and
the integral of the error ( 0 edt).
Rt Instead of taking all the
input information (e, e,
_ and 0 edt) to obtain the input of
the fuzzy mechanism (as in FC1 ), the fuzzy mechanism of
the fuzzy controller FC2 is designed to have the fuzzy
if-then rules with the input [10]:
x2 ¼ e  kd y_ ; ð6Þ
where kd is a positive gain factor. Note that since the
reference input r is a constant, e_ ¼ y.
_ Let the output

Fig. 3 Input and output membership functions of FC1

Table 1 The heuristic rules for the fuzzy mechanism (in FC1 )
with input variable x1
x1

nb nm ns ze ps pm pb
nb nm ns ze ps pm pb
Fig. 4 Structure of the PID-like fuzzy controller FC2
433

variable of the fuzzy controller FC2 be u2 , the universes where fi2 is one of the input membership functions of the
of discourse of x2 and u2 are fuzzily partitioned into fuzzy controller FC2 , and kI is a positive integral gain
seven fuzzy sets in Fig. 5. Then the jth fuzzy if-then rule factor. It is easy to see that the fuzzy mechanism of FC2
can be expressed as the following: has one dimensional fuzzy rule base. Thus, the fuzzy
controller FC2 is also complexity reduced. The output u2
Rule2j : ‘‘1If x2 is A2j , then u2 is Q2j .’’ in each region R2i , i ¼ 3; 2; . . . ; 2; of the input space
(see Fig. 5) is
where the A2j and Q2j are one of the seven fuzzy sets of the Z
input variable and output variable in FC2 , respectively. 2 x2  b2i
In order to have the outputR of the fuzzy controller FC2 u2 ¼ðciþ1 þ kI e dtÞ 2
t biþ1  b2i
include the information 0 edt, the q2j in Fig. 5 is Z
designed to be b2  x2
þ ðc2i þ kI e dtÞ 2iþ1
biþ1  b2i
Zt Z
q2j ¼ c2j þ kI edt: c2  c2i b2iþ1 c2i  b2i c2iþ1
¼ iþ1 x 2 þ þ k I e dt
0 b2iþ1  b2i b2iþ1  b2i
Z
with the fuzzy if-then rules in Table 2, the output u2 is ¼b2i x2 þ c2i þ kI e dt ; ð7Þ
calculated by using the centroid defuzzification method.
Then, as in our previous work [10], the output u2 of the where
fuzzy controller FC2 can be derived:
c2iþ1  c2i b2iþ1 c2i  b2i c2iþ1
P3 Rt b2i ¼ ; and c2i ¼ :
P3 2
i¼3 ðci þ kI edtÞfi2 ðxÞ b2iþ1  b2i b2iþ1  b2i
q2i fi2 ðxÞ 0
u2 ¼ Pi¼3
3
¼ P3 with the output of the fuzzy controller u2 in Eq. 7, the
2 2
i¼3 fi ðxÞ i¼3 fi ðxÞ Laplace transform of the output of the control system in
region R2i is
sP ðsÞC2i ðsÞ þ ðb2i s þ kI ÞP ðsÞ<ðsÞ
YiFC2 ðsÞ ¼
s þ P ðsÞðb2i kd s2 þ b2i s þ kI Þ
sgC2i ðsÞ þ gðb2i s þ kI Þ<ðsÞ
¼ ; ð8Þ
s3 þ ðd þ b2i kd gÞs2 þ ðm þ b2i gÞs þ kI g
c2
where C2i ðsÞ ¼ si is the Laplace transform of c2i , and c2i
is considered as a constant input in region R2i . Also,
<ðsÞ ¼ rs is the Laplace transform of the reference
input.

3 Definition of a robust controller and Kharitonov’s


theorem

In this Section, the definitions of the robust controller


and relative robustness factor are first provided and then
the Kharitonov’s theorem is introduced to enable the
analyses of the stability robustness of the system with
uncertain parameters [7].

Fig. 5 Input and output membership functions of FC2 3.1 Definition of a robust controller

Let Dl be the variation interval of the parameter dl ,


l ¼ 0; 1; . . . ; n  1; for the transfer function Tp ðs; dÞ of
Table 2 The heuristic rules for the fuzzy mechanism (in FC2 ) the plant p, and
with input variable x2
Kc
x2 Tp ðs; dÞ ¼ Pn1 l ; Kc is a constant gain: ð9Þ
n
s þ l¼0 dl s
nb nm ns ze ps pm pb The following definitions are provided to indicate the
nb nm ns ze ps pm pb
meaning of a robust controller.
434

Definition 1 (A family of plants) A family of plants, Qp , are stable. Note that the interval Di , i ¼ 1; 2; . . . ; n, is
is defined to be a collection of plants such that every plant allowed to be a lumping interval. For a lumping interval
p in Qp has the transfer function Tp ðs; dÞ, Dj ,
Dj ¼ f ðD0 ; D1 ; . . . ; Di ; . . . ; Dn Þ; j 6¼ i;
Kc
Tp ðs; dÞ ¼ Pn1 ; Kc is a constant gain; with f ðÞ is a function, the overbounding technique can
sn þ l¼0 dl sl be utilized for simplicity to find the bounds of Dj ,
and dj ¼ min ðDj Þ;
d0 2D0 ;d1 2D1 ;...;di 2Di ;...;dn 2Dn
dl 2 D l ; 8l:
and
djþ ¼ max ðDj Þ:
d0 2D0 ;d1 2D1 ;...;di 2Di ;...;dn 2Dn
Definition 2 A controller Gc is said to be a robust
controller if the controller Gc can stabilize every plant in In this case, only the sufficient condition of Kharitonov’s
the family Qp . condition is guaranteed [1].

3.3 Definition of the relative robustness factor


3.2 Kharitonov’s theorem
Since different controllers might have different robust-
To analyze the stability of the uncertain control system ness abilities to the parameter variations of the uncertain
with all the possible paramter variations considered, the plant, it is important to have the robustness comparison
characteristic polynomial of the uncertain system is for different controllers. In order to compare the
represented by a characteristic polynomial with interval robustness between different controllers, the range of
parameters, robustness and the relative robustness factor are defined.
X
n1 X
n1
~ ¼ sn þ
pðs; DÞ Dl sl ¼ sn þ ½dl dlþ sl ; ð10Þ
l¼0 l¼0 Definition 3 (Range of robustness RgGc ) If the control
system with the controller Gc is stable for every value of
where dl , dlþ are the lower-bound value and the upper- the plant parameter g 2 R~gGc , then RgGc ¼ maxðR~gGc Þ is a
bound value of the variation interval Dl , respectively. range of robustness of the controller Gc to the variation of
~
Associated with the characteristic polynomial pðs; DÞ, the plant parameter g.
four Kharitonov polynomials [7] are
K1 ðsÞ ¼d0 þ d1 s þ d2þ s2 þ d3þ s3 þ d4 s4 þ   
Definition 4 (Relative robustness factor RR) The relative
K2 ðsÞ ¼d0þ þ d1þ s þ d2 s2 þ d3 s3 þ d4þ s4 þ    robustness factor RRgGc1 =Gc2 of Gc1 to Gc2 is defined to be
K3 ðsÞ ¼d0þ þ d1 s þ d2 s2 þ d3þ s3 þ d4þ s4 þ   
K4 ðsÞ ¼d0 þ d1þ s þ d2þ s2 þ d3 s3 þ d4 s4 þ    ð11Þ SðRgGc1 Þ
RRgGc1 =Gc2 ¼ ;
SðRgGc2 Þ
If the coefficient of sn is not zero (as in Eq. 10), the
~ is a degree invariant polynomial. For
polynomial pðs; DÞ where SðRÞ is the size of the range R.
~ the
a degree invariant characteristic polynomial pðs; DÞ, With the Definition 3, we can define that the con-
Kharitonov’s theorem is as follows. troller Gc1 is more robust than the controller Gc2 to the
variation of the plant parameter g if the
RRgGc1 =Gc2 > 1:
Theorem 1 (Kharitonov’s theorem (1978)) If a cha-
~ with intervals is degree
recteristic polynomial pðs; DÞ
invariant, the system with the charateristic polynomial
~ is stable if and only if the systems with four
pðs; DÞ 4 Comparison of the stability robustness between
Kharitonov polynomials as charecteristic polynomials are controllers
stable.
Since all the structures of the classical PID controller,
In the Kharitonov’s theorem, the system with the PID-like fuzzy controllers FC1 and FC2 include the terms
~ being stable means
charateristic polynomial pðs; DÞ Z
that a family of systems which have the charateristic kI e dt and  kd y; _ ðkd e_ ¼ kd ðr_  yÞÞ
_ ;
polynomial pðs; dÞ,
X
n1 the gain coefficients kd , and kI are taken to be the same
pðs; dÞ ¼ sn þ d l sl ; dl 2 ½dl edlþ  8l; for the classical PID controller, FC1 and FC2 to denote
l¼0 that for a specific classical PID controller (with gain
435

coefficients determined and fixed), the corresponding It is known that the conditions
FC1 and FC2 can be designed to have the the same gain 
d þ kd g > 0;
coefficients (as the specific classical PID controller).
kI g > 0;
Moreover, the b1i of FC1 and b2i of FC2 ,
i ¼ 3; 2; . . . ; 2, are designed to satisfy are always satisfied for the designed classical PID con-
b¼ b1i ¼ b2i ; 8i : troller, since no changes will happen according to the
parameter variation. Therefore, in order to find the
To compare the stability robustness to the variation of range of robustness, we need to find the mþ and m to
the uncertain parameter m in the second order plant (Eq. satisfy the conditions
1) between classical PID controller, FC1 , and FC2 , the
ðmþ þ gÞðd þ kd gÞ  kI g > 0; ð16Þ
parameter d is assumed to be a constant for simplicity.
From the discussion in Sections 2 and 3, the ranges of and
robustness for different controllers are studied in the
ðm þ gÞðd þ kd gÞ  kI g > 0: ð17Þ
following paragraphs.
Based on the common sense that mþ  m , the range of
robustness can be obtained from Eqs. (16) and (17),
Range of robustness for the classical PID controller   
kI g
Rm
PID ¼  g 1 : ð18Þ
Let the gain factors kd ¼ kd , kI ¼ kI be designed for the d þ kd g
classical PID controller to have the stability robust (see
the Definitions 1 and 2) to the variation of the parameter
m in the plant model. Thus, to find the range of robustness
Rm  Range of robustness for the controller FC1
PID for the designed classical PID controller with kd ¼ kd ,
 m
kI ¼ kI is to find a largest interval RPID such that the
control system with the designed classical PID controller The fuzzy controller FC1 is designed to make the control
is stable for every m 2 Rm system be able to get through the regions R1I and R1II in
PID . The characteristic polynomial
of the control system with the classical PID controller can Fig. 3. Thus, the stability of the fuzzy control system is
be obtained from Eq. 2. With m being represented by an discussed without the consideration of the regions R1I
interval M ¼ ½m mþ  in the characteristic polynomial, and R1II . In order to show that for every classical PID
an interval characteristic polynomial, PPID ðs; MÞ, controller with gain coefficients determined and fixed,
the piecewise linear PID-like fuzzy controller can always
PPID ðs; MÞ be designed to be more robust than the specific classical
¼ s3 þ ðd þ kd gÞs2 þ ðM þ gÞs þ kI g PID controller, the same gain factors kd ¼ kd , kI ¼ kI (as
determined and fixed in classical PID controller) are
¼ s3 þ ðd þ kd gÞs2 þ ð½m mþ  þ gÞs þ kI g; ð12Þ assigned to the controller FC1 . The characteristic poly-
is generated and g; d are assumed to be constants. Then, nomial of the control system with the controller FC1 is
the four Kharitonov polynomials associated with the the modified denominator of Eq. (5). With m being
interval characteristic polynomial PPID ðs; MÞ are represented by an interval M ¼ ½m mþ  in the charac-
teristic polynomial, an interval characteristic polyno-
K1 ðsÞ ¼kI g þ ðm þ gÞs þ ðd þ kd gÞs2 þ s3 mial, PFC1 ðs; MÞ is generated, and
K2 ðsÞ ¼kI g þ ðmþ þ gÞs þ ðd þ kd gÞs2 þ s3 PFC1 ðs; MÞ ¼s3 þ ðd þ bkd gÞs2 þ ðM þ bgÞs þ bkI g
K3 ðsÞ ¼kI g þ ðm þ gÞs þ ðd þ kd gÞs2 þ s3 ¼s3 þ ðd þ bkd gÞs2 þ ð½m mþ  þ bgÞs
K4 ðsÞ ¼kI g þ ðmþ þ gÞs þ ðd þ kd gÞs2 þ s3 ð13Þ þ bkI g ; ð19Þ
with the Routh Hurwitz criterion, the conditions for the By following the similar procedure in the previous par-
systems with the Kharitonov’s polynomials as charac- agraph, the range of the robustness for the controller
teristic polynomials to be stable are FC1 is derived as
8 d þ k  g > 0;  
> bkI g
< ðmþ þd gÞðd þ k  gÞ for K ðsÞand K ðsÞ Rm
FC1 ¼  bg 1 : ð20Þ
d 2 4
ð14Þ d þ bkd g

:  kI g > 0;
>
kI g > 0;
and
8 d þ k  g > 0; Range of robustness for the controller FC2
>
< ðm þd gÞðd þ k  gÞ
d for K1 ðsÞand K3 ðsÞ Let the controller FC2 be designed to also have the same
 ð15Þ
:  kI g > 0;
> gain factors kd ¼ kd , kI ¼ kI . With m being represented
kI g > 0: by an interval M ¼ ½m mþ  in the characteristic poly-
436

nomial (which is the denominator of Eq. 8), an interval be more robust than the classical PID controllers to the
characteristic polynomial PFC2 ðs; MÞ is defined as variation of the parameter m. Likewise, the robustness
comparison between controllers to the variation of the
PFC2 ðs; MÞ ¼ s3 þ ðd þ bkd gÞs2 þ ðM þ bgÞs þ kI g
parameter d in the plant model (with the parameter m
¼ s3 þ ðd þ bkd gÞs2 þ ð½m mþ  þ bgÞs assumed to be a constant) can be implemented. If the b
þ kI g; ð21Þ for the fuzzy controllers FC1 and FC2 is designed to be
   
kI m
Then the range of the robustness for the controller FC2 b > max 1;  m g ; ð23Þ
can be found, kd ðm þ gÞ
 
kI g then
Rm
FC2 ¼  bg 1 ; ð22Þ
d þ bkd g SðRdFC2 Þ > SðRdFC1 Þ > SðRdPID Þ :
with the similar procedure in the previous paragraph. As Also, the controller FC2 is found to be always more
defined in Sect. 3, the controller Gc1 is more robust than robust than the fuzzy controller FC1 in both cases. It can
the controller Gc2 to the variation of the plant parameter be found that if the gain coefficient for the error of
g if the classical PID controller and the combinational coeffi-
RRgGc1 =Gc2 > 1: cient for the error of the fuzzy controller are equivalent
and are not equal to one, the similar results can be ob-
That is, if the size of the range of robustness, SðRgGc1 Þ, is tained for the robustness comparison.
larger than the size of the range of robustness, SðRgGc2 Þ,
then the controller Gc1 is more robust than the controller
Gc2 . From the discussion above, it is easy to know that if
the b for the fuzzy controllers FC1 and FC2 is designed to 5 Simulation results
be
     In this Section, the simulation results are provided to see
kI d the fact that the fuzzy controller FC1 and FC2 are more
b > max 1; d kd g ;
d þ kd g robust than the classical PID controller to the parameter
variation. Let the plant model be
then
10
SðRm m m
FC2 Þ > SðRFC1 Þ > SðRPID Þ:
P ðsÞ ¼ :
s2 þ ds þ 1
This robustness comparison leads to the result that (for with the variation interval D ¼ ½d  d þ  for parameter d,
the second order plant in Eq. (1)), we can design the b of where the lower-bound value d  ¼ 1 and the upper
the fuzzy controllers FC1 and FC2 to have FC1 and FC2 bound value d þ ¼ 1. It is known that the plant is stable

Fig. 6 Performances of fuzzy


control systems with different
controllers, (classical PID (dash
line), FC1 (dash-dotted line), FC2
(solid line))
437

whenever d > 0, and the system is unstable if d becomes FC2 (solid line)). From Fig. 6, it can be seen that the
negative. Without losing the generality, the kI , and kd are system with classical PID controller reaches marginal
selected as an example to have the classical PID con- stable condition, and the system with FC1 and FC2 ap-
troller be marginal stable when d ¼ d  ¼ 1. This leads proaches its final value without steady state error in a
to strictly stable manner. Thus, the fuzzy controllers are
illustrated to have larger robustness range than the
1 þ 10k
11
I
classical PID controller to the variation of the parameter
kd ¼ ffi 0:191 ; ð24Þ
10 d. It is known that the kd and kI can be selected to
with the arbitrary chosen kI ¼ 1. That is, if d ¼ improve the robustness of the system with classical PID
d  < 1, then the second order system with classical controller, however, the fuzzy controller FC1 and FC2
PID controller becomes unstable. To show that we can can be designed to have larger robustness range as long
always design piecewise linear PID-like fuzzy controllers as the kd and kI are determined and fixed. With
(FC1 and FC2 ) to be more robust than the specific clas- d ¼ 7:5345, Fig. 7 shows that controller FC2 is more
sical PID controller, the same kd and kI are used in the robust than controller FC1 to the variation of the
fuzzy controllers (FC1 and FC2 ). With the unit step ref- parameter d. Note that FC1 and FC2 have the same b.
erence input, the parameters in the input membership For a third order uncertain plant with transfer function
functions of the fuzzy controllers FC1 and FC2 are 1
determined as P3 ðsÞ ¼ ; a ¼ ½a aþ  ;
s3 s2
þ þsþa
½b13 b12 b11 b10 b11 b12 b13  the simulation is developed for the robustness compar-
¼ ½b23 b22 b21 b20 b21 b22 b23  ison between the classical PID, FC1 , and FC2 . The
robustness of the classical PID, FC1 , and FC2 for P3 ðsÞ
¼ ½1  2=3  1=3 0 1=3 2=3 1: ð25Þ
can be compared with the same derivation in Sect. 4.
Let the parameters in the output membership function Again, it can be shown that for the specific classical PID
of the fuzzy controllers FC1 and FC2 be controller with determined gain coefficients, we can de-
sign the corresponding FC1 and FC2 to have larger
½c13 c12 c11 c10 c11 c12 c13  robustness range. Let the gain coefficients be selected as
¼ ½c23 c22 c21 c20 c21 c22 c23  ki ¼ 1; kd ¼ 1:2;
¼ c3  ½1  2=3  1=3 0 1=3 2=3 1 ; ð26Þ as an example for the classical PID controller without
where c3 ¼ 5 is determined to satisfy the condition in losing the generality. From Fig. 8, it is seen that when
Eq. (23). Figure 6 indicates the unit step responses of the a ¼ 0:6, the three order plant with classical PID con-
control system with d ¼ d  ¼ 1 for different control- troller is unstable, i.e., aþ < 0:6 is necessary for the PID
lers (classical PID (dash line), FC1 (dash-dotted line), control system to be stable. However, if fuzzy controllers

Fig. 7 Robustness comparison


between the control systems with
different fuzzy controllers (FC2
(solid line), and FC1 (dash-dotted
line))
438

Fig. 8 Robustness comparison


between the control systems with
different controllers (FC1 (solid
line), classical PID (dash line),
and FC2 (dash-dotted line))

FC1 and FC2 (which are designed to be the same as those


used for the second order plant in this section) are ap- References
plied for the third order plant, the system becomes stable
1. Barmish BR (1994) New Tools for Robustness of linear syst-
(see Fig. 8). Thus, it indicate that the fuzzy control mes. McMillan, New York
systems can remain stable even a is larger than 0.6. That 2. Coleman CP, Godbole D (1994) A comparision of robustness:
is, fuzzy controllers FC1 and FC2 have larger robustness fuzzy logic, PID, and sliding mode control. Proc.: The third
range. Therefore, FC1 and FC2 are said to be more IEEE Conference on Fuzzy Syst. pp. 1654–1659
robust than the classical PID controller. 3. Driankov D, Hellendoorn H, Reinfrank M (1993) An Intro-
duction to Fuzzy Control. Springer, New York
4. Dubois D, Prade H (1980) Fuzzy Sets and Systems: theory and
Applications. Academic, New York
6 Conclusions 5. Kawaji S, Matsunaga N (1991) Fuzzy control of VSS type and
its robustness, Proc. IFSA Eng., pp. 81–84
6. Kosko B (1992) Neural Networks and Fuzzy Systems. Prentice-
In this paper, the stability robustness of the piecewise Hall, New Jersey
linear PID-like fuzzy controllers and the classical PID 7. Nguyen HT, Kreinovich V (1994) How stable is a fuzzy linear
controller is compared for a plant with second order system. Proc. The third IEEE Conference on Fuzzy Systems.
uncertain model. The comparison result indicates that pp. 1023–1027
8. Tanaka K, Ikeda T, Wang HO (1996) Robust stablilization of a
piecewise linear PID-like fuzzy controller can be more class of uncertain nonlinear systems via fuzzy control: quadratic
robust than the classical PID controller. This explains stabilizability, H 1 control theory, and linear matrix inequali-
one of the reason that fuzzy controller is preferable to ties. IEEE Trans. on Fuzzy Syst. 4(1): 1–13
the classical PID controller. Also, the piecewise linear 9. Tao CW, Taur JS (1999) Design of fuzzy controllers with
adaptive rule insertion. IEEE Trans. Syst. Man Cybern. SMC-
PID-like fuzzy controller FC2 is witnessed to be more 29(3): 389–397
robust than the piecewise linear PID-like fuzzy con- 10. Tao CW, Taur JS (1996) Design and analysis of a new PID
troller FC1 to the parameter variations of the second fuzzy controller. J. Chinese Inst. Eng. 19(2): 193–204
order plant. In this case, the piecewise linear PID-like 11. Taur JS, Tao CW (1997) Design and anlysis of region-wise,
fuzzy controller is better to be designed with the struc- linear fuzzy controllers. IEEE Trans. Syst. Man Cybern. 27 part
B(3): 526–532
ture of FC2 . Simulation results empirically indicate the 12. Yi SY, Chung MJ (1998) Robustness of fuzzy logic control for
correctness of the proposed robustness comparison. an uncertain dynamic system. IEEE Trans on Fuzzy Syst. 6(2):
216–225
Acknowledgements This work is partly supported by the the R.O.C. 13. Zadeh LA (1965) Fuzzy sets Inform. Contr. 8(3): 338–353
National Science Council through Grant NSC 90-2213-E-197-002.

You might also like