Performance Analysis of

Fuzzy Proportional-Derivative Control Systems

H u a i d o n g Li z H e i d a r M a l k i 2 and Guanrong C h e n 3

University of Houston

Abstract " I. Introduction

This paper analyzes the performance of a fuzzy One of the main advantages of fuzzy controllers
proportional-derivative (PD) controller in com- is that t h e y can be used to replace conven-
parison with the conventional PD controller. The tional controllers when the system under control
design of the fuzzy PD controller follows the struc- is highly complex and/or only linguistically de-
ture of the conventional digital PD controller, scribed. With its "rule-based" characteristics,
with additional fuzzy logic control rules. The re- many complex and ill-conditioned control require-
sulting controller, therefore, has the same linear ments can be implemented in a relatively efficient
structure as that of the conventional digital PD and inexpensive way.
controller, except that both the proportional and
the derivative parts have non-constant gains. The This paper is an extension of our earlier work
fuzzy proportional and derivative gains are non- on the design of the fuzzy PD controller [4] and its
linear functions of the control-input signals and stability analysis. On the other hand, this paper
hence have a self-tuning control capability. Thus, is in parallel to the studies of the fuzzy PI con-
the proposed fuzzy PD controller preserves the troller design and its stability analysis [1,2]. One
simple linear structure of the conventional P D of the significant differences of this design from
controller yet enhances its adaptive control ca- other fuzzy controllers is that the fuzzy logic is
pability. In. computer simulations, a set of linear used in this design to self-tune (or adjust) the pa-
systems, with or without time-delays, were used rameters of the controller. This design, however,
to test the performance of the fuzzy PD controller follows the standard procedure of fuzzy logic con-
in [4], and a set of nonlinear systems are used to trollers design, which consists of (1) fuzzification,
test the performance of the fuzzy PD controller in (2) fuzzy rule-base establishment, and (3) defuzzi-
this paper. The performance has been compared fication.
to the conventional PD controller for the same lin-
The design of the fuzzy PD controller will
ear and nonlinear systems. Computer simulation
be briefly described in the next section, while in
results have demonstrated the advantages of the
Section 3 the computer simulation results will be
fuzzy PD controller, particularly if the system to
demonstrated.
be controlled is nonlinear.
2. D e s i g n of t h e Fuzzy P D C o n t r o l l e r
1,3 supported in part by the Institute of Space
Systems Operations, University of Houston. The fuzzy PD controller was designed by the
present authors in [4] used the following notation:
supported in part by the Research Initia- Let T, sp(nT), and y(nT) denote the sampling pe-
tion Grant (RIG) and the Energy Labora- riod, the set-point, and the output of the system
tory, University of Houston. respectively. The error and the rate of change of
the error signals are defined by e(nT) := y(nT) -
Permission to copy without fc~ all or part of this material is gnmtc.d provided that
the copies arc not made or distributed for diroct commcqn~ial advanlag~ the A C M
copyright notice ~md the title" of the publication tad its date appctr, and notice is
given tim copying is by permission of the Association for Computing Machinery. 115
To copy othcrw'is¢, or to mTmblish, requires • fee zmd/or specific permbsion.

0 19'9,4 ACM 089791-.647,.,6/94./0003 $3.50

sp(nT) and r(nT) := (e(nT) - e(nT - T))/T, re- = ~[-L + Kpd(nT)] in I C l l , IC12, (5)
spectively. The fuzzy controller consists ,of two in-
•puts: e(nT) and r(nT). The output of the fuzzy = ~ [ , L - Kdr(nT)] in IC13, IC14, (6)
controller u(nT), which is also the input to the
process under control, is given by = ~[L+Kpd(nT)] in IC15, IC16, (7)
u(nT) = - u ( n T - T) + K~,&u(nT), (1) = 0 in IC17, IC19, ( 8 )
=• -L in IC18, (9)
Where K , is a constant control gain and Au(nT)
= L in IC20. (10)
is an incremental control output, both need to be
determined in the design. The block diagram of
the closed-loop fuzzy PD control system is illus- To snmmarize, we have determined all the
trated in Fig. 1, where .Af(,) is the process under control rules and formulas for the fuzzy PD con-
control which can be either linear or nonlinear. troller, with the control law (1) and the incre-
For more details, see [4]. mental control ~5u(nT) calculated by (2)-(10), ac-
cording to the different locations in Fig. 3 of the
The membership functions for the error, the error signal e(nT) and the rate of the change of
rate of change of the error, and the output of the the error signal r(nT). The initial conditions for
controller axe shown in Fig. 2. The constant L > 0 the overall control system are the following natu-
u.sed in the definition of the membership functions ral values: for the incremental control, ~u(0) = 0;
i s d e t e r m i n e d by the designer according to the for the system output, y(0) = 0; and for the er-
value ranges of the error, rate, and output, and ror and rate signals, e(O) = sp (the set-point) and
will be fixed after it is chosen. r(0) = 0.
In [4], it Was shown that the membership Finally, we note that in the steady-state situ-
functions of the error and the rate signals de- ation, le(nT)l = O, so that Ir(nT)l = Id(nT)t = 0
compose their value ranges into twenty adjacent in the denominators of the coefficients of Au(nT).
input-combination (!C) regions as shown in Fig. 3. Hence, we obtain the steady-state relations be-
It was also shown in [4] that when the locations tween the conventional PD control gains K~ and
of the error e(nT) and the rate r(nT) are in the K~ and the fuzzy PD control gains Kp and Kd as
regions IC1, IC2, IC5, IC6, we have, by the stan- follows:
dard "center of mass" defuzzification formula,
K,,Kp
K~= KuKd4 and K~ = - - 4 (11)
L
au(nT) =
2(2L - Kpld(nT)l)
[Kpd(nT)- Kdr(nT)], 3. S i m u l a t i o n R e s u l t s
(2)
This section shows some computer simulation re-
where d(nT) = [e(nT) + e(nT - T)]/T. Simi-
sults obtained by using the fuzzy PD controller for
larly, if e(nT) and r(nT) are located in the regions
•linear and nonlinear systems, with a comparison
IC3, IC4, IC7, IC8, then we have
to the conventional PD controller. In the case
L that the system under control is linear first- or
u(nT) second-order, computer simulations results show
2 ( 2 L - Kdlr(nT)I)
[Kpd(nT)- Kdr(nT)], (3) that the fuzzy PD controller works as well as the
conventional PD controller. It is not advanta-
and in the regions IC9 - IC20, we have geous to apply the fuzzy PD controller for these
cases. However, in the case that the system un-
au(nT) der control is linear but has time delay, and par-
ticularly for nonlinear systems, the fuzzy PD con-
= ~-[L- Kdr(nT)] in IC9, IC10, (4) troller is significantly better than the conventional

116
PD controller, as shown in the simulation results Using the set-point sp = 1.0, we have compared
described below. the two PD controllers. The fuzzy controller is de-
signed with Kp = 35.0, Kd = 1.0, K~ = 0.1, L =
In the following simulations, the x-axis is the 150.0, and T = 0.1, which produces a very good
discrete-time interval, T is the sampling period tracking response as shown in Fig. 6 (the lower
and n is the number of recursions shown on the curve). However, no m a t t e r how one designs the
z-axes of the figures. two constant gains of the conventional PD con-
We first consider a set of lower-order linear - troller, it does not show any reasonable results.
systems with time-delays that have been used for In Fig. 6, the upper curve is one system output
comparison. We observed t h a t the fuzzy PD con- example of the conventional PD control system,
troller produces lower overshots and faster conver- with K~ = 1.0, K~ = 0.01, and T = 0.1.
gence in general, as can be seen from the example
4. C o n c l u s i o n s
shown in Fig. 4. In this simulation, the system
transfer function is Our many computer simulation results clearly in-
1 dicate that the fuzzy PD controller works as welt
H(s) (100s + 1) 2e-2t " as the conventional PD controller for the first and
second order linear systems [4]. Furthermore, the
The conventional PD controller, with K~, = 50.0, fuzzy PD controller has remarkable performance
K~ = 20.0, and T = 0.1, produces the solid-curve. in the case of time-delay and nonlinear systems.
On the other hand, the fuzzy PD controller with It is worth mentioning that although still in its
T = 0.1, Kp = 51.8, Kd = 9.3, Ku = 0.6, and L = infancy, with many basic theories such as control-
16.0 yields the dashed-curve in the same figure. lability and stability remaining to be further ad-
dressed, fuzzy control technology has been shown
In the second case, the conventional and to be successful in many c o m p u t e r simulations
fuzzy PD controllers have been compared, using and industrial applications, which make the in-
two nonlinear systems. The computer simula- vestigation of various fuzzy control theories and
tion results demonstrate that the fuzzy PD con- techniques more challenging and interesting. As
troller outperforms the conventional one signifi- - part of the ongoing research, we have investigated
cantly. The first one has the simple nonlinear the stability of the closed-loop fuzzy PD control
model system in [3]. It is our belief that fuzzy control
y(t) = 0.0001[y(t)l + u(t). technology has a great potential in nontraditional
systems control and hence deserves further inves-
We used the constant set-point sp = 1.0 as the ref- tigation and development.
erence in this paper. Even for this constant set-
point case, the conventional PD controller can- References
not handle this nonlinear system no m a t t e r how
one changes its two constant gains. One con- 1 G.Chen and H.Ying, "Stability analysis of
trol performance is shown in Fig. 5 (a), with nonlinear fuzzy PI control systems," Pro-
K~ = 3.0, K~ = 0.1, and T = 0.1. In con- ceedings of the 3rd Int'l Conf. on Fuzzy Logic
trast, the fuzzy PD controller, with the param- Applications, Houston, TX, Dec. 1-3, 1993,
eters T = O.1,Kp = 19.5,Kd = O.5, Ku = 0 . t , and pp. 128-133.
L = 20.0, performed the tracking well, as shown 2 G.Chen, T . T . P h a m , and J.J.Weiss, "Fuzzy
in Fig. 5 (b). modeling of control systems," I E E E Transac-
tions on Aerospace and Electronic Systems,
The second example of nonlinear processes 1994, in press.
is shown in Fig. 6, where the nonlinear system is 3 G.Chen and H.Malki, "On the stability
of fuzzy Proportional-derivative control sys-
9(t) = - y ( t ) + y2(t) + u(t) . tems," 1993, submitted.
 4 H.MLIki, H.Li and G.Chen, "Design of
fuzzy proportional-derivative control sys-
tems," 1993, submitted.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. [ proc~a

. - - - . - . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ° . . . . . . . J

Figure I. The closed-loop fuzzy PD control

error negative 1 error positive
system.

r(nT)
m
IC18 l IC12 ICll ] IC17 -L 0 L error signal

IC13
,% rate negative I rate positive.
ICS N

ICI4 IC9
D
IC7 -L o L rate of change
, ,, of exrot si~l
IC19 IC15 IC16 IC20
output uegative output output positive
I

Figure 3. Regions of the fuzzy controller

input-combinatlon values.

'p

-L 0 L output $ig'n~

Figure 2. The membership functionsof e(nY),

,(nT) ~ad u(nT).

118
1,~ ¸ ~o,~

1.3i

1.10 0

1.08

EIO " "° I i

ole,0,31
0.?5

O.O0

0.4~

0.30

0.1S •- q

0
I1 ~cn 3nn

Figure 4. Time delay example. Figure 5a, Response of conventional PO

for nonlinear system.

1$ 1,1

V 0.$
t

0.?

O,I

O.S ,, "

0.4

t/""i
o '1o
..... i .....
20
i .....
30
i .....
40
i":
so
...........................
30 ?o 30 30 1Go
0
SO 100 130 2OO

Figure 6. Responses of both conventional and Figure 5b. Response of fuzzy PD for
fuzzy PD for nonlinear system. nonlinear system.

x-axis: # of sampling period, y-axis: reference,

119