Proceedings of the 12th IEEE

International Symposium on Intelligent Control

16-18 July 1997, Istanbul, Turkey

Fuzzy Controller Design for Parametric

Controllers
M u r a t Akgul f and G m e r M o r g u l $
Havelsan, Eskigehir yolu 7. k m , Ankara, Turkey
$ Bilkent University, Electrical and Electronics Eng. Dept. A n k a r a , Turkey

Abstract
In this study, fuzzy logic controller (FLC) design for
tuning some parametric controllers are investigated.
The objective in designing an FLC is t o determine
the rule base of the system and the data base which
includes membership functions, set operations, and inference engine. Two designs have been realized using
heuristic rule generation, one for a PID controller and
one for a lead-lag type controller. The FLC in these
designs sets the parameters of the PID and lead-lag
controller on-line. The rules and the corresponding
membership functions are constructed by observing
the effect of the changes in the parameters on the overall performance. Simulation results show that better
controller performance can be achieved in comparison
with the classical design methods.
Keywords : Fuzzy logic, fuzzy logic controllers,
membership function, parameter tuning.

Fuzzy Logic Controllers (FLCs) can be used in various ways. One of them is to use tjhe FLC directly as
the controller for the syst8emunder consideration. The
other choice is t o use a standard controller like PID
or lead-lag and tune the parameters of this controller
with a fuzzy logic system (FLS'). A few of the studies
on parameter tunings a.re [5], [3] and [4]. In [3] the
PID parameters are tuned according t o some rules derived by considering a typical system output. In [4]
some changes t o the model of fuzzy PI controller is
introduced t o reduce the overshoot even more.
In this study the implementation of a fuzzy controller which sets the parameters of a parametric controller has been investigat,ed. The rulebase were constructed by considering a typical output, and its relation t o the error and it,s derivative or by observing
the effect of the parameter changes t o the system output. We construct FLS for the standa.rd PID, and the
lead-lag type controllers.

1 Introduction

Over the past few years many researchers of various

fields and manufacturers of control equipment pay a
great a.tteiition to fuzzy control. Most of the inventors
of fuzzy control have a strong control engineering or
systems theory background. From their perspective,
fuzzy control can be seen as heuristic and modular
way for defining iionlinear, tabie based systems.
It is generally agreed that an important point in
t.he evolution of the modern concept of uncertainty was
the publication of a seminal paper by Lofti A. Zadeh
[lo], even though some ideas presented in the paper
were envisioned some 30 years earlier by the American
philosopher M a s Black in 1937. In his paper, Za.deh
iiit,roduced a theory whose objects-fuzzy sets-are sets
wit.11 Imundaries that a.re not precise. The membership
in a fuzzy sct. is not a niatt.er of affirmation or denial,
hut, ra.ther a niat,ter of dey7.ee.

The control problem which we invest,igat,e in t,liis study

is reference input, tracking. The system structure under investigation is a. unit feedback with a controller
and 1inea.r system which is shown in Fig.1
In this figure C ( s )is the transfer function of a. parametric controller, P(s) is the transfer function of the
plant under investiga.tion, I( t ) is t8hereference input,
y(t) is the output of t,he system, U ( L ) is t#heoutput of
the controller, e(t,)=r(t,)-y(t)is t<hcerror and, v(t) is
a vector fuiict,ioii in ?K"' representiiig the wa.veform for
the parameters of the controller. Here w is the number
of parameters of the controller C(s). These para,meters
are updated by the Frizzy Logic System (FLS) a.ccording to the error sigiial. 111 (,his st,udj. t,he problem is
iiivest,igat,ed for t,he uni
ep referenw signal.

0-7803-41 16-3 197 I $lO.OO" 1997 IEEE

67

Control Problem

,.

te@)+--++
de(t> - -

++

+
-+

f
-

Consider the poiat a given 011 t h e step response in

Fig 3. A t the heginning a big control signal is required
111 order t o achieve a fast rise time. To produce a
big control signal, the PID cout(rol1r s h s d d have a
large proportional gam, a large inkegral gain and a
miall derivative gain. Froin point a t o &I the emor is
pmifive and big The inc~emeiitof emx is negative
m d the magnitude seems t,o he big in &hemiddle of
these points.

Hence we obtain tlit follon-ing heuristic rules,

if eit) is PS aid de(t) is W W then LKpis BIG,
if e(t) is PB and de(t) is N F then I<,is BIG,
if e(&)
is P F and de($) is NB then *I<d is S U P .

(1)

under consideration is given

I(s) i h 1 1 1 transfer
~
function of the plant
iilltl 1
I:L.( 5 ilr( Iliv liizzy logic systeil1s tuning each
p;iraiwt(~rol llic 111) controller. The objective is t o
coiistriict ii rulc 1)ast i\lld t,o set the wlat,ed memberh FEC to get a satisfactory out11

1Ii(.

st riici 11rc

Here PB, NE3 stands for Positive Big and Negative

Big wrliich are fuzzy sets As it is clear the first letter
stands OT the sign and the second for the iuagnitvde.
Fuzzy sets defined on the ciomam of error and increment of the error are labeled with the following: NB,

I
.
NS, ZE, PS, PM, PB. Here S staxids for SniaA,
M for Medium, and ZE for zero

At poiiits closer t o b the error IS close to Eero hut

t h e increinent oE the error is negative a n d big. Si this
point t o avoid a large overshoot the PID coiitro3ler
should have a small proportional gain, a large derivative gain, and a small integral gam. Thus the following
rules are given-

for tiic PZD controller E derived from

iiiiiiiatioii of ii typical step response given in
Iii illis figure llir sign of error(e(t)) and increo f 1 1 1 ~c,rror(tlc(i)) are given in a tahular form

~ I I L Z \ Xritlcaq

iirtAnt

1ioi11ts where e ( t ) and tle(t) changes their slgns

if e(t) is ZEand -de(t) is NB then lil,

is SMALL,
if e(t) is Z E aiid cle(t) is N B then A , i s SMALL,
if e(t) is ZE aiid de(t) is NB then h , l 15 BIG
(2)

lal)c4cd 0 1 1 t h e stcp response with the letters a b ,

(
, J Tlic. followiiig argument for the derivation of
t h v ru1~qw i l l I)e wiiilai to the work done hy Toinizulia

68

membership functions for tlie ga.iiis because in their

work they relakd t,he int,egra.l gain t,o t,he proportiona,l
and derivative ga.in with a. constant. C Y . In our design
all PID gains are considered t,o be independent of the
each other and ea.ch of them have a. separa.te rule base.
In the simulations t>heranges of the iiieiiibership functions are found by first, a.pplying this met,liod and then
making a fine t,uning.

3.2 Simulations and Results

Rule Ta.ble

The fuzzy parameter tuning has been tested on a variety of processes. One of them is given here. The plant,
transfer function is given in Eyn.3. Ziegler-Nichols
PID tuning results in the following gains for this plant.

delt)

Pl(S)

ICp = 36.0000,

NB NM NS

PS

53

+ 65-2 + 11s + 6

(3)

ICz = 38.0047, I<d

PM PB

NB NM NS

-1

AL1

-2/3 -113

= 8.5253.

PS

PM PB

11.1 213

Figure 4: Membership functions for error and increment of error

By using similar arguments, a set of rules, as shown

in Tab.??, may be used to adapt the gains. In these
tables, S , M , and B refer to Small, Medium,
and Big, respectively. The membership functions for
the linguistic variables error and increment of error are
given in Fig.4. A normalization procedure is applied
to t,he increment of the error before the fuzzification
operat>ion. The scaling factor is chosen so that the
increment of error is large enough to span t8heinterval
[-I,11.
In designing t,he controller individual rule-based in,f e r m c e : is chosen. The fuzzy set operations used in
t!he design a,re t,he minimum, maximum, and standard
complement opera.tions. Mamdani implication is used
in firing each rule [8].
Rased on ext,ensive simulation study on various
s. a. rule of thumb for determining the range
nrl t,he range of l i d is given as
lil,,7,Lin
= 0.:321icT:,,,
Ji,j,7,)j7, = O.O81i,.,.P,.,,

I(,,,,,,,

Figure 5: Membership function for the va.riables ICp,

Ki,and K d
The membership functions for error and change of
error are given in Fig.4. The nornia,liza.tion c,onstant
for the increment of tlie error is chosen to be 20. The
membership functions for Ii,, Iif , and I<[l are given in
Fig.5. The output of the system wit,h Ziegler-Nichols
PID controller and the outtput of t,lie system where
the parameters a.re t.uned with FLCk a.re compared in
Fig.6.a. The PID para.met#ersca.lcula.ted by the FLCs
are given in Fig.6.b.
The designed coiit,roller is t8estjedfor a plant Pz(s)
giveii below, which may be consiclered as a slightly
perturbed version of the plant, giveii by Eqn.:3.

= 0.61CcT,
= 0.15K,,Pc,..

P?(S)=

.s3

+ LIs ? + I 1s + fi.

(4)

Tile oUt,pLltwa\-cforllls for t.ilr fuzzy PID alld t,ile

classical PIU arc co1111)ar(4iii Fig.7.a. Tile PID parainet,er variations a.rc g i w i i iii Yg.i.l).

wlirre ii,.T
aiitl &. are t81iep i n and the period of oscillat8iooat, t,hr stBahilitry
lilylit under P-cont8rol[2].Tomizulta.

t8his niet,hod t80 set the domain of t,he

69

Fuzzy Lead-Lag Controller

I'or t h i + tlt91g11 ~)rol)Ic~ni

t h e qtlucture of tlie overall
syhteni 15 ah h i l o ~ ~ 111
i i r i g h . 411 this figure P(s) is the
plant transfer function There are two parameters in
t h e lead-lag confroll~rto IF determined, iiainely n and

7'
1

311,

ib

Figure A: Syst,eiiistructure under investigakiou

--I

As in the case of PID controller we iiwd two FZC

blocks, one for a and one for T. The effect, of the variations of the parameters a and T o n the output of the
syst>eniseems t o be in tlie sane direction. Therefore
t li(>~ " x m e t e rTis fixed and only a is tuiied with FLC.
The value of T is determined with a classical design
niethod.

Figure 9: Membership functions of the linguistic variable a

ls .
I\

_I

'

..

-.

. .

..

I
5

Figiin

4.1 Structure of t h e Fuzzy Logic Controller

6
ime m S e m d s

10

To determine the effect of the parameter a on the step

response of the system, inaiiy simulations l i m p been
performed. The reason to choose tlie parameter a for
the tuiiiiig is that, for values of a le5s t h a n 1 the controller is a phase lag coiitroller and for U greater than
1 it is a phase lead coiitroller
Froiii the siinulatioiis i t is observed tlhat as the
value of a increases the rise time decreases ancl the
overshoot increases Tlir olcrrslioot of thP step respoiise is proportioiial to the parameter a
is an iiiverse relationslllp between the ri
system and the parameter II
The output of' a systeiii is desired to have a shoIt
rise time and no overshoot and osullations. To have

12

7 I'IU praii1eters eteriniiied hy the FLCs

(onti oller design has heen tested

111 siniulat,lon, and satisfactory reFroiii Fig.7-b it is seen that with
oller the overall systpm LS at

70

a short rise t,iiiie we have to chose a large value for a,

say a,,,,,.
But for large a we get a large overshoot and
rather an oscillatory response at the output. One possible way is to set the value of a t o a large value when
the error is big and reduce it when the system output is about t o be equal t o the reference signal. From
Fig.3 at, the beginning, the error will be a member of
the fuzzy set PB and its increment will be a member
of the fuzzy set NB (or NS). So using these arguments
one may state the following rules:

if e(t) is PB and de(t) is NB then a is LA7,

if e(t) is PB and de(t) is NS then a is LA7,
if e(t) is ZE and de(t) is NB then a is LAB.
(5)
The meaning of the linguistic values LA7, LA6, and
the rest a.re given in Fig.9. Using this discussion on the
effect of the variation of a t o the output a rule-base is
constructed as given in Tab.l.
Rule Table
for a
I NB
NM
e(t)

NS
ZE
PS
PM
PB

NB
LA3
LA4
LA4
LA5
LA6
LA7
LA7

NM I N S
LA3 I LA2
LA3 LA3
LA4 LA3
LA4 LA4
LA5 LA4
LA6 LA5
LA7 LA6

de(t)

ZE

PS

PM

I LA2 I LA1 I LA1 I

LA2
LA3
LA3
LA4
LA4
LA5

LA2
LA2
LA3
LA3
LA4
LA4

LA1
LA2
LA2
LA3
LA3
LA4

PB
LA1
LA1
LA1
LA2
LA2
LA3
LA3

om

004

006

om

01

012

014

0,5

028

hme I d

ib4

Figure 10: Comparison o t ie step responses of P ~ ( s )

with the two different controller.

Table 1: Rules for lead-lag parameter a

The membership functions for the parameter a are
chosen so that they all equally share the domain in the
intervals [0,1] and [l ama+].

4.2 Simulations and Results

This fuzzy parameter tuning scheme for the lead-lag
type controller is applied to various systems. Simulations showed that a variety of processes can be satisfact,orily controlled by the fuzzy parameter tuned lead-lag
controller. One of the simulations is performed with
a plant whose transfer function is given in Eqn.6 . A
phase-lead controller based on a classical design for
t,liis plant results in a controller whose transfer function is given in Eqn.7
P3(S)

55

5.
40.

35

2500
s ( s 25)

m 30

25

20

1510.

+
+

0.016s 1
C ( S )=
(7)
0.002s 1
In tjliis design the parameter a and T are 8 and
0.002 respect,ively[l]. In the proposed fuzzy parameter
t,iiiiiiig algorithm the value of T is not changed. It is

002

O M

006

008

01

012

014

016

018

02

71
8

[a] J . G . Zeigler, and

t , l i c > value obt.a.ined in the classical design. The

nortiialieiitioil constant. for the increment of error is
choscw t.o I)c (3. rIl~(,Illembership functions for error
i l t l d itss iiicrcmietrt. are as in Fig.4. The membership
fuiict,ioiis for 1.110 liiiguist,ic values of a are taken as
giveii i t 1 Fig.:). For this simulation U,,,, is set t,o 100.
Again Mitrrtdani Itnplication is used in interpreting the
iiimtiiiig o f the rules.
I I i c x st,(:l) rcspotises of the system P ~ ( swith
)
the
t,wo difrermt, cont,rollers , one classical and one fuzzy,
iI.r( givm i i i Fig.lO.(a). In Fig.lO.(b) t,he variation of
ILIir i ) i l . y ; l . l l i ( t J ( rN is givcw. From the figure it is seen that
at, t , I i v I)c.gittiiing a t,akes large values and as the output, al)l)roa,chesthe desired reference signal the FLC
twliiccts tlic parriiitcter a. This FLC has been tested
011various syst,eii-iswhich may be considered to be the
platit, P ~ ( Swit,h
)
distorted coefficients. One of these
plii,tit,s is given I)elow:

N . B. Nichols, Optimum settings for automatic controller , Trans. ASME,

vol. 64, PD. 759-765, 1942.
[3] Zhen-Yu Zhao, Masayoshi Tomizuka, and Satoru
Isaka, Fuzzy Gain Scheduling of PID Controller, IEEE Transactions On Systems, Man,
and Cybernetics, vol. 23, no. 5, pp. 1392-1398,
Sep/Oct 1993.
[4] Jihong Lee, On Methods for Iinproviiig Performance of PI-Type Fuzzy Logic Controllers ,
IEEE Transactions On Fuzzy Systems, vol. 1, no.
4, pp. 298-301, NOV. 1993.
.[5] Hao Ying, Practical Design of Nonlinear Fuzzy
Controllers with Stability Analysis for Regulating
Processes with Unknown Mathematical Models ,
Automntica, vol. 30, no. 7, pp. 1185-1195, 1994.
[GI Chuen Chien Lee, Fuzzy Logic in Control Systems: Fuzzy Logic Controller-Part 1,Part2 , IEEE
Trans. on System, Man, and Cybernetics, vol. 20,
no. 2, pp. 404-432, Mar./Apr. 1990.
[7] Dimiter Driankov, Hans Hellendoom, and
Michael Reinfrank, An Introduction t,o Fuzzy
Control, Springer- Verlag, 1993.
[8] Witold Pedrycz, Fuzzy Control and Fuzzy Systems, Research Studies Press Ltd., 1993.
[9] Li-Xin Wang, Adaptive Fuzzy System and Control, Design and Stability Analysis , Prentice
Hall, 1994.
[lo] L. A. Zadeh, Fuzzy Sets, Information and Control, vol. 8, pp. 338-353, 1965.
[11] B. ICosko, Neural Networks and Fuzzy Systems,
Englewood Cliffs, N J , Prentice Hall, 1992.

sc4 l o

P,,(S)

5000

s2+24s+1

I l i c ~ st,cq) rcymises

corresponding to the fuzzy tuned

Iwtl-lag cont.roller a n d the classical design is given in
Yig. 1 1 . ( a ) . I n M l i of t,he designs the rise time reduces
wliilc> a. sniidl iiicrea.se in the overshoot of the system is
o I ) s ( ~ v t 4l+oiii
.
t.he figure it is seen that even with the
tlist,rtrh(xl systy II t.hc fuzzy tuned lead-lag controller
ptdoriiis Iwt.t.er t.1ia.n t,lie classical design.

Conclusion

In this study, FLC design for tuning some parametric

coiit.ro1lcrs are investrigated. Two designs have been
redizcd i)y usitig heuristic rule generation, one for a
ID cont,roller atid one for a lead-lag type controller.
Sinidation result,s showed t81iatbetter controller perLortiiancr, ca.n lie achieved by FLCs in comparison with
t . 1 ~:lassicaI design methods.
t h e VL( !s iu these designs set the parameters of the
PID ;i.ii(l Ica~d-lagcont,roller on-line. The rules and the
corrcspoiiding membership functions are constructed
by observing the effect of the changes of the parameters on the overa.11performance. The simulation results
of t,lixe controllers show better control performance
relative t80their classical design. The disadvantage of
t(li(>heurist,ic design is t,liat, there is no analytical tool
t,o construct, the rule base. The designer should colIwt, as much information as possible on the system
and utilize t.hein to construct the rule base and the
relat,ed data ba.se including the membership functions,
set opera.tions a i d inference engine.

References

[I] B. C. Iiuo, Automatic Control System,

Prc lit ice- Hnll, 1987.

72