Stewart 2004
Stewart 2004
Stewart 2004
evolutionary algorithms
Electrical Machines and Drives Research Group, Department of Electronic and Electrical Engineering, University of
Sheffield, UK., e-mail: p.stewart(,ysheffield.ac.uk
3Department of Automatic Control and Systems Engineering, University of Sheffield, Sheffield, UK.,
Abstract Evolutionary algorithms are applied to the online generation of servo-motor control systems. In this paper,
the evolving population of controllers is evaluated at run-time
via hardware in the loop, rather than on a simulated model.
Disturbances are also introduced at run-time in order to produce robust performance. Multiobjective optimisation of both
PI and Fuzzy Logic controllers is considered. Finally an on-line
implementation of Genetic Programming is presented based
around the Simulink standard blockset. The on-line designed
controllers are shown to be robust to both system noise and external disturbances while still demonstrating excellent steadystate and dvnamic characteristics.
INTRODUCTION
to extract the knowledge base from experts or training examples to construct the input-output membership functions and
the fuzzy rule-base. These methods can be based on neural networks [I I], [ 14] or the application of fuzzy clustering techniques to construct a fuzzy controller from training
data sets [8]. It has been observed that the major drawback
of most fuzzy controllers and expert systems is the need to
predefine membership functions and fuzzy rules. In [1 1], a
1459
g.orithrn.
A. Multiobjective optimisation bv evolutionarv algorithlm
f(x),j 1.J.
(4)
The application consists of a brushed DC permanent magnet field motor fed by a four quadrant DC chopper drive
operating at 5kHz. Figure 1 shows a schematic of the on-
1460
velocity controller. The microcontroller also hosts the velocity and current feedback signals from the motor set and
chopper drive respectively. The multiobjective optimisation
Velocity Feedback
Current Feedback
(Brushed DC 1/
' Wound Field
Drive Motor;%.'_, DC Load Motor
I
TEL
48001961(0
,- D
v4 Quadrant
Microcontroller
Host PC
(MOGA Matlab)
Chopper
l. 200VDC}A
i`200V DC
I200V DC
Spl
~~~~~~~~Diturbac
I
NI
Data Acquisition
d
Demand )BIar
jtConco
Board
~Control
________________
Dema nd
Various methods exist to tune the gains of a PID controller off-line to attain the prescribed transient response and
steady state error criteria. These methods generally involve
some form of iterative approach to achieve performance criteria such as rise - tii-nt , o!eaIcrshoot and settling - time
[17]. In keeping with the development of the fuzzy controller designed later in this paper, the PID control scheme
was designed and tuned on-line. This on-line method has
been shown to be extremely effective in a variety of applications including active magnetic bearings [19]. In the ap'Objective Furrction)
FMeasured
Veloity
Erro
...
Measured
Multiobjedive
Genetc . - - Current
Algorithm 1
sDisturbance
Error
Profile
Control
Controller
g( Parameters
(Velocity
Demand
J PID VelocitvyI
Controller'
DC Load Motor>
(Disturbance
ri
PID Current
Controller
eRRelay
DC Chopper A
J Drive
Current
DMechanical
Feedback
Velocity Feedback
Fig. 2.
setup
1461
vC
1,l
t:_
14
U.;
"l 1E
ts)
Objectve Function
(Msiti.bjectioe
Evouionary
Agorithm
Rise time,
Steady-State error
(Membership fUncons
etc.)
Matlab
Fuzzy Logic
Toolbox
Rarndom
Demandl
profile
Fuzzy Lactic
t ler
Disturbance
Moor
Velocity
1462
put set are shown in figure 8 and are linked to the input set
by the rule base;
* if velocity error is negbig THEN current demand is neg-
big
big
possmall
7..
C
j'.
j! \,
[,
Si;
(I 8 L
tE
,{
u4L
\
',
_
-1S
-1
-,g
'
cuirrent aemand (4
of
1f.
Fig. 8. On-lilne designed Fuzzy Logic velocity controller output membership functions. A:negbig. B:negsmall. C:zero,
D:possmall, E:posbig.
(q,
-400
30
f,
211
2
O
30
Fig. 6. On-line designied Fuzzy Logic velocity controller performance with disturbance
v6
-3u~
,'
-2oi
'
-16
A symbolic approach was adopted to design the RST controller shown in block diagram form in figure 9.The decision variables which form the genetic material of the individuals in the population are the contents of the R, S and T
controllers. Each genome contains infomiation defining the
symbolic contents and associated parameters of each of the
controllers. The symbolic contents available are shown in
' -'
1O
3 20\
Fig. 7. On-line designed Fuzzy Logic velocity controller input membership functions. A:negbig, B:negsmall, C:zero,
D:possmall, E:posbig.
1463
.,
PlC
De rieative
PID
REFERENCES
PID
[1] Betin F.. Pinchon D. and Capolino G.A., Control of electrical dnrves
subject to large variations of load: a fuzzy logic approach., Electromotionl, 8(3), July-September 2001, 155-168.
[2] Chang C.H. and Wu Y.C., The genetic algorithm-based tuning method
for symmetric membership functions of fuzzy logic control systems.
Proceedings of the International IEEE/IAS C.onference on Industrial
Automation and Control. Emerging Technologies.. 1995, 421-428.
I Controller
L A.,
Integrator
Tra
F>
nlsport
elay
1
F1
T ransf er Fcn
[3] Chipperfield A.J., Fleming P.J. and Pohlheim H.P., A genetic algorithm toolbox for MATLAB, Proceedings ofthe International Conlference on Systems Engineering, 1994, pp.200-207,
III
C, a3Ir,
sachusetts.1989.
(a)
3(X)1
250
200_-
[11]
i,100t
50k
[12]
0L0
[13]
(b)
2
15
[14]
05_
[15]
0)
time (s)
Coventry U.K.
[4] Fonseca C.M. and Fleming P.J., An overview of evolutionary algorithms in multiobjective optim'sation. Evolutionary Computation.
-3(1). 199-5, 1-16.
[51 Fonseca C.M. and Fleming P.J., Multiobjective optimisation and multiple constraint handling with evolutionary algorithms - Part 1: A unified formulation and Part 2: Application example. IEEE Tiansactions
on Systems. Man and Cybernetics - Part A.: Sy,stems and Humanis.
28(1). 1998. 26-37 and 38-47.
[6] Fonseca C.M. and Fleming P.J.. Genetic algorithms for multiobjective
[16]
[17]
[18]
functions and rule sets for fuzzy controller using genetic algorithms,
IEEE Tr}ansactions on Fuzmzi Systems, 3(2), 1995, 129-139.
Hong T.P. and Lee C.Y... Induction of fuzzy rules and membership
functions from training examples. Fuzzv Sets and Systems. 84(1).
1996 33-47.
Hughes E.J., Evolutionary multi-objective ranking with uncertainty
and noise. Eouinr
Evolutionan. Multi-criterion Optimsation Zitzler E. (Ed.),
pp.329-343, ISBN 3-540-41745-1, Springer-Verlag. 2001.
Blumel A.L., Hughes E.J. and White B.A., Multi-objective evolutionary design of fuzzy autopilot controller. Evolutionary Mliltcriterion Optimsation Zitzler E. (Ed.). pp.669-680. ISBN 3-5404174--1. Springer-Verlag, 2001.
Ishibuchi H. and Tanaka H., Neural networks that learn from if-then
fuzzy rules, IEEE Transactions on Fu:n Svstems. vol.1, 1993. 85-97.
Jamshidi M.. dos Santos Coelho L., Krohling R.A. and Fleming P.J..
Robust control systems with genetic algorithms, CRC Press Control
Series, ISBN 0-8493-1251-5, 2003.
Koza J.R., Genetic Programing: On the programing of computers by
means of natural selection, MITPress Cambridge. 1992.
Kuo B.C. and Hanselman D.C.. Matlab tools for control system analysis and design. Prentice-Hall Interlnational. New .Jersev, 1994, ISBN
0-13-099946-6.
Mamdani E.H., Application of fuzzy algorithms for control of simple
[23] Zhu Z.Q., Shen Z.X. and Howe D.. Comparative study of alternative fuzzy logic control strategies of Permanent Magnet Brushless AC
1464