Professional Documents
Culture Documents
Keywords: Induction Motor Drive, Fuzzy Control, Neural Network, Hybrid Artificial Intelligent Controller, High
Performance
Rr + Lrp
Fig. 2 Research of hybrid artificial intelligent controller
0 LmP -(OiSLm WicoslLr 'dr
0 CosILm LmP lLr Rr LrP- iqr - (1 ) This paper is consisted of hybrid artificial intelligent
controller that is mixed adaptive control, fuzzy control
tsl =WMe - 0r slip angular speed
and neural network. This controller control induction
motor with high performance.
Ids' lqs d, q axis current
VdsI Vqs 4. DESIGN OF FUZZY-NEURO
d, q axis stator voltage
CONTROLLER
L
Rs Ls Resistance of stator, self inductance
Fuzzy-neuro controller is consisted of fuzzy control
Rr, Lr Resistance of rotor, self inductance and neural network. It has the softness of strong
Lm : Mutual inductance expression and numerical value processing ability. This
controller is consist that the first part of fuzzy rule and
The mechanical equation of induction motor is equal the second part of clustering method, multilayer neural
below that; network. And it has advantage of robust control that like
fuzzy control, and high adaptive ability that like neural
network.
Te =J m+ Bwm + T, To control of induction motor drive, structure of
dt (2) direct fuzzy controller is equal that Figure 3.
p
2 (3) ltrds
679
is characterized by membership function 1A(XJ), and Bi Indicated in figure 6, neural network between
is a constant that is composed of real. Fuzzy group of E, Al layer and A3layer is indicated realization of the first
CE and U is {NL, NM, NS, ZE, PS, PM, PL}. Figure 4 part about fuzzy rule. To effectively design membership
shows the membership function of fuzzy group. function of first part, first part is applied clustering
To control robust and high performance of induction method that is improved the contraction speed and is
motor drive, dynamic characteristic is required simple structure. Table 1 shows set of fuzzy rule
satisfactory performance that can be adjusted various
speed obeying ability, adapted load variation. Thus this Table I Fuzzy rule table
paper is consist of fuzzy-neuro controller by inter-mixed
NL NM NS ZE PS PM PL
fuzzy controller and neural network, because generally
fuzzy controller is not satisfied these demands. Figure 5 NL NL NL NL NL NM NS ZE
shows fuzzy-neuro controller structure for induction NM NL NL NL NM NS ZE PS
motor drive control.
NS NL NL NM NS ZE PS PM
',l(eCor) ZE NL NM NS ZE PS PM PL
NL NM NS ZEPS PM PL PS NM NS ZE PS PM PL PL
PM NS ZE PS PM PL PL PL
PL ZE PS PM PL PL PL PL
j
Sl(ceo) ) Input space is divided space of 49 at fuzzy rule base.
If rule is equal same action, doing clustering with same
NL NM NS ZEPS PM PL
input space. And then rule that united new clustering is
redesigned by expert and realized neural network. The
~Izv v \/ \ Lcewrlpu] number of fuzzy rule can be very decreased by
+1 C1 C2 0-cl -C2
(b)
1 clustering method. This realization can be designed the
nonlinear function and is divided input space of fuzzy
,uA(Aiqs) by sigmoid characteristic of neuron. Each output of
NL NM NS ZEPS PM PL
neuron at A3layer is truth value of fuzzy rule at each
part space. To realize fuzzy clustering and learning,
Ai*qAA] fuzzy-neuro controller that obtained the same effect at
+1 "Il i2 0 i-"I-2 -1 49 rules is needed function of error. Error function that
(c)
Fig. 4 Membership function of the fuzzy variables is determined through achievement of requirement
partition in fuzzy-neuro controller is frame as
2 2 (1T
=
i 6))
(6)
Ti =
1, (xI,x2)eR'
Fig. 5 Structure of fuzzy-neuro controller 0, otherwise (7)
Figure 6 shows organization of fuzzy-neuro controller.
Input variable of two is error e, variation of error Where, r is number of cluster, Ti is function that
ce and output variable is U. determine whether any input date is included required
cluster. And °i is output of neuron at A3 layer.
A1 A2 A3 A4 After defined error function, To minimize error, next
step is adjust weight Wik and Wi between Al layer
and A3 layer by back-propagation algorithm. Through
weight control, neural network can be completely
realized the first part of fuzzy rule.
A Wjk= (8)
Fig. 6 The construction of fuzzy-neuro controller a Wjk (8)
5. REALIZATION OF FUZZY-NEURO AWij = -77 8i -qOiOj
CONTROLLER (9)
5.1. The first part realization Where
680
(5. =
(T 0,)f'(U,) (10)
6j f'(Uj)Y 6,Wj
=
(11)
°i is output of neuron atA2layer, 17is learning rate,
f ( ) is differentiation of sigmoid function, Ui and U is Or ( kw (k)
~~T NN)LL-
FNN IntgrIo {Ss
indicated total input about each neuron at
0Or(k)
between A2 layer and A3 layer. Finally, to improve Fig. 7 Algorithm of proposed hybrid artificial intelligent
contraction speed and avoid vibration in learning course, controller
new adjustment can be assumed momentum term.
Wjk (t + 1) Wjk (t) + AWJk + [Wjk (t) -Wjk (t- 1)] (12) Output by fuzzy-neuro algorism is iqs(k) output
by AFC is iqs2(k) . After sum of two outputs, order q
A
W,j (t + 1) = WXj (t) + A\Wj + a[W,, (t) Wxy (t 1)]
- -
(13)
axle current obtained trough integrator.
5.2 Realization of the second part Output of motor 6r(k) is compared with output of
reference model Mm (k) and error enm(k) is
In Fig. 6, neural network between A3 layer accomplished by AFC. Reference model is used primary
andA4 layer is indicated second part of fuzzy rule. While system for requirement performance that is satisfied
design reference like overshoot and stability time.
learning, weight wci is adjusted minimize next error Figure 8 is indicated the AFC by reference model. AFC
function and fuzzy-neuro loop is connected in parallel.
E - (U* U)2 r
ceW (k)
2 (14)
O (k) AFC
+Xp ee m(k)
U and U is indicated demand value and real value COr(k)
of fuzzy-neuro controller. Fig. 8 AFC with reference model
Variation ciof Weight ci using general delta
rule can be minimized equation (14) and Output is occurred a Aqs2(k) by AFC. After the sum
The second part of fuzzy rule redefine and determine
the below the equation. Ai*I (k) and A*i2(k) , Ai qs (k) is obtained. And
of As
then iqS(k) is occurred by integrator, and apply the
AWCi = -aE -qdcO0 plant.
(15) AFC input use input of reference model, error of real
WJ(t+1)=WWi(t)+AWWi +a[WZi(t)-WZi(t-1)] (16) speed (eOJm(k)), variation of error (cean (k))
681
7. PERFORMANCE RESULT OF SYSTEM CH1 12, 00 t1 << Mn,in >>
CHZ 12~00Vt
1800
Figure 10 is indicated that manufactured induction
motor operation system that made for experiment in this
paper. ( [rpm]
CH1 -4: 00Vt
CHZ -4:. 00VW
CH2 6, 0 0Vt
20 ... ..
(b)iqsI[A] 0
..
20 ..
(c)Te[N m]
-5000 - 0"
CH4 00 V1
5.00
682
CH1 8-. 00 v << M.,m >)
REFERENCES
CH3 81 00 v
1200
[1] E. Cerruto, A. Consoil, A.Raciti and A. Testa,
(a) af[rpm] "adaptive fuzzy control of high performance
motion systems," in Proc. IEEE IECON Conf. Rec.,
San Diego, CA, Nov. 9-13, pp. 88-94, 1992.
-1200 CH1 -8: 00 v [2] H. Hong, et al., "A design of auto-tuning PID
CH3 -8-. 00 V
20
CH2
CH4
6-. 00 V
v
controller using fuzzy logic," in Proc, IEEE
IECON Conf, Rec., San Diego, CA, Nov. 9-13, pp
(b)'qs [A] 20 971-, 1992
20 [3] E. Cerruto, A. Consoil, P. Kucer and A. Testa, "A
(c) Te[N m]0 fuzzy logic quasi sliding mode controlled motor
CH2
CH4
.10 00 V
- 1O', 00 v
drive," in Proc. IEEE ISIE Conf. Rec, Budapest,
-5. 00 5. 0 0
Hungrary, June 1-3, pp. 652-657, 1993.
Fig. 15 Response characteristics of FNN controller with [4] K. J. Astrom and B. Wittenmark, "Adaptive
change of step command speed control," Addison-Wesley, 1989
[5] D. H. Chung, "Fuzzy control for high performance
CHI 8,:00 v <<Mni >>
vector control of PMSM drive system," KIEE, vol.
1200 47, no. 12, pp. 2171-2180, 1998.
[6] M. G. Simoes and B. K. Bose, "Meural network
based estimation of feedback sigmals for a vector
()[rpm] controlled induction motor drive," IEEE Trans. IA,
vol. 31, no. 3, pp. 620-629, 1995.
[7] M. T. Wishart and R. G. Harley, "Identification and
control of induction machines using neural
20 CHI -.
networks," IEEE Trans. IA, vol. 31, no.3, pp.
20 ....---------
..1.. 612-619, 1995
[8] D.H. Chung, et al., "MRAC fuzzy control for high
(c) Te [N m] performance control of induction motor," The
0 Trans. of KIPE, vol. 7, no. 3, pp. 215-223, 2002
683