F.J. Lin
Authorized licensed use limited to: Reva Institute of Tehnology and Management. Downloaded on December 14, 2008 at 01:10 from IEEE Xplore. Restrictions apply.
L, = magnetising inductance per phase
1, I]
L, 0 0
L, = stator inductance per phase
L, = rotor inductance per phase G=hjo!im (20)
or= rotor angular speed
The discrete time equivalent model of eqn. 1 is 0 0 k
+
x(k 1) = $ x ( k ) ru(k) + (7) H(X(k)) = [iqs(k) (21)
where
The process noise W(k) is characterised by
E { W ( k ) }= 0 (22)
E { W ( k ) W ( l ) T=
) Qdki Q 20 (23)
The measurement noise V(k) is characterised by
r= (1 h
eA"ds)B
E { V ( k ) }= 0
E { V ( k ) V ( l ) T=) R S k l
The initial state is characterised by
R20
(24)
(25)
where h is the sampling interval. One way to compute 4
and r is as follows [24]: E(X(0))= xo (26)
 20))
E ( ( X ( 0 ) XO)(X(O) = Po (27)
I 1
(&R,L, ) z I  L ~ , w ,s 2 + L , , , z 3 z ~  L , , L , W r m 4 + L r p * ~ where klk denotes a prediction at time k based on data
L2,w,z1+(&R5 L,)m2+L7,LL,w, z 3 + L , r , z 4 z 8 + L r p d , up to time k. Similarly, (k + 1)lk denotes a prediction
hn R b L r r , ~ i + L T r b L X3Zw
+ (, ~  ~ , ~ ~ ) ~ ~ + L c L r W ~ ~ 4  ~ at time k + 1 based on data up to time k. The block
~ r ~ ~ y r
Authorized licensed use limited to: Reva Institute of Tehnology and Management. Downloaded on December 14, 2008 at 01:10 from IEEE Xplore. Restrictions apply.
Y(k+l)
I HP(S) =
where
P = number of poles
iis = torque current command generated from the
speed controller
ii5 = flux current command
J = total mechanical inertia constant
B = total damping constant.
Fig.1 EKF block diagram
In the currentcontrolled PWM inverter of the indirect
fieldoriented inductionmotor drive, the current com
4 Modelling of fieldoriented inductionmotor mands in the synchronous reference frame, denoted by
drive i;, and i;,, must be transformed into the phase domain
to yield the reference currents. The unit vector (cos@,+
The block diagram of the indirect fieldoriented induc jsino,) used in the transformation matrix is generated
tionmotordrive system combined with an EKF esti by using the measured rotor angular velocity CO, and
mator is shown in Fig. 2, which consists of an the following estimated slip angular velocity wx1:
induction motor loaded with a DC generator, a ramp
comparison currentcontrolled PWM voltage source
inverter, a fieldorientation mechanism, a coordinate
translator and a speedcontrol loop. The induction Since Rr is sensitive for different operating conditions,
the estimated R, from the EKF estimator will be used
motor used in this drive system is a threephase Ycon
in eqn. 38 to guarantee a correct estimation of the slip
nected twopole 8OOw 60Hz 120Vi5.4A type.
frequency, and to preserve the decouple control charac
teristic.
The dynamic modelling, based on measurements 1251,
is applied to find the drive model offline at the nomi
nal case (aro= lOOOrpm, RL = On). The results are,
(the scaling is 5OradislV)

K t = 0.5756NmIA a = 0.538 b = 3.31

J = 6.04 x 10p3Nms2= 0.302Nmsrad/V (39)

B = 3.25 x 10p3Nms/rad = 0.1625Nm/V
The estimated drive parameters will be used in the
design of the proposed controller.
5 Design of IP controller
L _ _ _ _ _ _ _   _   _   I
Owing to the absence of zeros, the overshoot of the
Fig.3 Indirect fieldoriented inductionmotor drive M.ith EKF estimator step response in eqn. 6 is avoided by setting the damp
showing the control rystem block diagram
ing ratio 5 = 1. Then one can find the unitstep
response in eqn. 40 to be
By using the referenceframe theory and the linearisa
tion technique, the fieldoriented inductionmotor LJ,(~) = 1  eP'7Lt(l +writ) (42)
drive, shown in Fig. 2, can be reasonably represented For convenience in designing the IP controller quanti
by the control system block diagram shown in Fig. 3, tatively, the response time is defined as, the time
in which, G,(s) is an IP speed controller and required for the step response to increase from 0 to
Te K 'taqs
* 90% of its final value. By specifying the response time
(35) as t,,, the following nonlinear equation is yielded:
0.9 = 1  ew"t" (1 + wnhe) (43)
188 IEE Pro,.Electr. Power Appl., Vol. 143, No. 3. May 1996
Authorized licensed use limited to: Reva Institute of Tehnology and Management. Downloaded on December 14, 2008 at 01:10 from IEEE Xplore. Restrictions apply.
Solve the above nonlinear equation to obtain a,; then P(k  l)C(k  1)T
K ( k )=
from eqn. 40 one can find the parameters of the IP
controller as
1 + C(k  l ) P ( k  1 ) C ( k  1)T (49)
1 P ( k  1 ) C ( k  l ) T C ( k  1 ) P ( k  1)
KI = Jw:/K~ K p = ( 2 7 ~ B~) / K ,
1 + C ( k  1 ) P ( k  l)C(/k 1)T 1
(44) P(lc)=[P(kl)
01
Eqn. 40 also indicates that the tracking steadystate (50)
error is zero. It follows from the above analysis that where
the desired tracking specifications can be completely
achieved by using the simple IP controller. C ( k ) = [wT(k)&(W %/GI (51)
6 Proposed RLS estimator with torque observer @(IC) = [a1 (IC), b1 (IC)] (52)
The value of the forgetting factor a should be restricted
The block diagram of the proposed RLS estimator to 0 < a 2 1. After O(k) is obtained, the estimated val
with a torque observer is shown in Fig. 4. If online ues of J and B can be easily determined from eqn. 46.
parameter identification is available when the parame The convergence of the RLS algorithm with the torque
ter variations occur, the IP controller can be of online observer will be shown by simulation results.
design, according to eqn. 44, to preserve the tracking
The observed torque is also fed forward through a
performance. Though the RLS estimator is one of the
weighting factor W, to realise a robust speed control
most effective methods for online parameter identifica
system, as shown in Fig. 4. Ideally, the value of W is
tion, it is difficult to get unbiased results [22] in this
set at 1. For the system possessing nonlinearities, e.g.
application, owing to the dynamic model of the plant
being disturbed by the external load torque. As shown limiter, deadtime element etc., the weighting factor
in Fig. 4, the proposed RLS estimator is combined should be selected at less than 1 to preserve stability.
with a simple torque observer to resolve the above dif
ficulty. The torque observer uses the inverse dynamic 7 Design and simulation results
of the motor drive to obtain the observed torque,
which is denoted TLIK,. The torque current command 7. I Simulation of EKF estimator
minus this value results in the unbiased identified The sampling interval used in the simulation is 2ms.
parameters, J and B, which denote the estimated rotor The process noise covariance and measurement noise
inertia constant and damping constant, through the covariance are set as
RLS estimator. Then J A a n dB in the torque observer
are replaced by J and B. By this recursive process, the
identified J and B parameters and the observed load
torque will quickly converge to their real values.
The simulation result is shown in Fig. 5. Between 0 and
toraue observer
1s, the estimated rotor resistance quickly converges to
its real value, which is 1.3Q. Beginning at 1s, the real
rotor resistance is varied according to the following
equation:
R, = 1.3 (t  l ) f l + (54)
where t is the simulation time. The estimated rotor
resistance can trace the variance of the real value as
shown in Fig. 5.
o.5[ 1
where
O I ' , ~ ' ~ ~ '
a1 =  exp(T,B/J) bl = ( l y , / B ) ( l ~ e x p (  T s ~ / J ) ) 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
(46) time,s
Fig. 5 Simulation result ofthe EKF estimator
and T, is the sampling interval. The system model can a rotor resistor
be written as b estimated rotor resistor
Authorized licensed use limited to: Reva Institute of Tehnology and Management. Downloaded on December 14, 2008 at 01:10 from IEEE Xplore. Restrictions apply.
7.3 Simulation of RLS estimator with torque
observer
1120 1
I
In this simulation, the load torque is set at 1Nm. In the
beginning, J and B are set according to their nominal
values. Then, at 2s, J is abruptly increased by a factor
of three. From the simulation results shown in Fig. 6:
the estimated values of J and B, after a step change in
the speed command, are quickly converged to their true
values under contant load disturbance. Though the
value of the observed torque is influenced by the speed
lOZO~/
transient, the steadystate observed value is correct;
/ /
and this observed torque transient has little effect on
the desired control performance from the simulation
and experimental results owing to feedforward control.
   . . lo
0
 05 1 15 2
, 
25
time, s
,
3
'
35 4 45 5
Fig. 6 Simulation se.su1t.s (j"the RLS estiniator and the torque observer.
showing sotor s eed () and e.mmated values of'rotor speed (xi, observed
~(6
torque ( a i , and ~ ( c )
60' ' ,
7.4 Simulation of proposed controllers 0
"
0.2 04
1
06 0.8 1 1.2 14
1
1.6
'
1.8 2
To investigate the effectiveness of the proposed robust tirne,s
controller, suppose that the mechanical inertia constant Fig.8 Simulation results of proposed controller showing load regulation
N Nominal case with IP controller only
J is significantly changed to allow the transfer function b Case 1 with IP controller only
model HJs) to be changed to c Case 2 with IP controller only
d Nominal case with proposed controllers
e Case 1 with proposed controllers
( J = 5 xJ) (56)
+10.1625
Case 1 : f Case 2 with proposed controllers
HpL(S) = 1.51s
The control intervals are all selected at 5ms. To reduce
Case 2 ( J = 0.5 x 7) the calculation burden of the CPU and to increase the
(37) accuracy of the threephase command current, the
coordinate transformation in the fieldoriented mecha
In the nominal case, the step rotor speed tracking and
nism is implemented by an AD2S100 AC vector proc
step load regulating responses of the drive system
shown in Fig. 4, without and with the proposed online essor. The control interval of the vector processor is
selected at 0.2ms. The pure differentiator, which is
IP controller design and feedforward control, are
shown in Figs. 7 and 8 by curves a and d, respectively. shown in Fig. 4, may amplify the highfrequency noise,
The results are identical for the tracking conditions, so the operating stability of the closedloop controlled
and the regulating performance is improved. Suppose drive will be greatly affected. Thus in practical imple
that the mechanical inertia constant J is changed mentation, a filter is used as an alternative. It is
according to eqns, 56 and 57; then, the rotor speed designed to behave as a pure differentiator for the main
responses due to a step command change and a step lowfrequency dynamic signal, but it becomes a low
load torque change, without and with the proposed on pass filter for highfrequency signals.
line IP controller design and feedforward control, are Some experimental results are provided in the follow
also compared in Figs. 7 and 8. Significant perform ing to show the effectiveness of the estimators and the
ance improvements both in the tracking and regulating proposed controllers. The estimated rotor resistance
responses by the proposed controller are observed from obtained from the EKF estimator, which converges to
the results. its real value, 1.3!2, within 0.3s, is shown in Fig. 10.
This estimated value, after suitable filtering, is used on
8 Implementation and experimental results line in the indirect fieldoriented mechanism. To avoid
the numerical problem of dividing by zero during the
The block diagram of the coprocessor computer con implementation for the parameters J and B, the param
trol system for the indirect fieldoriented induction eters a and b (expressed in eqn. 37) are estimated
motor servo drive is shown in Fig. 9. The robust speed instead of J and B. Moreover, the load resistance of the
control algorithm is realised in a 486DX66 and the DC generator is set at 118R during load torque obser
EKF and RLS algorithms are realised in a DSP32C. vation. The estimated values of a, b and the observed
190 IEE Proc.Electr. Powes Appl., Vol. 143, No. 3, May I996
Authorized licensed use limited to: Reva Institute of Tehnology and Management. Downloaded on December 14, 2008 at 01:10 from IEEE Xplore. Restrictions apply.
L
~ +
t*
0 25;
Fig.11 Experimental r e d 3 of the RLS estimator and the torque
observer showing the valne~of the a and b parumeters
Fig.16 Measured rotor speed res onhe5 at the nominal case owin
step load resistance change with onfne IP controller design andjee&fi
ward control
9 Conclusions
Authorized licensed use limited to: Reva Institute of Tehnology and Management. Downloaded on December 14, 2008 at 01:10 from IEEE Xplore. Restrictions apply.
First, an EKF estimator was implemented to estimate KRISHMATU’,R., and BHARADWAJ, AS.: ‘A review of
the rotor resistance. The estimated rotor resistance was parameter sensitivity and adaptation in indirect vector controlled
induction motor drive systems’, IEEE Power Electronics Special
used to estimate the slip frequency. Next, an IP speed ist Conference Record, 1990, pp. 560566
controller was quantitatively designed, according to the MATSUO, T.; and LIPO, T.A.: ‘A rotor parameter identification
estimated nominal drive model and the prescribed scheme for vectorcontrolled induction motor drives’, IEEE
Trans.: 1985, IA21, pp. 624632
speedtracking specifications. The proposed RLS esti OHNISHI, K., UEDA, Y., and MIYACHI, K.: ‘Model reference
mator was implemented to estimate the rotor parame adaptive system against rotor resistance variation in induction
ters which are used to design the IP controller online. motor drive’, IEEE Trans., 1986, I C 3 3 , pp. 217223
The proposed RLS estimator was combined with a SUIMOTO, H., and TAMAI, S.: ‘Secondary resistance identifica
tion of an induction motor applied model reference adaptive sys
torque observer to obtain unbiased results. Then the tem and its characteristics’, I E E E Trans., 1987, TA23, pp. 296
observed torque was fed forward to obtain robust con 303
trol performance. KOWALSKA, T.O.: ‘Application of extended Luenberger
observer for flux and rotor timeconstant estimation in induction
motor drives’, I E E Proc. D,1989, 136, pp. 324330
 command f
T CHAN. C.C., and WANG, H.: ‘An effective method for rotor
t 00 T rpm resistance identification for highperformance induction motor
speed vector control’, IEEE Trans., 1990, 1IG37, pp. 477482
+
I / HUNG. K.T.. and LORENZ, R.D.: ‘A rotor flux errorbased,
c
4 : : ’’ ’
! ’ ‘ b

+. .
adaptive tunmg approach for feedforward field oriented induction
2000rpm I machine drives’; Proceedings of IEEE IAS Annual Meeting, 1990,
i
I
0 25s pp. 589594
Fi .I7 Mearured rotor speed iesponses at on operatiii condition of ZAI: L.C.. and LIPO, L.A.: ‘An extended Kalman filter
2080 rpnt owing to ~ t e ptommand change wtlz an IP conti ol& approach to rotor time constant measurement in PWM induction
motor drives‘. Proceedings of IEEE IAS Annual Meeting, 1987,
pp. 177183
10 ATKINSON, D.J., ACARNLEY, P.P., and FINCH, J.W.:
‘Observers for induction motor state and parameter estimation’,
f (R~=Ofl+24Ofl) I E E E Trans., 1991, IA27, pp. 11191127
11 MENDEL, J.M.: ‘Lessons in digital estimation theory’ (Prentice
Hall, Englewood Cliffs, NJ, 1987)
12 FUCCIO, M.L.: ‘The DSP32C : AT & T’s secondgeneration
4+
floatingpoint digital signal processor’, IEEE Micro, 1988, 8, (12),
60rpm 0 25s pp. 3 0 4 7
Fi . I 8 Measured rotor speed response, at an opercitrizg condition of 13 YEH. H.G.: ‘Realtime implementation of a narrowband
20%0 rpnt m i n g to step load reiistarzce change ~ 1 t an
h IP controllei Kalman filter with a floatingpoint processor DSP32’, IEEE
Trans., 1990, IC37, pp. 1318
1 command + 14 HO. Y.Y.; and SEN, P.C.: ‘A microcontrollerbased induction
motor drive system using variable structure strategy with decou
pling’, I E E E Trans., 1990, I C 3 7 , pp. 227235
15 HSIA. T.C.: ‘A new technique for robust control of servo sys
tem’, IEEE Trans., 1989, I C 3 6 , pp. 17
t t 16 OHISHI, K., NAKAO, M., OHNISHI, K., and MIYACHI, K.:
7 . . ”‘
” ’ t : :
+ ” ! ’Microprocessorcontrolled DC motor for loadinsensitive posi
it 2000rpm 0 25s tion servo system’, I E E E Trans., 1991, I C 3 8 , pp. 2125
L
17 LIAW, C.M., and LIN, F.J.: ‘A discrete adaptive induction posi
Fi . I 9 Measured rotor .speed responses at an operating condition oJ tion servo drive‘, IEEE Trans., 1993, EC8, pp. 350356
2080 rprn owing to step command change with orzline IP controiier design 18 LIN, F.J., and LIAW, C.M.: ‘Control of induction fieldoriented
and feedforward control induction motor drives considering the effects of deadtime and
parameter variarions’, I E E E Trans., 1993, I W O , pp. 486495
19 MATSUI. N., MAKINO, T., and SATOH, H.: ‘Autocompensa
tioii of torque ripple of direct drive motor by torque observer’,
IEEE T ~ N I ~1993,
s . , IA29, pp. 187194
20 IWASAKI, M.; and MATSUI, N.: ‘Robust speed control of IM
with torque feedforward control’, IEEE Trans., 1993, IGlO, pp.
553560
17rpm
fL
+c
0 25s 21 KO, J.S.. LEE, J.H., and YOUN, M.J.: ‘Robust digital position
control of brushless DC motor with adaptive load torque
Fi .20 Measured rotor s eed iesponses at iin operating condition of observer’, I E E Proc. Electr Power Appl., 1994, 141, pp. 6370
“n c e change irirh online IP controller
20%0 rpm owing to ste ~ o J ‘ r xc m 22 LAUDAU, J.D.: ‘System identification and control design’ (Pren
design ani~fc.edfor,ycirrontrol ticeHall, Englewood Cliffs, NJ, 1990)
23 BOSE, B.K.: ‘Power electronics and AC drives’ (PrenticeHall,
Englewood Cliffs, NJ, 1986)
10 References 24 ASTROM, K.J., and WITTEHMARK, B.: ‘Computer controlled
systems‘ (PrenticeHall, Englewood Cliffs, NJ, 1990)
I KRTSHMAN, R., and DORAN, F.C.: ‘Study of parameter sensi 25 LIAW, C.M.. OUYANG, M., and PAN, C.T.: ‘Reduced order
tivity in highperformance inverterfed induction motor drive sys parameter estimation for continuous system from sampled data’,
tems’, I E E E Trans., 1992, IA23, pp. 623635 Trans. ASME, J. Dyn. Syst. Meas. Control, 1990, 11, pp. 305308
Authorized licensed use limited to: Reva Institute of Tehnology and Management. Downloaded on December 14, 2008 at 01:10 from IEEE Xplore. Restrictions apply.