Anda di halaman 1dari 6

A Study on Adaptive Load Torque Observer for Robust Precision

Position Control of BLDC Motor

Jong Sun KO*,Sung Koo Youn* and Bimal K. Bose**

*Division of Electrical & Electronics Engineering, Wonkwang University,


344-2 Sinryong-dong, Iksan-city 570-749, Jeonbuk, South-Korea,
Phone : +82-653-850-6713, Fax : +82-653-857-3999,
Email :jsko0,wonnms.wonkwang.ac.kr
* * Department of Electrical Engineering, University of Tennessee
Knoxville, TN 37996-2100, USA

Abstract : A new control method for precision robust high cost and more complex controller because of
position control of a brushless DC(BLDC) motor using nonlinear characteristics [3]. The P-I (proportional-
asymptotically stable adaptive load torque observer is integral) controller is usually used in a BLDC motor
presented in the paper. Precision position control is ob- control, which is simple in realization but difficult to
tained for the BLDC motor system approximately line- obtain sufficiently high performance in the tracking
arized using the field-orientation method. Recently, application. It is, however, well-known that the
many of these drive systems use BLDC motors to avoid
tracking controller problem using state variable feed-
backlashes. However, the disadvantages of the motor
are high cost and complex control because of nonlinear back can be simply solved by the augmentation of the
characteristics. Also, the load torque disturbance di- state variables using the output error [4][5]. It is more
rectly affects the motor shaft. The application of the efficient to obtain the control gain using optimal
load torque observer is published in [l]using fixed gain. control theory than the trial-and-error method for the
However, the motor flux linkage is not exactly known P-I controller. For the unknown and inaccessible in-
for a load torque observer. There is the problem of un- puts, the observer technique was studied by [6]. Also,
certainty to obtain very high precision position control. the application of the load torque observer with fixed
Therefore, a model reference adaptive observer is con- gain was published [l]. However, the machine flux
sidered to overcome the problem of unknown parameter
linkage is not exactly known for a load torque ob-
and torque disturbance in this paper. The system sta-
bility analysis is carried out using Liapunov stability server that gives the problem of uncertainty. There-
theorem. As a result, asymptotically stable observer gain fore, a model reference adaptive observer was consid-
can be obtained without affecting the overall system ered to overcome the problem of the unknown pa-
response. The load disturbance detected by the asymp- rameter in this paper [7]. But that is for the case of
totically stable adaptive observer is compensated by bounded input and bounded [BIBO] stable. So to re-
feedforwarding the equivalent current which gives fast duce position error much more a new asymptotically
response. The experimental results are presented in the stable load torque is needed. That problem is solved
paper. in this paper. The comparison between the two sys-
tem which are fixed gain and adaptive gain system
I. INTRODUCTION responses has been done in detail.

Recently, DC motors have been gradually replaced 11. MODELING OF BLDC MOTOR
by BLDC motors since the industry applications re-
quire more powerful actuators in small sizes. The Generally, the behavior of a small horsepower
advantage of using a BLDC motor is that it can be BLDC motor used for position control is essentially
controlled to have the speed-torque characteristics the same as the permanent magnet synchronous ma-
similar to that of a permanent magnet DC motor. In chine. By means of field-oriented control, it is possi-
addition, the BLDC motor has low inertia, large ble to make i, become zero[3]. Therefore, the sys-
power-to-volume ratio, and low noise as compared to
tem equations of a BLDC motor model can be de-
permanent magnet DC servo motor having the same
scribed as
output rating [2]. However, the disadvantages are the

Dr. KOis currently Visiting Professor in the University of Tennessee, Knoxville, U.S.A.
0-7803-5735-3/99/$10.00 019% IEEE 1091
= k,i, To guarantee the time required for calculating the
load torque to be less than the overall system re-
sponse time and to compensate the load torque at
transient state, a dead beat observer is desirable. It
For implementation of field-orientation control, each then follows from the Cayley-Hamilton theorem that
of the three phase current control commands must be
@a = 0
A A

generated separately. where Q C = Q- L c . The pole placement


using Ackermann's formula can obtain the feedback
gain L as follows: L = P(Q)w;'[o o .... I].
111. CONTROL ALGORITHM Even though the observer feedback gain is obtained
by using the nominal parameter value, there is a cer-
The control reference in the drive is a step value as tain variation or uncertainty of the parameter, such as
in a tracking servo problem. The dynamic equation of the machine flux linkage. To overcome this problem,
a given system can be expressed as follows: an adaptive observer is considered [SI. In this system,
the reference model is the real plant with the aug-
x = Ax@)+ bu(t) mented state variable feedback controller. In a similar
Y = Mt) way, the adaptive load torque observer is considered
as an adjustable system. Eqn. (7) can be partitioned as
where the dimensions of the matrices A, b and c follows:
are n x n , n x l and l x n , respectively. Usually, a A
linear quadratic controller is used to solve the regu- X, =Ax,+B,u-B,TL (8)
lator problem resulting in a state variable feedback. A
new state is defined for the tracking controller as z = CIXl - y , (9)
z = y - y, where y, is the rotor position reference
A

where U = - K , x , - k,z + k, TL , K , ,k , and k, are


[ 5 ] . The control input becomes U = -kw - k,z . From 1x n vector and two scalars, respectively. In the same
this equation, the state feedback controller gain can way, the adaptive torque observer of (7) is described
be obtained by the optimal control law minimizing by
the performance index with the weighting matrices
= A x, + B, U - B2TL+ L, (clxl - c, i,)
A A

Q and R . However, a large feedback gain is needed i, (1 0)


for fast reduction of error caused by the disturbance
which results in a very large current command. If the
load torque TL is known, an equivalent current com- =I,(c,x, -c, ")
mand i,, can be expressed as TL = k&,. Then, the where the A means estimated values. In order to de-
load torque effect is compensated by feeding forward rive the adaptive scheme, Lyapunov theorem is util-
an equivalent q -axis current command to the output -
ized. From (8) (lo), the estimation errors of the rotor
controller. But in the real system, there are many speed and rotor position are described by the follow-
cases where some of the inputs are unknown or inac- ing equation.
cessible. For simplicity, a 0-observer is selected.
Thus, T, can be considered as an unknown and as-
e, =(A-L,c)e, +(B, - k l ) u - B 2 ( T L -6)
sumed to be a constant. The system equation can be
expressed as: (
=Ge,+ B,-B, U - B , T L - T L
A ) (
where e, = x, - x1 and G = A - L,c . NOW,a new Ly-
apunov function candidate V is defined as follows:
V = erPe, +-
a
( B, -Bi r ( B , - i i )

P
+'(TL -+L)2

where P and a are a positive definite matrix and


positive constant, respectively. The time derivative of

1092
V becomes obtained as

v = .{((A -L,C)TP + P(A -L,c))e,


(a') [
+2 eru+-ABi AB, - 2 eTB, +-TL
; A ] .ATL (14)
where a, and a , are the elements of the vector a .
The resultant block diagram of the model reference
adaptive observer is shown in Fig. 1.
A A

where AB, = B, - B, and AT, = TL - TL . From (14))


the adaptive mechanism can be obtained by equaliz-
ing the second term to zero as follows: 1 Reference Model
I -- I
I rqtor position
A
I Augmented State + Plant 1
BT =-aeTu I

The third term can be decreased to zero using the


following equation:

where p' = - Pp> 0 . Therefore, the new adaptive


2J
torque observer can be obtained as

L I

If we decide the observer gain matrix L, with optimal Fig. 1. Configuration of the model reference adaptive
theory, the first term of (14) can be negative semi- torque observer
defmite. Assuming that there exists a positive definite
matrix R such that
G ~ P PG + = -R (18) IV. CONFIGURATION OF OVERALL
the derivative of the Lyapunov function candidate can SYSTEM
be written as
The block diagram of the proposed controller is
V = -er(R)e, I O . (19) shown in Fig. 2 where the controller is composed of
Hence, e, is uniformly asymptotically stable. And the two parts. The position controller is composed of the
maximum error can be decreased by reducing the augmented state feedback. For the realization of the
estimated load torque error and selecting the gain L, augmented state z(k+l), the discrete form of this state
properly to make e, as zero. Discrete motor equation is approximately obtained by using a trapezoidal rule.
DSP TMS32OC31 is used as a digital controller. An
can be written as the following ARMA model: asymptotically stable load torque observer implement

where A,, A,, A3, B, and B, are 1, ala3lh,


-ala2I h, b, and a3bl, respectively. In this case,
B, and B2 are not exact values. So, only these
terms are expressed at the quasi term. Employing the
definition of the new quasi -output can reduce the
order of the estimated matrix, Then, quasi-output Y
and model output Y,,, can be obtained as the differ-
ence of real output and known values as follows:
A

Y(k + 1 ) = (y(k + 1) - (0,0,) = (0* 0, (21)


Y ,(k + 1) = b2iT( k ) + a3b,i, (k - 1). (22)
Using, E = Y(k + 1) - Y, ( k + 1) the gradient can be Fig. 2 . Block diagram of the proposed precision
position control system
1093
the other part. Estimated torque is used for compen- The hysteresis-band gap is chosen as 0.05 A , and the
sation of position error. Motor system is driven by sampling time Ts is determined as 0.1 ms . After
vector controller which make the system as linear some trial and error, the weighting matrix is selected
system. as Q = diagE0.1 60 10001, R = 1 and optimal gain
matrix becomes k = [0.0598 2.0810 7.35401. The
V. SIMULATION AND EXPERIMENTAL nominal deadbeat observer gain becomes
RESULTS L = [21762 2.8187 -1000.21. In the adaptation
mechanism, the adaptation ratesa, , a,,2222 and
The parameters of a BLDC motor used in this ex-
periment are given as follows: p' are obtained as 0.004,0.002 and 0.025 by trial and
error, respectively. The simulation results are in Fig.
Power 1 4OOwaft 3. The deadbeat observe has a quite good result in
Inertia I 0.363xlO4kgm2 point view of large scales as shown in Fig. 3(a). That
1.3 Nm current shows adaptive control action of reduction
effort of algorithm is bigger than dead beat system.
Rated current
Stator resistance 1.07nlphme
Time constant
150
100
I >b
0.50
0.00
I bb
-050 1
b >O

0 bb
I b
-b bb 1
-2.00
- I bb

- I 20
[AI
2.0 I I

-1.5 r - 1
I
r -
I

-2.0
180 185 190 195 200 205 210

(a) Dead beat control with load (a) Dead beat control with load

I bb 1.50
I Ob 100
C >b 0.50
C Ob 0 00
I C
- - b bb -0.50 1 0
- I Ob -1 00
- I bb

I
-1 50
-2.00 ' I

[AI

-2.0 '
(b) Adaptive observer with load 180 185 190 195 200 205 210

Fig 3. Simulation results of the rotor position


and q phase current command (b) Adaptive observer with no load

1094
adaptation to different condition. Current pattern is
2.00 changed as a position response. And the difference
1.50 can not be detected in position result between unload
1.oo case of Fig. 4(b) and loaded case of Fig. 4(c). As
0.50 shown in Fig. 4(c), the magnitude of current in tran-
0.00 sient is increased up to 2[A] which has more control
-050
effect to reduce the effects of inertia load. Zoomed
-1 .oo
-1 S O
results are depicted in Fig. 5. In conventional case,
-2.00 I J error magnitude is not increase as increase load. But
[AI there is a little position chattering. These results came
2 .o from parameter uncertainty about 20%. This small
1.5 chattering decrease as shown in Fig 33).A process-
1 .o ing of this algorithm is shown in Fig 6. It is started
0.5
fiom Fig. 5(a) as a nominal gain. At beginning we can
0 .o
-0.5 use this controller after about 3 minute which is a
-1 .o time for adapting. Next day, adapted data will be used
-1.5 for saving few minutes.
-2 .o
180 185 190 195 200 205 210 [rad3
1.75 r
(c) Adaptive observer with load

.
Fig 4. Experimental results of the rotor position
and q phase current command
..
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5
The experimental results are in Fig. 4, Fig. 5 and sec

Fig. 6. The deadbeat observe has also a quite good 1.75 r


result in point view of large scales as shown in Fig. 1.70
4(a) as same as simulation. That current shows con- 1.65
trol action of reduction effort of algorithm. But there 1.60
is a little chattering of position at bottom side. In Fig.
1.55
4(b) the different type control action is depicted.
1.50
There is very small position error in not only upper 60.0 60.5 61.0 61.5 62.0 62.5 63.0 63.5
side but also lower side. This indicate good sec
1.75 r
Padl 1.70
1.75 I 1 1.65
1.60
1.55
1.50
- 120.0 120.5 121.0 121.5 122.0 122.5 123.0 123.5
sec

1.75 r

1’70
1.65 i7
1 180.0 180.5 181.0 181.5 182.0 182.5 183.0 183.5
sec

1a 3 5 1845 185.5 186.5 Fig 6. Process of successive adaptation with no load


(b)
VI. CONCLUSION
Fig 5. Performance comparison with dead beat and
adaptive observers A load torque observer with the model reference
(a) results of conventional dead beat observer adaptive system is used to obtain better performance
(b) results of proposed adaptive observer from the BLDC motor in a precision position con-

1095
trol system. Also, the augmented state variable feed- [8] Y. D. Landau, Adaptive control : the model refer-
back is used in the digital experimental control sys- ence approach, Marcel Dekker, Inc., 1979, pp.
tem with an optimal gain. The system response com- 18-30.
parison between the fixed dead beat gain observer and
the adaptive observer has been done. The load torque
compensater based on the adaptive observer and the
feedfonvard injection can be used to cancel out the
steady state and the transient position error due to the
external disturbances, such as a various friction and a
load torque. The stability analysis is carried out using
Liapunov stability theorem. Under this analysis, the
new adaptive torque observer is proposed. In the pro-
posed scheme, the rotor position error caused by non-
exact parameter is decreased asymptotically. The total
adaptive control system is realized by a digital con-
troller DS1102(TMS320c31) with 0. lms sampling
time where the gain is obtained in z-domain using the
optimal theory.

ACKNOWLEDGEMENT

This work was supported by grant No. 981-0905-016-


1 from the Basic Research program of the KOSEF.

REFERENCES

[I] J. S . KO,J. H. Lee, S . K. Chung, and M. J. Youn,


“A robust position control of brushless DC motor
with dead beat load torque observer,” IEEE
Trans. on Industrial Electronics, vol. 40, no. 5,
1993, pp. 512-520.
[2] A. Kusko and S.M.Peeran , “Brushless DC mo-
tors using unsymmetrical field magnetization,”
IEEE Trans. on Industry Applications, vol. IA-23,
no. 2, 1987, pp.319-326.
[3] D. W. Novotny and R.D. Lorentz ”Introductionto
field orientation and high performance AC
drives”, IEEE-IAS Tutorial Course, 1986
0

[4] K. J. A strom and B. Wittenmark, Computer


Controlled System, Prentice Hall , 1984, pp.198-
212.
[ 5 ] E. J. Davison, “The output control of linear t h e -
invariant multivariable systems with unmeas-
urable arbitrary disturbances,” IEEE Trans. on
Automatic Control, AC-17, no. 5 , 1972, pp. 621-
630.
[6] J. S . Meditch and G. H. Hostetter, “Observer for
systems with unknown and inaccessible inputs,”
Int’l.. J. Control, vol. 19, no. 3, 1974, pp. 473-
480.
[7] Jong Sun KO, Young Seok Jung, and Myung
Joong Youn, “MRAC Load Torque Observer for
Position Control of Blushless DC Motor,” Int’l..
J. of Electronics, Vol. 80, No. 2, pp. 201-209,
June, 1996

1096

Anda mungkin juga menyukai