Download as pdf or txt
Download as pdf or txt
You are on page 1of 5

Proceedings of the American Control Conference Chicago, fllinois June 2000

Discrete-Time Robust Tracking Control Using A State Space Disturbance Observer


Seung-Hi Lee, Young-Hoon Kim, Sang-Eun Baek Electro-mechanics Lab. Samsung Advanced Institute of Technology P.O.Box 111, Suwon 440-600, KOREA s h l , younghoon, sebaek@sait.samsung.co.kr
for high frequency sensor noise rejection. Thus, the Q filter can be interpreted as a complementary sensitivity function of the feedback loop. For further information, the interested readers are referred to [3][4] and references therein. Conventional approach in the design of state space disturbance observers is t o model disturbances and to augment it into the estimator model (e.g. bias and periodical disturbance [l]). However, it is in general impossible to build models for arbitrary disturbances. Therefore, state space disturbance observers are hardly used except for simple disturbances which can be modeled easily. This paper considers design of a discrete-time robust tracking control system which consists of a state space disturbance observer, a state estimator, and a state feedback controller. In this paper, we propose a new discrete-time state space disturbance observer not requiring modeling of disturbances. An application example is presented to show applicability and effectiveness of the proposed disturbance observation method. Throughout the paper, a transfer matrix in terms of state-space is denoted by

Abstract
This paper considers desigll of a discrete-time robust tracking control system which consists of a state space disturbance observer, a state estimator, and a state feedback controller. A new discrete-time state space disturbance observer is proposed not requiring modeling of disturbances. Through the results of applications, i t is shown that the proposed method is very effective to compensate variety of disturbances and t o improve the performance of a tracking control system in the existence of external disturbances. It is also observed that the proposed state space disturbance observer allows more accurate state estimation in the existence of modeling error and disturbances.

1 Introduction
External disturbances have significant impact on the performance of a tracking controller. Accordingly, there has been much work on the problem of rejecting the disturbances effectively. Modeling errors in t h design of control systems can also cause tracking error. Modeling errors include unmodeled dynamics, which represent high order flexible modes, as well as plant parameter uncertainties/variations during operation. The difference between the output of the plant and the output of the nominal model is regarded as an equivalent disturbance applied to the nominal model. Disturbance observers are used to estimate the equivalent disturbances. Thus, a disturbance observer can be used to make the plant behave like the nominal model in the existence of the equivalent disturbances. In [3], the idea of a disturbance observer was proposed. In [4], the disturbance observer theory was refined based on the design of two degrees of freedom sevo controllers and the factorization approach. The design of a disturbance observer is dependent on the design of so called the Q filter, which determines robustness and disturbance rejection performance. In the design of the Q filter, unit low frequency gain is required for disturbance rejection, while high frequency roll-off is required
0-7803-5519-9/00 $10.00 0 2000 AACC

p ( M ) denotes the spectral radius of a matrix M . It is


assumed that all the vectors and matrices have appropriate dimensions.

2 Problem Statement
To begin with, consider a continuous-time model of a plant described by

x = A,x+B,u+w y = C,x+n

(1)

where w is the disturbance and n is the measurement noise, A,, B, and C, are matrices with appropriate

4194

dimensions. I t is assumed that ( A p , p ,C,) is stabilizB able and detectable. In addition, we assume the disturbance w is bounded. However, no time/frequency characteristics of w are assumed. Performance of a control system is significantly affected by disturbances. Thus, it is very important to reject the disturbance effectively. As mentioned before, the transfer function approach has many drawbacks. It requires filtering of output measurements. Thus, the performance of the disturbance observer is determined by the Q filter. In addition, the transfer function approach can not be used in the multi-input-multi-output (MIMO) cases, while the state space approach can be. In the state space approach, specific disturbance models are augmented in the estimator model to estimate the disturbances. The models are developed from the timelfrequency characteristics of disturbances (for example, constant bias and periodic disturbance) [l]. When the time/frequency characteristics of the disturbance are unknown, the assumption of piece-wise constant bias is used with sufficiently fast eigenvalue assignment to the states associated the disturbance. However, fast varying disturbances can not be observed effectively by this approach. In addition, performance of a state estimator can be affected.

where

and T, is the sampling period. Here, (@,I',C)is assumed to be stabilizable and detectable. For the simplicity of formulation we assume no computation time delay. However, the time delay can be handled easily by augmenting the delayed control as an additional state [I].
3.2 N e w State S p a c e Disturbance Observer The basic structure of the proposed state space disturbance observer (shown in Figure 1) is expressed by

where % is the state of the model, z is the state of the plant, Kdist is the disturbance observer gain to be determined. If all the state measurements are not available, we use the estimated states ?. In this case, we use

udist(k) = Kdzst(e(k) - ?(k)).

(4)

It should be noted that even a well designed disturbance observer can not compensate disturbances perfectly. In this paper, the remaining disturbance is regarded as a residual disturbance. This residual disturbance is considered as an exogenous signal in the design of a controller. Thus, the disturbance observer is to be designed to suppress the equivalent disturbance, while the controller is to be designed such that the feedback loop is less sensitive to the residual disturbance.
Udist

Plant Figure 1: Disturbance observer scheme

3 Control S y s t e m Design
In this section, we design a discrete-time control system which consists of a state space disturbance observer, a current state estimator, and a state feedback controller. Firstly, the concept of the new state space disturbance observer is introduced. Subsequently, assuming that the disturbance observer is working properly, we design a state estimator and a state feedback controller for the system with the residual disturbance. Finally, we design a disturbance observer to compensate equivalent disturbances and to make the closed loop system stable.

The disturbance observer gain Kdist determines the convergence rate to the disturbance w and affects the stability of the feedback loop. A condition for proper gain &ist will be addressed in Disturbance Observer Design subsection. Not like the transfer function a p proach in the s domain, there is nothing like the Q filter. Thus, there is no additional dynamics associated with the disturbance observer itself. Given the states 3 and 2, no additional dynamics computation is required to observe the disturbance. The states 3 and 2 are to be obtained from a state estimator which is to be addressed in the next subsection. The disturbance observation zLdist ( k ) is injected into the feedback control loop (as shown in Figure 2) to compensate disturbances. As a result, we have a fictitious system in which only the residual disturbance is acting on (as shown in Figure 3). We design a state estimator and a state feedback controller for this system in the following subsections.

3.1 Discrete Time Model

TO begin with, consider a discrete-time model described


bY

Gm =

[*]
41 95

Figure 2: Control system with disturbance observer


d

where K , is the feedback gain to be determined such that both tracking performance and disturbance rejection are satisfactory. In order to attain sufficient robustness, the loop transfer recovery (LTR) design technique is applied in the design of state feedback gain K,. Given the target loop transfer function Gtl, the state feedback gain K , is determined using the LTR methodology at the plant output such that the loop transfer function approaches to the target loop transfer function. By solving the DARE, with control weighting R2 and state weighting R1 = qCTC for recovery,

x -G

~ + aTxr + rTxr)-'rTxdr x ~ , (1-2~ -RI = O (10)

one can determine the state feedback gain from

K,
Figure 3: System for controller design
3.3 State Estimator Design The current state estimator [l]is expressed by

(rTxr R2)-*rTx+ +
(10).

(11)

where X

2 0 is the solution of

3.5 Analysis of Feedback Loop The sensitivity transfer function is expressed by

q k + 1)

i ( k ) = f ( k ) L,(y(k) - C z ( k ) ) = dri(lc) rzL(Ic)

+ +

(5.)
where

where y is the output measurement, 2 and f are the, state update and the state prediction, respectively. Although the time/frequency characteristics of the residual disturbance is unknown, it is reasonable to assume that the residual disturbance is bounded. Considering this fact, we design a current state estimator

-(a

dr - rK,L,C - rK,)L,C

r ( K , - KAC) dr - r K , - (dr - rK,)L,C

'

in the 'H2 point of view. The current estimator gain L, that minimizes the estimation error in the 'H2 sense with given weightings can be obtained by solving the discrete algebraic Riccati equation (DARE)

The complementary sensitivity transfer function is


Gcs=

I*[
(Kc - K&C) 6, - I'K, - (6,- TK,)

- drYV

+ drYCT(DIDT + CYCT)-'CYdrT
-rlrT =o.
(6)

where

a,,

dr - rK,L,C - (6,- rK,) L,C

L,C

The estimator gain L, is computed by

'

L,

= YCT(CYCT

+ DIDT)-'

(7)

where Y > 0 is the solution of ( 6 ) . Then, from (5) we obtain the transfer function of the state estimator
Gtl

[ ] F
c
CL,
(9)

It should be noted that the complementary sensitivity function is the Q filter equivalent in the design of state space disturbance observer based control systems.
3.6 Disturbance Observer Design Using (5) we can rewrite the disturbance observer model (4) shown in Figure 4-(a) as
Udist(k)

which is used as a target loop transfer function

3.4 State Feedback Design To begin with, consider a system described in Figure
3. Here, d = w - U & t , the residual disturbance. State feedback control law is described by
u ( k ) = KC(ZT(k) - W ) )

=KdistLc

(-Y(k) -tC z ( k ) )

(14)

which is depicted in Figure 4-(b). The disturbance observation u d i s t ( k ) is represented in terms of the estimation error y ( k ) - Cf(lc). So far, we have assumed that an appropriate disturbance observer gain Kdist is

4 96 1

given. To complete design we need t o determine the gain Kdzst. In general, larger gain Kdzst results in more accurate disturbance observation. However, it also affects stability of the closed loop system. Thus, the performance of disturbance observation a well as the s stability of closed loop should be considered in the determination of the gain Kdzst. The following theorem provides a condition for the disturbance observation gain Kdzst.

is considered. Here, n denotes measurement noise and w denotes unknown but bounded disturbance of the system.

4.1 Control System Design For a sampling time of 300 psec, a discrete model described by

a=

1 2.9429e-41 0 9.6216e-1

r=

9.5445e-41 6.3224

Theorem 1 Suppose that p (a - (aL,C) < 1 and p (a - FK,) < 1. Then, there always exist some Kdist such that p ( ( a - @L,C - rKdiStL,C) < 1, i.e. the feedback loop is stable. Moreover, in this case, there Kdist) > 0 and ks (Ts, Kdist) > 0 such that exist b (Ts, ( ( W- U d i s t ( k ) l l < 6 (T~,Kdist) for > ks (T.,Kdist). Proof:
U

Not presented due to space limitation


3

Model

is obtained. We first determine a state estimator gain L, = [ 0.75 0.5 IT and a state feedback gain K , = [ 107.42 0.3 ] such that the sensitivity and complementary sensitivity of the control system are satisfactory. The state estimator has p((a - (aL,C) = 0.25, 0.96, and the state feedback controller has p ((a - FK,) = 0.895, 0.93. For sufficient compensation of the equivalent disturbance we choose a disturbance observer gain Kdzst = [ 18 3.75 1. Then, we have p ((a - @L,c - rKdistLcC)= 0.28, 0.92. As stated in Theorem 1, the closed loop is stable and the disturbance observer is convergent. Figure 5 shows sensitivity and complementary sensitivity of the feedback loop.

4
Model

Udist

(b)

Figure 4: Current state estimator and disturbance observer ((a) and (b) represent the disturbance observer models (4) and (14),respectively.)
The performance of a disturbance observer is highly dependent on the sampling frequency. As the sampling frequency becomes higher, the disturbance observer can reject higher frequency disturbances. In the case of limited sampling frequency, multi-rate implementation of the proposed disturbance observer is effective.

Figure 5: Sensitivity and complementary sensitivity 4.2 Simulation Results Disturbances of 0.2 sin(8nt) and 0.2 sin(8nt) O.lsin(24nt) are injected into the plant input in the simulations of step command tracking and sinusoidal command tracking, respectively. For more realistic simulations, we also introduce a control saturation of Iul 5 umax= 0.7 amp, the quantization effect of ADC and DAC, and measurement noises.

4 Example

A continuous time system described by


x =
y =

[
[

0 0

1
-128.59

] [

z - t 2.1484e4

]
4197

1 Olzfn

Figure 6 shows time response of the plant for a step position command. Comparisons of the model output with the plant output are shown. It is shown that the plant output is very close t o the model output following the step command with disturbance compensation. Figure 7 shows that the disturbance observer effectively compensates the sinusoidal disturbance. Figure 8 shows time response of the plant for a sinusoidal

position command. Comparisons of the model output with the plant output are shown. It is shown that the plant output is very close to the model output and to the sinusoidal command with disturbance compensation. Figure 9 shows very good compensation of the sinusoidal disturbances by the disturbance observer.

1 0

~ 01

02

"

03

"

" 04

05

"

06

'

07

08

09

01

02

03

04

05 06 Tim (58~1

07

06

09

Figure 8: Sinusoidal position command tracking

2 1
I

I
06t

01

02

03

04

OS

08

07

08

09

04-

lime [SRI

Figure 6: Step position command tracking

-08

01

02

03

04

05 lime [%cl

08

07

08

09

Figure 9: Disturbance and disturbance observation in sinusoidal position command tracking


frequency is fast enough. In addition, accurate state estimation was attained even in the existence of modeling uncertainties as well as external disturbances.
-08

01

02

03

04

nMiw i

05

08

07

08

09

Figure 7: Disturbance and disturbance observation in step position command tracking

References
[I] G.F. Franklin, J.D. Powell, and M.L. Workman, Digital Control of Dynamic Systems, Addison Wesley, 1990.

5 Conclusions

This paper has presented a new discrete-time state space disturbance observer not requiring modeling of disturbances. A discrete-time robust tracking control system is proposed, which consists of the state space disturbance observer, a state estimator, and a state feedback controller. Through the results of simulations, it is shown that the proposed method is very effective to improve the tracking accuracy of control systems in the existence of external disturbances. It is observed that the proposed disturbance observer effectively estimate and compensate any disturbances if the sampling
41 98

[2] T. Murakami and K. Ohnishi, Observer-based Motion Control: Application to Robust Control and parameter Identification, Proc. of the IEEE Industrial Electronics Society: Asia-Pacific Workshop on Advances in Motion Control, July 15-16, 1993, pp. 1-6.

[3] K. Ohnishi, A New Servo Method in Mechatronzcs, Trans. of Japanese Society of Electrical Engineers, Vol. 107-D, 1987, pp. 83-86.

[4] T. Umeno and Y. Hori, Robust Speed Control of D C Servomotors using Modern Two Degrees-ofFreedom Controller Design, IEEE Trans. on Industrial Electronics, Vol. 38, No. 5, 1990, pp. 363-368.

You might also like