2009 Fourth International Conference on Innovative Computing, Information and Control

The H ∞ Optimal Control of Linear Systems with Actuators Magnitude and

Rate Saturation

Pang-Chia Chen1,*, Shi-Jaw Chen1, Chih-Hua Hsu2

1,*
Department of Electrical Engineering, Kao Yuan University
No.1821, Chung-Shan Rd., Lu-Chu Hsiang, Kaohsiung County 821,Taiwan, R.O.C.
[email protected]
2
Department of Information Management, Chang Jung Christian University
No. 396 Chang Jung Rd., Sec. 1, Kway Jen Hsiang, Tainan County 711, Taiwan, R.O.C.

Abstract of the integrator present the rate of the control efforts. The
augmented dynamics formulate the systems control efforts as
The proposed controller design algorithm in this paper can additional states. A state feedback control law is expressed in
take into account independently both the magnitude and rate terms of LMIs to achieve optimization of H ∞ performance
constraints on actuator dynamics. The control law is while subject to the specified constraints on the
constructed in terms of linear matrix inequalities. A suitable magnitude and rate of the actuators. The suitable set of
chosen initial state condition, which can represent the desired initial states representing desired ranges of tracking
set of tracking commands, is treated as a parameter for the commands is assumed and addressed in the LMI
controller design algorithm. By the example of satellite attitude algorithm. The feedback of the new state, the original
control, the effectiveness of the proposed algorithm is verified control efforts, in the design scheme forms an inner
through the achievable H ∞ performance and the time loop of the controlled systems and represents the
response simulations with specified constraints on the suitable actuator dynamics by the chosen feedback gain
magnitude and rate of the control torque. such that the desired H ∞ performance can be
optimized and the magnitude the rate constraints can be
satisfied.
1. Introduction

The magnitude and rate saturation on actuator dynamics are 2. H ∞ Optimal Control via LMIs
prevailing in physical control systems and substantially affect
the behavior of the controlled systems. Therefore, the issues of PROBLEM STATEMENT Consider the linear system
the magnitude and rate constraints on the actuator saturation x =Ax +B1 w +B 2 u (1a)
must be taken care well during controller design, otherwise, the with regulated output,
designated performance can be deteriorated and even the z =Cx+D1 w +D2 u (1b)
system stability can possibly be destroyed. The design objective is to achieve a certain level of
Among the works that have been done for systems with
convergence rate β > 0 and a minimized level of
both magnitude and rate constraints, [1] are relied on nonlinear
control techniques. The works in [2] addressed the magnitude H ∞ performance γ from disturbance w to regulated
and rate saturation as an anti-windup problem. A design output z while satisfy the actuator magnitude and rate
approach with low gain controller with respect to the actuator constraints,
magnitude and rate saturation was proposed in [3]. The u i ≤ ui,max , u i ≤ u i,max , (2)
approaches in [4] synthesized the magnitude and rate saturation
compensators using the frequency domain techniques. The for given initial condition x 0 .
works in [5] used the convex computation methods, such as the In order to address the actuator rate saturation, let
linear matrix inequalities (LMIs), to handle the control system the derivative of the control effort denoted as, u :=v .
designs with actuator magnitude and rate saturation. Considering the control effort as augmented state, the
This paper considers the controller design for linear systems linear dynamic systems and the regulated output are
with actuator magnitude and rate constraints by augmented written as,
diagonal integral dynamics to the systems actuator. The inputs

978-0-7695-3873-0/09 $29.00 © 2009 IEEE 561

 ⎛ x ⎞ ⎛ A B 2 ⎞⎛ x ⎞ ⎛ B1 ⎞ ⎛0⎞ controlled system. Let the Lyapunov matrix P satisfy,
⎜  ⎟=⎜ ⎟⎜ ⎟ + ⎜ ⎟ w + ⎜ ⎟ v (3a) x T0Px 0 < 1 , then it can be written as the following LMI
⎝ u ⎠ ⎝ 0 0 ⎠⎝ u ⎠ ⎝ 0 ⎠ ⎝I⎠
⎛x⎞ in terms of the synthesis matrix variable Q = P −1 > 0 ,
z = ( C1 D2 ) ⎜ ⎟ +D1w (3b)
⎝u⎠ ⎛ 1 x T0 ⎞
⎜ ⎟>0 (9)
It is noted that the control effort u has been treated as ⎝ x 0 Q ⎠
part of the state variables, while the derivative of the If the controlled closed-loop system is stabilized, then
control effort v is formulated as the new control effort we have the condition on the state x ,
to be manipulated in the augmented dynamics. With x T Px ≤ x 0TPx 0 < 1 , (10)
the denotation, The magnitude constraints on the control efforts of
⎛ x ⎞  ⎛ A B 2 ⎞  ⎛ B1 ⎞  ⎛ 0 ⎞ 
x = ⎜ ⎟ , A =⎜ ⎟ , B1 = ⎜ ⎟ , B 2 = ⎜ ⎟ , C1 = ( C1 D2 ) (4) the augmented dynamics v , that is, the rate constraints
⎝u ⎠ ⎝0 0 ⎠ ⎝0⎠ ⎝I⎠ on the original control effort u , is known as
and the state feedback control law,
vi = u i ≤ u i,max from (2), which is equivalent to,
⎛x⎞
v = − ( Fx Fu ) ⎜ ⎟ := −Fx
 , (5) x Τ F i Τ F i x ≤ u i.max
2
(11)
⎝u⎠
By (10), the rate constraints in (11) can be established
with Fx ∈ R m× n , Fu ∈ R m× m , where n is the dimension if the following condition is true,
of the state x and m is the dimension of the control 1
effort u . 2
x Τ Q −1LΤi L i Q −1x ≤ x Τ Q −1x , (12)
u i,max
The closed-loop controlled systems can be written as,
⎧⎪ x =A x +B w which is equivalent to,
cl 1
⎨  (6) 1
⎪⎩ z = C1

x + D1w
LΤi L i ≤ Q ,
2
(13)
u i.max
with A =A  − B F .
cl 2 and can be written as the following LMI in the
According to the Bounded-Real Lemma [6], the synthesis variables Q and L ,
H ∞ performance Twz ∞ < γ can be satisfied if the ⎛ Q LΤi ⎞
following matrix inequality holds under the positive ⎜⎜ ⎟⎟ ≥ 0 , (14)
⎝ Li u i,max ⎠
Lyapunov matrix P ,
where Li denoted the i-th row of L .
⎛A P +PA  PB 1 CΤ ⎞ The control effort of the original dynamics is
cl cl 1
⎜ Τ ⎟ reformulated as the augmented states, that is,
⎜ B1 P -γ I D1Τ ⎟ <0 (7)
ui = A i+n x , i = 1, 2,3," , m . (15)
⎜ 
C1 D1 -γ I ⎟⎠
⎝ As stated in (2), the given magnitude constraint on i-th
−1
By denoting P := Q , then multiplying the original control effort, u i ≤ ui,max , is equivalent to,
congruence matrices, diag(Q, I, I ) ,from left and right x Τ A Τi+n A i+n x ≤ u i,max
2
. (16)
side of the block matrix, and let L := FQ , the matrix According to (10), the magnitude constraint in (16) can
inequality can be written as be established by the following,
⎛ QA T +AQ
 + LT B + B L B QCΤ⎞ 1
⎜
2 2 1 1 x T A Τi+n A i+n x ≤ x Τ Q −1x , (17)
 Τ Τ ⎟ 2
u i,max
⎜ B 1 -γ I D1 ⎟ <0 (8)
⎜ 
C1Q D1 -γ I ⎠ ⎟ which can be written as,
⎝
1
which is already in a LMI formulation with matrix QA Τi+n A i+n Q ≤ Q , (18)
variables Q > 0 and L . Once the matrix variables u 2i,max
Q and L are solved, the state feedback law can be and expressed as the following LMI in terms of the
synthesis variable Q ,
computed as F = LQ −1 .
If the matrix inequality can be established for system ⎛ Q QA Τi+n ⎞
 , then the ⎜⎜ 2 ⎟⎟ ≥ 0 . (19)
matrix A replaced by Aβ = A + β I in A ⎝ A i+n Q u i,max ⎠
close-loop controlled system will achieve a PROPOSITION 1 Consider the linear system with
convergence rate β * > 0 . disturbance input w and regulated output z defined in
Assumed the initial condition is given as x 0 , which (1a) and (1b), and given initial condition x 0 . The
may represent desired tracking command for the state-feedback controlled system can achieve a

562
H ∞ performance level Twz ∞
< γ * with convergence considered are x 0,1 = (1, 1, 0, 0, 0)T and
rate β > 0 if there exist synthesis matrix variables
* T
x 0,2 = (−1, − 1, 0, 0, 0) for representing the satellite
Q > 0 and L such that the LMI (8) can be established normalized tracking commands θ1,cmd = θ 2,cmd = ±1 .
for γ * > 0 with system matrix A replaced by Under the actuators magnitude and rate constraints,
 . Moreover, the magnitude and rate
A = A + β I in A u i,max = {10−1 ,1,10} and u i,max = {10−2 ,10−1 ,1,10,102 } ,
β

constraints for control effort can be satisfied for the given respectively, and the designated convergence rate
initial condition x 0 if the LMIs in (9),(14), and (19) can β * = 0.05, the obtained optimized H ∞ performance
be established for given constraints u i,max , u i,max , and the levels by Proposition 1 are depicted in Figure 1.
Two designs are adopted for demonstration in the
augmented initial state condition x 0 .
time response simulations for the proposed algorithm.
Both of them are constructed under magnitude
3. Numerical Example constraint ui,max = 1 . One design with rate constraint
The proposed controller design algorithm is u i,max = 10−2 , the achieved H ∞ performance level is
demonstrated by the satellite attitude control problem. γ * = 69.0450 and the state-feedback control law is,
The satellite model is considered to be two rigid bodies
F1,10 −2 = [ 0.0011 0.0070 0.0359 0.0352 0.2236] . (23a)
*
connected by a flexible link. The rigid bodies consist of
a main body and an instrumentation module Another design with rate constraint u i,max = 1 , the
represented by inertias J1 and J 2 , respectively. The
achieved H ∞ performance level is γ * = 5.8035 and
flexible link is modeled as a spring with torque the state-feedback control law is,
constant k and viscous damping f . The dynamic
F1,1
*
= [ 0.3076 0.1144 1.1068 0.6283 0.5259] (23b)
equation is
The time responses are conducted under normalized
⎧⎪ J1θ1 + f (θ1 − θ2 ) + k (θ1 − θ 2 ) = T + w tracking commands θ1,c (t ) = θ 2,c (t ) = 1 from the origin
⎨ (20)
⎪⎩ J 2θ2 + f (θ2 − θ1 ) + k (θ 2 − θ1 ) = 0 , and the band-limited white noise w(t ) . Figures 2 and 3
where θ1 and θ 2 are the attitude angles for the main show the time responses for the controlled systems
body and the instrumentation module, T is the control with control law F1,1
*
. Figure 2 shows the time
torque, and w is the torque disturbance acting on the responses with the disturbance input imposed. The
main body. The control objective is to minimize the consistent unsteady ripples are found in the time
H ∞ performance level from w to θ 2 and the closed- response of θ 2 (t ) but the overall time response is still
loop system preserves certain level of convergence rate. satisfactory. In both the simulations in Figure 2, the
The derived state-space representation in the form of designated constraints on magnitude and rate saturation
the generalized system (1) is written as are well met.
⎧ x = Ax + B1 w + B 2 u Figure 3 shows the time response that the rate
⎨ (21) constraint is decreased further to the more restricted
⎩ z = Cx ,
u i,max = 10−2 . As shown, the time response is restrained
where x = (θ , θ , θ , θ )T , u = T , and
1 2 1 2
by the over-bounded rate constraint and the system
⎛ 0 0 1 0 ⎞ stability can not be maintained even without the
⎜ ⎟
0 0 0 1 ⎟ disturbance input w(t ) . It is noted that the vertical
A=⎜ , (22a)
⎜ − k / J1 k / J1 − f / J1 f / J1 ⎟ scale has been modified as –10 to 10.
⎜ ⎟ Figure 4 shows the time responses for the
⎝ k / J2 −k / J 2 f / J2 − f / J2 ⎠
controlled systems with control law F1,10
*
−2 when the
⎛ 0 ⎞ ⎛ 0 ⎞
⎜ ⎟ ⎜ ⎟ disturbance input is imposed. Although the ripples can
0 ⎟ , B = ⎜ 0 ⎟ , C = ( 0 1 0 0 ) . (22b)
B1 = ⎜ be observed in the time response of θ 2 (t ) , the rate
⎜ 1 / J1 ⎟ 2 ⎜ 1 / J1 ⎟
⎜ ⎟ ⎜ ⎟ constraint u i,max = 10−2 has not been reached in all the
⎝ 0 ⎠ ⎝ 0 ⎠
time and the overall time response is still acceptable.
The satellite parameter values are J1 = 1 , J 2 = 0.1 ,
k = 0.2450 , f = 0.0188 . The state initial conditions

563
4. Conclusion Transactions on Automatic Control, vol. 41, no. 9, pp.
1256-1270, 1996.
The proposed controller design algorithm in this paper can take [2] L. L. Giovanini, “Model Predictive Control with
Amplitude and rate Saturation,” ISA Transactions, vol.
into account both the magnitude and rate constraints on actuator
42, pp. 227-240, 2003.
dynamics. The control law is constructed based on the initial [3] Z. Lin, “Semi-Global Stabilization of Linear Systems
state conditions. It is demonstrated that the allowable with Position and Rate-Limited Actuators,” Systems &
change rate must be well handled; otherwise, not only Control Letters, vol. 30, pp. 1-11, 1997.
the designated performance level can not be fulfilled, [4] K. Hui, “Design of Rate and Amplitude Saturation
but also even the stability can not be preserved. Compensators Using three Degrees of Freedom
Controllers,” Int. J. Control, vol. 74, pp. 91-101, 2001.
5. Acknowledgements [5] V. S. Chellaboina, W. M. Haddad, and J. H. Oh, “Fixed-
Order Dynamic Compensation for Linear Systems with
Actuator Amplitude and Rate Saturation,” Int. J.
This work was partly sponsored by the National Science Control, vol. 73, pp. 1087-1103, 2000.
Council under Projects NSC 98-2221-E-244-015 and NSC [6] S. Boyd, L. El Ghaoui, E. Feron, and V. Balakrishnan, Linear
97-3114-E-244-001 Matrix Inequalities in System and Control Theory. SIAM,
Philadelphia, 1994.
6. References
[1] A. Teel, “A Nonlinear Small Gain Theorem for the
Analysis of Control Systems with Saturation,” IEEE

Figure 3. Time response with controller designed

for umax = 1 and umax = 1 without
disturbance while working in an rate
Figure 1. H ∞ performance level under various constraint umax = 0.01
actuator magnitude and rate constraints

Figure 4. Time response with controller designed

for umax = 1 and umax = 0.01 under
Figure 2. Time response with controller designed band-limited white noise disturbance
for umax = 1 and umax = 1 under band-
limited white noise disturbance

564