Proceedings of the 2016 5th International Conference on WeBB.

5
Systems and Control, Cadi Ayyad University, Marrakesh,
Morocco, May 25-27, 2016

Improved results on observer-Based Control for uncertain

Continuous-time systems
Abdelaziz Hmamed1, Badreddine El haiek1 , Mohammed Alfidi2 and Fernando Tadeo3

Abstract— This paper is concerned with the design [13]. Available techniques that have been devoted to
of robust observer based controllers for continuous- observer-based stabilization of uncertain linear systems
time linear systems with parameter uncertainties. The can be classified into three categories: Lyapunov-based
proposed method allows one to compute simultane-
ously the observer and controller gains by solving a techniques as in [14], iterative linear matrix inequalities
single bilinear matrix inequalities (BMI), which be- (ILMIs) procedures as proposed in [15], and convex
comes an linear matrix inequalities (LMI). Numerical optimization-based algorithms with equality constraints
examples are given to illustrate the validity and the as recently discussed in [16].
effectiveness of the proposed method. Furthermore, considerable efforts have been assigned to
keywords: Linear matrix inequalities(LMIs)approach, the robust stability and stabilization of linear systems
Observer-based control, Uncertain linear systems. with parameter uncertainties. For recent works, we refer
the readers to [17,18,19,20,21,22].
I. Introduction
Motivated by the above discussions, a new design
The name of observer is referred to as a dynamical methodology is proposed. This leads to a quite simple
system that uses the information of the system inputs LMI condition that is numerically tractable with any
and outputs to reconstruct the unmeasured states of LMI software. It is important to underline that the
the system under consideration. For deterministic and proposed LMI condition is solved without any additional
stochastic linear systems, the theory of observers is well restrictive conditions, namely the a priori choice of the
developed thanks to the pioneer works of Kalman [1] and Lyapunov matrix and the equality constraint [16].
Luenberger [2]. However, for uncertain linear systems, This note is organized as follows. In section II we
there is no generic procedure to solve the observation present the problem formulation and some preliminaries;
issue, which motivates the research in this area for the in section III we provide the main results concerning the
past decades, see for example [3]. When parts of the design of robust observer based controllers via LMI ap-
system dynamic is not completely known and the state proach for continuous−time linear systems with param-
vector is not entirely available for feedback, the available eter uncertainties. Finally, in section IV The illustrative
results are limited to some cases including matched example is proposed to illustrate the effectiveness of the
uncertainties [4] norm-bounded uncertainties [5,6] and proposed results.
uncertainties of dyadic types [7,8]. Notaions:Throughout this note, we use the follow-
In some real models, state feedback control might fail ing notations: Rn denotes the n−dimentional Euclidean
to guarantee the stabilizability when some of the sys- space. ∗ is used for the blocks induced by symmetry. I
tem states are not measurable. This is why a state is the identity matrix with appropriate dimensions. AT
observer is required and included in the feedback control represents the transpose matrix of A. P >0 means that
[9,10,11,12]. Observer-based controllers are often used P is real symmetric and positive definite, and He(M ) is
to stabilize unstable systems or to improve the system defined as He(M )=M+M T .
performances. The observer-based control of uncertain
linear systems is recognized to be a non-convex issue II. System description and problem statement
since the computation of the observer and the controller Consider the continuous-time uncertain linear systems
gains is usually conditioned by the solution of some described by state-space equations:
matrix inequalities which are not numerically tractable
ẋ(t) = (A + ∆A(t))x(t) + Bu(t)
(1)
1 Abdelaziz Hmamed is with Department of Physics Fac- y(t) = (C + ∆C(t))x(t)
ulty of Sciences Dhar El Mehraz B.P. 1796 Fes-Atlas Morocco
hammed− [email protected] where x(t)∈Rn is the state vector, y(t)∈Rp is the
1 Badreddine El Haiek is with the Department of Physics Fac- output measurement, and u(t)∈Rm is the control input
ulty of Sciences Dhar El Mehraz B.P. 1796 Fes-Atlas Morocco vector.The nominal matrices A∈Rn×n , B∈Rm×n , and
[email protected]
2 Mohammed Alfidi is with laboratory of engineering of systems C∈Rp×n are known and constant. First, we consider the
and applications, National school of applied sciences, Fes university following assumptions:
alfidi− [email protected]
3 Fernando Tadeo is with Universidad de Valladolid, Depart.
de Ingenieria de Sistemas y Automatica, 47005 Valladolid, Spain • The pairs (A,B) and (A,C) are assumed to be
[email protected] stabilizable and detectable, respectively.

978-1-4673-8953-2/16/$31.00 ©2016 IEEE 146

• There exist matrices Mi ,Ni ,Fi , i=1,2, of appropriate Proof. Now, let us associate to the dynamics (6) the
dimensions. following Lyapunov function:
T 
The uncertain terms ∆A(t)∈Rn×n and ∆C(t)∈Rp×n
  
x(t) P 0 x(t)
V (x(t), e(t)) = (9)
are unknown matrices that account for time-varying e(t) 0 R e(t)
parameter uncertainties and are assumed to be
structured under the form: the time derivative of V (x(t),e(t)) along the trajectories
of (6), is given
 by T 
∆A(t) = M1 F1 (t)N1 , ∆C(t) = M2 F2 (t)N2 (2)
 
x(t) Π11 Π12 x(t)
V̇ (x(t), e(t))=
e(t) ∗ Π22 e(t)
Remark 1 The parameter uncertainties ∆A(t) and Evidently, V̇(x(t),e(t))<0, ∀x(t)6=0 and e(t)6=0 if the
∆C(t) will represent the impossibility for exact mathe- matrix inequality Π<0. 
matical model for a dynamic system due to the system
The difficulty to have an LMI stems from the off-diagonal
complexity. The uncertainty has been widely used in many
bilinear term P BK. Indeed, there is no congruence
practical systems which can be either exactly modeled or transformation eliminating this bilinearity.
overbounded by the condition: We propose a novel manner to overcome the obstacle of the
coupling P BK without equality constraint or an a priori
FiT (t)Fi (t) ≤ I, i = 1, 2. (3) choice of the matrix P . The following preliminary lemmas
are needed to derive the new results.
The matrices F1 (t) and F2 (t) contain the uncertain pa-
rameters, and constant matrices Mi and Ni , i=1,2, spec-
ify how the uncertain parameter Fi (t) affect the nominal Lemma 2 .[28](Schur Lemma): Let Q1 , Q2 and Q3 be three
matrix of (1). matrices of appropriate dimensions such that Q1 =QT1 and
T −1 T
Q  Q1 <0 and Q1 −Q2 Q3 Q2 <0 if and only if
 3 =Q3 . Then,
The Luenberger-type observer-based controller that we Q1 Q2
< 0.
consider in this paper is under the form: QT2 Q3

˙
x̂(t) = Ax̂(t) + Bu(t) + L(y(t) − C x̂(t)) (4) Lemma 3 .[29] Let X, Y , and ∆ be real matrices with
n×p appropriate dimensions and ∆T ∆ ≤ I the following
where L∈R is the observer gain to be determined and inequality holds:
x̂(t)∈Rn×n is the estimate of x(t). Let e(t) = x(t) − x̂(t)
be the estimation error. Then, under the feedback X∆Y + Y T ∆T X T ≤ 1 XX T + Y T Y (10)

u(t) = −K x̂(t) (5) Lemma 4 .[25]: For matrices T, Q, U, and W with appro-
priate dimensions and scalar ξ. The inequality
m×n
where K∈R is the controller gain, the closed loop
dynamics is described by T + W T QT + QW < 0 (11)
    
ẋ(t) (A − BK + ∆A(t)) BK x(t) is fulfilled if the following condition holds:
=
ė(t) ∆A(t) − L∆C(t) A − LC e(t)
(6)
 
T ∗
<0 (12)
ξQT + U W −ξU − ξU T
The objective consists in finding the parameters L and
K so that the closed-loop system (6) is asymptotically Proof. Pre-and post-multiplying (12) by the full row rank
matrix I ξ1 W T

stable. and its transpose, respectively, the
inequality (11) can be obtained. 

Lemma 1 System (1) is asymptotically stabilizable by

(4)−(5) if there exist two positive definite matrices P >0, III. Main result
R>0, K, L̂ and L=R−1 L̂ satisfying the following matrix In this section, we shall present a numerically efficient
inequality: technique to find the observer-based controller gains in such
  a way to stabilize the system in closed loop, i.e., to ensure
Π11 Π12 that (6) is asymptotically stable. We summarize the resulting
Π= <0 (7) LMI design conditions in the following theorem.
∗ Π22
where
Π11 = (A + ∆A)T P + P (A + ∆A) − P BK − K T B T P Theorem 1 System (1) is asymptotically stabilizable by
(4)−(5) if there exist a known scalar ξ, two positive definite
Π12 = P BK + ∆AT R − ∆C T L̂T matrices P ∈Rn×n , R∈Rn×n and U ∈Rm×m , N ∈Rm×n ,
Π22 = AT R + RA − L̂C − C T L̂T L̂∈Rn×p satisfying the following linear matrix inequality:
(8)

978-1-4673-8953-2/16/$31.00 ©2016 IEEE 147

 He(P (A + ∆A) − BN ) ∗ ∗
" #
Proof. We rewrite the matrix Ω as a sum of two matrices,
Ω = N T B T + R∆A − L̂∆C He(RA − L̂C) ∗ <0 one contains the uncertainties and another one without the
ξ(P B − BU )T − N N Υ uncertainties, that is:
where
Υ = −ξU − ξU T (13) Ω = Ω1 + Ω∆
where "
Proof. Suppose that inequality Ω holds. The feasible solution He(P A − BN ) ∗ ∗
#
of this inequality satisfies −ξU − ξU T < 0 , Ω1 = N T BT He(RA − L̂C) ∗
which implies that the matrix U is  nonsingular. Applying ξ(P B − BU )T − N N Υ
−1
Lemma
 4 with W =U  −N N , where Υ define in (13).
(P B − BU )
Q= , and ∆AT P + P ∆A ∗ ∗
" #
0

He(P (A + ∆A) − BN ) ∗
 Ω∆ = R∆A − L̂∆C 0 ∗
T= 0 0 0
N T B T + R∆A − L̂∆C He(RA − L̂C)
By developing ∆A(t) and ∆C(t) we obtain:
the inequality in Ω leads to
#T #T
N1T N1T
" # " " # "
  P M1 P M1
He(P (A + ∆A) − BN ) ∗
Ω∆ = 0 F1 (t) 0 + 0 F1T (t) 0
N T B T + R∆A − L̂∆C He(RA − L̂C) 0 0 0 0
#T " #T
N1T N1T
" # " # "
    0 0
(P B − BU ) −1
 
+ RM1 F1 (t) 0 + 0 F1T (t) RM1
+He U −N N
0 0 0 0 0
" T #T " T # #T
By defining K = U −1 N 0 0
" # "
, we obtain N2 N2
T
  + −L̂M2 F2 (t) 0 + 0 F2 (t) −L̂M2
He(P (A + ∆A) − BN ) ∗ 0 0 0 0
(14)
N T B T + R∆A − L̂∆C He(RA − L̂C) Consequently, using the lemma 3 and (3), we have
 
−P BK + BN P BK − BN Ω ≤ Ω1 + −1 T T −1 T T
+He 1 X1 X1 + 1 X2 X2 + 2 X3 X3 + 2 X2 X2
0 0 +−1 X X T
+  X X T
3 4 4 3 5 5

From (14), it implies that the inequality (7) in lemma 1 (16)

holds. 
for any 1 >0,2 > 0 and 3 >0, where
 
X1T = M1T P 0 0 ,
Remark 2 By using the matrix inequality technique, theorem
1 introduces a new criterion for continuous-time systems. In X2T =

N1 0 0

,
the new criterion, the appearance of crossing terms P BK
appearing in (7) has been avoided, it enables us to give LM I X3T =

0 M1T R 0

,
conditions for designing observer based control.
 
Notice that the theorem 1 contains the uncertainties, so X4T = 0 M2T L̂T 0 ,
now, we linearize the last theorem 1 with respect to the  
uncertainties, the following theorem hold. X5T = N2 0 0 .

Using the Schur complement lemma 2 in the right hand

side of (16), we obtain that Ω<0 holds if the LMI (15) is
Theorem 2 System (1) is asymptotically stabilizable by satisfied.
(4)−(5) if there exist a known scalar ξ, and positive scalars
1 >0, 2 >0 and 3 >0, two positive definite matrices P ∈Rn×n ,
R∈Rn×n and U ∈Rm×m , N ∈Rm×n , L̂∈Rn×p satisfying the
following linear matrix inequality:
Remark 3 :The author in [16] introduced a new matrix P̂
 Ψ11 ∗ ∗ ∗ ∗ ∗ 
satisfying the additional strong equality condition P B=B P̂
 N T BT Ψ22 ∗ ∗ ∗ ∗  to linearize the BMI (7). However, it is clear that our
 Ψ31 N Ψ33 ∗ ∗ ∗ 
<0 theorem is obtained without any equality constraint and
M1T R

 0 0 −2 I ∗ ∗  enjoy the property to be numerically tractable by any convex
0 M2T L̂T 0 0 −3 I ∗
 
optimization software.
M1T P 0 0 0 0 −1 I
(15)

Hence, the stabilizing observer−based control gains are given Remark 4 : Consider the LMI (15) in theorem 2 and
by K=U −1 N and L=R−1 L̂. pre−multiplying by F and post−multiplying by F T , with
where
I 0 0 0 0 0
 
Ψ11 = AT P + P A − BN − N T B T + (1 + 2 )N1T N1 + 3 N2T N2
 0 I 0 0 0 0 
Ψ22 = AT R + RA − L̂C − C T L̂T
Ψ33 = −ξU − ξU T  0
F = 0 0 0 0 I 
0 0 0 I 0 0

Ψ31 = ξ(P B − BU )T − N 0 0 0 0 I 0

978-1-4673-8953-2/16/$31.00 ©2016 IEEE 148

one gets the [LMI (10) theorem 2 in [16] with ρ=0, U =P̂ and 3
x 10
5

x1
N =K̂]. x2
x3
2.5

2
Remark 5 : The conditions in Theorem 2 is LMI for a given
scalar ξ [23]. In order to avoid any optimization technique 1.5

to search for such an ξ, a logarithmically spaced search has

states
been proposed in [24]. They use a finite set of LMI constraint 1

problems with: ξ ∈ E = [10−6 , 10−5 , ..., 106 ] . Thus, ξ is fixed

and it is not a decision variable. This logarithmically spaced 0.5

search has been tested intensively in [26,27]. 0

−0.5
IV. Numerical application. 0 1 2 3 4 5
time
6 7 8 9 10

In this section, a numerical example is given to illustrate

the effectiveness of the proposed method. Figure 1:Trajectories of the open−loop system.

A. Example 6
x1
5 x2
Consider the following system in [16]: 4
x3

" # " #
1 1 1 1 0 3

A= 0 −2 1 , B= 0 1 2

1 −2 0 0

States
−5 1
   
C= 1 0 1 ∆C = 0 d(t) 0 0

" # −1
0 0 a(t)
−2
∆A = 0 b(t) 0
c(t) 0 0 −3

−4
where a(t) ≤ α, b(t) ≤ β, c(t) ≤ γ, d(t) ≤ δ, 0 1 2 3 4 5
time s
the uncertainties can be rewritten under the form (2) with
" #
a(t)/α 0 0 Figure 2:State responses of the closed-loop system.
F1 = 0 b(t)/β 0 ,
" 0 #0 c(t)/γ
0 0 α 8
e1
N1 = 0 β 0 6
e2
e3
γ 0 0
4
 
M1 =M2 =I, F2 (t)=d(t)/δ N2 = 0 δ 0
2

The simulation results of the observer-based controller

Errors

0
are given in table 1. In fact for the α = β = γ = δ the LMIs
theorem 2 remain solvable until the maximum allowable −2

αmax = 1.6 . In other word, it is shown that the approaches −4

proposed in [16,22] are not able to achieve the maximum
allowable value αmax =1.6 obtained via the method proposed −6

in this work. Which prove that the developed LMIs are less −8
0 1 2 3 4 5
conservative than those presented in [16,22]. time s
Using LMI Toolbox Software, there is a set of feasible
solutions of the LMIs given in Theorem 2 with " ξ=0.1, we #
get Figure 3:State error of the closed-loop system.
  247.8902
18.1474 4.0936 3.6397
K= , L= 55.2508 table 1:
2.8939 6.8989 0.5175
13.0312 Calculated maximum allowable αmax for different methods.
with
Method Th1 [16] Th1 [22] Th2 [16] Th2 (ξ=0.1)
" # " #
2 5
x(0) = 3 ; x̂(0) = −1 αmax 1 1.3 1.35 1.6
−4 3
a(t) = cos(3t) b(t) = 1 − 0.3sin(2t) V. Conclusion
In this paper, the problem of designing observer−based
c(t) = 1.3sin(4t) d(t) = −1.2sin(3t) controllers for continuous−time linear systems has been stud-
ies. Sufficient condition for designing robust observer−based
Figure 1, plots the uncontrolled states trajectories of controllers have been given of solutions to a set of linear
the open−loop system. The state responses and state errors matrix inequalities. In particular , it has been proved that the
are shown in figure 2 and figure 3 respectively from which new proposed conditions are more relaxed than the existing
we see that the closed−loop system is asymptotically stable. ones with equality constraints. Finally, a numerical example
Thus, simulation results demonstrate the effectiveness of the has been given to illustrate the effectiveness of the proposed
proposed observer-based controller. method in this paper.

978-1-4673-8953-2/16/$31.00 ©2016 IEEE 149

 References tems with parameter uncertainties”. Automatica 49 (2013)
3700−3704.
[1] R.E. Kalman, A new approach to linear filtering and predic- [23] Mozelli, L. A., Palhares, R. M., Avellar, G. S. C. (2009). A
tion problems, Transactions of the ASME, Journal of Basic systematic approach to improve multiple Lyapunov function
Engineering 82 (D) (1960)35−45. stability and stabilization conditions for fuzzy systems. Infor-
[2] D.J. Luenberger, An introduction to observers, IEEE Trans- mation Sciences, 179, 1149−1162.
actions on Automatic Control AC−16 (6) (1971) 596−602. [24] Oliveira, R.C.L.F., de Oliveira, M.C., Peres, P.L.D. (2011).
[3] D.-W. Gu, F.W. Poon, A robust state observer scheme, IEEE Robust state feedback LMI methods for continuous−time
Transactions on Automatic Control 46 (12) (2001) 1958−1963. linear systems: discussions, extensions and numerical com-
[4] F. Jabbari, W.E. Schmitendorf, Effects of using observers on parisons. In 2011 IEEE int. symposium on computer−aided
stabilization of uncertain linear systems, IEEE Transactions control system design (pp. 1038−1043)
on Automatic Control 38 (2) (1993) 266−271. [25] Xiao−Heng Chang,Liang Zhang,Ju H.Park.0 Robust static
[5] I.R. Petersen, C.V. Hollot, High-gain observers applied to output feedback H∞ control for uncertain fuzzy systems0 fuzzy
problems of stabilization of uncertain linear systems, dis- set and systems,http://dx.doi.org/10.1016/j.fss.2014.10.023
turbance attenuation and H∞ optimization, International [26] Jaadari, A., Guerra, T.M., Sala, A., Bernal, M., Guelton, K.
Journal of Adaptive Control and Signal Processing 2 (1988) (2012). New controllers and new designs for continuous−time
347−369. Takagi−Sugeno models. In IEEE int. conf. on fuzzy systems
[6] Y.H. Chen, J.S. Chen, Combined controller-observer design for (pp. 1−7). Brisbane, Australia.
uncertain systems using necessary and sufficient conditions, in: [27] Marquez, R., Guerra, T.M., Kruszewski, A., Bernal, M.
Proceedings of the 29th CDC, 1990, pp. 3452−3454. (2014). Non−quadratic stabilization of second order continu-
[7] I.R. Petersen, A Riccati equation approach to the design of ous Takagi−Sugeno descriptor systems via line−integral Lya-
stabilizing controllers and observers for a class of uncertain punov function. In IEEE int. conf. on fuzzy systems (pp.
systems, IEEE Transactions on Automatic Control 30 (9) 2451−2456). Beijing, China.
(1985) 904−907. [28] Boyd S, El Ghaoui L, Feron E et al (1994) Linear matrix
[8] B.R. Barmish, G. Leitmann, On ultimate boundness control inequalities in system and control theory, studies in applied
of uncertain systems in the absence of matching assumptions, mathematics. SIAM, Philadelphia.
IEEE Transactions on Automatic Control AC−27 (1) (1982) [29] Petersen IR (1987) A stabilization algorithm for a class of
153−158. uncertain linear system. Syst Control Lett 8:351−357.
[9] Atassi, A., Khalil, H. (2000). Separation results for the stabi- [30] Kheloufi, H., Zemouche, A., Bedouhene, F., Boutayeb, M.
lization of nonlinear systems using different high-gain observer (2013). On LMI conditions to design observer-based con-
designs. Systems and Control Letters, 39(3), 183−191. trollers for linear systems with parameter uncertainties. Au-
[10] Hendricks, E., Luther, J. (2001). Model and observer based tomatica, 49, 3700−3704.
control of internal combustion engines. In Proceedings of the
international workshop on modeling,emissions and control in
automotive engines (MECA) (pp. 9−21).Fisciano,Italy.
[11] Karafyllis, I., Kravaris, C. (2005). Robust output feedback sta-
bilization and nonlinear observer design. Systems and Control
Letters, 54(10), 925−938.
[12] Pagilla, P., King, E., Dreinhoefer, L., Garimella, S. (2000).
Robust observer-based control of an aluminum strip process-
ing line. IEEE Transactions on Industry Applications, 36(3),
865−870.
[13] L. Xie, Output feedback h1 control of systems with parameter
uncertainty, International Journal of Control 63 (4) (1996)
741−750.
[14] S.K. Yang, Observer−based robust controller design for a
linear system with time−varying perturbations, Journal of
Mathematical Analysis and Applications 231 (1997) 641−642.
[15] H.H. Choi, M.J. Chung, Observer−based H∞ controller design
for state delayed linear systems, Automatica 32 (7) (1996)
1073−1075.
[16] C.−H. Lien, Robust observer−based control of systems with
state perturbations via LMI approach, IEEE Transactions on
Automatic Control 49 (8) (2004) 1365−1370.
[17] Arcak, M., Kokotovic, P. (2001). Observer-based control of
systems with slope-restricted nonlinearities. IEEE Transac-
tions on Automatic Control, 46(7), 1146−1150.
[18] Gao, H., Wang, C., Zhao, L. (2003). Comments on: 0 An LMI-
based approach for robust stabilization of uncertain stochastic
systems with time-varying delays0 . IEEE Transactions on Au-
tomatic Control, 48(11), 2073−2074.
[19] Heemels, W., Daafouz, J., Millerioux, G. (2010). Observer-
based control of discrete-time LPV systems with uncertain
parameters. IEEE Transactions on Automatic Control, 55(9),
2130−2135.
[20] Ibrir, S. (2008). Design of static and dynamic output feed-
back controllers through Euler approximate models: uncertain
systems with norm-bounded uncertainties. IMA Journal of
Mathematical Control and Information, 25(3), 281−296.
[21] Kheloufi, H., Zemouche, A., Bedouhene, F., Boutayeb, M.
(2013). On LMI conditions to design observer-based con-
trollers for linear systems with parameter uncertainties. Au-
tomatica, 49, 3700−3704.
[22] Shenquan Wanga, Yulian Jiang.Comment on ”On LMI con-
ditions to design observer-based controllers for linear sys-

978-1-4673-8953-2/16/$31.00 ©2016 IEEE 150