Proceedings of the 18th World Congress

The International Federation of Automatic Control

Milano (Italy) August 28 - September 2, 2011

Nonlinear Unknown Input Observer Design

for Nonlinear Systems: A New Method
T. Sedighi ∗ A. J. Koshkouei ∗∗ K. J. Burnham ∗∗∗
∗
Control Theory and Applications Centre, Coventry University, Priory
Street, Coventry CV1 5FB, UK (e-mail: [email protected]).
∗∗
(e-mail: [email protected])
∗∗∗
(e-mail: [email protected])

Abstract: This paper provides new developments in the design of observers for nonlinear
systems with unknown inputs. The proposed methods guarantee the error system stability and
yield many additional degrees of freedom available to the designer. The algorithms for designing a
nonlinear observer for nonlinear systems with unknown inputs is derived in detail. The proposed
observers may be used for fault detection and isolation. A nonlinear mass-spring-damper model
is given in order to highlight the efficiency of the proposed method.

Keywords: Nonlinear control systems; Nonlinear observer; Unknown input observer; Error
stability analysis; Mass-spring-damper system;.

1. INTRODUCTION algorithm allows the user to recover the state and the
unknown inputs in finite time.
Observer design is an important problem that has various Under less restrictive conditions several approaches for
applications such as output feedback control, system mon- designing reduced order observers have been proposed
itoring, process identification and fault detection. by Guan and Saif (1991), Hou and Müller (1992), and
The observer was first proposed and developed by Lu- Mehta et al. (2010) and for full order observers many
enberger (1971) in the early 1960s. After the pioneer- methods have been presented including Hou et al. (1999)
ing work of Luenberger on estimation problems, various and Darouach (2009). Achieving less restrictive existence
types of estimators have been developed for both linear conditions and more direct design procedures has been a
and nonlinear systems. An observer for nonlinear sys- challenge in this area.
tems whose nonlinearity is globally Lipschitz has been Since the 1990s, many attempts have been made to extend
developed by many researchers including Abbaszadeh and the existing UIO design from linear systems to nonlinear
Marquez (2010). The full order Luenberger-like observer systems. UIOs for bilinear systems were designed by many
for nonlinear systems whose nonlinearity satisfy the well researchers including Zasadzinski et al. (1998), and Hamidi
known Lipschitz condition has been proposed in recent et al. (2008). UIOs for more general nonlinear systems were
years, see for example Raghavan and Hedrick (1994). also proposed in Seliger and Frank (1991), Yang and Saif
It is of importance to design observers for multivariable (1996), Ha et al. (2003), Barbot et al. (2007) and Barbot
linear or nonlinear systems partially driven by unknown et al. (2009).
inputs. In many control systems, the measurement of all More recently, observer architectures utilising the concept
the input signals in the system is impossible. In some cases, of sliding mode control/observer for uncertain systems, see
the uncertainties of certain parameters of the system can for example, Zak (2003), Koshkouei and Zinober (2004),
be modelled as unknown inputs for designing a robust and Barbot and Floquet (2009). Also a direct extension of
observer. For that type of system, an observer, which has the linear results to the nonlinear case was referred to as
the capability of estimating the states in the presence of nonlinear unknown input observer (NUIO) and considered
unknown inputs, is required in order to design an appro- systems with nonlinearities that are functions of inputs
priate control for the system. Such a problem arises in and outputs.
systems subject to disturbance or with inaccessible inputs An UIO for nonlinear systems using H∞ approach has
and it may appear in many applications such as fault been designed by Pertew et al. (2005) and a nonlinear
detection and isolation or parameter identification, see observer for descriptive type of nonlinear systems with
Mondal et al. (2009). unknown inputs based on linear matrix inequality (LMI)
Design of observers for linear systems subject to unknown approach has been presented by Koeing (2006), Chen
inputs has attracted considerable attention in the last and Saif (2006a), and Chen and Saif (2006b). This pa-
three decades. Different approaches have been considered per present an alternative method to design the NUIOs
to design unknown input observers (UIO) for linear sys- for nonlinear systems under certain conditions. Section
tems following the conventional Luenberger design pro- 2 presents a mathematical description of the nonlinear
cedure or using sliding mode control theory. For both system. Design of the NUIO for the nonlinear system
approaches the existence conditions are exactly the same. along with theorems and error dynamics and stability
In Floquet and Barbot (2005) the proposed observation analysis are addressed in Sections 3 and 4, respectively. A

978-3-902661-93-7/11/$20.00 © 2011 IFAC 11067 10.3182/20110828-6-IT-1002.02509

18th IFAC World Congress (IFAC'11)
Milano (Italy) August 28 - September 2, 2011

numerical example is provided in Sections 5. Conclusions defined as follows:

are presented in Section 6.
ex (t) = x(t) − x̂(t) = x(t) − z(t) + Ey(t) (3)
2. SYSTEM DESCRIPTION
By substituting the system output defined in (1) into
Consider the following nonlinear system the error equation (3), the time-derivative of the error
equation (3) will have the following form:
ẋ(t) = Ax(t) + Bu(t) + Dµ(t) + Sg(x, u, t) + Ka fa (t)
y(t) = Cx(t) + Ks fs (t) (1) e˙x (t) = ẋ(t) − ż(t) + EC ẋ(t) (4)
n m p
where x ∈ R , u ∈ R and y ∈ R represent the Then substituting (1) and (2) into (4), where H = In +EC
state, input and output vectors respectively. A ∈ Rn×n , and H = H ∗ , yields
B ∈ Rn×m , C ∈ Rp×n , S ∈ Rn×s , Ka ∈ Rn×r and
Ks ∈ Rp×s are known matrices, fa and fs are the actuator
and sensor faults. D ∈ Rn×q , is referred to as the known e˙x (t) = N ex (t) + (HA − N H − LC)x(t) + (HB − G)u(t)
input disturbances distribution matrix and µ(t) ∈ Rq is an +HS(g(t, u, x) − g(x̂, u, t)) + HD(µ(t) − µ̂(t))
unknown bounded vector. The function µ(t) describes the
+HKa (fa (t) − fˆa (t)) (5)
disturbances and any kind of modelling uncertainties such
as noise, time-varying terms, and parameter variations. If the following conditions hold,
In system (1), the function g(x, u, t) ∈ Rs represents the HD = E0 (6)
known nonlinearity function.
HB − G = 0 (7)
Prior to nonlinear observer design, the following assump-
tion is made: HA − N H − LC = 0 (8)

where E0 ∈ Rn×m has a small norm, then the error

Assumption 1: equation (5) is presented as follows,
• There is no fault in the output, fs (t) = 0.
• CD = 0 is of full rank. e˙x (t) = N ex (t) + Ag HSeg + Ad HDeµ (t) + Aa HKa ea (t)
• D, S, and Ka are full column rank matrices. (9)
• The output y(t) and its derivative ẏ(t) are available.
where eg (x, u, t) = g(x, u, t) − g(x̂, u, t), eµ (t) = µ(t) − µ̂(t)
3. NONLINEAR UNKNOWN INPUT OBSERVER and ea (t) = fa (t) − fˆa (t) represent the nonlinearity, dis-
DESIGN (NUIO): turbance and actuator fault errors respectively. Matrices
Ag , Ad , and Aa are the design matrices.
To obtain matrices N , E, H, G and the observer gain L
In this section the design procedure of an observer is
the following steps should be followed:
considered. The conditions which eliminate some terms
from the error system are given. These conditions are • First substitute H = In + EC into (6), the matrix E
sufficient conditions for designing the observer. may be presented as follows:
E = (E0 − D)(CD)+ + Γ(I − (CD)(CD)+ ) (10)
3.1 Observer design +
where (CD) is the pseudo-inverse of (CD), Γ is an
An unknown input observer (UIO) for the system (1) is arbitrary matrix.
designed such that its state estimation error vector e(t),
approaches zero asymptotically, regardless of the presence • Then by substituting E into H = In − EC, matrix H
of the unknown input term in the system. will be obtained.

• After that substitute H into (7) to obtain matrix G.

3.2 NUIO Design
• Assume that (HA, C) is an observable pair and P0 is
An full-order NUIO could be presented in the following the symmetric positive definite (s.p.d.) solution of the
form: following algebraic Riccati equation, then the gain K
is selected such that the matrix N = HA − KC is
ż(t) = N z(t) + Ly(t) + Gu(t) + H ∗ Sg(x̂, u, t) + H ∗ Dµ̂(t) stable.
+H ∗ Ka fˆa (t) (HA)T P0 + P0 (HA) − P0 C T R−1 CP0 = −Q0 (11)
x̂(t) = z(t) − Ey(t) (2) where Q0 ∈ Rn×n and R ∈ Rp×p are arbitrary s.p.d.
matrices. By selecting K = P0 C T R−1 the matrix N
where z ∈ Rn is the state observer, g(x̂, u, t), µ̂(t) and fˆa (t) will be an stable matrix.
are the estimate of g(x, u, t), µ(t) and fa (t) respectively.
Matrices N ∈ Rn×n , L ∈ Rn×p , G ∈ Rn×m , H ∗ ∈ Rn×n • Finally by substituting N , K and (6) into (8) the
with E ∈ Rn×p , and x̂ is an estimate of x. It is desired to observer gain L will be obtained:
design the observer such that x̂ eventually tends to x.
The error equation for system (1) and observer (2) is L = HAE + K(Ip − CE). (12)

11068
18th IFAC World Congress (IFAC'11)
Milano (Italy) August 28 - September 2, 2011

4. STABILITY ANALYSIS OF THE ERROR SYSTEM 4.2 The actuator fault error ea (t) stability analysis

In this Section the behaviour of the error system (9) is The actuator fault fa (t), could also be estimated from the
considered. The desire is to obtain the stability of eµ (t), available known signals as follows:
ea (t) and eg (x, u, t), to show that each of these errors are
stable, then the stability of error (9) is guaranteed. fˆa (t) = F11 y(t) + F12 ẏ(t) + F13 x̂(t) + F14 x̂(t)
˙
+F15 Sg(x, u, t) + F16 u(t) (19)
4.1 The disturbance error eµ (t) stability analysis where matrices F11 ∈ Rr×p , F12 ∈ Rr×p , F13 ∈ Rr×n ,
F14 ∈ Rr×n , F15 ∈ Rr×n and F16 ∈ Rr×m need to be
The unknown input (disturbance) µ(t), could be estimated designed. From (1), fa (t) is giving by:
from the available known signals as follows:
fa (t) = Ka+ ẋ(t) − Ka+ Ax(t) − Ka+ Bu(t) − Ka+ Sg(x, u, t)
˙
µ̂(t) = M11 y(t) + M12 ẏ(t) + M13 x̂(t) + M14 x̂(t) − Ka+ Dµ(t) (20)
+M15 Sg(x, u, t) + M16 u(t) (13) Using (19) and (20), ea (t), the actuator fault error is
obtained:
where matrices M11 ∈ Rq×p , M12 ∈ Rq×p , M13 ∈ Rq×n ,
M14 ∈ Rq×n , M15 ∈ Rq×n and M16 ∈ Rq×m will be
designed to obtain the observer µ̂(t). ea (t) = fa (t) − fˆa (t)
µ(t) is obtained from equation (1) as follows: = −(Ka+ A + F11 C)x(t) + (Ka+ − F12 C)ẋ(t)
− (Ka+ B + F16 )u(t) − (Ka+ + F15 )Sg(x, u, t)
µ(t) = D+ ẋ(t) − D+ Ax(t) − D+ Bu(t) − D+ Sg(x, u, t) ˙
− F13 x̂(t) − F14 x̂(t) − K + Dµ(t)
a (21)
− D+ Ka fa (t) (14) From ex (t) = x(t) − x̂(t) substitute x̂(t) into (21), yields:
Using (13) and (14), the disturbance error eµ (t) can be
rewritten as: ea (t) = −(Ka+ A + F11 C − F13 )x(t) + (Ka+ − F12 C − F14 )
ẋ(t) − (Ka+ B + F16 )u(t) − (Ka+ + F15 )Sg(x, u, t)
eµ (t) = µ(t) − µ̂(t)
+ F13 e(t) + F14 ė(t) − Ka+ Dµ(t) (22)
+ +
= −(D A + M11 C)x(t) + (D − M12 C)ẋ(t) Thus if the following conditions hold:
− (D+ B + M16 )u(t) − (D+ + M15 )Sg(x, u, t)
˙
− M13 x̂(t) − M14 x̂(t) − D+ Ka fa (t) F14 = Ka+ − F12 C

(15) F13 = −Ka+ A − F11 C

F16 = −Ka+ B
Since ex (t) = x(t) − x̂(t) then by substituting x̂(t) into
(15), the following is obtained: F15 = −Ka+
0 = Ka+ D
+ +
eµ (t) = −(D A + M11 C − M13 )x(t) + (D − M12 C −
(23)
M14 )ẋ(t) − (D+ B + M16 )u(t) − (D+ + M15 ) then the fault error (22) is presented as:
+Sg(x, u, t)M13 e(t) + M14 ė(t) − D+ Ka fa (t) ea (t) = (−Ka+ A − F11 C)ex (t) + (Ka+ − F12 C)e˙x (t) (24)
(16)
Equation (24) shows that the actuator fault error ea (t) is
Therefore, if the following conditions hold: a function of ex (t), and tends to zero if ex (t) goes to zero.

M14 = D+ − M12 C 4.3 The nonlinearity error eg (t) stability analysis

M13 = −D+ A − M11 C The nonlinearity g(x, u, t), could be estimated from the
+
M16 = −D B available known signals as follows:
M15 = −D+ ˙
g(x̂, u, t) = G11 y(t) + G12 ẏ(t) + G13 x̂(t) + G14 x̂(t)
+
D Ka = 0
+G15 u(t) + G16 fˆa (t) (25)
(17) where matrices G11 ∈ Rs×p , G12 ∈ Rs×p , G13 ∈ Rs×n ,
then (16) can be written as follows: G14 ∈ Rs×n , G15 ∈ Rs×m and G16 ∈ Rs×q need to be
designed. From equation (1), g(x, u, t) is obtained as:
eµ (t) = (−D+ A − M11 C)ex (t) + (D+ − M12 C)e˙x (t) (18)
g(x, u, t) = S + ẋ(t) − S + Ax(t) − S + Bu(t) − S + Ka fa (t)
Equation (18) shows that the disturbance error eµ (t) is a
function of ex (t), and if ex (t) is stable then eµ (t) tends to − S + Dµ(t) (26)
zero. Using (25) and (26), eg (x, u, t), the nonlinearity error is
giving by:

11069
18th IFAC World Congress (IFAC'11)
Milano (Italy) August 28 - September 2, 2011

eg (x, u, t) = g(x, u, t) − g(x̂, u, t) freedom E0 , Mij , Fij and Gij and nonuniquness of Ag , Ad
+ + and Aa also provides extra design degrees of freedom.
= −(S A + G11 C)x(t) + (S − G12 C)ẋ(t)
Since error (32) is a function of ex (t), the error is eventu-
− (S + B + G15 )u(t) − (S + Ka + G16 )fa (t) ally stable and tends to zero.
˙
− G13 x̂(t) − G14 x̂(t) − S + Dµ(t) (27) 5. EXAMPLE SYSTEM
Substituting ex (t) = x(t) − x̂(t) into (27), eg (x, u, t)
becomes:

eg (x, u, t) = −(S + A + G11 C − G13 )x(t) + (S + − G12 C

−G14 )ẋ(t) − (S + B + G15 )u(t) − (S + Ka
+G16 )fa (t) + G13 e(t) + G14 ė(t) + G16 ea (t)
−S + Dµ(t) (28)
Then if the following conditions hold:

G13 = −S + A − G11 C
G14 = S + − G12 C
Fig. 1. The mass-spring-damper system.
G15 = −S + B
G16 = −S + Ka Consider the mass-spring-damper (M-S-D) system in Fig-
ure 1 where two masses, springs and dampers are con-
S+D = 0 nected together serially (Roch-Cozalt et al. (2005)). x1 and
(29) x2 are the position and velocity of the first mass and x3
then the nonlinear error estimation eg (x, u, t) is giving by: and x4 are the position and velocity of the second mass,
respectively. Anl is a nonsingular damping device whose
damping force is FAnl = Cnl sign(x2 ) ln(1+ | x2 |), where
eg (x, u, t) = (−S + A − G11 C)ex (t) + (S + − G12 C)e˙x (t)
Cnl ≥ 0. On the other hand, an arbitrary and unknown
+ G16 ea (t) (30) force µ(t) is applied on the second mass. u1 and u2 are the
Now by substituting ea into (30), the error eg is giving by: known input forces applied to masses 1 and 2, respectively.
The state equations of the system are presented as
   
eg (x, u, t) = (−S + A − G11 C − G16 Ka+ A − G16 F11 C)ex (t) 0 1 0 0 0
 k1 + k2 b1 + b2 k2 b2   1 
+ (S + − G12 C + G16 Ka+ − G16 F12 C)e˙x (t) (31) − −   
ẋ = 
 m1 m1 m1 m1  x +  m1  u
Equation (31) shows that the nonlinearity error eg (x, u, t)  0 0 0 1   0 

is a function of ex (t), so it tends to zero if ex (t) goes to  k2 b2 k2 b2   1 
zero. − −
m2 m2 m2 m2 m
Substituting (18), (24) and (31) into equation (9) yields:    2
0 0
0
 
1   0 
−  0.02 

M̄ ėx (t) = Ḡex (t) (32) +  m1  g(x2 ) + 

fa (t) +  0  µ(t).(36)
 
 0  0 
 1 
with 0
0 m2
M̄ = In − Ad HD(D+ − M12 C) − Aa HKa (Ka+ − F12 The actuator fault fa (t) is any unwanted change in the
+
C) − Ag HS(−S A − G11 C − G16 Ka+ A − G16 length of the springs. The outputs of the system are
y1 = x3 and y2 = x4 . Using the following values for
F11 C) (33) parameters: m1 = 5kg, m2 = 1kg, k1 = 30N/m, k2 =
where In is an identity matrix of size n, and, 10N/m, b1 = 4N s/m, b2 = 2N s/m, Cnl = 5N , and
µ(t) = 0.04 sin(t) + 2N , will give:
Ḡ = N + Aa HKa (−Ka+ A − F11 C) + Ad HD(−D+
0 1.0 0 0 0 0
   
−M11 C) + Ag HS(S + − G12 C + G16 Ka+ − G16  −8.0 −1.2 2.0 0.4   0.20 0 
A= , B=
F12 C). (34) 0 0 0 1.0  0 0 
Since M̄ is nonsingular, Ag ∈ Rn×n , Aa ∈ Rn×n and 10.0 2.0 −10.0 −2.0 0 1.0
Ad ∈ Rn×n should be selected to make M̄ −1 Ḡ Hurwitz. 0 0
   

If M̄ was not singular and M̄ −1 does not exist, then the 0 0 1 0 0  −0.2 
C= , D =  , S = 
singular value decomposition (SVD) technique needs to be 0 0 0 1 0 0 
used. Hence the state error equation (32) is represented as: 1 0
0 0 0 0
     
ėx (t) = M̄ −1 Ḡex (t) (35)
 0.2  0 0  0 
Ka =  , E=  , E0 =  0 

which satisfies the asymptotic stability of the state error 0  0 0
estimation (9). In this method sufficient design degrees of 0 0 −0.9990 0.0001

11070
18th IFAC World Congress (IFAC'11)
Milano (Italy) August 28 - September 2, 2011

error estimation error estimation

100 0.1
0 0
−100 −0.1
0 5 10 15 20 25 30 0 5 10 15 20 25 30
t(s) t(s)
disturbance error, emu disturbance error, emu
500 0.5
0 0
−500 −0.5
0 5 10 15 20 25 30 0 5 10 15 20 25 30
t(s) t(s)
actuator fault error, ea actuator fault error, ea
2000 1
0 0
−2000 −1
0 5 10 15 20 25 30 0 5 10 15 20 25 30
t(s) t(s)
nonlinearity error, eg nonlinearity error, eg
0.2 0.2
0 0
−0.2 −0.2
0 5 10 15 20 25 30 0 5 10 15 20 25 30
t(s) t(s)

Fig. 2. The behaviour of state errors ex (t), disturbance Fig. 3. The behaviour of state errors ex (t), disturbance
error eµ (t), actuator fault error ea (t) and nonlinearity error eµ (t), actuator fault error ea (t) and nonlinearity
error eg (x, u, t). error eg (x, u, t) where there is no fault on the system.
−0.20 0 0 0
 
−12.574 0.0012 −12.57 1.20
   
 3.179 0.399   3.17 −0.41   0 −0.20 0 0 
L= , K= Ag =  .
3.5250 1.000  3.52 0.99  0 0 −0.20 0 
0.9915 −0.0006 0.99 1.40 0 0 0 −0.20
1.0 0 0 0 0 0
   
Since the matrix M̄ −1 Ḡ is hurwitz with following eigen-
 0 1.0 0 0   0.2 0  values :
H= , G=
0 0 1.0 0  0 0 
( −0.603 + 2.76i −0.603 − 2.76i −1.4033 −3.5207 ) .
0 0 0 0.001 0 0.001

0 1.00 12.57 −1.207
 then the error system (32) is asymptotically stable.
 −8.00 −1.20 −1.17 0.819  Figure (2) shows that in the time of no fault, the errors
N = tend to zero asymptotically. When the fault occurs, the
0 0 −3.52 0.0085 
0.010 0.002 −1.001 −1.404 error responses are quite far from zero. If the fault was
small or occurs in a small period of time then the system
The matrix M̄ , in equation (33) is needs sufficient time to isolate the fault and reduce the
effect of the fault on the system. Otherwise the system
1.0 0 0 0
 
with fault is not stable unless the fault is bonded, then
 0 2.0 0 −0.040  the error under some conditions is eventually ultimately
M̄ = 
0 0 1.0 0  bounded (see Sedighit et al (2008)).
0 0 0 1.0008 Figure (3) shows the error behaviour of states, disturbance,
fault and nonlinearity where there is no fault in the system.
which is invertible In this case, the errors are stable and converge to zero
1.0 0 0 0 asymptotically. Figure (4) shows that the error estimation
 
 0 0.50 0 0  of the system in the absence of faults is stable regardless
M̄ −1 = . of any unknown disturbances.
0 0 1.0 0 
0 0 0 0.9992
6. CONCLUSIONS
Also the matrix Ḡ, in (34) is:
The robust nonlinear UIO has been designed for nonlinear
0 1.0 12.57 −1.20
 
systems and the stability of the error systems have been
 −16.00 −2.40 0.82 1.37  demonstrated. Sufficient existence conditions were derived
Ḡ =  .
0 0 −3.52 0.0085  for the nonlinear UIO. In this method sufficient design
0.020 0.0040 −1.011 −1.406 degrees of freedom (E0 , M1i , F1i and G1i for i = 1, · · · , n+
2) and nonuniquness of Ag , Ad and Aa also provides extra
The design matrices Ad , Aa and Ag are selected as follows design degrees of freedom. This observer may be useful for
to make M̄ −1 Ḡ hurwitz: state estimation and fault diagnosis for nonlinear systems.
−1 0 0 0 −1 0 0 0 The effectiveness of the observer is shown with a numerical
   
 0 −1 0 0   0 −1 0 0  example. The simulation results show that the designed
Ad =  , Aa =  NUIO guarantees the asymptotically convergence of the
0 0 −1 0  0 0 −1 0 
0 0 0 −1 0 0 0 −1 state to zero in the presence of the unknown inputs.

11071
18th IFAC World Congress (IFAC'11)
Milano (Italy) August 28 - September 2, 2011

Hou M., Pugh A. C., and Müller P. C. (1999). Distrib-

error response in absence of fault

0.1 ance decoupled functional observers. IEEE Trans. on

Automatic Control, volume 44, No.2, 382–386.
0.05
Koeing D. (2006). Observer design for unknown input
0 nonlinear descriptor systems via convex otimization.
IEEE Trans. on Automatic Control, volume 51, 1047–
−0.05 1052.
−0.1
Koshkouei A. J. and Zinober A.S.I. (2004). Sliding mode
0 5 10 15 20 25 30 state observation for nonlinear systems. Int. J. Contr.,
t(s)
volume 77, No.2, 118–127, .
4 Luenberger D. G. (1971). An introduction to observers.
unknown disturbance

3
IEEE Trans. on Automatic Control, volume AC-16,
No.6, 596–602.
2 Mondal S., Chakraborty G., and Bhattacharyya K. (2009).
Unknown input nonlinear observer for component fault
1
detection and isolation of dynamic systems. Int. J. of
0 Automation and Contr., volume 3, 154–170.
0 5 10 15 20 25 30 Mehta A. J., Bandyopadhyah B., and Inoue A. (2010).
t(s)
Reduced-order observer design for servo system using
duality to discrete-time sliding-surface design. IEEE
Fig. 4. The behaviour of error estimation and the distur- Trans. on Industrial Electronics, volume 57, 3793–3800.
bance. Pertew A. M., Maeqez H. J., and Zhao Q. (2005). synthe-
REFERENCES sis of unknown input observers for nonlinear Lipshitz
systems. International Journal of Control, volume 78,
Abbaszadeh M., Marquez H. J. (2010). Robust H-infinity 1155–1165.
filter design for Lipschitz nonlinear systems via multi- Raghavan A. and Hedrick J. K. (1994). Observer design for
objective optimization. System and control, volume 1. a class of nonlinear systems. Int. J. Contr., volume 59,
Barbot J. P., Fliss M., and Floquet T. (2007). An algebric No.2, 515–528.
framework for the design of nonlinear observers with Roch-Cozatl E., Moreno J., and Zeitz M. (2005). Cun-
unknown inputs. IEEE Trans. on Decision and Control, structive design of unknown input nonlinear observers
USA. by dissipativity and LMI’s. IFAC..
Barbot J. P. and Floquet T. (2009). Iterative higher Seliger R., and Frank R. M. (1991). Robust component
order sliding mode observer for nonlinear system with fault detection and isolation in nonlinear dynamic sys-
unknown inputs. tems using nonlinear unknown input observers. Proceed-
Barbot J. P., Fliss M., and, Floquet T. (2009). An ob- ings of the IFAC/IMACS Symp. SAFE-PROCESS’91.,
servation algorithm for nonlinear system with unknown 277-282, Baden-Baden, Germany.
inputs. Automatica, volume 45. Sedighi T., Koshkouei A. J., and Burnham K. J. (2008).
Chen W. , and Saif M. (2006a). Fault detection and Nonlinear Observer-Based Residual Design for the Non-
isolation based on novel unknown input observer design. linear Systems. UKACC2008., Manchester, UK.
Proceedings of the 2006 American Conference. Yang H. and Saif M. (1996). Monitoring and diagnosis
Chen W. and Saif M. (2006b). Unknown input observer of a class of nonlinear systems using nonlinear unknown
design for a class of nonlinear systems: an LMI approach. input observers. Proceedings of the IEEE Conference on
Proceedings of the 2006 American Conference. Control Applications, 1006-1011, Dearborn, Michigan.
Darouach M. (2009). Complements to full-order observer Zak S. H. (2003). Systems and Control. New York: Oxford
design for linear systems with unknown inputs. Applied University press.
mathematics letter, volume 22, 1107–1111. Zasadzinski M., Rafaralahy H., Mechmeche C., and
Floquet T. and Barbot J. P. (2005). State and unknown Darouach M. (1998). On disturbance decoupled observer
input estimation for linear discrete time systems. IFAC. for a class of bilinear systems. Trans. of the ASME, J.
Praha.. of Dyn. Syst.,Meas and Cont., volume 120, 371-377.
Guan Y. and Saif M. (1991). A novel approach to the
design of unknown input observers. IEEE Trans. on
Automatic Control., volume 36, No.7, 632–635.
Ha Q. P., Trinh H. , Nguyen H. T., and Tuan H. D. (2003).
Dynamic output feedback sliding-mode control using
pole placement and linear functional observers. IEEE
Trans. on Industr. Electron., volume 50, No.5, 1030–
1037.
Hamidi M.C., Yahoui H., and Lab H. (2008). State
observer for bilinear systems subjected to unknown
inputs. 5th int. multi-conf. on systems, signals and
decices., 1–6.
Hou M. and Müller P. C. (1992). Design of observers for
linear systems with unknwon inputs. IEEE Trans. on
Automatic Control, volume 37, No.6, 871–875.

11072