Proceedings of the 19th World Congress

The International Federation of Automatic Control

Cape Town, South Africa. August 24-29, 2014

Continuous observer design for nonlinear

systems with sampled and delayed output
measurements ?
Daoyuan Zhang ∗ Yanjun Shen ∗∗ Xiaohua Xia ∗∗∗
∗
College of Science, China Three Gorges University, Yichang, Hubei,
443002, China,
∗∗
College of Electrical Engineering and New Energy, China Three
Gorges University, Yichang, Hubei, 443002, China,
∗∗∗
Department of Electrical, Electronic and Computer Engineering,
University of Pretoria, Pretoria 0002, South Africa

Abstract: We design observers for nonlinear systems with sampled and delayed output
measurements. The observers are of continuous and hybrid in nature. Based on an auxiliary
integral technique, the exponential stability of the estimation errors is achieved, and the
sampling period and the maximum delay are also given. Finally, numerical examples are provided
to illustrate the design methods.

1. INTRODUCTION [2008], Shen and Huang [2009], Li et al [2013a,b]). In

recent years, the observer design for continuous nonlinear
In this paper, we consider the following system systems with sampled and delayed measurements have
 attracted more and more attention. For linear systems,

 ẋ1 (t) = x2 (t) + f1 (x1 (t)),

 ẋ2 (t) = x3 (t) + f2 (x1 (t), x2 (t)), observers can be designed based on the discrete time

 model of the continuous systems. However, the design
 ..
. (1) method cannot be extended to nonlinear systems because

 ẋ (t) = xn (t) + fn−1 (x1 (t), x2 (t), · · · , xn−1 (t)), it is difficult to obtain the exact discrete time model. In


n−1
 such a case, two main approaches are proposed to deal
 ẋn (t) = fn (x1 (t), x2 (t), · · · , xn (t)) + u(t),

y(t) = x1 (t), with this problem. The first one is to design a discrete
observer by introducing a consistent approximation of the
where the state x(t) ∈ Rn , the input u(t) ∈ R. We exact discretized model (Arcak and Nešić [2004]). The
assume that the output y(t) is sampled at instants tk second one is based on a mixed continuous and discrete
and is available for the observer at instants tk + τk , design. For instance, a high gain exponential observer is
where {tk } denotes a strictly increasing sequence such that presented in (Deza and Busvelle [1992]). In (Raff et al
limt→∞ tk = ∞, and τk ≥ 0 represents the transmission [2008]), the observers for Lipschitz nonlinear continuous
delay. The sampling interval T = tk+1 − tk is a positive time systems with nonuniformly sampled measurements
constant. The transmission delays τk are unknown, but are designed based on LMI and sampled data control tech-
have an upper bound τ̄ , that is, max{τk } ≤ τ̄ for all niques. In (Ahmed-Ali and Lamnabhi-Lagarrigue [2012]),
k = 0, 1, · · · , ∞. We also make the assumption: τ̄ ≤ T , that some conditions on the maximum allowable transmission
is, the measures sampled at instants tk are available for interval are given to guarantee an exponential convergence
the observer before the next measures sampled at instants of the observation error for networked control systems.
tk+1 . In addition, fi (·) (i = 1, · · · , n) satisfy the following The authors proposed a sampled-data nonlinear observer
globally Lipschitz condition design by using a continuous-time design coupled with
| fi (x1 , x2 , · · · , xi ) − fi (x̂1 , x̂2 , · · · , x̂i ) | an inter-sample output predictor (Karafyllis and Kravaris
[2009]). In (Nadri et al [2013]), the problem of observer
≤ l(| x1 − x̂1 | + | x2 − x̂2 | + · · · + | xi − x̂i |), (2) design was investigated for uniformly observable systems
with sampled output measurements. There are some re-
where l is a positive constant. sults on observer design for nonlinear uniformly observ-
There are some results on observer design for (1) without able systems with time-varying delayed output measure-
sampled and time delayed measurements, for example, ment (Van Assche et al [2011], Ahmed-Ali et al [2013a,b]).
(Deaza [1991], Gauthier et al [1992], Gauthier and Kupka For example, in (Ahmed-Ali et al [2013b]), the authors
[1994], Praly [2003], Andrieu et al [2009], Shen and Xia addressed two classes of global exponential observers of
? This work was supported by the National Science Foundation of continuous systems with sampled and delayed output mea-
surements. There are two parts: a prediction part and
China (No. 61174216, 61273183, 61374028, 61304162), the Grant
National Science Foundation of Hubei Province (2013CFA050), the
a correction part. Sufficient conditions on the maximum
National Science Foundation of Hubei Province (2011CDB187), the allowable delay and the maximum allowable sampling pe-
Scientific Innovation Team Project of Hubei Provincial Department riod are also given to ensure exponential stability of the
of Education (T201103). observation error.

978-3-902823-62-5/2014 © IFAC 269

19th IFAC World Congress
Cape Town, South Africa. August 24-29, 2014

In this paper, we consider high gain observer design for the is only updated at time instants tk +τk , that is, the current
system (1) with sampled and delayed measurements. The evolution process is used until the new evolution process
observer is continuous and hybrid. By using an auxiliary is available. Therefore, the dynamics of observer (3) is of
integral technique, sufficient conditions are presented to continuous and of hybrid nature.
ensure the global exponential stability of the observation Remark 2. Even though τk and τk+1 are unknown, e1 (tk )
error. Compared with the existing results, the systems we is updated automatically in the observer whenever sam-
considered are more general, and the design method is pled and delayed measurement arrives. In (Ahmed-Ali
simpler. et al [2013b]), the sampled and delayed measurement is
This paper is organized as follows. In Section 2, we present available at instant tk + τk , however, e1 (tk ) is updated at
our main results: the continuous observer is designed instant tk + τ , that is, there exists time delay τ − τk .
for nonlinear systems with sampled and delayed output Remark 3. In this paper, the sampling period T and the
measurements. In Section 3, two examples are used to maximum allowable delay τ̄ depend on the Observer (3).
illustrate the validity of the proposed design method.
Finally, the paper is concluded in Section 4. From (1)-(3), for k ≥ 0, we obtain the observation error:


 ė1 (t) = e2 (t) − La1 e1 (tk ) + f˜1 ,
2. CONTINUOUS OBSERVER DESIGN FOR THE 


 ė2 (t) = e3 (t) − L2 a2 e1 (tk ) + f˜2 ,
SYSTEM (1) 
 ..
. (5)
We will design a continuous observer for the system (1), 
 ėn−1 (t) = en (t) − Ln−1 an−1 e1 (tk ) + f˜n−1 ,


and present the upper bound of the maximum allowable 
 n ˜
sample period and the delay, so that the observation  ėn (t) = −L an e1 (tk ) + fn ,

t ∈ [tk + τk , tk + T + τk+1 ),
errors are globally exponentially convergent. The following
lemma is useful for our main results. where ei (t) = xi (t)− x̂i (t), f˜i = fi (x1 (t), x2 (t), · · · , xi (t))−
Lemma 1. (Liu et al [2006] ) For any positive definite fi (x̂1 (t), x̂2 (t), · · · , x̂i (t)), 1 ≤ i ≤ n.
matrix M ∈ Rn×n , scalar γ > 0, vector function ω :
[0, γ] → Rn such that the integrations concerned are well Consider the following change of coordinates
defined, the following inequality holds: ei (t)
εi (t) = i−1+b , i = 1, 2, · · · , n, (6)
 γ >  γ   γ  L
Z Z Z
 w(s)ds M  w(s)ds ≤ γ  w(s)> M w(s)ds . where b is a positive real number. Then, we have


 f˜
0 0 0  ε̇1 (t) = Lε2 (t) − La1 ε1 (tk ) + 1b ,


 L

 f˜2
For the system (1), we design the following observer, 

 
 ε̇2 (t) = Lε3 (t) − La2 ε1 (tk ) + 1+b ,
 x̂˙ (t) = x̂2 (t) + La1 e1 (tk ) + f1 (x̂1 (t)), 
 L

 ˙1  ..

 x̂2 (t) = x̂3 (t) + L2 a2 e1 (tk ) + f2 (x̂1 (t), x̂2 (t)), . (7)

 

 .. 
 f˜n−1

 . 
 ε̇ (t) = Lε (t) − La ε (t ) + ,
 

n−1 n n−1 1 k
Ln−2+b
x̂˙ n−1 (t) = x̂n (t) + Ln−1 an−1 e1 (tk ) (3) 
 ˜
 
 fn

 +fn−1 (x̂1 (t), x̂2 (t), · · · , x̂n−1 (t)), 
 ε̇n (t) = −Lan ε1 (tk ) + n−1+b ,

 
 L


 x̂˙ n (t) = Ln an e1 (tk ) + fn (x̂1 (t), x̂2 (t), · · · , x̂n (t)) t ∈ [tk + τk , tk + T + τk+1 ).

 +u(t),


t ∈ [tk + τk , tk + T + τk+1 ), k ≥ 0, Further, (7) can be rewritten as follows:

 ε̇1 (t) = Lε2 (t) − La1 ε1 (t) + La1 (ε1 (t) − ε1 (tk ))
where L ≥ 1 is a constant, e1 (tk ) = x1 (tk ) − x̂1 (tk ), ai > 0 

(i = 1, · · · , n) are given such that there exists a symmetric

 f˜1

 + b,
positive definite matrix P such that 
 L

 ε̇2 (t) = Lε3 (t) − La2 ε1 (t) + La2 (ε1 (t) − ε1 (tk ))


A> P + P A ≤ −I, (4) 

 f˜2

 + ,
  
 L1+b
−a1 1 ··· 0 ..
 .. .. .. ..  . (8)
. 

where A = 
 −a
. . . . 
 ε̇n−1 (t) = Lεn (t) − Lan−1 ε1 (t) + Lan−1 (ε1 (t)
n−1 0 ··· 1 

 f˜n−1


−an 0 ··· 0 
 −ε1 (tk )) + n−2+b ,

 L

 ˜
Note that e1 (tk ) is a constant on [tk + τk , tk + T + τk+1 )  ε̇n (t) = −Lan ε1 (t) + Lan (ε1 (t) − ε1 (tk )) + fn ,


for k ≥ 0 and fi (·) (i = 1, · · · , n) are continuous and 
 L n−1+b
satisfy the condition (2), then, limt→tk +T +τ − x̂i (t) = t ∈ [tk + τk , tk + T + τk+1 ).
k+1
limt→tk +T +τ + x̂i (t) (i = 1, · · · , n). Therefore, x̂i (t) (i = Now, we will give our main results.
k+1
1, · · · , n) are continuous on [t0 , ∞). Theorem 1. Consider the system (1) with the condi-
Remark 1. Note that the state estimate is described in tion (2). The output y(t) is assumed to be sampled at
continuous time while the evolution process x1 (tk )− x̂1 (tk ) instants tk and is available for the observer at instants

270
19th IFAC World Congress
Cape Town, South Africa. August 24-29, 2014

κ1 κ2
tk + τk . Let T = L 3 , τ̄ = L3 (κ1 > κ2 > 0 are two positive Then, there exists L2 > 1 such that when L > L2 , we have
constants). If, ai > 0 (i = 1, · · · , n) are selected such that Lb+1 > l.
the condition (4) holds, and L ≥ 1 is given such that T + τ̄
satisfies If follows from (12) and (13) that
n L − 2λ ρ̂ 1 d 1
T + τ̄ ≤ min
1
, , V1 (t)|(8) ≤ − Lε(t)> ε(t) + 12nL3 ā21 p̄2 (t − tk )2 ε1 (tk )2
L 2
12nL ā1 p̄2 + ρ̂ dt 2
ρ̂λ2 o Zt
, τ̄ < T, (9) 3
+12nL ā1 p̄2 (t − tk ) [ε1 (s)2 + ε2 (s)2 ]ds,
12nL3 ā21 p̄2
then, the state observation error system (5) is globally tk
exponentially stable, where p̄2 = λmax (P√> P ), ā1 = t ∈ [tk + τk , tk + T + τk+1 ).
L−1
max{a2i }, λ1 = λmax (P ), λ2 = λmin (P ), ρ̂ = 2λ .
1
Note that when t ∈ [tk + τk , tk + T + τk+1 ), we have
Proof: Consider the positive definite function t − tk − τk ≤ T + τk+1 − τk , that is, t − T − τk+1 < tk .
Then,
V1 (t) = ε(t)> P ε(t), (10)
d 1
where ε(t) = [ε1 (t), · · · , εn (t)]> , then the derivative of V1 (t) | (8) ≤ − Lε(t)> ε(t) + 12nL3 ā21 p̄2 (t − tk )2 ε1 (tk )2
V1 (t) along the system (8) is given as follows: dt 2
Zt
d 3
V1 (t)|(8) = ε̇(t)> P ε(t) + ε(t)> P ε̇(t) +12nL ā1 p̄2 (t − tk ) [ε1 (s)2 + ε2 (s)2 ]ds,
dt
n X n t−T −τk+1
X
≤ −Lε(t)> ε(t) + 2l | εi (t)Pij | t ∈ [tk + τk , tk + T + τk+1 ). (14)
i=1 j=1 Construct the following auxiliary integral function
| Lb ε1 (t) | + | L1+b ε2 (t) | + · · · + | Lj−1+b εj (t) |
× Zt Zt
Lj−1+b

a1 (ε1 (t) − ε1 (tk ))
 V2 (t) = [ε1 (s)2 + ε2 (s)2 + · · · + εn (s)2 ]dsdρ,
 a2 (ε1 (t) − ε1 (tk ))  t−T −τ̄ ρ
+2Lε(t)> P   .. 
 t ∈ [tk0 , ∞), (15)
.
an (ε1 (t) − ε1 (tk )) where k0 = min{k : T + τ̄ < tk }. Then,
3 d
≤ (− L + 2np̄1 l)ε(t)> ε(t) + 4Lnā1 p̄2 (ε1 (t) − ε1 (tk ))2 , V2 (t) = (T + τ̄ )[ε1 (t)2 + ε2 (t)2 + · · · + εn (t)2 ]
4 dt
t ∈ [tk + τk , tk + T + τk+1 ), (11)
Zt
where Pij is the element of P at the ith line and jth − [ε1 (s)2 + ε2 (s)2 + · · · + εn (s)2 ]ds. (16)
column, p̄1 = max{|Pij |}.
t−T −τ̄
Note that there exists L1 > 1 such that Now, we consider the following Lyapunov-Krasovskii func-
L > 8np̄1 l, tion
V(t) = V1 (t) + LV2 (t). (17)
for L > L1 . Then,
Calculate the derivative of V (t) defined in (17) along the
d 1 system (8), we have
V1 (t)|(8) ≤ − Lε(t)> ε(t) + 4nLā1 p̄2 (ε1 (t) − ε1 (tk ))2 ,
dt 2
d 1
t ∈ [tk + τk , tk + T + τk+1 ). (12) V (t)|(8) ≤ (− + (T + τ̄ ))Lε(t)> ε(t) + 12nL3 ā21 p̄2
dt 2
By Lemma 1, we have Zt
Zt 2 2 2
(T + τ̄ ) ε1 (tk ) + (12L nā1 p̄2 (T + τ̄ ) − 1)L [ε1 (s)2
2
| ε1 (t) − ε1 (tk ) | = | ε̇1 (s)ds |2 ≤ (t − tk ) t−T −τ̄
tk

Zt Zt +ε2 (s)2 + · · · + εn (s)2 ]ds,

2 f˜1 2 (18)
× | ε̇1 (s) | ds ≤ L(t − tk ) [ε2 (s) − a1 ε1 (tk ) + ] ds t ∈ [tk + τk , tk + T + τk+1 ), k ≥ k0 .
Lb+1
tk tk Note that
Zt Zt
l2
≤ 3(t − tk )L 2 2
[ε2 (s) + 2 2
ε1 (s) + ā1 ε1 (tk ) ]ds, V2 (t) ≤ (T + τ̄ ) [ε1 (s)2 + ε2 (s)2 + · · ·
L2(b+1)
tk t−T −τ̄

t ∈ [tk + τk , tk + T + τk+1 ). (13) +εn (s)2 ]ds, t ∈ [tk + τk , tk + T + τk+1 ). (19)

271
19th IFAC World Congress
Cape Town, South Africa. August 24-29, 2014

Thus, (18) and (19) imply for L > L4 . Thus,

d 1 e−ρ̂(T −τ̄ ) (1 + ρ) < 1, 2(T + τ̄ ) < ρ (23)
V (t)|(8) ≤ −( − (T + τ̄ ))Lε(t)> ε(t)
dt 2
2
1 − 12L nā1 p̄2 (T + τ̄ ) Therefore, we can select L > max{L1 , L2 , L3 , L4 } such
− LV2 (t) + 12nL3 ā21 p̄2 (T + τ̄ )2 that η = max{e−ρ̂(T −τ̄ ) (1 + ρ), 2(Tρ+τ̄ ) } < 1. Therefore, it
(T + τ̄ )
follows from (22) that
1 − 12L2 nā1 p̄2 (T + τ̄ )
×ε1 (tk )2 ≤ − LV2 (t)
(T + τ̄ ) V (tk+1 + τk+1 ) + ρV (tk + T )
1 12nL3 ā21 p̄2 (T + τ̄ )2 = V (tk + T + τk+1 ) + ρV (tk + T )
− (1 − (T + τ̄ ))LV1 (t) + V1 (tk ),
2λ1 λ2
≤ η[V (tk + τk ) + ρV (tk )], k ≥ k0 . (24)
t ∈ [tk + τk , tk + T + τk+1 ), k ≥ k0 . (20) Applying iteratively (24), for k ≥ k0 , we have
Since T + τ satisfies (9), then, we have V (tk + τk ) + ρV (tk ) ≤ η k−k0 [V (tk0 + τk0 ) + ρV (tk0 )].(25)
d
V (t)|(8) ≤ −ρ̂V (t) + (T + τ̄ )ρ̂V (tk ), It follows from (21) and (25) that
dt
V (t) ≤ V (tk + τk ) + ρV (tk )
t ∈ [tk + τk , tk + T + τk+1 ), k ≥ k0 .
≤ η k−k0 [V (tk0 + τk0 ) + ρV (tk0 )],
From the above differential inequality, we obtain t ∈ [tk + τk , tk + T + τk+1 ), k ≥ k0 .
V (t) ≤ e−ρ̂(t−tk −τk ) V (tk + τk ) + (T + τ̄ )V (tk )
For any t > tk0 +τk0 , there exists k ≥ k0 such that t ∈ [tk +
−(T + τ̄ )e−ρ̂(t−tk −τk ) V (tk ), t−tk0 −τk0 t−tk0 −τk0
τk , tk +T +τk+1 ). Note that T −1 ≤ k ≤ T .
t ∈ [tk + τk , tk + T + τk+1 ), k ≥ k0 . (21) Then,
Let t = tk + T + τk+1 and t = tk + T , respectively, we have t−tk −τk
0 0 −1
V (t) ≤ η T [V (tk0 + τk0 ) + ρV (tk0 )]
−ρ̂(T +τk+1 −τk )
V (tk + T + τk+1 ) ≤ e V (tk + τk ) t
−tk −τk
0 0 −1
=η η T T [V (tk0 + τk0 ) + ρV (tk0 )],
+(T + τ̄ )V (tk ) − (T + τ̄ )e−ρ̂(T +τk+1 −τk ) V (tk ),
t ∈ [tk + τk , tk + T + τk+1 ), k ≥ k0 . (26)
and, That is, the system (8) is globally exponentially stable.
V (tk + T ) ≤ e−ρ̂(T −τk ) V (tk + τk ) + (T + τ̄ )V (tk ) Remark 4. The value of η has an effect on the convergent
time. The state observation error has a faster convergent
−(T + τ̄ )e−ρ̂(T −τk ) V (tk ).
speed with a smaller value of η. A large value of L can be
Thus, selected to make η small. However, it may yield a small
sampling period T . In practical application, we can select
a proper L (on T and τ̄ ) to achieve the desired effect.
V (tk + T + τk+1 ) + ρV (tk + T )
Remark 5. The change of coordinates (6) differs by the
≤ e−ρ̂(T −τ̄ ) (1 + ρ)V (tk + τk ) + 2(T + τ̄ )V (tk ), (22) positive real number b from the standard change of co-
where 0 < ρ < 1 is a positive constant and will be ordinates used in high-gain observers where b = 0. If
κ1 κ2 the lipschitz constant l is very large, we can choose a
determined later. Note that T = L3 , τ̄ = L3 (κ1 > κ2 ),
√
L−1 proper constant b such that L is not very large. Then,
and ρ̂ = 2λ , then, there exists L3 > 1 such that when
1 the parameter b gives more flexibility on selection of the
L > L3 , we have high gain L.
√
( L − 1)(κ1 − κ2 )
2(κ1 + κ2 ) < . 3. EXAMPLE AND SIMULATION
2λ1
Since In this section, we use two examples to show the effective-
√
( L − 1)(κ1 − κ2 ) ness of our high gain observer design for nonlinear systems
ρ̂(κ1 − κ2 ) = , with sampled and delayed measurements.
2λ1
Example 1. An academic bioreactor is reactor in which
0
and microorganisms grow by eating a substrate. Let x1 (t) and
3 0
eρ̂(κ1 −κ2 )/L > 1 + ρ̂(κ1 − κ2 )/L3 , x2 (t) denote the concentrations of microorganisms and
substrate, respectively. Then, the following standard equa-
then, tions for the bioreactor (Contois [1959]) can be obtained:

2(κ1 + κ2 ) ρ̂(κ1 − κ2 ) 3

0 0
a1 x1 (t)x2 (t)
3
< 3
< eρ̂(κ1 −κ2 )/L − 1. 

 ẋ
0
(t) =
0
− u(t)x1 (t),
L L 
 1 0 0
a2 x1 (t) + x2 (t)
0 0
Therefore, there exist L4 > 1, 0 < ρ < 1 such that 0 a3 a1 x1 (t)x2 (t) 0 (27)

 ẋ (t) = − − u(t)x2 (t) + u(t)a4 ,
2(κ1 + κ2 )  2
 a 2 x
0
(t) + x
0
(t)
3 
 1 2
< ρ < eρ̂(κ1 −κ2 )/L − 1, 0
y(t) = x1 (t),
L3

272
19th IFAC World Congress
Cape Town, South Africa. August 24-29, 2014

where the control u(t) is in the interval Θu = [umin , umax ] ⊂ Example 2. A single-link robot arm system can be mod-
(0, 1) and we choose a1 = a2 = a3 = 1, a4 = 0.1. eled by (Isidori [1995]) or (Marino and Tomei [1995])
In (Gauthier et al [1992]), it is observed that the set 
 ż1 (t) = z2 (t),
0 0 0 0
Θx0 = (x1 (t), x2 (t)) ∈ R2 : x1 (t) ≥ γ1 , x2 (t) ≥ γ2 , 


 K F2 (t) K
0 0 
 ż2 (t) = z3 (t) − z2 (t) − z1 (t)
x1 (t) + x2 (t) ≤ 1 is forward invariant, where γ1 = (1 − 
 J2 N J2 J2
umax γ2 /umax ) and umin = γ2 /(0.1 − γ2 ). This means that 
 mgd

 − cos(z1 (t)),
the bioreactor state solutions are bounded and actually 
 J2
remain in a known compact set. After the state transfor- ż3 (t) = z4 (t), (28)
0 0 0 0 0 
 1 K K
mation x1 (t) = x1 (t), x2 (t) = x1 (t)x2 (t)/(x1 (t) + x2 (t)), 
 ż4 (t) = u(t) + z1 (t) − z3 (t)
we have 


 J1 J1 N J2 N
 
 F1 (t)
ẋ1 (t) = x2 (t) − u(t)x1 (t), 
 − z4 (t),

 
 J1
x3 (t) + (x1 (t) − x2 (t))2 (0.1u(t) − x2 (t)) 
ẋ2 (t) = 2 y(t) = z1 (t),

 x21 (t)
−x2 (t)u(t). where J1 , J2 , K, N , m, g, d are known parameters, F1 (t)
and F2 (t) are viscous friction coefficients that are not
We get the following equation for the observer: precisely known. Suppose F1 (t) and F2 (t) are bounded
 by an unknown constant C > 0. Consider the change of

 x̂˙ 1 (t) = x̂2 (t) − u(t)x̂1 (t) + 3L(y(tk ) − x̂1 (tk )),

 coordinates
˙ x̂3 (t) + (x̂1 (t) − x̂2 (t))2 (0.1u(t) − x̂2 (t))
x̂2 (t) = 2 x1 (t) = z1 (t), x2 (t) = z2 (t),
 x̂21 (t)

 2
−x̂2 (t)u(t) + 2L (y(tk ) − x̂1 (tk )), K mgd K

 x3 (t) = z3 (t) − , x4 (t) = z4 (t)
t ∈ [tk + τk , tk + T + τk+1 ), k ≥ 0. J2 N J2 J2 N

In the following simulation, we apply the input u(t) = 0.08, and the pre-feedback
for t ≤ 10, u(t) = 0.02, for 10 < t, x0 = [0.7, 0.3]> , K ¡1 mgd ¢
υ= u(t) − ,
x̂0 = [0.2, 0.5]> , L = 1, the sampling period T = 0.1s and J2 N J1 J2
the delay τk satisfying 0 ≤ τk ≤ 0.05s. The trajectories
of the error states are shown in Fig. 1. In practice, the which transforms (28) into
measure is often disturbed by a white noise. Fig. 2 also 
 ẋ1 (t) = x2 (t),
shows the trajectories of the errors with white noise. 
 F2 (t) K

 ẋ2 (t) = x3 (t) − x2 (t) − x1 (t)



 J 2 J 2
0.6 
 mgd
e1(t)
− (cos(x1 (t)) − 1),
e2(t) J2
0.4

 ẋ3 (t) = x4 (t),



 K2 K F1 (t)
0.2

 ẋ (t) = υ + x1 (t) − x3 (t) − x4 (t),

 4
J1 J2 N 2 J2 N J1
0

y(t) = x1 (t).
−0.2
Construct the following observer:

x̂˙ 1 (t) = x̂2 (t) + 4L(y(tk ) − x̂1 (tk )),
−0.4



 F2 (t) K
−0.6 
 x̂˙ 2 (t) = x̂3 (t) − x̂2 (t) − x̂1 (t)



 J2 J2
−0.8 
 mgd
0 5 10
t/s
15 20 
 − (cos(x̂1 (t)) − 1) + 6L2 (y(tk ) − x̂1 (tk )),
J2
Fig. 1 Trajectories of the errors e1 (t) and e2 (t). 
 x̂˙ (t) = x̂4 (t) + 4L3 (y(tk ) − x̂1 (tk )),
 3


 K2 K F1 (t)

 x̂˙ 4 (t) = υ + x̂ (t) −
2 1
x̂3 (t) − x̂4 (t)
0.6

 J J
1 2 N J2 N J1
e1(t)

 4
+L (y(tk ) − x̂1 (tk )),
0.4
e2(t) 

t ∈ [tk + τk , tk + T + τk+1 ), k ≥ 0.
0.2
In the following simulation, we apply the system pa-
0 rameters: K/J2 = 5, mgd/J2 = 4, K 2 /(J1 J2 N 2 ) =
2, K/(J2 N ) = 3. The initial conditions of the whole
−0.2
system are (x1 (0), x2 (0), x3 (0), x4 (0), ) = (−5, −1, 4, 2)
−0.4 and (x̂1 (0), x̂2 (0), x̂3 (0), x̂4 (0), ) = (5, 3, −1, −4), a =
(4, 6, 4, 1)> , and L = 1. The sampling period T and the
−0.6 delay τk are chosen as T = 0.1s and 0 ≤ τk ≤ 0.05s.
−0.8
The simulation results is shown in Fig. 3. In practice,
0 5 10
t/s
15 20 the measure may be disturbed by a white noise. The
trajectories of the errors with white noise are also shown
Fig. 2 Trajectories of the errors e1 (t) and e2 (t) with white noise. in Fig. 4.

273
19th IFAC World Congress
Cape Town, South Africa. August 24-29, 2014

D. Contois, Kinetics of bacterial growth relationship be-

15
e1(t) tween population density and specific growth rate of
e2(t) continuous cultures, J. Genet. Macrobio, vol. 21, pp. 40-
10
e3(t)
e4(t)
50, 1959.
F. Deza, Contribution to the synthesis of exponential
5
observers, PH. D. Dissertation, INSA Rouen, France,
1991.
F. Deza, E. Busvelle, J. Gauthier, and D. Rakotopora,
0
High gain estimation for nonlinear systems, Syst. Contr.
Lett., vol. 18, pp. 295-299, 1992.
−5 J. Gauthier, H. Hammouri, and S. Othman, A simple ob-
server for nonlinear systems applications to bioreactors,
−10 IEEE Trans. Automat. Contr., vol. 37, no. 6, pp. 875-
0 5 10
t/s
15 20
880, June 1992.
J. Gauthier, and I. A. Kupka, Observability and observers
Fig. 3 Trajectories of the errors ei (t) (1 ≤ i ≤ 4) .
for nonlinear systems, SIAM J. Contr. and Optimiz.,
vol. 32, no. 4, pp. 975-994, 1994.
15
e1(t) Isidori, Nonlinear Control Systems. New York: Springer,
e2(t) 1995.
10
e3(t)
I. Karafyllis, and C. Kravaris, From continuous-time de-
e4(t)
sign to sampled-data design of observers, IEEE Trans.
5
Automat. Contr., vol 54, no. 9, pp. 757-762, Sep. 2009.
Y. Li, X. Xia, and Y. Shen, A high-gain-based global finite-
time nonlinear observer, Int. J. Contr., vol. 86, no. 5, pp.
0
759-767, 2013.
Y. Li, Y. Shen, and X. Xia, Global finite-time observers
−5 for a class of nonlinear systems, Kybernetika, vol. 49,
no. 2, pp. 319-340, 2013.
−10 Y. Liu, Z. Wang and X. Liu, On global exponential
0 5 10
t/s
15 20
stability of generalized stochastic netural networks with
mixed time-delays, Neurocomputing, vol. 70, pp. 314-
Fig. 4 Trajectories of the errors ei (t) (1 ≤ i ≤ 4) with white noise.
326, 2006.
R. Marino, and P. Tomei, Nonlinear Control Design. Uk:
4. CONCLUSION
Prentice Hall International, 1995.
In this paper, we designed observers for nonlinear systems H. Nadri, H. Hammouri, and R. Mota, Observer design
with sampled and delayed output measurements. The for uniformly observable systems with sampled measure-
observers were continuous and hybrid in nature. Based on ments, IEEE Trans. Automat. Contr., vol 58, no. 3, pp.
an auxiliary integral technique, the exponential stability 757-762, Mar. 2013.
of the estimation errors was achieved, and the sampling L. Praly, Asymptotic stabilization via output feedback
period and the maximum delay were also given. Finally, for lower triangular systems with output dependent
numerical examples were provided to illustrate the design incremental rate, IEEE Trans. Automat. Contr., vol. 48,
methods. pp. 1103-1108, 2003.
T. Raff, M. Kögel, and F. Allgöwer, Observer with
REFERENCES sample-and-hold updating for lipschitz nonlinear sys-
tems with nonuniformly sampled measurements, in
T. Ahmed-Ali, and F. Lamnabhi-Lagarrigue, High gain ACC’08 (American Control Conference), Seattle, Wash-
observer design for some networked control systems, ington, USA, 2008.
IEEE Trans. Automat. Contr., vol. 57, no. 4, pp. 995- Y. Shen, and X. Xia, Semi-global finite-time observers for
1000, Apr. 2012. nonlinear systems, Automatica, vol. 44, no. 12, pp. 3152-
T. Ahmed-Ali, I. Karafyllis, and F. Lamnabhi-Lagarrigue, 3156, 2008.
Global exponential sampled-data observers for nonlin- Y. Shen, and Y. Huang, Uniformly observable and
ear systems with delayed measurements, Syst. Control glaobally lipschitzian nonlinear systems admit global
Lett., vol. 62, pp. 539-549, 2013. finite-time observers, IEEE Trans. Automat. Contr., vol.
T. Ahmed-Ali, V. Van Assche, J. Massieu and P. Dorleans, 54, no. 11, pp. 995-1006, Nov. 2009.
Continuous-discrete observer for state affine systems V. Van Assche, T. Ahmed-Ali, C. Ham, and F. Lamnabhi-
with sampled and delayed measurements, IEEE Trans. Lagarrigue, High gain observer design for nonlinear
Automat. Contr., vol. 58, pp. 1085-1091, Apr. 2013. systems with time varying delayed measurements. in
V. Andrieu, L. Praly, A. Astolfi, High gain observers with 18th IFAC World Congress, Milan, Italy, 2011.
updated high-gain and homogeneous correction terms,
Automatica, vol. 45, no. 2, pp. 422-428, 2009.
M. Arcak and D. Nešić, A framework for nonlinear
sampled-data observer design via approximate discrete-
time models and emulation, Automatica, vol. 40, pp.
1931-1938, 2004.

274