Professional Documents
Culture Documents
1
Department of Control and Automation Engineering, Yildiz Technical University, Istanbul, Turkey
ocean_blue04@yahoo.com
2
Department of Control and Automation Engineering, Yildiz Technical University, Istanbul, Turkey
beklan@yildiz.edu.tr
1. Introduction
Fig.1. Block diagram of proposed observer
Time-delay system (TDS) is a system having delays in its states,
inputs or outputs and occurs in many natural and engineering
To design an observer for TDSS we use simple Luenberger
events. Time-delay is commonly encountered in chemical
approach, but we introduced here two feedback line instead of
processes, biological systems, long transmission line of
one. The first feedback line contains a proportional gain matrix
pneumatic, hydraulic system, steel rolling mills, space missions
and second feedback line has a gain matrix (given) followed by
and usually a very common source of instability.Time-delay
a differentiator block. So here we are considering not only the
systems actually belongs to the class of functional differential
difference between real states and estimator states or error
equation (FDE),which has infinite dimensions making it more
signals but also the rate of change of error signals. Taking into
complex. Both to analysis and designs taking into the
consideration both error and rate of change of error data would
consideration of deviating arguments or differential difference
make the observer more reliable than simple Luenberger type
term is necessary for engineers to make models to behave like
one.
more to real process.
ܪஶ state observer problem has been studied for many years in 2. Problem Formulation
order to improve satisfactory observer action under exogenous
disturbance. Several methods for ܪஶ observer design for time Consider the following linear time-delay system,
delay system such as Lyapunov–Krasovskii approach, algebraic
Riccati Equation approach, Fattouh et el method has been ሶ (t)=Ax(t)+Adx(t-h)+Bu(t)+Nw(t)
discussed in [1].The delay-dependent design methods which are y=Cx(t)
suitable for systems with time delay being of known size have x(t+ߠ) =߮ (ߠ) ߳ߠሾെ݄ǡ Ͳሿ (1)
been proposed in [2,3] based on Riccati Equation approach.A
Lyapunov approach to design an observer for discrete time where ,
system in terms of Riccati Equation has been proposed in [4]. xሺݐሻ߳Rn : The State vector.
Based on Lyapunov stability theory, the design of observer with w(t)߳Rq : The exogenous disturbance input which belongs to
internal delay and unknown input is formulated in terms of L2[0,λ).
Linear Matrix Inequality (LMI) in [5] and authors developed a y(t)߳Rp : The output vector.
delay independent matrix representation. A reset observer A, Ad߳܀ൈ , B߳܀ൈ , N߳܀ൈ , C߳܀ൈ .
framework has been proposed in [6] for linear time-delay The above matrices are constant and known system matrices.
systems to improve settling time and overshoot performance. It
is well known that ܪஶ filtering problem is dual to the ܪஶ h 0 :a positive scalar denoting the time delay.
control one for linear systems without uncertainty.ܪஶ ߮ (Ǥ) : a continuously differentiable function on [-݄ǡ Ͳ]
Controller (observer) design procedure has been proposed and representing the initial condition.
848
3. Main Results L1,L2߳܀ൈ :The constant observer gain matrix to be selected
appropriately.
Let us formulate an observer dynamics as follows, ݕො(t)߳Rp :The estimated output vector.
Subtracting equation (2) from equation (1) we get, e(t)0 as t if the following conditions are satisfied:
ሶ (t) -ොሶ(t) = Ax(t)+Adx(t-h)+Bu(t)+Nw(t) - Fො(t)-Gො(t-h) (1) The system is stable and observable.
(2) ( I+ L2C) is invertible.
-Hu(t)-Mw(t)-L1(y(t)-ො(t))-L2(ሶ (t)-ොሶ(t)) (3) A=F,
Ad=G,
ሶ (t) =Ax(t)+Adx(t-h)+Bu(t)+Nw(t) - Fො(t)-Gො(t-h)-Hu(t) B=H,
N=M, then the error dynamics reduces to,
-Mw(t)-L1(y(t)-ො(t))-L2(ሶ (t)-ොሶ(t))+ Fx(t)+Gx(t-h)
݁ሶ (t) =( I+ L2C)-1[(F-L1C)e(t)+Ge(t-h)]
-Fx(t)-Gx(t-h)
݁ሶ (t) = Z[(F-L1C)e(t)+Ge(t-h)] (4)
ሶ (t) = (A -F)x(t)+(Ad–G)x(t-h)+(B-H)u(t)+(N-M)w(t)
We will utilize following the Leibniz rule
+F(x(t)െො(t))+G(x(t-h)-ො(t-h))-L1(Cx(t)- Cො(t))
௧
Lemma 1: A(t-h) = A(t)-௧ି ܣሶሺȽሻ Ƚ
-L2(Cሶ (t)-Cොሶ(t))
ሶ (t) = (A -F)x(t)+(Ad–G)x(t-h)+(B-H)u(t)+(N-M)w(t) We will also use the following lemma in our proof
849
݁ሶ (t) = Z[(F-L1C)+G]e(t) e(t)T[FTPZ-1 -CTL1TPZ-1+GTPZ-1+ Z-TPF- Z-TPL1C+ Z-TPG]e(t)
+ h e(t)TZ-TPGZR1-1ZTGTPZ-1e(t)+ h e(t)TZ-TPGZR2-1ZTGTPZ-1e(t)
െܼ ܼܩන ሾሺ െ ܮଵ ሻሺ Ƚሻ
ሺ Ƚ െ ሻሿ Ƚ
ି + h e(t)T(F-L1C)TR1(F-L1C)e(t)+ h e(t)TGTR2Ge(t)
(5) Now applying Schur complement we get,
݁ሶ (t)՜0 as t՜ λ means error in (5) tends to ‘0’ as time evolves. e(t)T
ߗ ݄ܼ ି் ܼܲܩ ݄ܼ ି் ܼܲܩ ்݄ ܩ ݄ሺ െ ܮଵ ሻ்
Delay-dependent approach: Consider the following Lyapunov– ି ் ் ۍଵ ې
ێ ݄ܼ ܩ ܼܲ െ݄ܴଵ Ͳ Ͳ Ͳ ۑ
Krasovskii functional
ି ܼܲ ் ܩ ் ܼ݄ێଵ Ͳ െ݄ܴଶ Ͳ Ͳ ۑ
ێ ۑ
V(e,t)= e(t)T Z-TPZ-1 e(t) ێ ݄ܩ Ͳ Ͳ െ݄ܴଶିଵ Ͳ ۑ
ିଵ
݄ ۏሺ െ ܮଵ ሻ Ͳ Ͳ Ͳ െ݄ܴଵ ے
௧
+ି ௧ା ሺɅሻ ሺ െ ܮଵ ሻ ଵ ሺ െ ܮଵ ሻሺɅሻ ݀ɅǤ ݀ݏ
e(t)
௧ି
+ି ௧ିା ሺɅሻ
ܴଶ
ሺɅሻ ݀ɅǤ ݀ݏ Here, ߗ= (FTPZ-1 -CTL1TPZ-1+GTPZ-1+ Z-TPF- Z-TPL1C+ Z-TPG)
<0 (6)
= e(t)T[(F-L1C)+G]TZT Z-TPZ-1e(t)
+ e(t)T Z-TPZ-1Z[(F-L1C)+G] e(t) Pre and post multiplying (6) by diag {I,I,I,I,P} and replacing
-2e(t)TZ-TPZ-1ZGZିሾሺ െ ܮଵ ሻ݁ሺ ݐ Ʌሻ ݁ܩሺ ݐ Ʌ െ ሻሿ݀Ʌ
–hR2-1 by h(R2-2I) as we know –R2-1<(R2-2I) ,we get
+ h e(t)T(F-L1C)TR1(F-L1C)e(t)
௧
-௧ି ሺɅሻ ሺ െ ܮଵ ሻ் ܴଵ ሺ െ ܮଵ ሻሺɅሻ Ʌ ߗ ݄ܼ ି் ܼܲܩ ݄ܼ ି் ܼܲܩ ்݄ ܩ ݄ሺ െ ܮଵ ሻ் ܲ
ି ் ் ۍଵ ې
ێ ݄ܼ ܩ ܼܲ െ݄ܴଵ Ͳ Ͳ Ͳ ۑ
௧ି
+ h e(t)TGTR2Ge(t)-௧ିି ሺɅሻ
ଶ
ሺɅሻɅ ି ܼܲ ் ܩ ் ܼ݄ ێଵ Ͳ െ݄ܴଶ Ͳ Ͳ ۑ
ێ ۑ
ێ ݄ܩ Ͳ Ͳ െ݄ܴଶିଵ Ͳ ۑ
ିଵ
݄ܲۏሺ െ ܮଵ ሻ Ͳ Ͳ Ͳ െ݄ܴܲଵ ܲ ے
e(t)T[(F-L1C)+G]TPZ-1e(t)+ e(t)T Z-TP[(F-L1C)+G] e(t)
<0
+ି ሺሻ ܼ ି
ିଵ ் ் ିଵ
ଵ ܼ ݁ ܼܲ ܩሺݐሻdɅ
+ି ሺሻ ܼ ି்
ܴଶିଵ ܼ ் ି ܼܲ ் ܩଵ ݁ሺݐሻdɅ hR1-hP-hP+ hPR1-1P>0 then –hPR1-1P<h(R1-2P)
+ି ݁ሺ ݐ Ʌ െ ሻ் ܴ ் ܩଶ ݁ܩሺ ݐ Ʌ െ ሻ݀Ʌ
ߗ ݄ܼ ି் ܼܲܩܲ ்ି ܼ݄ ܼܩ ்݄ ܩ ሺ െ ܮଵ ሻ் ܲ
ି ் ் ۍଵ ې
ܼܲ ܩ ܼ݄ ێ െ݄ܴଵ Ͳ Ͳ Ͳ
T T T T
+ h e(t) (F-L1C) R1(F-L1C)e(t) + h e(t) G R2Ge(t) ۑ
ି ܼܲ ் ܩ ் ܼ݄ ێଵ Ͳ െ݄ܴଶ Ͳ Ͳ ۑ
ێ ۑ
െ ି ݁ሺ Ʌሻሺ െ ܮଵ ሻ் ܴଵ ሺ െ ܮଵ ሻሺ Ʌሻ Ʌ ێ ݄ܩ Ͳ Ͳ ݄ሺܴଶ െ ʹܫሻ Ͳ ۑ
݄ܲۏሺ െ ܮଵ ሻ Ͳ Ͳ Ͳ ݄ሺܴଵ െ ʹܲሻ ے
െ ି ሺ Ʌ െ ሻ
் ଶ
ሺ Ʌ െ ሻɅ
<0
850
Now let PL1=X and defining the matrix (right hand side of the equation ) as ȭ,
ஶ
ሾܸሶ ሺǡ ሻ+z(t)Tz(t)-ߛ2w(t)Tw(t)]dt<0 (7)
According to necessary condition, replacing F=A and G=Ad we get the following final LMI
851
4. Numerical Example
ͲǤͲͻͳͺ ͲǤͷͲͻͷ The transfer function from exogenous disturbances to error state
X=ቂ ቃ L1 =ቂ ቃ
ͲǤʹͳͷ ͲǤʹͺͷ outputs meets the prescribedܪஶ norm upper bound
constraintȁȁܪ௬௪ ሺݏሻȁȁஶ 0.8
Here in the example plant initial state is [5;-2] and estimator Here, we chose ߛ=0.3.
initial state is [0;0].
Solving the LMI, we can get the values as follows,
ʹǤ͵Ͷ
L1 =ቂ ቃ
ͲǤͶͺʹ
852
6. References
Here in the example plant initial state is [4;-3] and estimator
initial state is [0;0]. [1] Oliver Sename, New trends in design of observers for time-
delay systems, Kybernetika, Vol 37 no. 4(2001) 427-458.
Fig.5. Trajectories of state ݔଶ (t) and ݔොଶ (t) [11] Cao, Y.-Y., Sun, Y.-X., & Lam, J. ,Delay dependent robust
ܪஶ control for uncertain systems with time varying delays,
IEE Proceedings: Control Theory and Applications, (1998).
From the simulation result shown on graphs,we can see that the 143, 338–344.
trajectories of plant states and observer states converge within
few seconds,which is pretty good performance by the observer [12] Vladimir B. Kolmanovskii , On the Liapunov-Krasovskii
designed using the method developed in this paper. functionals for stability analysis of linear delay systems,
International Journal of Control,(1999) 72:4, 374-384.
853