AAA Abstract and Applied Analysis 1687-0409 1085-3375 Hindawi Publishing Corporation 10.1155/2014/935216 935216 Research Article Krein Space-Based H Fault Estimation for Discrete Time-Delay Systems http://orcid.org/0000-0003-4690-5588 Song Xinmin 1 Yan Xuehua 2 Shen Bo 1 School of Information Science and Engineering Shandong Normal University Jinan Shandong 250014 China sdnu.edu.cn 2 School of Electrical Engineering University of Jinan Jinan Shandong 250022 China ujn.edu.cn 2014 1482014 2014 27 05 2014 21 07 2014 14 08 2014 2014 Copyright © 2014 Xinmin Song and Xuehua Yan. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

This paper investigates the finite-time H fault estimation problem for linear time-delay systems, where the delay appears in both state and measurement equations. Firstly, the design of finite horizon H fault estimation is converted into a minimum problem of certain quadratic form. Then we introduce a stochastic system in Krein space, and a sufficient and necessary condition for the minimum is derived by applying innovation analysis approach and projection theory. Finally, a solution to the H fault estimation is obtained by recursively computing a partial difference Riccati equation, which has the same dimension as the original system. Compared with the conventional augmented approach, the solving of a high dimension Riccati equation is avoided.

1. Introduction

Krein-space theory has proven to be an effective tool in dealing with the indefinite quadratic control/filtering problems. Some recent researches on H filtering have led to an interesting connection with Kalman filtering in Krein space [1, 2]. Comparing with the linear estimation approaches in Hilbert space, the Krein-space theory can lead to not only less conservative results but also computationally attractive algorithms. It has been shown in  that a finite horizon linear estimation problem can be cast into a problem of calculating the minimum point of a certain quadratic form. By applying linear estimation in Krein space, one can calculate recursively the minimum point via Riccati equation. In  the authors consider the H prediction problem for time-varying continuous-time systems with delayed measurements in the finite horizon case. The necessary and sufficient condition for the existence of an H predictor is obtained by applying a reorganized innovation approach in Krein space.

On the other hand, fault estimation is one of the most important issues. The paper  designs a fuzzy fault detection filter for T-S fuzzy systems with intermittent measurements, and all the results are formulated in the form of linear matrix inequalities. In , a sufficient condition for the existence of a fault filter is exploited in terms of certain linear matrix inequality. Reference  is concerned with the robust fault detection problem for a class of discrete-time networked systems with distributed sensors. The existence of the desired fault detection filter can be determined from the feasibility of a set of linear matrix inequalities. The paper  addresses the fault detection problem for discrete-time Markovian jump systems. The characterization of the gains of the desired fault detection filters is derived in terms of the solution to a convex optimization problem that can be easily solved by using the semidefinite program method. As for the fault estimation problem, the Krein-space approach has received much attention so far . A Krein-space approach is presented in  to H fault estimation for LDTV system, where the augmented approach  is also used. Different from , a more further result is obtained by a Krein-space approach and nonaugmented approach for the same problem in . Recently, by applying Krein-space approach and reorganized innovation approach,  considers the finite-horizon H fault estimation for linear discrete time-varying systems with delayed measurements . Finite-horizon H fault estimation for uncertain linear discrete time-varying systems with known inputs is considered in .

Recently, we note that time-delay systems have received much attention . For optimal estimation problem, when the delay appears in state, the reorganized innovation approach is not suitable for estimation problem. Motivated by this point, we consider the finite-horizon problem H fault estimation for linear discrete systems with time delay, where the delay appears in both state and measurement, which contain  as a special case. To the best of our knowledge, this problem has not yet been investigated, and this constitutes the primary motivation for our research. On the other hand, we desire to obtain the necessary and sufficient condition for the existence of an H fault estimator. A natural idea is to use Krein space to deal with the finite-horizon H fault estimation for linear time-delay systems, and this gives rise to another motivation of our work. The main contributions of the paper are highlighted as follows. (i) The necessary and sufficient condition will be derived for fault estimation problem with time delay. (ii) Compared with the augmentation approach , our result on estimation is given based on a partial difference Riccati equation, and hence the solving of an high dimension Riccati equation is avoided.

The organization of this paper is as follows. The problem statement is given in Section 2. Section 3 presents the fault estimator design in terms of a partial difference Riccati equation. A numerical example is given to demonstrate the effectiveness of the approach in Section 4, and the paper is concluded in Section 5.

Notation. Throughout this paper, a real symmetric matrix P>0 (≥0) denotes P being a positive definite (or positive semidefinite) matrix. I denotes an identity matrix of appropriate dimension. The superscripts “-1” and “T” represent the inverse and transpose of a matrix. Rn denotes the n-dimensional Euclidean space. Rn×m is the set of all n×m real matrices. δij=0 for ij and δii=1. For stochastic vectors α and β, inner product α,β equals the covariance matrix of α and β. θ(k)l2[0,N] means k=0NθT(k)θ(k)<. Matrices, if the dimensions are not explicitly stated, are assumed to have compatible dimensions for algebraic operations.

2. Problems Statement

Consider the following linear systems with time delay: (1)x(k+1)=i=0dAix(k-hi)+Bdd(k)+Bff(k),(2)y(k)=i=0dCix(k-li)+Dff(k)+v(k), where x(k)Rn, d(k)Rp, and f(k)Rr are the state, the driving disturbance, and the fault to be estimated, respectively. Also, y(k)Rm and v(k)Rm are measurements and noises, respectively. Without loss of generality, the delays are assumed to be of an increasing order: 0=h0<h1<<hd, 0=l0<l1<<ld. Moreover, it is assumed that d(k), f(k), and v(k) belong to l2[0,N]. For simplicity of presentation, we assume that Ai, Bd, Bf, Ci, and Df are constant matrices even though the later development and results can be easily adapted to the time-varying case.

Problem. Given the observation {y(0),,y(N)}, seek a fault estimator r(k) (k=0,,N) such that (3)sup(x0,w)0k=0Nr(k)-f(k)2(x0-x^0)TΠ0-1(x0-x^0)+k=0Nw(k)2<γ2, where γ is a given positive scalar, Π0 is a positive definite matrix, and (4)w(k)=[dT(k)fT(k)vT(k)]T. Without loss of generality, the initial state estimator x^0 is assumed to be zero. The value of x(-k) is assumed to be zero, where 1kτ, τ=max(hd,ld), E{x(-i)xT(-j)}=0.

Define the fault estimation error between r(k) and f(k) as (5)vf(k)=r(k)-f(k), and introduce the following quadratic form: (6)JN=x0TΠ0-1x0+k=0Nw(k)2-γ-2k=0Nvf(k)2. Obviously, the H performance (3) is satisfied if and only if JN>0 for all (x0,w(k))0.

3. Main Result

We consider constructing an equivalent Krein-space problem to the minimum for JN. To do so we need to introduce the following stochastic systems in a Krein space: (7)x(k+1)=i=0dAix(k-hi)+Bdd(k)+Bff(k),(8)y(k)=i=0dCix(k-li)+Dff(k)+v(k),(9)r(k)=f(k)+vf(k), with (10)[x(0)d(k)f(k)v(k)vf(k)],[x(0)d(j)f(j)v(j)vf(j)]=diag(Π0,Iδkj,Iδkj,Iδkj,-γ2Iδkj). Let yr(k)=[yT(k)rT(k)]T; then the linear space generated by the measurements in the Krein space up to time N can be written as (11)L{yr(k),0kN}. It is readily known that yr(k) satisfies (12)yr(k)=i=0dCrix(k-li)+Dfrf(k)+vr(k),(13)Cri=[Ci0],Dfr=[DfI],vr(k)=[v(k)vf(k)].

In the sequel, we denote the Krein-space projection of m(k) onto L{{yr(j)}j=0s} by m^(ks). Construct the innovations (14)y~r(k)=yr(k)-y^r(kk-1). Defining x~(kj)=x(k)-x^(kj), we further have (15)y~r(k)=i=0dCrix~(k-lik-1)+Dfrf(k)+vr(k). Furthermore, we define the cross-covariance matrices of the state estimation error: (16)P(i,j,k)=E{(x(i)-x^(ik))(x(j)-x^(jk))T}. By employing the Krein-space theory, a necessary and sufficient condition for the existence of the desired H fault estimator is given in the following.

Lemma 1.

Consider the stochastic systems (7)–(9). For a given γ>0, a fault estimator r(k) achieving the performance (3) exists if and only if (17)Θ(k)=i=0dj=0dCiP(k-li,k-lj,k-1)CjT+I+DfDfT>0,Ξ(k)=(1-γ2)I-DfTΘ-1(k)Df<0. Furthermore, if the above conditions are satisfied, the desired H fault estimator is given by (18)r(k)=DfTΘ-1(k)y~(k), where y~(k)=y(k)-y^(kk-1), and the minimum of the quadratic form JN is (19)minJN=k=0Ny~T(k)Θ-1(k)y~(k).

Proof.

It can be seen from (12) and (15) that (20)Ry~r(k)=y~r(k),y~r(k)=i=0dj=0dCriP(k-li,k-lj,k-1)CriT+DfrDfrT-γ2I=[Θ(k)DfDfT(1-γ2)I]. Equation (20) can be further written as (21)M(k)Ry~r(k)MT(k)=Λ(k), where (22)M(k)=[I0-DfTΘ-1(k)I],Λ(k)=[Θ(k)00Ξ(k)]. We can draw the conclusion from (17) and (21) that Ry~r(k) and Rvr(k) have the same inertia. Therefore following the same line as in , the minimum value of JN can be obtained as follows: (23)minJN=k=0Ny~rT(k)Ry~r(k)-1y~r(k)=k=0N[y~T(k)rT(k)-y~T(k)Θ-1(k)Df]×[Θ-1(k)00Ξ-1(k)][y~(k)r(k)-DfTΘ-1(k)y~(k)]. Thus the rest of the proof is clear.

In the following, we are devoted to the estimator design in terms of the solution to a partial difference Riccati equation.

Lemma 2.

The state estimate is recursively calculated as (24)x^(k+1k)=i=0dAix^(k-hik),x^(k-jk)=x^(k-jk-1)+Kj(k)y~r(k)=x^(k-jk-1)+i=0dP(k-j,k-hi,k-1)CiTΘ-1(k)y~(k),j=0,,l, where the initial values are x^(-j-1)=0 and Kj(k) can be calculated by (25)Kj(k)=i=0dP(k-j,k-hi,k-1)CriTRy~r(k)-1,Ry~r(k)=[Θ(k)DfDfT(1-γ2)I], while P(,,) is calculated by the partial difference Riccati equation as (26)P(k-i,k-j,k)=P(k-i,k-j,k-1)-Ki(k)Ry~r(k)KjT(k),0ijτ,(27)P(k+1,k-j,k)=i=0dAiP(k-hi,k-j,k),0jτ,(28)P(k+1,k+1,k)=i=0dj=0dAiP(k-hi,k-hj,k)AjT+BdBdT+BfBfT,(29)P(k-i,k-j,k)=PT(k-j,k-i,k),(30)P(-i,-j,-1)=P0(-i,-j),0ij,0jτ.

Proof.

Applying projection theory, we have (31)x^(k+1k)=i=0dAix^(k-hik),(32)x^(k-jk)=x^(k-jk-1)+Kj(k)y~r(k), where Kj(k) is given as (33)Kj(k)=E{x(k-j)y~rT(k)}Ry~r(k)-1. Noting that x(k-j)=x^(k-jk-1)+x~(k-jk-1), then based on (15) one has (34)Kj(k)=E{x~(k-jk-1)y~rT(k)}Ry~r(k)-1=E{x~(k-jk-1)[i=0dCrix~(k-lik-1)i=0d+Dfrf(k)+vr(k)]Taaaaaaaaaaaaaaaaaaaai=0d+Dfrf(k)+vr(k)]T}Ry~r(k)-1=i=0dP(k-j,k-li,k-1)CriTRy~r(k)-1. Then one has (35)x^(k-jk)=x^(k-jk-1)+Kj(k)y~r(k)=x^(k-jk-1)+i=0dP(k-j,k-li,k-1)CriTRy~r(k)-1y~r(k)=x^(k-jk-1)+i=0dP(k-j,k-li,k-1)×[CiT0]MT(k)Λ-1(k)M(k)y~r(k)=x^(k-jk-1)+i=0dP(k-j,k-li,k-1)CiTΘ-1(k)y~(k).

Next it follows from (7) and (32) that (36)x~(k-jk)=x~(k-jk-1)-Kj(k)y~r(k),x~(k-ik)=x~(k-ik-1)-Ki(k)y~r(k). Based on (36), we have the estimation error covariance matrices (37)P(k-i,k-j,k)=E{x~(k-ik)x~T(k-jk)}=E{x~(k-ik-1)x~T(k-jk-1)}+E{Ki(k)y~r(k)y~rT(k)KjT(k)}-E{x~(k-ik-1)y~rT(k)}KjT(k)-Ki(k)E{y~r(k)x~T(k-jk-1)}=P(k-i,k-j,k-1)-Ki(k)Ry~r(k)KjT(k), which is (26). Combining (7) and (31), we obtain (38)x~(k+1k)=i=0dAix~(k-hik)+Bdd(k)+Bff(k). Furthermore, one has (39)P(k+1,k-j,k)=E{x~(k+1k)x~T(k-jk)}=E{[i=0dAix~(k-hik)+Bdd(k)aaaaaaaai=0d+Bff(k)]x~T(k-jk)}=i=0dAiP(k-hi,k-j,k),P(k+1,k+1,k)=E{x~(k+1k)x~T(k+1k)}=E{[i=0dAix~(k-hik)+Bdd(k)i=0d+Bff(k)]i=0d+Bdd(k)+Bff(k)]Taaaaaaaa×[i=0dAix~(k-hik)aaaaaaaaaaaai=0d+Bdd(k)+Bff(k)]T}=i=0dj=0dAiP(k-hi,k-hj,k)AjT+BdBdT+BfBfT. Finally (29) is straightforward by virtue of the definition of (16). Thus the proof is completed here.

Theorem 3.

Consider the system (1)-(2). For a given γ>0, a fault estimator r(k) that achieves the performance index (3) exists if and only if Θ(k)>0 and Ξ(k)<0, where Θ(k) and Ξ(k) are defined in Lemma 1. In this case, one possible finite-time H fault estimator is given by (40)r(k)=DfTΘ-1(k)y~(k)=DfTΘ-1(k)[y(k)-i=0dCix^(k-lik-1)], where (41)x^(k+1k)=i=0dAix^(k-hik),x^(k-jk)=x^(k-jk-1)+i=0dP(k-j,k-li,k-1)CiTΘ-1(k)y~(k),aaaaaaaaaaaaaaaaaaaaaaaaaaaaaj=0,,l.

Proof.

According to , we can see that the fault estimation problem addressed for the deterministic systems (1) with (2) and (5) is partially equivalent to that for the stochastic systems (7) with (8) and (9) in a Krein space, and therefore the proof is readly and we omitted here.

Remark 4.

In fact, the problem mentioned in this paper can be converted into the problem in [7, 8] by applying augmented approach. However, due to the existence of time delay, we need to solve a high dimension Riccati equation. Here the solutions to the fault estimator can be obtained by solving partial difference Riccati equations (26)–(28), which have the same dimension as the original system (1). Therefore solving a high dimension Riccati equation is avoided. Here we present a simple explanation. Because the multiplications and divisions cost much more in computation than additions, hence we only use the number of multiplications and divisions as the operation count. Denote Caug and Cnew as the numbers of multiplications and divisions for augmentation method and our proposed approach in one step, respectively. According to , one can see that the order of hd in Caug is 3, while the order of hd in Cnew is 2. Therefore if hd is large enough, CaugCnew.

Remark 5.

Recently, by applying Krein-space approach and reorganized innovation approach,  has considered the finite-horizon H fault estimation for linear discrete time-varying systems with two-channel single measurement delay. In this paper, we have investigated the finite-horizon problem H fault estimation for linear discrete systems with time delay, where the delay appears in both state and measurement, which contain  as a special case.

4. Numerical Example

Consider the linear discrete-time system: (42)x(k+1)=[0.30.500.4]x(k)+[0.20.1-0.050.2]x(k-1)+[0.40.1-0.50.3]x(k-2)+[0.50.4]d(k)+[1.21.8]f(k),y(k)=[-0.50.5]x(k)+[0.50]x(k-1)+[0.7-0.3]x(k-2)+2.5f(k)+v(k).

The finite time horizon concerned here is [0,100]. The driving disturbance and measurement noise are selected as d(k)=0.4cos(k), v(k)=0.6sin(k). The fault to be estimated is assumed to be (43)f(k)={1,10k25,50k70,0,others. Set initial values as x0=[1-0.5]T, P(i,j,-1)=0, -2i-1, P(i,j,-1)=0, -2j-1, P(0,0,-1)=I. By using the result given in Theorem 3, the desired fault estimator is designed with γ=0.85. Figure 1 shows the fault and its estimate, which confirm that the designed estimator performs very well.

Fault and its estimation.

5. Conclusion

The finite-time H fault estimation problem for linear time-delay systems has been investigated. The design of finite horizon H fault estimation has been converted into a minimum problem of certain quadratic form. Then an stochastic system in Krein space has been proposed, and a sufficient and necessary condition for the minimum has been derived by applying innovation analysis approach and projection theory. Finally a solution to the H fault estimation has been obtained by recursively computing a partial difference Riccati equation. Compared with the conventional augmented approach, the presented approach lessens the computational demand when the delay is large. In the further study, we will consider the H fault estimation problem for linear time-delay systems with multiplicative noise.

Conflict of Interests

The authors declare that there is no conflict of interests regarding the publication of this paper.

Acknowledgments

This work was supported in part by the National Natural Science Foundation of China (61104126, 61304013, and 61170145), the Doctoral Fund of Ministry of Education of China (20113704120005), and the Excellent Young Scholars Research Fund of Shandong Normal University.

Hassibi B. Sayed A. H. Kailath T. Indefinite Quadratic Estimation and Control: A Unified Approach to H2 and H Theorie 1998 SIAM Studies in Applied Mathematics Series Zhang H. Zhang D. Xie L. An innovation approach to H prediction for continuous-time systems with application to systems with delayed measurements Automatica 2004 40 7 1253 1261 10.1016/j.automatica.2004.02.016 MR2148321 2-s2.0-2442541007 Zhao Y. Lam J. Gao H. Fault detection for fuzzy systems with intermittent measurements IEEE Transactions on Fuzzy Systems 2009 17 2 398 410 10.1109/TFUZZ.2009.2014860 2-s2.0-65549138999 Du D. Jiang B. Shi P. Fault detection for discrete-time switched systems with intermittent measurements International Journal of Control 2012 85 1 78 87 10.1080/00207179.2011.639087 MR2879988 ZBL1282.93255 2-s2.0-84855943261 He X. Wang Z. Ji Y. D. Zhou D. H. Robust fault detection for networked systems with distributed sensors IEEE Transactions on Aerospace and Electronic Systems 2011 47 1 166 177 10.1109/TAES.2011.5705667 2-s2.0-79551642005 Dong H. Wang Z. Gao H. Fault detection for Markovian jump systems with sensor saturations and randomly varying nonlinearities IEEE Transactions on Circuits and Systems. I. Regular Papers 2012 59 10 2354 2362 10.1109/TCSI.2012.2185330 MR2983257 2-s2.0-84867317492 Zhong M. Liu S. Zhao H. Krein space-based H fault estimation for linear discrete time-varying systems Acta Automatica Sinica 2008 34 12 1529 1533 10.3724/SP.J.1004.2008.01529 MR2502863 2-s2.0-58149235364 Zhong M. Zhou D. Ding S. X. On designing H fault detection filter for linear discrete time-varying systems IEEE Transactions on Automatic Control 2010 55 7 1689 1695 10.1109/TAC.2010.2046921 MR2675833 2-s2.0-77954553372 Shen B. Ding S. X. Wang Z. Finite-horizon H fault estimation for linear discrete time-varying systems with delayed measurements Automatica 2013 49 1 293 296 10.1016/j.automatica.2012.09.003 MR2999975 2-s2.0-84871307738 Shen B. Ding S. X. Wang Z. Finite-Horizon H fault estimation for uncertain linear discrete time-varying systems with known inputs IEEE Transactions on Circuits and Systems II: Express Briefs 2013 60 12 902 906 10.1109/TCSII.2013.2285969 Shen B. A survey on the applications of the Krein-space theory in signal estimation Systems Science and Control Engineering 2014 2 1 143 149 Ding S. X. Shen B. Wang Z. Zhong M. A fault detection scheme for linear discrete-time systems with an integrated online performance evaluation International Journal of Control 2014 10.1080/00207179.2014.930183 Xiao L. Hassibi A. How J. P. Control with random communication delays via a discrete-time jump system approach 3 Proceedings of the American Control Conference June 2000 Chicago, Ill, USA 2199 2204 2-s2.0-0034540045 Chen B. Yu L. Zhang W. Robust Kalman filtering for uncertain state delay systems with random observation delays and missing measurements IET Control Theory & Applications 2011 5 17 1945 1954 10.1049/iet-cta.2010.0685 MR2906241 2-s2.0-80755150107 Lam J. Zhang B. Chen Y. Xu S. Reachable set estimation for discrete-time linear systems with time delays International Journal of Robust and Nonlinear Control 2013 10.1002/rnc.3086 Zhang B. Zheng W. X. Xu S. Filtering of Markovian jump delay systems based on a new performance index IEEE Transactions on Circuits and Systems I: Regular Papers 2013 60 5 1250 1263 10.1109/TCSI.2013.2246213 MR3055140 2-s2.0-84877765359 Sun S. Linear minimum variance estimators for systems with bounded random measurement delays and packet dropouts Signal Processing 2009 89 7 1457 1466 10.1016/j.sigpro.2009.02.002 ZBL1178.94153 2-s2.0-62749199111 Liu S. Xie L. Zhang H. Infinite horizon LQR for systems with multiple delays in a single input channel Journal of Control Theory and Applications 2010 8 3 368 374 10.1007/s11768-010-0039-2 MR2834031 2-s2.0-77954925562 Zhang H. Song X. Xie L. ‘Stage-by-stage’ optimization approach to optimal control for general time-delay systems Proceeding of the 8th IEEE International Conference on Control and Automation (ICCA '10) June 2010 Xiamen, China 154 159 10.1109/ICCA.2010.5524117 2-s2.0-77957875735