Distributed fault detection for a class of network systems: Optimal unknown input observer design

Abstract

This paper is concerned with the fault detection problem for a class of network systems composed of multiple clusters with unknown system matrices. Each cluster consists of multiple subsystems and the connections between clusters are unmeasurable. For these unmeasurable connections in network systems, traditional system identification and fault detection methods may be difficult to be directly applied. To solve this problem, the subspace instrumental variable method is proposed under the distributed framework, which utilizes the intersection of subspaces on local observations as the states of connections to further identify the local cluster subsystem matrices. Based on the result of identification, the unknown input observer (UIO) is then designed to detect the faults of local cluster systems. However, these connections also lead to the rank conditions for designing UIO not being satisfied. Thus, the unknown input decomposition approach is presented to address this problem, such that the decouplable part is eliminated from the error system and the impact of undecouplable part is attenuated by robust performance index. Finally, the effectiveness and advantages of the proposed fault detection scheme are verified via numerical simulation and comparative analysis.