Decentralized Fault Detection of Discrete-Event Systems Using Identified Models

The classical methods of fault diagnosis assume that the fault-free and the post-fault system behavior dynamics are known, which can be difficult to be obtained for complex and large systems. To overcome this problem, techniques for system Identification with the objective of fault detection have been proposed in the literature. In this case, the main idea consists in computing a monolithic fault-free model by observing the input and output signals of the system controller. However, if the system information is decentralized, then it is necessary to implement a decentralized fault diagnosis architecture. In this paper, we adapt the classical decentralized fault detection architecture, which is based on the a priori knowledge of the complete system model, by considering that the local observation models are not known and are obtained by using Identification techniques. We present a practical example to illustrate the proposed method and we show that, in this example, the decentralized fault detection of DES using identified models can be used without reducing significantly the fault detection efficiency.