A fault detection scheme for a class of distributed discrete-time non-linear systems with abrupt and incipient faults

Abstract

In this note, a consensus fault detection architecture for non-linear dynamical systems is investigated in discrete-time framework. Here, the discussed faults include abrupt and incipient faults. Moreover, the unstructured modelling uncertainty is taken into account in a large-scale system, which can be modelled as a set of subsystems, and the physical interaction between subsystems is described by non-linear functions, which can be learned on-line by neural networks. A network of local fault detectors (LFDs) is built so that each LFD monitors a single subsystem. For some overlapping components of subsystems, a co-operative estimator (CE) is embedded to avoid the consensus-less or non-co-operative case with respect to the LFDs. With the help of the designed CE, LFDs can be allowed to decide collectively on the presence of faults and the capability of detecting faults affecting overlapping variables may be improved by receiving more local diagnostic information. In addition, the derivation of rigorous analytical results for detectability properties is provided. Simulation results are given to show the effectiveness of the proposed scheme.