Distributed fault detection in large-scale systems using hybrid extended information filter

A fault monitoring system plays an important role to ensure and improve the reliability of an industrial plant to operate safely and efficiently. For large-scale systems, their high-dimensional nature impels the designer to develop a lower-order algorithm to overcome the practical limitations. To reduce the order of computations and the amount of communication, the system should be decomposed into low-dimensional sub-systems. For this purpose, a distributed fault detection algorithm was presented based on hybrid extended information filter. This algorithm has the advantages of two existing approaches, the hybrid extended Kalman filter and the information filter. In the proposed method, several local fault detectors were employed to monitor the local sub-system instead of a single centralized monitoring node. For each detector, a limited number of measurements were accessible based on the generated local state and residual vectors using a local filter. The local detector extracted the faults' information content of the residual signals via CUSUM statistical hypothesis testing. Moreover, it could communicate the processed information of its neighbors if needed. The algorithm was implemented on alkylation of benzene process plant. The results obtained from the distributed algorithm showed a considerable decrease in computational burden and communication bandwidth.