An Expert System Coupled With a Hierarchical Structure of Fuzzy Neural Networks for Fault Diagnosis

An on-line fault diagnosis system, designed to be robust to the normal transient behaviour of the process, is described. The overall system consists of an expert system cascade with a hierarchical structure of fuzzy neural networks, corresponding to a multi-stage fault detection and isolation system. The fault detection is performed through the expert system by means of fault detection heuristic rules, generated from deep and shallow knowledge of the process under consideration. If a fault is detected, the hierarchical structure of fuzzy neural networks starts and it performs the fault isolation task. The structure of this diagnosis system was designed to allow for the diagnosis of single and multiple simultaneous abrupt and incipient faults from only single abrupt fault symptoms. Also, it combines the advantages of both fuzzy reasoning and neural networks learning capacity. A continuous binary distillation column has been used as a test bed of the current approach. Single, double and triple simultaneous abrupt faults, as well as incipient faults, have been considered. The preliminary results obtained show a good accuracy, even in the case of multiple faults.