Fault detection and diagnosis of complex engineering systems based on a NNARX multi model applied to a fossil fuel electric power plant

In this paper, we present a Fault Detection and Diagnostic method for complex engineering systems and its application to a Fossil Fuel Electric Power Plant. It is a model-based approach with a multi-variable generation of residuals of the main fault situations, organized in a characteristic matrix of fault signatures, which establishes a reference pattern to continuously evaluate the residuals generated on-line in normal operation conditions. Residuals are calculated as the difference between the measured dynamics of the plant and a reference given by a simulated multi NNARX model presented in a companion paper. In our proposal, false detections are reduced by the introduction of hierarchic memories that filter spurious faults and threats compensated by the internal loops of control. The system identifies the kind of fault and the severity of the abnormal behavior in a four level scale, from threats to imminent faults. We describe the main procedures of the method and we illustrate them with examples obtained using data of the Steam Generation and Reheating/Super-heating Subsystem from the Electric Power Plant. We also present some results of the real time application implemented in a co-simulation architecture using a high performance simulator under the main faults situations of a Steam Generator sub-system.