Multiple fault diagnosis by signature recognition of time-varying residuals

A Fault Detection and Diagnosis scheme able to deal with concurrent, incipient, sensor and actuator faults is presented. The architecture allows the diagnosis whenever the system's outputs are less than the number of faults. Residual generation is achieved by taking advantage from observer-based design, whereas the classification is performed by extending the concept of directional residual towards a time-varying setting. The scheme is designed to leverage the power from both model-based and data-driven approaches while mitigating their inherent drawbacks. The performances of the proposed strategy are evaluated by employing real data coming from the TEKOB1 Thermal Plant of Kostolac, Serbia