System identification and fault diagnosis using dynamic functional-link neural networks

The paper investigates the application of new neural networks with internal dynamics to Fault Detection and Isolation (FDI). The suggested dynamic functional-link structures are used to design efficiently different neural observes schemes by means of system identification. Structured sets of residuals are thus generated based on the one-step ahead prediction errors. A first case study refers to the component FDI of a three-tank laboratory system. A second investigation concerns the sensor FDI of an evaporator sub-process from a sugar factory. In both experimental applications, static neural networks are used to classify the generated symptoms.