Parametric Identification for Robust Fault Detection

The work presents some simulation results concerning the application of robust model–based fault diagnosis to an industrial process by using identification and disturbance de–coupling techniques. The first step of the considered approach identifies several equation error models by means of the input–output data acquired from the monitored system. Each model describes the different working conditions of the plant. In particular, the equation error term of the identified models takes into account disturbances (non–measurable inputs), non–linear and time–invariant terms, measurement errors, etc. The next step of this method exploits state–space realization of the input–output equation error models allowing to define several equivalent disturbance distribution matrices related to the error terms. Moreover, in order to achieve good robustness properties for a process normally working at different operating points, a single optimal equivalent disturbance distribution matrix is selected. Finally, eigenstructure assignment method for robust residual generation and disturbance de–coupling can be successfully exploited for the fault diagnosis of the dynamic process. The fault diagnosis procedure is applied to a benchmark simulation of a gas turbine process