Model-based fault diagnosis and prediction for a class of distributed parameter systems

This paper deals with a novel model-based fault diagnostics and prognostics scheme for distributed parameter systems (DPSs) expressed by a series of partial differential equations (PDEs). Assume that system states are available, an observer is developed based on the PDE model of the system and to compare the detection residual, which is characterized as the different value between the output of the physical system and the observer, with a predefined threshold a fault can be detected. Then, the fault dynamics is approximated and its parameters are learned by a proposed update law using system state information. The parameter magnitudes together with the tuning update law are used to estimate the time to failure (TTF). Two output filters and one input filter are proposed to relax the demand of system state measurable. Finally, the act of the state and filter based diagnosis and prognosis scheme is demonstrated by using a heated rod with an actuator fault.