A parameter estimation method for the diagnosis of sensor or actuator abrupt faults

This paper describes a method for additive abrupt fault detection and isolation. Parameter estimation is applied off line to obtain a discrete model. The contribution of sensor or actuator faults on the output error residual is analysed. Fault occurrence is detected using a Page-Hinkley algorithm. Then the parameters of several output error models of the residual are estimated. The analysis of these models allows fault isolation. As soon as a fault is diagnosed, the measurements provided to the fault detection and isolation algorithm are modified so that the detection of a consecutive fault is allowed even if the first fault is not corrected.