Automated fault accommodation for discrete-time systems using online approximators

Fault detection and accommodation (FDA) is still a challenging and difficult task for complex systems even when an acceptable model is available. Different model based fault detection and accommodation schemes in continuous-time recently have been developed in this literature. By contrast, in this paper, the FDA framework is developed for nonlinear discrete-time systems. In this framework, the changes in the system dynamics due to incipient failures are modeled as a nonlinear function of state and input variables while the time profile of the failures is assumed to be exponentially developing. A fault is detected by monitoring the system states and reconstructing the failure dynamics using online approximators. The online approximator output is used for detection and fault accommodation. A stable adaptation law in discrete-time is developed not only to characterize the faults but also for controller reconfiguration. The stability of the FDA algorithm is shown by extensive mathematical analysis.