A control-theoretical fault prognostics and accommodation framework for a class of nonlinear discrete-time systems

Fault diagnostics and prognostics schemes (FDP) are necessary for complex industrial systems to prevent unscheduled downtime resulting from component failures. Existing schemes in continuous-time are useful for diagnosing complex industrial systems and no work has been done for prognostics. Therefore, in this dissertation, a systematic design methodology for model-based fault prognostics and accommodation is undertaken for a class of nonlinear discrete-time systems. This design methodology, which does not require any failure data, is introduced in six papers. In Paper I, a fault detection and prediction (FDP) scheme is developed for a class of nonlinear system with state faults by assuming that all the states are measurable. A novel estimator is utilized for detecting a fault. Upon detection, an online approximator in discrete-time (OLAD) and a robust adaptive term are activated online in the estimator wherein the OLAD learns the unknown fault dynamics while the robust adaptive term ensures asymptotic performance guarantee. A novel update law is proposed for tuning the OLAD parameters. Additionally, by using the parameter update law, time to reach an a priori selected failure threshold is derived for prognostics. Subsequently, the FDP scheme is used to estimate the states and detect faults in nonlinear input-output systems in Paper II and to nonlinear discrete-time systems with both state and sensor faults in Paper III. Upon detection, a novel fault isolation estimator is used to identify the faults in Paper IV. It was shown that certain faults can be accommodated via controller reconfiguration in Paper V. Finally, the performance of the FDP framework is demonstrated via Lyapunov stability analysis and experimentally on the Caterpillar hydraulics test-bed in Paper VI by using an artificial immune system as an OLAD --Abstract, page iv