A model based fault detection and prognostic scheme for uncertain nonlinear discrete-time systems

A new fault detection and prognostics (FDP) framework is introduced for uncertain nonlinear discrete time system by using a discrete-time nonlinear estimator which consists of an online approximator. A fault is detected by monitoring the deviation of the system output with that of the estimator output. Prior to the occurrence of the fault, this online approximator learns the system uncertainty. In the event of a fault, the online approximator learns both the system uncertainty and the fault dynamics. A stable parameter update law in discrete-time is developed to tune the parameters of the online approximator. This update law is also used to determine time to failure (TTF) for prognostics. Finally a fourth order translational oscillator with rotating actuator (TORA) system is used to demonstrate the fault detection while a mass damper system is used for demonstrating the prognostics scheme.