Robust fault diagnosis and compensation in nonlinear systems via sliding mode and iterative learning observers

This thesis deals with issues of robust Fault Detection and Isolation (FDI) and compensation in uncertain nonlinear systems using second order sliding mode and iterative learning observers.; The problem of detecting and diagnosing actuator faults using a variable structure adaptive observer (VSAO) is first discussed. The VSAO is constructed directly based on the uncertain nonlinear system itself. The VSAO-based FDI can achieve robust fault detection and estimation. Furthermore, a second order sliding mode observer (SOSMO)-based robust fault detection in uncertain nonlinear systems is addressed. The SOSMO has the property of sharply filtering unwanted high frequency signals due to unmodelled dynamics, as the sliding surface dynamics forms a low-pass filter. The SOSMO is then extended to an uncertain constrained nonlinear system (UCNS) for fault detection and estimation, where the SOSMO can directly supply fault estimation. This makes fault isolation become easier.; An Iterative Learning Observer (ILO), which is updated online by immediate past system output errors as well as inputs, is constructed for the purpose of fault diagnosis. An automatic control reconfiguration scheme for fault accommodation using iterative learning strategy is then suggested. It is shown that the effects of disturbances can be attenuated by ILO inputs. The ILO is applied to excite an adaptive law in order to generate an additions control input to the nonlinear system. The additional control input can annihilate the effect of faults on system dynamics. ILO-based adaptive fault compensation strategy is independent from any existing strategies. It can supply fault detection, estimation, and compensation at the same time, and does not need a fault detection and isolation subsystem.; The last chapter is concerned with the design of a sliding mode observer (SMO) for a class of uncertain nonlinear differential-algebraic systems (DAS). An algorithm is developed to reconstruct the algebraic variables with a singular distribution matrix. An SMO is then designed based on the reconstructed algebraic variables to compensate the effect of disturbances on estimation error dynamics such that the estimated states including both the differential and algebraic variables can track the actual ones.