Observer Based Parameter Estimation With Applications To Model Based Diagnostics

Presented in this manuscript is an observer-based method for consistent parameter estimation and the utility of these estimates in real-time model based diagnostics. Parameter estimation is realized through a calculation involving the state estimates produced from the combination of two infinite time Luenberger observers with a delay. The significance of this approach is the ability of the coupled observers to have their state estimates converge in fi nite time to the true state values. Unique to this formulation is that consistent parameter estimates can be obtained using data limited to one sampling event provided the converged estimated states at that sampling interval meet a full rank condition similar to persistancy of excitation. The observer based parameter estimation method is then integrated with Information Synthesis to create a one-step-ahead fault detection, isolation and estimation knowledge base. The utility of the proposed method is demonstrated on a numerical example and an experimental application of engine onboard sensor diagnostics.