This thesis describes a new approach to fault detection and diagnosis using a sliding mode observer with no chattering. A sliding observer is designed to estimate the state of the system and generate residuals for fault detection and diagnosis. To achieve this purpose, the residuals generated by the sliding mode observer must be robust with respect to parametric and model uncertainties or errors, and reduce the effects of noise and disturbance. We used a sliding observer from the literature that guarantees such robustness. Earlier fault detection studies using sliding mode observers require n-order differentiation of input u, which is not easily realized in a practical system. A detailed description of the sliding mode observer for fault detection and diagnosis is given in the thesis including: (1) Conventional structure observer combined with smooth switching within a thin boundary layer around the sliding surface to reduce chattering. (2) Residual generation based on an estimation error sliding surface for fault diagnosis and detection.Design, analysis, and simulation results demonstrate the effectiveness of the new fault detection methodology. The results of the study indicate that fault detection and diagnosis using a sliding mode observer are realized in the presence of model uncertainties, noise and disturbances. |