Distributionally Robust Active Fault Diagnosis

The main objective of active fault diagnosis is the design of separating input signals that enhance the detection and isolation of faults in modern technological systems. A major consideration when evaluating active fault diagnosis methods is robustness in the presence of modeling uncertainties. The presence of modeling inaccuracies will typically compromise the performance of the separating input signals designed for effective fault diagnosability. This work investigates a distri-butionally robust active fault diagnosis approach for nonlinear systems, which takes into consideration variation or ambiguity in the uncertain parameters of the models. A probabilistic approach is presented using total variation distance as an information constraint, and as a measure for the separation of multiple models based on the similarity of their output probability density functions. The effectiveness of the proposed approach is demonstrated through an application to a three-tank system.