A Hybrid Qualitative and Quantitative Diagnosis Approach

Abstract

Parametric faults are detected and isolated using parameter tracking algorithms based on optimization algorithms or filtering techniques (e.g., Kalman filter, particle filter). Online, simultaneous tracking of all parametric faults can fails since there may be too many combinations of parameter values that explain the observed behavior. Hence, a correct diagnosis solution is not obtained. An alternative in the single fault case is to track separately each parametric fault in parallel and choose the one that best explains the observed behavior according to some chosen metric (e.g., mean square error). This approach is feasible but computationally expensive, since there may be too many tracking algorithms running in parallel. We propose using analytic redundancy relations (ARRs) to reduce the number of parametric faults that are tracked simultaneously. ARRs qualitatively point to a set of possible explanations but usually require a large number of sensors to achieve good isolability of fault causes. They induce a fault signature matrix (FSM) that can be derived offline. The parameter tracking algorithms will be instantiated for the faults in the set of possible explanations produced by the ARRs. By combining ARRs with online parameter tracking algorithms we can obtain a good tradeoff between computational effort and fault isolability. We demonstrate our approach by diagnosing faults in a rectifier circuit.