Active fault diagnosis based on consistencies to a class of hybrid systems by using genetic algorithms and markov decision process

Systems are growing in complexity and the fault diagnosis process requires using interdisciplinary methods to improve the diagnosis performance in continuous and hybrid systems, particularly when uncertainties affect the diagnosis results. In this work we apply an active diagnosis to a class of hybrid system, which is designed in two parts. First, we use a Genetic Algorithm (GA) to find the proper Analytical Redundancy Relations (ARR) based on the minimal test equation support and structural model analysis over a bipartite graph. These ARR are used as residual generation in a consistency based diagnosis. In the second part, an active diagnosis based on a Markov Decision Process (MDP) is used to get an optimal policy of actions driving the system to the most informative operation points, to minimize the possible ambiguity in the passive fault diagnosis due to existing uncertainties in the system. The active diagnosis scheme is verified on a two interconnected tanks system under a set of faults, controlled by a model matching strategy. Several discrete faulty states were identified; some of them being ambiguous states. The obtained ARR fulfills the diagnosis properties and the confidence in the diagnosis under ambiguity situations was improved.