Technical Condition Monitoring Methods to Manage the Redundancy of Systems. Part III: Nonclassical Models in Fault Diagnosis

Abstract. Redundancy management in a technical system involves a monitoring procedure to reconfigure the system as needed. The four-part survey presents modern diagnosis methods for dynamic systems as an integral function of monitoring. Part III is devoted to diagnosis methods employing neural networks, fuzzy models, structural models, set-based models, and a statistical approach. The fundamentals of creating and training neural networks to perform diagnostic functions are considered. The approach with fuzzy models is described, including general modeling rules and the features of their use in diagnosis tasks. The approach with structural models is demonstrated, including its features in failure detection. The fundamentals of set-theory methods, particularly the formalism of zonotopes, are presented. Finally, the approach based on statistical pattern recognition is briefly discussed.

Keywords: artificial neuron, neural network, fuzzy models, membership function, fuzzy clustering, structural models, Dulmage­–Mendelsohn decomposition, diagnosability, zonotopes, gradations in discrete feature space, empirical likelihood ratio estimate.