Information theoretic fault detection

In this paper we propose a novel method of fault detection based on a clustering algorithm developed in the information theoretic framework. A mathematical formulation for a multi-input multi-output (MIMO) system is developed to identify the most informative signals for the fault detection using mutual information (MI) as the measure of correlation among various measurements on the system. This is a model-independent approach for the fault detection. The effectiveness of the proposed method is successfully demonstrated by employing MI-based algorithm to isolate various faults in 16-cylinder diesel engine in the form of distinct clusters.