Fuzzy-based refinement of the fault diagnosis task in industrial devices

This paper first describes a fuzzy classifier to be used for fault diagnosis. Then, the paper presents a refinement of the diagnosis task performed with this fuzzy classifier. For each fault, a number of 20 levels of fault strength have been considered. In previous work, more than one single category per fault has been used to improve the classifier performance, i.e. distributing the strength levels into small, medium and, respectively large strength subsets. However, this distribution scheme is too rigid. This paper introduces a flexible distribution scheme that takes into account the (di)similarities between different strength levels. The refinement proposed here offers better insight on the behavior of each fault and it increases separation between overlapping faults, which improves the final outcome of the diagnosis process.