Hierarchical architecture and its implementation oriented toward multi-faults diagnosis

When multiple faults arise simultaneously in a machine, their fault symptoms are quite complicated, not simple superposition of symptoms of multiple single faults. Moreover, faulty vibration signals often show themselves nonstationary. All these make multi-faults diagnosis difficult. In this paper, a hierarchical architecture for multi-faults diagnosis was proposed. Based on combined of mutual information maximization and gray relation analysis, relational characteristics between multi-faults to be recognized and single potential faults were quantitatively evaluated to obtain a set of candidate faults. Then wavelet transformation energy moments were extracted from the candidate faults and the multi-faults. They were used for training and testing a self-organizing map network. Finally, coupled features were decoupled and separated in feature space by the self-organizing map network with powerful performance on topologically mapping. At the same time, multi-faults was recognized. Experimental results show that the proposed architecture is effective on multi-faults diagnosis. It is capable of sorting the recognized single faults according to their severity, under concurrency of multiple faults, which helps making a reasonable maintenance decision. © 2018, Science Press. All right reserved.