Gearbox health condition identification by neuro-fuzzy ensemble

A neuro-fuzzy ensemble (NFE) model has been investigated for machinery health diagnosis. The proposed diagnosis system was illustrated by discriminating between various gear health conditions of a motorcycle gearbox. Four different health scenarios were considered in this work: normal, slight-worn, medium-worn and broken-teeth gear. Experimental results show the NFE model performs better than single neuro-fuzzy (NF) model with respect to classification accuracy, sensitivity and specificity, while the computational complexity is not increased significantly. In addition, the NF-based models are able to interpret their reasoning behavior in an intuitively understandable way as fuzzy if-then rules, which allows users to gain a deep insight into the data.