Intelligent fault diagnosis of rotating machinery based on multiple relevance vector machines with variance radial basis function kernel

Abstract

Automatic and accurate fault diagnosis is very important for condition-based maintenance. In this study, an intelligent fault diagnosis method based on relevance vector machines (RVM) is proposed for automatic fault diagnosis of rotating machinery. First, the global optimal features from all node energies of full wavelet packet tree are obtained by combining wavelet packet transform with an improved Fisher feature selection method. Individual salient feature subsets are selected for each pair of classes separately. Then, RVM method is adopted to train the intelligent fault diagnosis model. The multi-class RVM classifier is constructed by combining several RVM binary classifiers using ‘max-probability-win’ strategy. Moreover, improved from Gaussian radial basis function, a new kernel function denoted variance radial basis function is developed and used for RVM to adaptively balance the difference between the scales of different features. The proposed method was carried out to develop a multi-class bearing fault diagnosis model under varying load conditions, resulting in high accuracy around 99.58 per cent. Experimental results demonstrate that the proposed method is promising for intelligent fault diagnosis of rotating machinery.