Gear fault detection based on adaptive wavelet packet feature extraction and relevance vector machine

Abstract

In this article, a novel fault detection method based on adaptive wavelet packet feature extraction and relevance vector machine (RVM) is proposed for incipient fault detection of gear. First, ten statistical characteristics in time domain and all node energies of full wavelet packet tree are extracted as candidate features. Then, Fisher criterion is applied to evaluate the discrimination power of each feature. Finally, two optimal features from time domain and wavelet domain, respectively, are selected to be used as inputs to the RVM. Furthermore, moving average is applied to each feature to improve accuracy for online continuous fault detection. By combining wavelet packet transform with Fisher criterion, it is able to adaptively find the optimal decomposition level and select the global optimal features. The RVM, a Bayesian learning framework of statistical pattern recognition, is adopted to train the fault detection model. The RVM was compared with the popular support vector machine (SVM) with the increase of training samples. Experimental results validate the effectiveness of the proposed method, and indicate that RVM is more suitable than SVM for online fault detection.