An improved image processing approach for machinery fault diagnosis

Abstract

Wavelet analysis has been proven to be effective in analysing non-stationary vibration signals. However, the interpretation of the wavelet analysis results, such as a wavelet scalogram, requires high levels of knowledge and experience, which remains a great challenge to practitioners in the field. Recently, the rapid development and advancement of image processing technologies have shed new light on this challenge. In this study, image features such as Harris Stephens(Harris);speeded-up robust features(SURFs);and binary, robust, invariant, scalable keypoints (BRISKs)were obtained from a red, green, and blue (RGB) colour-filtered wavelet scalogram. Each colour filter generates a set of image features from an RGB-filtered wavelet scalogram. Then, the features were utilised as inputs to the fault classifier, namely the support vector machine (SVM),for fault classification. However, there will be a situation where the classification results from the fault classifier, based on the image generated from the different colour filters, are contradictory to each other. No conclusion can thus be made in these situations. This paper employed the Dempster-Shafer (DS) theory to refine the contradicting results and provide an ultimate conclusion to the machine condition. Therefore, the proposed method has improved the fault classification accuracy from 69% to 78%.