Bearing Fault Diagnosis Based on Wavelet Transform and Residual Shrinkage Network

Abstract

Rolling bearings are frequently employed in machinery equipment, and the safely production of such equipment depend greatly on the properly work of bearing. Fault diagnosis of bearing is usually performed by means of vibration signal analysis, which are further divided into traditional analysis methods and machine learning based analysis methods. Traditional methods rely on expert knowledge, which makes it difficult to extract fault features, machine learning methods rely on huge volumes of data and need to eliminate the distraction of surrounding noise. This paper proposes a fault diagnosis method based on wavelet transform and deep residual shrinkage network. Firstly, transform the vibration signal into wavelet time-frequency map by wavelet transform method, and then expand data set by augment to create enough training samples, next, build a deep residual shrinkage network, and train the model for fault classification, thus implement the fault diagnosis task. Experiment shows that the accuracy of this model is superior to other comparable models, which reached 99.3% in low-noise environment, and 93.6% in high-noise environment.