Bearing Fault Diagnosis Method Based on EMD-CNNs

Abstract

The fundamental purpose of our newly proposed approach is to extract distinguishing features because feature extraction is one of the most important steps in mechanical fault diagnosis. In the feature extraction process, time domain feature analysis is a traditional feature extraction method of statistics. When the vibration signals are no stationary and nonlinearly, EMD[1] techniques may have better performance than traditional techniques. Also, EMD is a self-adaptive processing method, which means less manual work. Unfortunately, engineering and interpreting such features requires a significant level of human expertise. To enable non-experts in vibration analysis, the overhead of feature engineering for specific faults needs to be reduced as much as possible. In this paper, an empirical mode decomposition (EMD) transform method with time do maim feature extraction is used, which is combined with convolutional neural networks (CNNs)[2] is proposed to establish a new fault diagnosis method based on EMD-CNNs. Furthermore, the new combined feature-learning fault diagnosis method is compared to a feature-engineering approach based using the EMD and time domain feature extraction method, the results of our proposed approach are also compared with works in some other papers, which illustrates that, our method based on EMD-CNNs is more effective and the accuracy of fault diagnosis is higher.