Rolling Bearing Fault Feature Extraction and Diagnosis Method Based on MODWPT and DBN

Abstract

In order to enhance the performance of bearing fault identification, a fault feature analysis and recognition model of rolling bearings, combined with maximum overlap discrete wavelet packet transform (MODWPT) and deep belief network (DBN), is proposed in this paper. The execution of the proposed procedure is divided into four steps, including signal processing, features extraction, features reduction and patterns recognition. In the first step, by using Maximum overlap discrete wavelet packet transform (MODWPT) method, the initial statistical characteristics, which can represent the fault condition under different frequency bands, were calculated from reconstructed signals obtained by decomposing original vibration signal. Due to the information redundancy of the initial statistical characteristics, in the following steps, this paper presents a novel features selection method that combines the Dunn Validity Index (DVI) and Standard deviation (STD) based on the selection of the most sensitive features. In addition, with the advantage of deep learning methods in high-dimensional data processing and nonlinear data analysis, this paper used deep belief network (DBN) method to analyze the fault characteristics of rolling bearings and identify the fault status. Finally, four cases with 12 working conditions respectively have been employed to evaluate the performance of the proposed procedure, where vibration signals were freely provided by Bearing Data Center of Case Western Reserve University. In our cases study, the results show that the FSDVSD method can effectively screen out the fault-sensitive features. The DBN model can improve the accuracy of fault status recognition. To summarize, the effectiveness, adaptability and superiority of the proposed method of which the fault diagnosis thought can serve as an intelligent bearing fault diagnosis system.