Study On The Rolling Bearing Fault Diagnosis Based On EMD And SVD

Fault feature extraction is directly related to the reliability and accuracy in the fault diagnosis of rolling bearings. However, the fault signal is mixed with complex background noise and presents nonlinear and non-stationary characteristics, how to accurately extract the fault feature is always the key point and difficulty in related research. Apply the means of empirical mode decomposition and singular value decomposition, improvement and supplement them on the basis of the existing methods, using the multiple noise reduction method to extract the fault feature in this paper. The main research works can be described as follows:Firstly, the study of rolling bearing fault diagnosis based on empirical mode decomposition. It has studied the theory and process of EMD, some issues must be resolved especially the end effects of EMD. It has analyzed the causes of the end effects of EMD and found the most effective way to inhibit it is the boundary extension by the waveform matching. Then the pre-processing method of stationarity test for boundary signals is put forward to increase the scope of application of the boundary extension method of waveform match based on distance. At last, in the process of waveform matching, it has improved the method by considering the waveform’s appearance and the extension’s trend. It has be verified the validity of the improved EMD method through the experiment.Secondly, the study of rolling bearing fault diagnosis based on singular value decomposition. It has studied the theory and process of SVD, analyzed the difference spectrum theory and found its defectiveness on the selection of singular values. It has proposed a selected double singular value method to reconstruct the signal that based on studying the properties of singular values in specific Hankel matrix. It has be verified the validity of the improved SVD method through the experiment.At last, the study of extracting the fault feature based on EMD-SVD. It has combined the EMD and the SVD, it can remove the most distracting information and extract the fault feature accurately by applying different ways to reduce noise twice. It has be verified the validity of the method in this paper.