Fault diagnosis of Rolling Bearing for Motor Based on LSTM-EEMD and Genetic Optimization

Abstract

In the process of ensemble empirical mode decomposition (EEMD) for motor rolling bearing time series, if the classifier is trained directly using the eigenvalues extracted from the pattern components, there are two shortcomings leading to the reduction of fault identification accuracy as follows: decomposition has serious endpoint effects; the correlation between extracted features lead to the confusion of the fault feature vector classification boundary. Aiming at the problems, in this paper, a fault diagnosis model was built. Firstly, LSTM is used to extend the original data to reduce the divergence degree of the endpoint. Secondly, correlation analysis and dimensionality reduction were carried out for the extracted features to reduce the feature dimension. Finally, the eigenvalues were weighted and genetically optimized to enhance the boundary of different eigenvectors. The experimental results validated by data sets showed that the accuracy of fault diagnosis would be improved by using LSTM extension and genetic optimization feature vector.