Fault diagnosis of rolling bearings based on improved multiscale diversity entropy (MDE)

In multiscale diversity entropy (MDE), after multiscale coarse-grained reorganization, the data distance between the reconstructed subsequences is too close, and the difference decreases after average operation, the feature extracted by this coarse-grained method is not conducive to fault classification. To solve this problem, in this study, an improvement on the multiscale coarse-graining process by adding a sliding factor is proposed, and a reasonable range of sliding factor value is set according to the value of scale factor. Under the same scale factor, the original sequence is coarsely reconstructed by using multiple values within the value range of the sliding factor, then multiple entropy values are calculated, finally, average the MDE entropy value calculated for many times as the result. Such coarsening method avoids the distance between data too close or too far, and makes entropy calculation more accurate. The fault diagnosis framework of coarse-grained improved MDE combined with extreme learning machine (ELM) and support vector machine (SVM) respectively is constructed, and the algorithm is verified with German Paderborn bearing data set. Compared with MDE, the diagnosis accuracy of coarsegrained improved MDE combined with the same machine learning method has increased significantly.