A multilevel recovery diagnosis model for rolling bearing faults from imbalanced and partially missing monitoring data.

Abstract

As an indispensable part of large Computer Numerical Control machine tool, rolling bearing faults diagnosis is particularly important. However, due to the imbalanced distribution and partially missing of collected monitoring data, such diagnostic issue generally emerging in manufacturing industry is still hardly to be solved. Thus, a multilevel recovery diagnosis model for rolling bearing faults from imbalanced and partially missing monitoring data is formulated in this paper. Firstly, a regulable resampling plan is designed to handle the imbalanced distribution of data. Secondly, a multilevel recovery scheme is formed to deal with partially missing. Thirdly, an improved sparse autoencoder based multilevel recovery diagnosis model is built to identify the health status of rolling bearings. Finally, the diagnostic performance of the designed model is verified by artificial faults and practical faults tests, respectively.