Rolling Bearing Diagnosis Based on CNN-LSTM and Various Condition Dataset

Abstract

Flaking is typical failure mode in rolling bearings. Therefore, flaking diagnosis plays a critical role in condition monitoring of general rotating machinery. In recent years, there has been an increasing interest in deep learning technique for bearing flaking diagnosis, because it can learn the flaking induced vibration features with no information of bearing specifications nor that of rotating speed. However, most of the studies have only focused on laboratory data using one test rig as well as a small dataset under the limited operating condition. Accordingly, no discussion has been found on the generalization performance of the diagnostic model, i.e., availability for actual rotating machinery, in which vibration feature is affected by various operating conditions and unknown disturbance. In this study, more than 21,000 timeseries waveforms of normal and bearing flaking induced machine vibration were prepared from three types of test rig and three bearing types under various operating condition. And deep learning such as Convolutional Neural Network - Long Short-Term Memory (CNN-LSTM) models were applied to recognize flaking bearing vibration. The applied models trained with various condition data showed higher accuracy of various condition test data diagnosis than other models trained using single condition data. Furthermore, the applied diagnostic models also showed less accuracy degradation for test data in which additional artificial noise was imposed, than the models trained with single condition data.