Wear State Classification of Bearing Based on Deep Belief Network Using Combined Online Oil and Vibration Monitoring Data

With the purpose of improving the wear state classification accuracy of bearing based on combined online oil and vibration monitoring data, this paper proposes a new method to classify the wear state of bearing based on the K-means clustering and deep belief network (DBN). First, we carried out several groups of full life test for bearings on a bearing life test bench to obtain the online oil parameters such as wear metal debris, oil viscosity, temperature, density, and permittivity and vibration data during the whole test. The data are pretreated and normalized to form the wear state classification input data. Then, K-means method is used to cluster the input data to obtain the training label with validation for further wear state classification. A five-layer DBN feature extraction model is constructed and trained using the labeled input data. The model is used to extract features from combined online oil and vibration monitoring data and to classify the wear status of other tested bearings. Finally, the experimental results indicate that the DBN model can effectively classify the wear state of bearings. Compared with the support vector machine (SVM) method previously used, the accuracy of classification has slightly improved.