Detection of abnormal vibration in multiworking condition bearings based on variational autoencoder

In view of the lack of abnormal samples in the abnormal detection of bearing vibration and the difficulty in improving the accuracy of abnormal detection due to insufficient consideration of working conditionsIn this paper, a multi-condition bearing vibration unsupervised anomaly detection method based on variational autoencoder is proposed. In this method, the correlation analysis between working condition data and vibration data is realized by constructing bearing state characteristic vectors, and the probability distribution model parameters of bearing vibration data are used as vibration characteristics to reduce the calculation and complexity of anomaly detection. The variational autoencoder is trained by using the normal bearing state eigenvector sample to learn the internal law between the vibration features and the working condition features in the normal bearing state eigenvector sample. Based on the similarity of the probability distribution of the vibration data of the two bearings represented by the vibration characteristics in the eigenvector before and after reconstruction, the anomaly detection index anomaly is designed to measure the difference between the internal law between the working condition characteristics and vibration characteristics of the sample to be tested and the normal sample, so as to realize the unsupervised anomaly detection of bearing vibration under multiple working conditions. Simulation results show that the proposed method can fully consider the influence of working conditions on the bearing vibration signal, and realize the unsupervised anomaly detection of bearing vibration under multiple working conditions based on the unlabeled normal bearing state data, which has good anomaly detection accuracy and anti-interference ability.