Research On Rolling Bearing Fault Diagnosis Methods Under Multiple Operating Conditions And Imbalanced Sample Conditions

As one of the most important components in rotating machinery and equipment,rolling bearings usually work under high speed,variable load and strong noise.Whether their working state is normal will directly affect the operation of machinery and equipment,so it is necessary to carry out accurate and in-depth fault diagnosis research on rolling bearings.With the rapid development of deep learning technology,its powerful feature extraction ability and fault classification ability overcome the shortcomings of traditional fault diagnosis relying on manual feature extraction and expert experience.Therefore,based on deep learning-related methods,three neural network models are proposed in this paper,aiming at three application scenarios of rolling bearings under noise interference,variable working conditions and unbalanced data.The specific research content is as follows:(1)Aiming at the problem that the collected rolling bearings vibration signals are easily interfered by environmental noise,and in order to extract rich fault features in vibration signals and ensure the accuracy of rolling bearings fault diagnosis,a diagnostic method by Multi-Scale Convolutional Long and Short Term Memory Neural Network is proposed.Firstly,the convolution kernel with different sizes is used to extract the feature information of different scales in the bearing vibration signal.Then it is passed to the Long Short-term Memory Networks to learn the time series information and complete the fault classification by softmax.Finally,the validation was carried out through the open data set of Western Reserve University and the Machinery Fault Simulator data.The experimental results show that the diagnostic accuracy,recall rate,and F1 score of the MSCLNN model are higher than those of the horizontal comparison model under normal and noisy conditions,reflecting the superiority of the MSCLNN model in rolling bearings fault diagnosis under noisy backgrounds.(2)Aiming at the problem that the vibration data collected by rolling bearings under variable working conditions are quite different,and the original vibration data only represent the time domain characteristics but ignore the frequency domain characteristics,resulting in low diagnostic accuracy,a diagnostic method for Multi-Scale Convolutional Attention Neural Network is proposed.Firstly,the continuous wavelet transform is used to convert the vibration signal into a time-frequency map,enhancing the fault feature representation ability.Then the hybrid attention mechanism is then used to optimize the Convolutional Neural Networks.At the same time,a multi-scale feature extraction module was constructed to adaptively extract the key spatial feature information of different scales and complete the fault classification through softmax.Finally,it was verified by the time-frequency graph dataset made by the public data of Western Reserve University and the data of the MFS testbed.The experimental results show that the diagnostic accuracy of the MSCANN model with a time-frequency graph as input is better than that of the comparison model under normal and variable conditions,which verifies the fault diagnosis ability of the MSCANN model under variable conditions.(3)Aiming at the real production of rolling bearings produces much more normal data than faulty data,resulting in the construction of an imbalanced dataset,which is difficult to take advantage of in the deep learning model.Therefore,an improved fault diagnosis method of Auto-Encoder Wasserstein Deep Convolutional Generative Adversarial Network is proposed.The encoder is first introduced to pre-train real sample data distribution in this method and then shared with Generator parameters for encoding and decoding.Secondly,Wasserstein distance is introduced to solve the problem of instability and easy mode collapse in the original DCGAN training process.Finally,the imbalanced dataset was set up by the public dataset of Western Reserve University and the MFS testbed data for verification.The experimental results show that the fault diagnosis accuracy of the AE-WDCGAN balanced dataset is significantly higher than that of the unbalanced dataset and higher than that of the SMOTE and DCGAN balanced dataset,thus verifying the effectiveness of the model.