Rasearch On Fault Diagnosis And Remaining Life Prediction Of Rolling Bearings Based On Deep Learning

Rolling bearings are an important part of ensuring the efficient and reliable operation of rotating machinery,and are the core components of mechanical equipment.Its status largely determines whether the production process is efficient or even safe.Therefore,it is of great practical significance to evaluate the state of health of bearings and predict the remaining life for mechanical equipment.In this paper,the deep learning method is used to diagnose faults and predict the remaining useful life of rolling bearings,so as to achieve efficient monitoring of bearing health.The main research contents are as follows:Aiming at the problem of bearing fault diagnosis,this paper builds the basic model by stacking the double-layer bidirectional gated neural network Bi GRU as the shallow feature extractor of the deep learning model,and then optimizes the Bi GRU model through the self-attention mechanism.After testing on two datasets,it is shown that the model can accurately judge the fault of the bearing,and the accuracy rate reaches more than 99%.This paper also compares the effects of different optimizers,dense layers and dropout layers on the final prediction results,which has guiding significance for the construction of large-scale neural network models in the future and their application in bearings.Aiming at the problem of bearing remaining life prediction,this paper combines the initial feature vector obtained by the feature extraction of the one-dimensional convolution kernel and the Embedding vector encoded by the Time2 Vec method into a complete feature vector,which is used as the input of the overall deep learning framework.Control the recurrent neural network,and pass through the multi-head attention layer,and then through the dense layer.A residual module is used in the dense layer to avoid the disappearance or explosion that may occur during the gradient backpropagation process,and the model is trained in a warmup manner to accelerate the convergence of the global optimal solution of the model.All bearing conditions under all operating conditions in the XJTU-SY data set are verified.The overall results show that,for the prediction of bearing remaining life,the model proposed in this paper has extremely high prediction accuracy and relatively good stability in time series.It provides a solution paradigm for the remaining life prediction problem of subsequent unknown bearings.A multi-classification task for bearing fault identification and a regression task for bearing remaining life prediction are integrated into a health state monitoring model.According to the actual task requirements,choose to perform fault diagnosis or predict the remaining useful life of the bearing.To achieve fault diagnosis and life prediction for new bearings,subsequent tasks can load a pretrained model to train the parameters of a deep neural network.Furthermore,ensemble learning is used to optimize the model algorithm.For the above two tasks,two ensemble strategies,the majority voting method and the simple average method,are used,respectively,so that multiple models jointly determine the prediction output of the final model.The comparative experimental results show that the ensemble learning method not only improves the prediction stability and prediction accuracy,but also improves the generalization ability of the model to completely unknown distributions,which provides a basis for future applications in other industrial occasions.