Research On Bearing Fault Diagnosis And Remaining Useful Life Prediction Based On Transformer

Bearing is an important component of various mechanical equipment.Its health status will affect the stability and life cycle of the whole mechanical equipment.The damage of bearing in the operation process will lead to safety accidents and huge economic losses.Therefore,it is of great significance to study the fault diagnosis and residual life prediction of bearing.As one of the most widely used deep learning models at present,Transformer has high computational efficiency and strong feature extraction ability,which brings new solutions for bearing fault diagnosis and life prediction.Aiming at the problem of bearing fault diagnosis and life prediction,this thesis carried out the research on fault diagnosis and life prediction methods based on Transformer model,designed and implemented an improved Transformer model suitable for bearing vibration signal feature extraction,in order to extract more sufficient fault features and obtain better fault diagnosis and life prediction results.The main research contents include:(1)Research on bearing fault diagnosis based on FD-Transformer.A Transformer model for fault diagnosis task is designed and implemented.The onedimensional vibration signal is directly used as the input data of the model,and the Encoder structure of the Transformer model is used as the main feature extraction module to extract the automatic bearing fault features,and identify different fault types.This model mainly uses the multi-head attention mechanism in Transformer model to extract features,which makes up for the lack of information between fault features of different time series in traditional neural network,and better realizes the fault diagnosis task.(2)Research on bearing fault diagnosis based on lightweight Vision Transformer.A lightweight Vision Transformer model suitable for bearing fault diagnosis task is proposed.Firstly,wavelet transform is performed on bearing data to obtain wavelet time-frequency map,and then the improved lightweight Vision Transformer model is used to extract fault features.The model adopts a lightweight convolution structure to reduce the parameters of the model,and makes full use of the advantages of Transformer model in feature extraction,so that the model can get better fault diagnosis results with less parameters.(3)Research on residual life prediction based on improved Transformer.To solve the problem of low computational efficiency in time series feature extraction of cyclic neural networks such as LSTM and RNN,an improved Transformer model for residual life prediction is proposed.First,the feature information of the data is extracted by convolution operation,then the multi-head attention mechanism is used to extract the features between different time series data,and different weights are assigned to the extracted feature information,which more fully represents the bearing degradation process.The multi-head attention mechanism can also further improve the calculation efficiency of the model.The experimental results on the PHM 2012 dataset show the effectiveness of the method.