Research And Application Platform Development Of Bearing Fault Diagnosis And Life Prediction Method Based On Deep Learning

Rolling bearing is a key component in the normal operation of mechanical system.However,due to the complex working environment,large discrete life,easy damage and other problems,it will seriously affect the relevant components and even the entire mechanical system,bringing incalculable losses.With the trend of fault diagnosis and remaining useful life(RUL)prediction technology towards intelligence,and mechanical components such as bearings will produce regular vibration signals in the process of operation,the deep learning intelligent fault diagnosis and RUL prediction technology based on vibration signals has become a current research hotspot.In this paper,the XJTU-SY rolling bearing data set is taken as the research object,the advanced technology of deep learning is applied to carry out the research on bearing fault diagnosis and RUL prediction methods,and an application platform is developed to display the diagnosis and prediction results.The main research contents of this paper are as follows:(1)Extract RMS,Kurtosis and STFT SUM from XJTU-SY data set,construct bearing health indicator,and evaluate each health factor index through trendability,monotonicity and robustness.The RMS index is taken as the bearing degradation index,and the first-order differential mutation point under the threshold condition is taken as the starting point of bearing fault degradation.Therefore,the health status of bearings can be divided into two stages: health stage and failure stage.(2)The first layer of multi-scale large convolution kernel and GRU network are proposed to improve the WDCNN network model.FFT signal processing is performed on the divided vibration signals of health state and fault state as the input data of the model.The experimental results verify that the improved network model has good fault classification and noise resistance.Finally,the superior performance of the proposed model algorithm is verified by visualizing the classification process of the network layer of the training set and the test set of the T-SNE algorithm.(3)Through unsupervised learning and supervised learning,the bearing RUL prediction is studied.1DCNN-VAE model is selected for unsupervised learning,1DCNN convolution network is used in the coding stage,and one-dimensional hidden layer is selected.The data of the hidden layer is unified as the current RUL of the bearing through the sigmoid activation function,and its prediction and fitting results are also better than those of AE 、DAE and other encoder models The supervised learning selects convolutional neural network as the backbone network model,and improves the network by adding BIGRU network layer and attention mechanism.The experimental results verify that the added network model improves the prediction effect.(4)A bearing health management system application platform is developed based on Py Qt5.Its functions include user management,signal processing,fault diagnosis RUL prediction,analysis of status causes and maintenance suggestions.The above theoretical research contents are applied in a systematic way to promote the application research of rolling bearing operation process health management.