Intelligent Fault Detection Of Rolling Bearings Based On Deep Learning

As the core part of mechanical equipment,the performance of rolling bearing affects the safety of the whole production equipment.Therefore,it is of great significance to study the intelligent fault detection technology of rolling bearing.Because the traditional intelligent fault detection requires rich experience and tedious feature extraction engineering,deep learning can extract features independently,so as to achieve end-to-end fault diagnosis.In this paper,one-dimensional bearing vibration data and two-dimensional bearing image data are used as input to study the bearing diagnosis problem of deep learning model in different scenes.In the intelligent fault detection of rolling bearing,the proposed deep learning model is superior to the traditional machine learning model.The one-dimensional convolution model is designed by using the Pytorch frame to make it conform to the characteristics of bearing vibration data.Considering the waveform characteristics of rolling bearing,the first layer convolution kernel of one-dimensional convolution model is designed as a larger convolution kernel 64 to obtain a larger receptive field.In order to solve the problem that the loss value is difficult to decrease during the experiment,the batchnorm layer and dropout layer are added to the one-dimensional convolution model.The results show that the accuracy of deep learning model is higher than that of traditional machine learning model.For the two-dimensional rolling bearing image data,it is proposed that improving the resolution of the bearing image is helpful to improve the accuracy;in the bearing fault diagnosis,the full convolution model is better than the general convolution model.The traditional data stitching method is improved to the fixed step data stitching method,and the one-dimensional vibration data is transformed into two-dimensional image data.In the original p-net model,a global adaptive pooling layer is added to receive image pyramid images.Improve the output port structure of R-Net and o-net models.The results show that improving the resolution of bearing image can improve the accuracy of bearing fault diagnosis;in the intelligent fault diagnosis of rolling bearing,the full convolution model is better than the general convolution model.Based on the case of small samples,a diagnosis method using feature migration and packet convolution to optimize the network model is proposed.Feature migration solves the problem that RESNET model does not converge when rolling bearing data is insufficient.Compared with ordinary convolution,block convolution has fewer parameters,so it can reduce the model complexity.The results show that feature transfer and RESNET model can improve the accuracy of rolling bearing fault diagnosis,and the memory space of RESNET model optimized by group convolution is smaller.Aiming at the problem of rolling bearing fault diagnosis when the data is unbalanced,a cascade model is proposed to solve the problem.Aiming at the problem that the loss value of Gan network does not converge,the cascade model is improved by improving the loss function of GAN network.The results show that the detection accuracy of the cascade model is higher than that of the single model in the case of unbalanced data,and the effect of the improved cascade model is better.