Research On Bearing Fault Diagnosis Technology Based On Convolutional Neural Network In Noisy Environment

Bearings are a key component of CNC machine tools,and their health status will affect the machining accuracy of the machine tool,the life of the machine tool and the probability of accidents.Therefore,it is very important to monitor and diagnose bearing faults to ensure the smooth operation of the bearings.This paper takes rolling bearings as the object,and studies the rapid diagnosis technology of bearing faults under noise scenarios based on convolutional neural networks.The main contents of the paper are as follows:An end-to-end fault diagnosis model(CNNDM-1D)based on a one-dimensional convolutional neural network was established to realize the synchronization of bearing vibration signal feature extraction and fault classification.Carry out experiments to optimize the model structure and hyperparameters,introduce batch normalization improvements,and verify that this method can make model training faster and more stable.The performance advantage of the CNNDM-1D model is proved by comparing the classic shallow model with other deep learning fault diagnosis models.A rapid anti-noise diagnosis model based on lightweight CNNDM-1D is established.In view of the large number of parameters of the CNNDM-1D model,a lightweight CNNDM-1D model that uses local sparse structure convolution and global average pooling to replace ordinary convolution and fully connected layers respectively is proposed.Aiming at the problem of reduced model diagnosis accuracy in noisy environments,a onedimensional convolutional denoising autoencoder was designed,and the strategy of randomly destroying input data was studied to enhance the anti-noise performance of the lightweight CNNDM-1D model.Combining the lightweight CNNDM-1D model and the one-dimensional convolutional denoising autoencoder model,a fast anti-noise diagnosis model is proposed.The anti-noise fast diagnosis model proposed in this paper is used to classify and verify the local bearing data set.Based on the bearing fault diagnosis experiment system independently built by the laboratory,a local measured data set was established,and the optimal structure scheme of the lightweight CNNDM-1D model was determined using this data set.Based on the local measured data set,it was verified that the one-dimensional convolution denoising autoencoder model has good denoising performance,and the "onedimensional convolution denoising autoencoder + lightweight CNNDM-1D" anti-noise fast diagnosis model and other anti-noise models were developed.The comparison experiment of the model proves the advantages of the model in the diagnosis performance under highnoise scenes.