Research On Rolling Bearing Fault Diagnosis Method Based On Improved Convolutional Neural Network

Rolling bearings are one of the most frequently used components in machinery and equipment,and its complex working environment causes it to be prone to failure,if a rolling bearing failure is not diagnosed in time,it may cause very serious consequences.Traditional fault diagnosis requires the use of manual experience to extract fault characteristics,and fault diagnosis accuracy is low.As a typical network in deep learning,convolutional neural network has a strong ability to extract information features,but the nonlinear mapping ability of convolutional neural network in processing feature information is insufficient;it is vulnerable to noise interference and different working conditions;and the hyperparameters of convolutional neural network need to be set artificially will bring uncertainty.The following work has been done in this paper to address the above issues:Aiming at the complex diagnosis process and low diagnosis accuracy of traditional fault diagnosis methods;the insufficiency of nonlinear mapping of convolutional neural network,this chapter proposes a rolling bearing fault diagnosis model based on MOE-CNN.The model is a single-channel model with a hybrid expert network layer added after the fully connected layer of the model,where multiple experts will learn the input data features simultaneously and each expert will choose their own weights based on the gating module.After experimental validation,the nonlinear mapping capability of the model is enhanced by adding a hybrid expert network layer,and the model can directly implement end-to-end bearing intelligent fault diagnosis without complex manual feature extraction.For the weak noise immunity and low generalization ability of the convolutional neural network model,this chapter proposes a rolling bearing fault diagnosis model based on CBAM-MSCNN.The convolutional neural network model structure of a single channel is extended to multiple channels,and the key features of the original signal of the bearing are extracted more comprehensively by convolutional kernels of multiple sizes on multiple channels,and the CBAM attention mechanism is added to each channel to strengthen the model’s attention to important information and avoid the interference of irrelevant information through the CBAM attention mechanism,and a gating module is equipped behind each channel to reduce the diagnostic errors caused by the excessive weight of a particular channel.The experimental results show that the model has strong generalization ability as well as noise immunity.A rolling bearing fault diagnosis model based on FSSA-optimized CBAMMSCNN is proposed for the problems of difficult hyperparameter determination of convolutional neural network model and unstable network diagnosis performance.Firstly,the vibration signal is time-frequency transformed by continuous wavelet transform,then the FSSA optimization algorithm is used to find the optimal hyperparameters such as the size,number,learning rate and batch size of the convolution kernel in the model,and finally the optimal parameters found by the optimization algorithm are used to reconstruct the model and conduct fault diagnosis experiments.The experimental results show that the model can avoid the uncertainty of artificially set hyperparameters and has good diagnostic performance.