Research On Bearing Fault Diagnosis Model Based On Deep Residual Network

As an important component of rotating machinery equipment,the health status of rolling bearing directly affects the safe operation of the entire mechanical system.When the rotating machinery equipment runs at high speed for a long time,due to the complex and changeable working environment,rolling bearings are prone to failure,which may affect the performance of the equipment at a light level,or cause safety accidents and property losses at a serious level.Therefore,it is of great practical significance to carry out the research on rolling bearing fault diagnosis.In the actual work of rolling bearing,it is often accompanied by the change of working condition and noise interference,which leads to the change of vibration signal characteristics.However,model-based and traditional data-driven diagnosis methods rely on expert experience and complex feature extraction,which is difficult to adaptively solve the problem of rolling bearing fault diagnosis under working condition change and noise interference.On the basis of analyzing and summarizing the previous research work,this paper studies the fault diagnosis model of rolling bearing under variable working conditions and noise interference based on the deep residual network.The main research contents of this paper are:1.Aiming at the problem of low fault diagnosis accuracy due to insufficient generalization ability of deep network model under variable working conditions,a Res Net diagnostic model of underlying parameters transfer based on Inception feature extension and attention mechanism is proposed.The Inception feature extension network is introduced into the deep residual network to extract the rich multi-scale features in the bearing vibration signal.The channel attention module is embedded in the residual block to improve the model’s ability to extract similar features in the same type of vibration signal under different working conditions.The two-dimensional gray images for model input are made by data preprocessing and overlapping sampling methods,and a fault diagnosis method combined with underlying parameters transfer is proposed.This method uses the source condition data to train the model,then transfers the underlying parameters to the target working condition,and uses a small amount of labeled data of the target condition to fine-tune the parameters of the full connection layer and softmax layer to realize the fault diagnosis under variable conditions.The experimental results show that the fault diagnosis performance of this method under variable working conditions is significantly improved compared with the method of directly testing the target working condition data with the model trained by the source working condition data.2.Aiming at the problem that the vibration signals of rolling bearing are interfered by noise and the global information abatement in deepened networks leads to the reduction of fault diagnosis accuracy,a DRSN diagnostic model based on hybrid attention mechanism diagnosis and dilated convolution feature extension is proposed.Firstly,the channel attention network in the residual shrinkage module is improved and a spatial attention network is introduced to construct a hybrid attention mechanism considering both the inner-channeled and cross-channeled characteristics.Through the comprehensive evaluation of the feature map,a threshold containing more comprehensive information is provided for soft thresholding operation,and the anti-noise performance of the model is improved.Secondly,the dilated convolution feature extension network is used to extract multi-scale context information.By fusing the features extracted by residual shrinkage module and dilated convolution,the global information of the bearing fault is strengthened and preserved as the fault diagnosis network is deepened.Experimental results show that compared with common convolution neural networks,the proposed diagnosis model provides a higher identification accuracy and better robustness under noise interference.In order to improve the performance of bearing fault diagnosis,this paper improves the model from two aspects of feature extension and attention mechanism based on deep residual network,and proposes two diagnosis models.Experimental results show that the proposed two fault diagnosis models can effectively improve the fault diagnosis performance under variable working conditions and noise interference respectively,which has important reference value for the deep network model to deal with the fault diagnosis task in complex environment.