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

With the gradual advancement of the national “intelligent manufacturing” strategy,mechanical systems began to develop towards the direction of large-scale,complex,and intelligent.As one of the indispensable mechanical components of rotating machinery equipment,rolling bearings can be called the joint of rotating machinery.Due to long-term and high-speed operation in a complex and changeable environment,rolling bearing failure is inevitable,which may lead to the shutdown of the rotating machinery system,performance decline,and even a serious threat to people’s life and property safe.Therefore,timely and accurate fault diagnosis of rolling bearings plays an important role in ensuring the safe and reliable operation of many complex mechanical systems.Based on deep learning theories and taking rolling bearings as the research object,this paper studies and discusses the key problems existing in the field of rolling bearing fault diagnosis.The major research contents are listed below.(1)For the problem that many rolling bearing fault diagnosis methods can not make full use of the spatial and temporal characteristics of vibration signal,an intelligent fault diagnosis method based on spatial-temporal feature fusion is proposed,which is mainly composed of one-dimensional convolutional neural networks and gated recurrent neural networks.This method maximizes the spatial features of the signal through a one-dimensional convolution module,and then further extracts the temporal features of the signal using stacked gated recurrent units and completes the spatial-temporal feature fusion.In addition,the loss of information contained in the signal is reduced by the improved activation function.Through experimental verification,the proposed method has certain advantages in feature extraction ability,generalization performance,and classification ability.(2)Aiming at the problems of many parameters and difficult training of the fault diagnosis model,a rolling bearing fault diagnosis method based on the improved lightweight network is proposed.By introducing the squeeze and excitation mechanism and depth-wise separable convolution into the network,the model parameters are reduced and the useful features are enhanced.In the process of network training,the use of the generalized cross-entropy function enables the network to treat each type of sample equally,speeds up the training speed,and improves the stability of the network.Through extensive experimental analysis and verification,the proposed method has fewer model parameters,stronger robustness,and higher fault diagnosis accuracy.(3)Aiming at the issues of big sample demand,complicated model structure,and high requirements for hardware equipment in fault diagnosis,a rolling bearing fault diagnosis method based on an improved two-stream compression convolution network is put forward.By constructing a simple and compact double-branch structure,the method can fully extract signal features and reduce the complexity of the model.Using the spatial dropout mechanism reduces the interdependence between output feature maps and enhances network stability.A large number of comparative experiments have shown that this method has superior fault diagnosis performance on small samples under different loads.