Research On Fault Diagnosis Method Of Rolling Bearing With Small Samples Based On Deep Learning

Rotating machinery is one of the most important components in modern industrial production and is widely used in modern industrial production equipment such as aerospace and high-speed trains.Therefore,online monitoring and real-time diagnosis of rolling bearings,the core component of such devices,is of vital great significance to ensure the safety and reliability in the operation of the devices.In recent years,datadriven intelligent fault diagnosis methods have been widely used.However,in actual industrial production,rotating machinery usually operates in a fault-free state,and the collected fault samples are extremely limited,resulting in the over-fitting or underfitting of the trained fault diagnosis model with low reliability.Therefore,how to make the model achieve better and stable classification results under small sample conditions has become an urgent problem to be solved.Based on the above problems,this paper carries out research work on the classification method of rolling bearing fault data based on convolution neural network and transfer learning with rolling bearing as the research object.The main contents are as follows:(1)To address the problem of low accuracy of traditional fault diagnosis methods for rolling bearings under small sample conditions or variable working conditions,a fault diagnosis model based on Markovian transfer field and multi-dimensional convolutional neural network is proposed.Firstly,this paper uses Markov transition field to perform dimensional transformations,which preserve the temporal correlation in time-series images;then,a multi-dimensional attention mechanism is proposed to extract feature channel information and location information simultaneously,and it is embedded into the multi-dimensional convolutional neural network together with the proposed E-Relu activation function to construct the MDCNN model,which can accelerate the convergence of the model while fully extracting the fault feature information;finally,the two-dimensional images are input into the model and the classification results are output.The proposed method is experimentally verified on the Case Western Reserve University bearing data set and the MFS bearing data set,and the experimental results show that the proposed method has higher fault diagnosis accuracy and stronger generalization ability under variable-working conditions and small sample conditions.(2)Aiming at the problem of poor fault diagnosis performance of rolling bearings under small samples and variable working conditions,a small sample rolling bearing fault diagnosis method based on self-calibrated coordinate attention mechanism and multi-scale convolution neural network is proposed.The self-calibrated coordinate attention mechanism generates different weight values on the channel dimension and the spatial dimension through two parallel information generation paths,and increases the receptive field of the network through self-calibrated operation,which effectively improves the feature recognition,and eliminates the problem of location information loss caused by global average pooling operation;At the same time,a novel multi-scale convolution neural network structure is constructed,which greatly reduces the number of parameters of the neural network model while extracting multi-scale feature information,thus saving the training time of the model.The proposed method is verified on two datasets for fault diagnosis,and the experimental results show the effectiveness of the method.(3)In order to solve the serious shortage of rolling bearing fault training samples in practical applications,a small sample rolling bearing fault diagnosis model based on improved residual neural network and transfer learning is proposed.Firstly,the model embeds squeeze-and-excitation network into one-dimensional residual neural network,which increases the feature extraction ability of the model;Secondly,the improved residual neural network model is trained using source domain data to determine the structure and parameters,and L2 regularization and Dropout mechanism are used to suppress over-fitting problem;Then,transfer learning is introduced to freeze some of the model parameters that have been trained using source domain data,and a small amount of target domain data is used to fine-tune the fully connected layer parameters of the model;Finally,the samples of different faults are classified.The method is verified by experiments on two different source datasets,and the experimental results show that the proposed method has higher fault diagnosis accuracy and stronger robustness ability compared with the experimental results of other methods under different experimental conditions.