Research On Rotating Machinery Fault Diagnosis Method Based On Convolutional Neural Network

With the continuous improvement of rotor speed in modern equipment,rotating machinery,as the "joint" of machine equipment,has a more obvious impact on the efficiency and reliability of equipment.Therefore,the further research on the fault diagnosis method of rotating machinery is of great significance to ensure the safety and effectiveness of industrial production.The study of fault diagnosis problems is based on practical problems.At present,the manufacturing industry has stepped into the era of intelligence.One of the problems caused by this is that the operation difficulty of the traditional diagnosis method based on model analysis and qualitative empirical knowledge analysis is greatly increased.At present,the overall development trend of fault diagnosis technology is to take the road of intelligent industrial big data decision based on the guidance of data science.Based on the above development trend,this paper takes rotating machinery as the research object and proposes three fault diagnosis models based on convolutional neural network:(1)In view of the difficulty of traditional methods in fault extraction and low recognition accuracy,an improved one-dimensional convolutional neural network based on Alex Net was proposed.On the one hand,a normalization layer is added after each convolution layer of Alex Net network to improve the operation speed and accuracy of the network.On the other hand,the redundant structure generated by adding normalization layer to the network is removed to prevent the network from overfitting.The experimental results show that the proposed model has better performance.(2)One-dimensional convolutional neural network with multi-scale fusion is proposed.Based on Alex Net’s improved model,this network integrates multi-scale fusion convolution structure to extract complex fault features of rotating machinery more completely.Then,the performance of the model is obviously improved by using the bearing data of Western Reserve University.Then,the proposed model has good recognition ability under multiple faults and variable working conditions,and the generalization ability of the proposed model needs to be improved through 197726 wheelset bearing fault data.(3)A one-dimensional convolutional neural network fused with residual blocks is proposed,which fuses the convolutional residual blocks on the basis of multi-scale fusion convolutional neural network to improve the accuracy and anti-noise performance of the network.The performance of the optimized network was verified by 197726 wheelset bearing fault data,and the simulated signal with noise interference was established based on the bearing data of Western Reserve University.The simulated fault signal with weak noise interference was preprocessed as the test set data,and the noise resistance and identification accuracy of the proposed network model were compared.