Research On Bearing Fault Diagnosis Algorithm Based On One-dimensional Convolutional Neural Network

As a key component of rotating machinery,rolling bearings have been widely used in industrial manufacturing,especially in some heavy machinery with rotating machinery.Traditional bearing intelligent diagnosis algorithms,such as statistical analysis methods,mostly use prior knowledge and expert experience to carry out personalized analysis of bearing fault signals.Most of today’s bearing fault diagnosis methods use intelligent sensors to obtain bearing vibration signals as the data source of the diagnosis algorithm.Smart sensors are usually deployed at the free and fixed ends of rotating machinery.With the continuous advancement of industrialization and digitization,smart sensors have begun to be deployed to terminal equipment on a large scale in the field of industrial manufacturing,and the scale of data collected by smart sensors is getting larger and larger.In contrast,some traditional bearing intelligent diagnosis algorithms have been unable to cope with the current bearing fault diagnosis needs in the era of big data.With the further development of computer technology,many experts and scholars have turned their attention to the algorithm model based on deep learning and applied it to bearing type fault diagnosis.The method proposed in this paper also adopts the algorithm model based on deep learning.This paper studies the bearing fault diagnosis algorithm from the following three aspects: The first is to process the original vibration signal of the bearing,without relying on artificial prior knowledge and expert experience,directly input the time domain signal collected by the sensor into the model and combine it with the introduction of regularization,batch normalization,and dropout neural network.To optimize the method,a singlechannel 1D convolutional neural architecture is designed.Experiments show that the model has good diagnostic performance in small samples and small batches of data.The second is to combine the design ideas of a multi-scale convolutional neural network and first-layer wide convolutional neural network to design a multi-channel model based on the pyramid concept for feature recognition.The model uses the same number of convolution kernels with different widths in the first convolutional layer to extract the global information and local details of the convolutional regions,respectively,and then perform feature fusion.Experiments show that this network structure has better performance than the single-channel network structure.better one.Third,in order to enhance the robustness of the model,based on the multichannel network design,the design concept of a residual network is introduced.Finally,it is proposed that the one-dimensional convolutional neural network model based on residual and multi-channel designed in this paper can achieve a 100% fault diagnosis and recognition rate for the public data of Case Western Reserve Bearing Data Center.And achieve 100% fault diagnosis and recognition rate in the case of additive white Gaussian noise.The model has good anti-noise ability and generalization performance in bearing fault diagnosis experiments under mixed conditions.