Research On Bearing Fault Diagnosis Based On Siamese Neural Network

With the continuous development of the industrial society,the health inspection of large-scale industrial equipment is becoming more and more important.Bearings are important components in large-scale equipment,and their health status is related to the overall status and safety of the equipment.At present,there are many researches on the diagnosis of rolling bearing faults,and the methods used include deep learning methods such as convolutional neural networks.Common deep learning methods mainly rely on a large amount of data to train a classifier with better performance,but in the actual operating environment,it is difficult to collect a sufficient number of fault sample data.Therefore,how to use fewer samples to train a stable classifier is very important.Aiming at this problem,this paper proposes a bearing fault diagnosis model based on the twin neural network.This method can obtain a model with better generalization performance by training a small amount of sample data.This paper firstly proposes a 1D convolutional neural network based on time domain signals and a two-dimensional convolutional neural network that takes 2D grayscale images as input.Based on the verification on the CWRU dataset,the 1D convolutional neural network has overfitting,while the 2D convolutional neural network has high accuracy and stable performance.Therefore,the 2D convolutional neural network is selected as the sub-network of the Siamese neural network.Aiming at the problems of small number of bearing fault data samples and large number of fault categories in the actual data collection process,this paper proposes a Siamese bearing fault diagnosis model based on 2D convolutional neural network.The model consists of 2D convolutional neural networks with the same structure.The CWRU data set is divided into different training sets and test sets in combination with the fault categories.The Siamese neural network has a test accuracy of 97.86% for the fault category set A that participates in training;for the fault category set B that does not participate in training,there are also good Discrimination ability,the accuracy rate is79.82%;for all fault category set C,the test set accuracy rate is 88.94%.Bearing fault diagnosis model based on Siamese neural network can better solve the problem of lack of sample size in actual bearing fault diagnosis.However,in the actual operating environment of the bearing,there are noise interference and load changes.In order to better measure the performance of the model,based on the Siamese neural network,this paper tests the generalization ability of the model through the data of different working conditions,and adds noise to the data to test the anti-noise performance of the model by adding noise.After testing on the CWRU dataset,it can be found that the Siamese neural network model has better noise resistance and generalization.