Research On Bearing Fault Diagnosis Based On Convolutional Neural Network

With the rapid development of intelligent manufacturing,mechanical equipment has become the key equipment in the field of production.As an essential part of most mechanical equipment,bearing status significantly affects the normal operation of mechanical equipment.Therefore,accurate detection and identification of rolling bearing status is crucial to the normal operation of mechanical equipment.In this paper,bearing is taken as the research object,and features of bearing signals encoded as images are extracted and classified through convolutional neural network,so as to carry out research on bearing fault diagnosis methods.The main research contents of this paper are as follows(1)In view of the low test accuracy and insufficient data set in the multiclassification of bearing fault diagnosis by traditional fault diagnosis methods,a bearing fault diagnosis model based on Gramian Angular field(GAF)method and convolutional neural network(CNN)is proposed,and a 20% overlapping sampling method is adopted to expand the data set.The bearing vibration signals were transformed by the Gram angular field method to construct the data set,which was imported into the constructed six-layer convolutional neural network with batch normalization and random inactivation to realize fault classification.The test accuracy and anti-noise performance of different bearing data sets and different data lengths were tested in the constructed CNN.The results show that,in the testing of different data sets,the highest test accuracy of the constructed model can reach 100% in the five classification of bearing faults.The constructed CNN has good performance and high accuracy in multiple classification problems.According to the comparison of test results before and after data set expansion,the method of expanding data set is feasible and can effectively improve the test performance of the model.(2)The bearing fault diagnosis model of CWT-CNN is proposed to solve the problems such as slow convergence speed and poor anti-noise performance of GADFCNN method.The influence of the size of the time-frequency graph on the training was analyzed.The maximum test accuracy of the time-frequency graph dataset with 500-hour sampling length could reach 100% in the ten classification of bearing faults in the established CNN.When the signal to noise ratio of 6d B Gaussian white noise was added into the original data,the maximum accuracy of the data set constructed by conventional methods still reached 99.01% in 100 rounds of testing,which has stronger anti-noise ability compared with the classic CNN bearing fault diagnosis model based on CWTCNN.In this paper,the rolling bearing data set was expanded by three times vertical random clipping of the time-frequency graph generated by continuous wavelet transform,and then introduced into an improved convolutional neural network for feature extraction to realize bearing fault classification.In order to test the model performance,the bearing data set of Case Western Reserve University was used for testing.The experimental results show that: Compared with the data set constructed by conventional methods,the data set constructed by the proposed method converges faster in the training of the constructed convolutional neural network,and the performance of the trained model is more stable.Compared with the data set constructed by conventional methods,the test accuracy is improved,which proves the feasibility of the data set expansion method.Figure 38 Table 10 Reference 70...