Research On Fault Diagnosis Method Of Rolling Bearing Based On Integrated Deep Learning

With the improvement of industrial automation and the development of sensor technology,more and more attention has been paid to the condition monitoring of complex rotating machinery.Rolling bearing is one of the most important parts of rotating machinery.Its running state is very important to the normal work of mechanical equipment and even the personal safety of the operator.Therefore,it is of great practical significance to carry out the fault diagnosis research of rolling bearing.In the background of industrial big data,data acquisition and storage are more convenient,traditional fault diagnosis technology feature extraction ability is insufficient,it is difficult to effectively use the deep features of massive data.At the same time,in the actual industrial production,the rolling bearing fault data is rare and the proportion of data samples is unbalanced,which challenges the deep learning algorithm which depends on data quality.Therefore,in this paper,rolling bearing fault diagnosis technology based on integrated deep learning is studied.The main work is as follows:(1)An improved Le Net-5 model is proposed to solve the shortcomings of the traditional Le Net-5 model,such as slow training speed and poor generalization ability.Aiming at the defect of single feature of one-dimensional vibration signal,the data is transformed into two-dimensional image and input into Le Net-5 model for classification and recognition.An improved Le Net-5 model is proposed for the training speed and generalization performance of the network model.In this model,RLU function is used to replace sigmoid function to improve the convergence speed of the model,and two dropout layers are added to reduce the over fitting phenomenon of the model in high noise data sets.The experimental results show that the training speed of the improved Le Net-5 model is faster and the accuracy of the test set is improved by 15% under the bearing data set with high noise.(2)A Bagging ensemble learning algorithm based on out of bag data voting is proposed to solve the problem of insufficient generalization ability of single learner and poor performance of traditional voting strategy of Bagging algorithm.The traditional voting strategy only considers the overall performance of the base learner and ignores the difference of the classification ability of each category.In the improved voting strategy,the reliability of all the base learners for each category is obtained by calculating the evaluation index of the base learners in the out of bag data,which is used as the final voting basis.This strategy makes full use of the differences between the base learners to improve the generalization ability of the whole model and give full play to the advantages of Bagging ensemble learning algorithm.The experimental results show that: in the bearing data set with high noise,the accuracy of the improved voting strategy Bagging ensemble learning algorithm test set reaches 90%.(3)WGAN-GP is proposed to enhance the highly unbalanced sample set to improve the phenomenon that the number of fault samples is far less than that of healthy samples in actual industrial production.In this paper,a highly unbalanced sample set is constructed based on the bearing data set of Western Reserve University.On this basis,the traditional GAN,WGAN and WGAN-GP are used to expand the fault samples,and the extended sample set is used to train and improve Bagging integrated deep learning model.The experimental results show that: WGAN-GP network has the best training speed and stability,and the model has the highest test accuracy on the sample set enhanced by WGAN-GP,and the generalization performance of the model is significantly enhanced.(4)Using the mainstream rapid development framework Django,front-end framework layui and mainstream database mysql,the rolling bearing fault diagnosis system based on B/S architecture is designed.The function modules of user management,bearing data management and bearing fault diagnosis are developed to meet the basic requirements of bearing vibration data fault diagnosis.