Intelligent Bearing Fault Diagnosis Based On Deep Learning Under Multiple Conditions

Rolling bearings are one of the most widely used mechanical parts in rotating machinery,and they play a very important role in the operation of the machinery.According to statistics,about 30% of mechanical failures in rotating machinery are caused by rolling bearing faults.The rolling bearing fault will reduce equipment performance in the short term,and its long-term development may cause serious damage to the equipment and result in safety accidents.Therefore,Carrying out equipment fault diagnosis has very high engineering application value and significance.This paper propose three effective rolling bearing fault diagnosis models which are based on the deep learning method in the field of machine learning.And these models have considered the background of a variety of common working conditions in practical applications.These models consider a variety of common operating conditions in practical applications.Since deep learning methods can provide end-to-end intelligent diagnosis modes without expert experience in manual feature extraction,many scholars have introduced deep learning methods into the field of fault diagnosis and achieved high accuracy.However,the prerequisite for these methods to achieve good results is often that the working conditions during data collection are relatively stable,and the amount of data is sufficient and ideal.In fact,there are various working conditions of rotating machinery in industrial applications,among which speed changes and even continuous fluctuations are common.Moreover,the amount of data in practical applications is often not as easy to obtain as in experiments,and it is often necessary to carry out fault diagnosis in the case of small samples.In order to improve the actual industrial application value of the models,this paper proposes three fault diagnosis models to solve the fault diagnosis problems in the above situations.The main works of this paper is as follows:(1).An Intelligent Fault Diagnosis Method Based on Industrial Small Sample Data:In order to deal with the insufficient sample size in actual industrial fault diagnosis,this paper proposes a 1D-MDense Net(One Dimensional-Multiscale Densely Connected Convolutional Networks)for intelligent fault diagnosis of small samples of bearing data under stable conditions.In this method,the multi-scale design of the convolution kernel of each densely connected block of Dense Net is carried out,and the stability of the network connection structure is adjusted,so that the network has strong generalization ability and stability.The overall network parameters are small and can be well adapted to one-dimensional vibration signal analysis in industrial engineering applications without manual feature extraction.Through the comparison of experimental design and three other methods,this paper verifies the superiority of the proposed 1D-MDense Net in the case of stable working conditions and small samples,and its practical industrial application prospects are better.(2).Intelligent fault diagnosis method under variable speed conditions: In order to accurately diagnose the faults of bearings running under variable speed conditions in practical engineering applications,this paper proposes an intelligent fault diagnosis model based on RR-ACNet(Resampling and Reorganization Asymmetric Convolution Network).This model can effectively reduce the difference of the signal model caused by the speed change,and more emphasizes the rotation position information of the shaft during rotation.And the ACNet used in this model strengthens the spatial information recognition ability of the original network without consuming additional resources.Such a combined model can better handle fault diagnosis tasks under changing speed conditions.The entire process of the model is automatically realized by code,without manual pre-processing,and it is completely dependent on data-driven and computer real-time processing for intelligent fault diagnosis,which can meet the needs of intelligent fault diagnosis in industrial applications.(3).Cross-domain intelligent fault diagnosis method under fluctuating speed conditions: Focusing on solving the problem of cross-domain fault diagnosis under the condition of fluctuating speed,this paper proposes a cross-domain intelligent fault diagnosis method based on RM-Res Net(Rotating Speed Normalization Multiscale-Residual Network).This method preprocesses the collected fluctuating speed signals for speed normalization,and inputs the processed data into the multi-scale fusion Res Net with good domain adaptability for fault diagnosis to obtain better fault diagnosis performance.This paper proves through experiments that this method has better domain adaptability and higher accuracy than traditional deep learning methods under fluctuating speed conditions,and has certain practical engineering application value.