Research On Intelligent Diagnosis Method Of Rolling Bearings Under Varying Conditions Based On Improved Convolutional Neural Network

Rolling bearing is one of the most frequently used mechanical parts in rotating machinery.Long-term operation at high speed and full load makes it one of the most vulnerable mechanical parts.The operating condition of the bearing has an extremely important impact on ensuring the normal operation of the entire mechanical system.Therefore,accurate fault diagnosis is essential to ensure the safe and reliable operation of mechanical equipment.In the past,traditional machine learning theories began to weaken the contribution to humans and brought the era of artificial intelligence into fault diagnosis.In recent years,data-driven faults diagnosis technology based on deep learning has received more and more attention due to its powerful feature learning capabilities.However,the fault diagnosis of bearings has problems such as small samples,strong noise and variable working conditions.How to overcome these problems is the key to accurate fault diagnosis of bearings.This paper takes the bearing as the research object,based on the deep learning method to enhance the bearing fault feature extraction ability and sample data as the main line,aiming to solve the problem of bearing fault diagnosis under small samples,strong noise and variable operating conditions.The main research contents of this paper are as follows:(1)In order to enhance the ability of deep learning methods to extract bearing fault features,this paper proposes two fault diagnosis methods.One is the fault diagnosis method based on STFT and CNN.This method inputs the two-dimensional image transformed by STFT into CNN,and the result shows that the combination of traditional method and deep learning can effectively identify bearing faults.The second is to build an improved convolutional neural network based on Dense Net and expanded convolution.This model adds direct connections to different network layers,which greatly improves the feature transfer of different network layers.In addition,the introduced expansion convolution also improves the convolution receptive field without increasing the amount of calculation,and strengthens the extraction of fault features.Compared with the current fault diagnosis method,this method has better anti-noise ability and domain adaptability.(2)In order to solve the problem of insufficient number of fault samples and unbalanced samples,a data generation method based on GAN is proposed.The model uses the excellent feature extraction capabilities of CNN to generate simulated data that conforms to the original sample distribution of the generator.The combined experiment of simulated data and original data shows that the original data can be identified by training the simulated data,indicating that the data generated by this method conforms to the distribution of the original sample,and can be used as an extended sample of the original sample to solve the problem of insufficient sample.(3)In order to further to verify the fault diagnosis ability of the method proposed to this paper under variable operating conditions,a fault data collection experiment was carried out with the wheelset bearing as the object.The experimental data set a number of different working conditions,including vertical loads,lateral load and speed.Under different vertical load conditions,the domain adaptability of the improved convolutional neural network of variable load conditions is verified.In addition,the sample generation ability of CGAN was verified again under different load conditions.The data sets used in this article include the bearing data set and wheelset bearing data set of Western Reserve University,and all methods have been verified through experiments.This article concludes with a summary of existing problems and future work.