Research On Fault Diagnosis And Prediction Methods Of Rolling Bearing Of Hub Testing Mechanism In Automobile Production Line

The hub testing mechanism is the equipment used to test the performance changes of vehicles at different speeds in the automobile production line.Its stability is not only related to the operation state of the whole production line,but also related to whether the vehicle performance and safety factor can be truly feed back to the testers.Rolling bearing is an important and fault prone part in the hub testing mechanism.Therefore,it is necessary to study the fault diagnosis and prediction methods of rolling bearing.The research results can guide the post maintenance work of the maintenance team.When the bearing is abnormal,it can judge the damaged part and even degree of the bearing through the accurate fault diagnosis results.Appropriate maintenance methods can effectively avoid the situation of "over maintenance" and "under maintenance",make it "maintain according to the situation",and even find the deterioration trend of the bearing in advance,so as to prevent the accident,reduce the maintenance cost and improve the production line rate,ensure production safety.This paper studies the fault diagnosis method of bearing,the prediction of bearing performance degradation and the calculation method of remaining service life.The main research contents are as follows:(1)Focusing on the problem of sample imbalance caused by less fault samples in actual production and the phenomenon that the dimension of the initial feature set extracted by the traditional fault diagnosis method is too high,which leads to the low accuracy of fault diagnosis,a bearing fault diagnosis method based on semi supervised Laplace score(SSLS)feature selection is proposed.This method uses a small number of labeled samples to reduce the workload of sample labeling on the premise of ensuring a certain accuracy of fault identification.At the same time,a rolling bearing fault diagnosis model based on SSLS,kernel principal component analysis(KPCA)and particle swarm optimization support vector machine(PSO-SVM)is proposed.Firstly,SSLS is used for feature selection of the original feature set,then KPCA is used for secondary mining of effective features,and PSO-SVM is used for fault pattern recognition.The results show that the model can effectively deal with the fault feature selection under the condition of sample imbalance.While reducing the workload of sample marking,it can still maintain a high accuracy in rolling bearing fault classification,and can accurately distinguish different fault types in bearing fault diagnosis(2)In view of the insufficient reflection of traditional fault characteristics on bearing fault information,the diagnosis results of bearing fault degree are relatively inaccurate,a fault feature extraction method based on generalized composite multi-scale diversity entropy(GCMDE)is proposed.This method can quantify the dynamic complexity of any time series,estimate the dynamic complexity of time series more accurately,and enhance the robustness of calculation results.It is an effective nonlinear dynamic feature extraction method.On this basis,a fault diagnosis model based on GCMDE,Laplacian score(LS)and PSO-SVM is established.Finally,the model is applied to the process of experimental data analysis to realize the accurate identification of different fault types and fault degrees of bearings,and verify the effectiveness and engineering practicability of the model.(3)Aiming at the problem of insufficient sensitivity of fault features to variable working condition environment and the low calculation efficiency of deep learning model in big data environment,a fault diagnosis model based on feature fusion is designed,which is called multi granularity convolution denoising auto encoder(MGCDAE).This method adds the concept of multi granularity convolution kernel to auto encoder(AE),which can increase the sparsity of the network and ensure the diversity of features.In addition,adding random Gaussian noise to the original data to increase the robustness of the feature.Then combining MGCDAE with the idea of deep learning,a stacked multi granularity convolution denoising auto encoder(SMGCDAE)is designed.In addition,in view of the phenomenon that industrial big data has higher and higher requirements for storage equipment and operation framework,the compressed sampling(CS)method is adopted to replace the data acquisition method based on Nyquist,which can greatly reduce the workload of data acquisition while retaining most of the information of the original signal,and then use the proposed SMGCDAE to automatically identify the fault mode of the compressed data.Finally,the feasibility of the compression acquisition method is verified by the experimental data,and the fault diagnosis of the type and degree of the bearing of the hub testing mechanism under different rotating speed and load conditions is realized.(4)Focusing on the scene of frequent cross work between equipment and complex mechanism in the production line,which leads to frequent concurrent faults,a two-level fault diagnosis model of rotating machinery based on support vector machine(SVM)and improved deep extreme learning machine(DELM)is proposed.Firstly,SVM is used to classify normal data and fault data.Then carry out fault pattern recognition on the fault data.Among them,the extreme learning machine auto encoder(ELM-AE)in DELM combines ELM and AE to significantly improve the training efficiency of the network.At the same time,the concept of feature space(FS)is introduced into ELM-AE,and ELM-FS-AE is proposed,which makes the extracted features have better classification effect.In addition,in the ELM classification stage,the output weight is learned through elastic network regularization.Finally,through the data collected on the high-speed rotor fault simulation comprehensive experimental platform,it is verified that the model can effectively identify the single type fault and concurrent fault of bearing,and significantly improve the training speed and test accuracy of the network.(5)In view of the problems that the bearing performance degradation index does not accurately reflect the bearing health state and there are many deviations in the estimation of the remaining service life of the bearing,and considering that the permutation entropy is sensitive to the sudden change of the signal and has high calculation efficiency,a high order difference mathematical morphological gradient spectrum permutation entropy(HMGSPE)is proposed as the performance degradation index of bearing,and the health threshold is established by Chebyshev inequality to evaluate the performance degradation state of bearing.On this basis,using the long-term dependence hidden in the time series data and fully considering the information before and after the vibration of the bearing at a certain time,this paper uses bi-directional long short term memory(Bi-LSTM)to train the remaining life model of the bearing.To sum up,this chapter puts forward the performance degradation evaluation and residual life prediction model of rolling bearing based on HMGSPE and Bi-LSTM.The experimental results show that the model can not only judge the degradation trend of the bearing as soon as possible,but also calculate the remaining service life of the bearing more accurately.Finally,the bearing fault diagnosis and prediction methods used in this paper are summarized,and the future research work is prospected.