Resrarch On Fault Diagnosis Method Of Packing Machine Rolling Bearing Based On Transfer Autoencoder

With the continuous development of manufacturing industry in our country,the packaging industry gradually becomes intelligent and modernization.The stable operation of packaging machine directly affects the speed and precision of packaging,and further affects the quality of packaged products.So the maintenance of the packaging machine and its related components is particularly important.Rolling bearing is the core component of the transmission system of packaging machine,and bearing fault account for the highest proportion of packaging machine fault.This article takes the rolling bearings of packaging machines with variable operating conditions and speed changes as the research object,and collectes its vibration data,a method for fault diagnosis of packaging machine rolling bearings is studied.Bearing faults can be timely detected and diagnosed which type of fault has occurred,and maintenance can be arranged in a timely manner to avoid further deterioration of the fault and greater economic losses.It is of great significance to improve the packaging speed of packaging machine and reduce the maintenance loss of mechanical equipment.The main research contents and results of this paper are as follows:(1)Packaging machine rolling bearing vibration signal characteristics and failure mechanism analysis.Based on the operating characteristics and working environment of the packaging machine and the frequency components contained in the vibration signal,the characteristics of the vibration signal caused by the packaging machine production and bearing faults are analyzed respectively.Using the basic rolling bearing parameters,the theoretical characteristic frequency of the outer ring fault,inner ring fault,roller fault and cage fault is calculated.By analyzing the collected rolling bearing vibration data of the packaging machine test,the characteristic frequency component of rolling bearing faults in the vibration signal is verified from the perspective of timefrequency analysis.(2)Research on noise cleaning and multi-domain primary feature extraction processing of rolling bearing vibration data.Based on the vibration signal characteristics of packaging machine rolling bearings,the application efficiency of wavelet denoising method in vibration data cleaning is studied,and preliminary denoising processing is carried out on vibration data;The feature information of vibration data is extracted from multiple fields,including primary feature parameters in time domain,frequency domain,and time-frequency domain,reducing the masking of interference factors on rolling bearing fault feature signals and improving the accuracy of feature extraction of rolling bearing vibration data.(3)Research on fault diagnosis method for packaging machine rolling bearing based on stacked convolutional autoencoding network.For the packaging machine rolling bearing fault diagnosis difficult problem,a deep learning network based on 5-layer convolutional autoencoder and Soft Max classifier(named as SCAE)in series is proposed and used to achieve deep feature extraction and fault classification diagnosis of the data.The experiments are validated with CWRU bearing dataset which operating conditions is similar to the packaging machine bearing’s.And the SCAE model is trained with multiple feature quantities in time domain,time-frequency domain,and time-frequency domain,and the parameters are optimized by combining with BP backward transfer algorithm.Experimental results show that the SCAE model has the best classification performance when the frequency domain features are used as input.And the diagnostic accuracy remains above 90% under noise interference.The results indicate that SCAE has good fault classification ability and anti-interference capability.(4)Research on the fault diagnosis method of rolling bearing of packaging machine based on transfer stacked convolutional autoencoding model.As for the complexity of the actual working conditions of the packaging machine,there is a problem of uneven spatial distribution between actual rolling bearing vibration signal and the data obtained from the experimental platform.Therefore,based on the SCAE classification network,transfer learning is introduced to build a transfer learning stack convolutional autoencoding network(TL-SCAE),which maps the source domain data feature volume and the destination domain data feature to the high-dimensional Hebert space,and aligns feature distribution differences with the goal of minimizing the maximum mean difference,so as to reduce the distribution differences between source domain features and destination domain features.Comparing the classification performance of the SCAE model and the TL-SCAE model,it is found that the TLSCAE network has a significant effect in dealing with the problem of fault diagnosis and classification of rolling bearings in actual working conditions of packaging machines when the amount of data is insufficient.After migration and improvement,the TL-SCAE model has better generalization ability and adaptability than the SCAE model.The research has important theoretical significance and practical application value for realizing the fault diagnosis of the rolling bearing of the automatic packing machine and ensuring the safe and stable operation of the automatic packing machine.