Research On Fault Diagnosis Method Of Rolling Bearing Based On Transfer Learning

Rolling bearings are very common and important in People’s Daily life.The safety of their running state directly determines production efficiency and profitability.Rolling bearings generally run in harsh environments,and their running state is always unstable,and the changes of working conditions are more difficult to distinguish.Time-frequency analysis can better classify the faults of bearings.Faced with the following two questions:(1)against the variable condition of the problems of unstable,we through continuous wavelet transform and multi-scale neural network algorithm to carry out the solution,(2)in the deep learning of the past,training a good neural networks often need a lot of good data to achieve a good network.However,in the actual industrial environment,it is difficult to collect enough label data,especially the lack of fault data,so the problem of small samples also needs to be discussed.Transfer learning algorithm is becoming more and more popular and more and more important in the field of rolling bearing fault diagnosis.The following introduces the relevant contents of transfer learning algorithm research in the case of rolling bearing under varying conditions and small samples.For the problems of variable working conditions and small samples,the transfer learning algorithm has unique advantages,so it is very necessary to carry out the research of rolling bearing fault diagnosis methods.(1)Research on multi-representation depth domain adaptive rolling bearing fault diagnosis method based on transfer learningBecause of the changes of rotational speed and load of rolling bearings in bad conditions,the vibration signals of bearings are often not stable,which makes fault feature extraction very difficult.Both traditional machine learning and deep learning require data features to be independent and distributed,but in actual production and life,it is difficult to ensure their operation under the same working conditions.Due to different working conditions,fault samples collected from the two fields show great differences in distribution,resulting in a decrease in classification accuracy.Therefore,it is still challenging to develop an effective fault diagnosis model that eliminates the assumption of the same distribution.In order to solve the problem,a deep neural network is proposed.Will first failure data input to the said more depth domain adaptive network(MDAN)for feature extraction,and through the said network to get information,different scale through multi-scale network access to the source domain and target domain,the characteristics of the two characteristics of the input to the alignment algorithm based on Hilbert space,then introduce the MK-MMD loss characteristics of distance measurement and calculation domain,Finally,Adam was introduced to optimize the network and accelerate the convergence of the model.Experimental verification was carried out on the bearing data set of the famous Western Reserve University,and the model was evaluated through several migration schemes.Experimental results show that the proposed method has excellent performance in variable working conditions.(2)Research on Small sample Rolling bearing fault diagnosis Method based on transfer learningWith the rapid progress of computer computing power,the deep learning method has gradually replaced the traditional machine learning method,and a large number of university professors use deep learning to conduct bearing fault diagnosis research,and a large number of research achievements have been achieved.However,deep learning methods usually require a large amount of data,but in practice it is difficult to collect enough fault data,which will reduce the generalization performance of deep learning methods.In addition,in a real industrial environment,label data is very valuable.Due to expensive equipment and personnel safety issues,it is difficult to obtain large quantities of high-quality fault marking data.For this,in the article,this paper proposes a multiple source domain of fault diagnosis method for small sample problems,first of all,dense convolution neural network is used to extract features of multiple source domain and target domain,will extract the source domain and target domain characteristics in high dimensional space for one-to-one correspondence,then using domain specific decision boundary to calculate boundary losses,Finally,the characteristics of rolling bearings in the target domain are classified.Experimental verification shows that such modeling results in robustness to perturbations,and comparison with other methods proves the superiority of this method in small sample cases.To sum up,this article through to the researches of the rolling bearing fault diagnosis theory and fault reason and mechanism of the formation has carried on the detailed description and analysis of the time domain analysis,frequency domain analysis and time-frequency domain analysis mode and index data,the convolutional neural network of the operation principle and structure of a component,A targeted neural network algorithm is designed for the research content,and experiments and verification are carried out through public data sets and laboratory data sets,which proves the validity and rationality of the proposed method.