Research On Bearing Fault Diagnosis And Remaining Useful Life Prediction Based On Transfer Learning

With the development of technology in the mechanical manufacturing industry,mechanical equipment has become more integrated,automated,complex,and intelligent,and uncertainties in the operation process will also increase.This makes traditional fault diagnosis and life prediction methods difficult to meet the needs.With the development of artificial intelligence technology,deep learning has been widely applied in this field.However,in practical engineering,on the one hand,it is difficult to obtain sufficient fault data,and on the other hand,due to the continuous changes in working conditions,training and testing data no longer follow the assumption of independent and uniform distribution.Therefore,existing machine learning methods are difficult to train effective universal models.The transfer learning method can effectively complete the task of identifying and predicting the target fields with missing tags by using the source domain data training model with complete tags.Aiming at the problem that the current bearing fault diagnosis and life prediction model can not train an effective general model,this paper studies the bearing fault diagnosis and life prediction based on transfer learning.Feature extraction is realized through deep learning network model,and the extracted source domain and target domain features are aligned using transfer learning technology to improve the accuracy of diagnosis and prediction of the model in the target domain.The specific research content of this article includes:(1)Joint intra class and inter domain distribution adaptation for rolling bearing fault diagnosis.Propose a domain adaptation method that combines intra class and inter domain distribution adaptation for cross working condition bearing fault diagnosis experiments.Using a one-dimensional convolutional neural network model as a feature extractor to extract features from both the source and target domains,the extracted source domain features are input into the source domain classifier to enable the network to achieve good classification results on the source domain data.A joint maximum mean difference of intra class density is proposed to align the source domain features with the target domain features,ultimately obtaining the transferred diagnostic accuracy.Design cross working condition fault diagnosis experiments and compare them with other transfer methods.The results show that this method improves the intra class density,narrows the classification boundary,and can better align the distribution of the source and target domains,thereby improving the transfer effect across working conditions.(2)Bearing fault diagnosis based on deep domain adversarial migration network.A deep domain adversarial transfer network model is proposed.Firstly,a onedimensional convolutional neural network is used as a feature extractor to extract features.Secondly,an adaptation method based on envelope spectrum is proposed as a metric of the domain adaptation network to adapt source and target domain data in the domain.The knowledge of source domain data is transferred to unlabeled target domain tasks.Finally,domain adversarial methods are used to train and optimize,Make the domain discrimination network unable to effectively distinguish between the feature input source domain and the target domain.We designed and implemented migration tasks across operating conditions and test benches,and compared them with the MMD method.The results showed that the proposed method in this chapter can better align inter domain features and improve diagnostic accuracy in the target domain.(3)The research of bearing remaining useful life based on transfer learning.Convolutional neural networks and gated recurrent units are used as feature extractors to obtain source and target domain feature information.Then,local maximum mean difference is used as a metric for domain adaptation,while aligning global and local distributions,effectively overcoming the limitations of current domain adaptation methods and increasing transfer efficiency.The cross working condition bearing health indicator fitting experiment and life prediction experiment were conducted on the PHM2012 dataset,and the experimental results showed that this model can improve the accuracy of migration life prediction.