Font Size: a A A

Research On Prediction Method Of Rolling Bearing Remaining Life Based On Data-driven And Transfer Learning

Posted on:2024-08-18Degree:MasterType:Thesis
Country:ChinaCandidate:X L HuangFull Text:PDF
GTID:2542307133493224Subject:Mechanical engineering
Abstract/Summary:PDF Full Text Request
Recently,Through the collection,analysis and modeling of fault data,the remaining useful life(RUL)prediction of mechanical equipment under actual working conditions is realized.It is an important research topic in the field of Prognostics and Health Management(PHM).However,the existing RUL prediction methods based on machine learning do not fully consider the distribution difference between data,and do not fully mine the deep features of fault data,which limits the improvement of prediction accuracy and stability.Therefore,aiming at the problem of residual life prediction of rolling bearings under multiple working conditions and multiple failure modes,this paper utilizes deep learning theory and constructs two kinds of residual life prediction methods based on data-driven and transfer learning,in order to obtain more accurate prediction results.The main research contents are as follows:(1)A multi-scale feature extraction method is used to address the problem of insufficient degradation information due to single-scale feature extraction.In the study of rolling bearing RUL prediction,the feature extraction of bearing vibration signal usually adopts only a singlescale feature extraction method,and the importance of multi-scale feature extraction is ignored.By dividing the bearing vibration signal data into multiple scale sub-segments,the multi-scale feature extraction of the signal data is realized,and the characteristics of the bearing signal are better understood and analyzed.Through multi-scale feature extraction,not only more degraded data information can be obtained,but also the global and local characteristics of the sampling point data can be considered,which effectively improves the RUL prediction accuracy.(2)In order to make better use of the time dependence of the degradation characteristics of rolling bearings in the degradation process,a RUL prediction method based on MSARCNN is used.Firstly,the initial features related to the time domain and frequency domain of the vibration signal are extracted.Secondly,the bearing degradation features were adaptively weighted through the attention mechanism to obtain more important degradation features.Finally,the cyclic convolutional neural network was used to extract the time-dependent features of the degradation process to achieve the RUL prediction of bearings.Through experimental verification and analysis,the prediction performance based on MSARCNN model is improved by 29.9% compared with that based on RCNN model,and the bearing RUL prediction is accurately achieved.(3)The remaining life prediction method of bearings based on transfer learning was established,which could effectively solve the problem of non-uniform data distribution in the bearing degradation process under different working conditions.Therefore,the unsupervised domain adaptive transfer learning method combined with domain countermeasure technology can effectively suppress the distribution difference of different data and realize the RUL prediction of bearings under different working conditions.The RUL prediction model based on transfer learning is composed of three parts.The domain adaptive module reduces the data distribution difference between source domain and target domain by domain antagonism.The feature extractor is composed of attention convolutional neural network and Long-short term memory network,which can extract the deep degradation features in the process of bearing degradation.The RUL prediction module further obtains the RUL of the corresponding bearing through the extracted deep degradation features.Through the interaction between modules,the RUL of rolling bearings under different working conditions can be predicted.
Keywords/Search Tags:rolling bearings, data-driven, transfer learning, remaining useful life prediction
PDF Full Text Request
Related items