Research On Fault Diagnosis And Remaining Useful Life Estimation Of Vehicle Motor Bearing Based On Machine Learning

In the 21st century,the automotive industry begins to transform to electrification,and electric vehicles become the hot spot of the times.As the key vulnerable parts of the power heart of electric vehicles,the drive motor bearing bears complex alternating stress and thermal stress.Therefore,it is of great practical significance to monitor its operation status and estimate its remaining useful life,which can effectively reduce the risk of safety accidents Risk.In this paper,based on the data-driven method,taking the bearing vibration signal as the main monitoring index,the methods of bearing fault diagnosis and remaining useful life prediction based on machine learning are studied and compared,and a new network architecture is proposed.First of all,this paper studies the bearing fault diagnosis method,and completes the bearing fault diagnosis under the traditional machine learning method and deep learning method.On the one hand,feature extraction and selection are carried out in time domain and frequency domain for the bearing one-dimensional vibration signal after noise reduction,and the SVM network optimized by genetic algorithm is applied to complete the bearing fault diagnosis and classification model;on the other hand,the deep learning fault diagnosis model based on CNN algorithm is constructed.CNN has the characteristics of convolution and weight sharing,which can automatically extract the deep features of vibration signals,avoiding the problem that traditional machine learning methods rely too much on expert experience.The two methods are verified and compared on CWRU bearing data set.The verification results show that the deep learning fault diagnosis model based on CNN algorithm has advantages in diagnosis accuracy and generalization performance.Secondly,this paper explores the bearing failure mode and the evaluation method of health state,and studies the prediction method of remaining useful life.On the one hand,CNN classification network is used to complete the evaluation of adaptive bearing health state without expert experience,and the bearing samples are divided into three health states: normal,degradation and failure,on which the failure mode of bearing is judged.For the bearing with degenerative failure,the health status of the bearing is regarded as the key feature,and the time-dependent CLDNN deep fusion network is used to estimate the remaining service life.For the bearing with sudden failure,an early warning is generated to remind the engineering personnel to replace it in time.On the other hand,the same CLDNN network is used to build the remaining life estimation model without health assessment.Through comparison,the advantages of the remaining life estimation method combined with health assessment are verified.Finally,the accuracy and feasibility of the models are verified by using the FEMTO open data set of bearing life,and the results are analyzed and compared through the evaluation index.The results show that the remaining useful life estimation model combined with health status assessment can effectively judge the failure mode of bearing,and has better prediction effect for the bearing with degenerate failure.The research shows that in the aspect of bearing fault diagnosis,convolutional neural network does not need feature engineering,and it is better than SVM model in terms of diagnosis accuracy and generalization performance;in the aspect of bearing remaining useful life prediction,the accuracy of RUL estimation model combined with bearing health assessment is better than the same type of model without health assessment.