Research Of A Method For Predicting The Remaining Useful Life Of Bearings Under Variable Loads

Rolling bearings,as key components of rotating machinery,have a wide range of applications in the field of rail transport.As failures of rolling bearings in equipment in rail transport often lead to long-term equipment shutdowns and thus huge economic losses.It is therefore particularly important to anticipate impending failures by estimating the remaining service life of rolling bearings,so that maintenance and timely repairs to rolling bearings can be planned.Rolling bearings are often operated under variable loads due to their complex and variable operating conditions in practice.However,the data generated by rolling bearings under variable loads vary considerably.To address this problem,this paper starts from the selection and extraction of the features of the full life vibration signal of rolling bearings,based on machine learning and deep learning methods,and researches the remaining life prediction of rolling bearings under variable loads,the main contents are as follows.(1)In order to classify the performance degradation state of rolling bearings more accurately and efficiently,a new adaptive performance degradation state classification method based on T-SNE and K-Means clustering algorithm is proposed.Noise reduction is achieved by compressing the data dimensionality through T-SNE.The K-Means clustering algorithm is used to accurately determine the performance degradation state of rolling bearings according to the different distributions of their feature sets in space.(2)A feature selection method based on Pearson Product-Moment Correlation Coefficient(PPMCC)and Kullback-Leibler divergence(KLIC)feature selection is proposed(2)A new feature selection method is proposed to reduce the redundancy between features by ensuring that the feature set has a high degree of information and a low correlation between them.(3)A new feature extraction method based on One-Dimensional Multi-Scale Convolutional Neural Networks(O-D-MSCNN)is proposed to further exploit the information of features at different scales.(4)Finally,this paper uses the LSTM network to experimentally validate the method under variable load conditions using the IMS rolling bearing dataset in the USA and the PRONOSTIA rolling bearing dataset in France,and to verify the superiority of the proposed method in predicting the remaining service life under different loads by comparison with other methods.