| As my country’s industrial field gradually develops towards high-end and technological development,people have higher and higher requirements for safety,stability and reliability in the production and use of industrial equipment.Bearings are an important part of mechanical equipment,and how to improve their safety and reliability has become the primary task.How to evaluate the health status of the bearing,provide a certain theoretical basis for the replacement and maintenance of the bearing,and its life prediction has become the main work.Before the life prediction of the bearing,the health indicator HI must be constructed first.The construction method is divided into supervised learning method and unsupervised learning method.Although the unsupervised learning method has the characteristics of high training model efficiency and no need for labels,However,the research field lacks different supervised learning methods to construct HI curve for bearing life prediction,and compare and analyze the difference of life prediction accuracy.Taking the solution to this problem as the starting point,this paper deeply researches the construction method of bearing HI curve in life prediction,and conducts the research of remaining life prediction method after obtaining the HI curve.This paper starts from the perspective of constructing bearing HI curve in two ways:supervised learning and unsupervised learning.Taking bearing vibration data as an example,the research on the construction of bearing HI curve and remaining life prediction is carried out.First,taking the supervised learning method as the research object,based on the analysis of the characteristics and application scenarios of the existing supervised learning method,a bearing HI construction method based on 1DCNN_LSTM is proposed,and the performance of the algorithm is verified on the PHM2012 data set.Secondly,taking the unsupervised learning algorithm as the research object,a bearing HI construction method based on SDAE+PCA is proposed,and the same bearing vibration data is used for experimental verification.Finally,this paper uses the BiLSTM bearing remaining life prediction method,and on the basis of the HI curve constructed by different supervised learning,the difference in life prediction accuracy is verified.The research results show that,compared with other literature methods,the proposed method based on 1D_LSTM and SDAE+PCA can construct a smoother and less noisy performance degradation curve,with better.trend and monotonicity.On the basis of different supervised learning methods to construct HI curves,the BiLSTM model life prediction is used.The results show that the 1D_LSTM method in this paper is better than the SDAE+PCA method to construct the HI curve to predict the remaining life.In most cases,the prediction accuracy is higher.The type of failure that has a large fluctuation before it actually enters the decay period,and the prediction accuracy is reversed.Finally,through the bearing life prediction trend change chart,it is further explained that different methods of constructing HI curve have certain differences in the accuracy of predicted life. |
