基于Pearson-KPCA和LSTM的伺服电机滚动轴承剩余寿命预测  

Residual Life Prediction of Servo Motor Rolling Bearings Based on Pearson-KPCA and LSTM

作　　者：李子涵 张营[1,2] 左洪福[2] LI Zihan;ZHANG Ying;ZUO Hongfu(College of Automobile and Traffic Engineering,Nanjing Forestry University,Nanjing Jiangsu 210037,China;College of Civil Aviation,Nanjing University of Aeronautics and Astronautics,Nanjing Jiangsu 211106,China)

机构地区：[1]南京林业大学汽车与交通工程学院,江苏南京210037 [2]南京航空航天大学民航学院,江苏南京211106

出　　处：《机床与液压》2023年第14期221-226,共6页Machine Tool & Hydraulics

基　　金：国家自然科学基金与民航联合基金重点基金(U1933202)。

摘　　要：针对伺服电机滚动轴承的寿命预测,提出一种基于皮尔逊相关系数及核主成分分析的长短时记忆网络预测方法。提取滚动轴承的时、频域信号,通过移动平均法进一步获取相关特征,并采用皮尔逊相关系数筛选高度相关特征指标,利用KPCA提取高度相关特征指标中的若干主成分;将第一主成分作为长短时记忆网络模型的输入对滚动轴承进行剩余寿命预测。采用IMS轴承数据集进行验证,得到的轴承寿命预测RMSE值和可决策系数值分别为0.0543和0.989。将其与长短期记忆网络模型和BP神经网络的预测结果进行对比,证明所提方法具有较高的精度。Aiming at the life prediction of servo motor rolling bearings,a long-term short-term memory network prediction method based on Pearson correlation coefficient and nuclear principal component analysis was proposed.The time and frequency domain signals of rolling bearings were extracted,the relevant features were further obtained by the moving average method,and the Pearson correlation coefficient was used to screen the highly correlated feature indexes,and several principal components of the highly correlated feature index were extracted by KPCA.The first principal component was used as input to the long-short-term memory network model to predict the remaining life of rolling bearings.The IMS bearing data set was used for verification,and the RMSE value and decisionable coefficient value of bearing life prediction were 0.0543 and 0.989,respectively.Compared with the prediction results of long short-term memory network model and BP neural network,the results show that the proposed method has high accuracy.

关 键 词：皮尔逊相关系数 核主成分分析 长短时记忆神经网络 滚动轴承 剩余寿命预测 

分 类 号：TH133.33[机械工程—机械制造及自动化]