一种基于特征聚类和评价的轴承寿命预测新方法  

A new method of bearing life prediction based on feature clustering and evaluation

作　　者：李海浪 邹益胜[1] 曾大懿 刘永志 赵市教 宋小欣[1] LI Hailang;ZOU Yisheng;ZENG Dayi;LIU Yongzhi;ZHAO Shijiao;SONG Xiaoxin(School of Mechanical Engineering,Southwest Jiaotong University,Chengdu 610031,China)

机构地区：[1]西南交通大学机械工程学院,成都610031

出　　处：《振动与冲击》2022年第5期141-150,共10页Journal of Vibration and Shock

基　　金：国家重点研发计划资助项目(2020YFB1708000);重庆市教委科学技术研究项目(KJZD-K201805801)。

摘　　要：在预测轴承寿命时,使提取的特征和剩余寿命保持高相关性,并使不同的特征之间保持低相关性,是有利于提升轴承寿命预测精度的。为解决单一的特征评价方法对后者考虑不足的问题,提出了一种基于相关性改进Kmeans聚类算法(correlation-based improved Kmeans cluster algorithm, Corr-Kmeans)和初始聚类中心确定方法,并与特征评价相结合,最终提出一种基于特征聚类和评价的轴承寿命预测新方法。首先利用卷积自编码对频域信息提取初始特征,用Corr-Kmeans对初始特征按相关性进行聚类,使得聚类后的特征类内相关性高,而类间相关性低;其次,使用相关性、单调性和鲁棒性3个指标来综合评价每一类中的特征,按照筛选阈值将得分较高的特征从每一类中分别选出,组成用于训练与预测的特征子集;最后采用LSTM(long short-term memory, LSTM)网络对轴承剩余寿命进行预测。在一个轴承加速寿命试验的公开数据集上使用留一法进行验证,利用对比试验证明了该方法在预测轴承剩余寿命上的有效性。When predicting bearing life, it is helpful to improve the accuracy of bearing life prediction by maintaining high correlations between the extracted features and the remaining life, and low correlations between different features. Here, to solve the problem of a single feature evaluation method not considering the latter mentioned above, a correlation-based improved Kmeans cluster algorithm(Corr-Kmeans) and an initial cluster center determination method were proposed, then combined with feature evaluation, a new bearing life prediction method based on feature clustering and evaluation was finally proposed. Firstly, convolution self-encoding was used to extract initial features from the frequency domain information, and Corr-kmeans was used to cluster initial features according to correlations to make intra-class correlations between the clustered features be high and correlations between features be low. Secondly, 3 indexes of correlation, monotonicity and robustness were used to comprehensively evaluate features in each class. The features with high scores were chosen from each class, respectively according to the screening threshold to form feature subsets for training and prediction. Finally, LSTM(long short-term memory) network was used to predict the remaining life of bearing, the results were verified by using the leave one method on a public data set of bearing accelerated life tests. The effectiveness of the proposed method in predicting the remaining life of bearing was proved with contrastive tests.

关 键 词：轴承 寿命预测 相关性改进Kmeans聚类算法(Corr-Kmeans) 聚类 特征评价 

分 类 号：TH17[机械工程—机械制造及自动化]