基于T_SNE和AW-SVR的滚动轴承退化状态预测  

Degradation State Prediction of Rolling Bearing Based on T_SNE and AW-SVR

作　　者：吕明珠[1,2] LV Ming-zhu(School of Automatic Control,Institute of Liaoning Equipment Manufacturing Professional Technology,Liaoning Shenyang 110161,China;Liaoning Radio and TV University,Liaoning Shenyang 110034,China)

机构地区：[1]辽宁装备制造职业技术学院自控学院,辽宁沈阳110161 [2]辽宁广播电视大学,辽宁沈阳110034

出　　处：《机械设计与制造》2022年第4期83-87,共5页Machinery Design & Manufacture

基　　金：高校应用性研究专项课题(2019YYYJ-1)。

摘　　要：针对滚动轴承存在故障提取信息冗余、非平稳突变故障预测效果不佳的问题,提出了一种基于t分布随机近邻嵌入算法(T_SNE)和自适应加权支持向量回归(AW-SVR)相结合的滚动轴承退化状态预测方法。该方法首先提取了滚动轴承的时域和频域指标,组成原始特征向量;然后通过T_SNE算法进行特征降维,获得二维退化特征集,并利用相对均方根值(RRMS)划分滚动轴承的退化阶段,以此来描述轴承的退化状态;最后将权函数引入支持向量回归机(SVR)中,并通过自适应地改变突变点的权重增强SVR的学习能力,以提高对突变故障的预测精度。通过实验数据对比分析,结果表明T_SNE和AW-SVR相结合比ISOMAP+SVR、LLE+SVR和T_SNE+SVR方法在轴承退化状态预测方面具有更好的效果。Aiming at the problem that the fault extraction information of rolling bearing is redundant and the prediction effect of non-stationary abnormal fault is not good,a new prediction method was proposed base on t-distribution stochastic neighbour embedding(T_SNE)and adaptive weighted support vector regression(AW-SVR)to describe the rolling bearing degradation state.Firstly,the time-frequency domain features of the rolling bearing was extracted to construct the initial feature sets.Then,the twodimensional degraded feature sets was obtained by using the T_SNE algorithm.The relative root mean square value(RRMS)to describe the degradation state of the bearing was used to classify the degradation stage of rolling bearings.Finally,the weight function was introduced into the support vector regression machine(SVR),and the learning ability of SVR was enhanced by adaptively changing the weight of the abnormal point,so as to improve the prediction accuracy of the abrupt faults.Through the comparative analysis of the test data,the results show that the combination of T_SNE and AW-SVR has a better effect on predicting the degradation state of rolling bearing than the ISOMAP+SVR,LLE+SVR and T_SNE+SVR methods.

关 键 词：t分布随机近邻嵌入算法 自适应加权支持向量回归 滚动轴承 退化状态预测 

分 类 号：TH16[机械工程—机械制造及自动化] TH165.3