基于Kmeans-DBSCAN融合聚类的轴承异常温升诊断模型  

The Abnormal Temperature Rise Diagnosis Model of Bearing Based on Kmeans-DBSCAN Fusion Clustering Algorithm

作　　者：罗怡澜 邹益胜[1] 王超[1] 邓佳林 LUO Yi-lan;ZOU Yi-sheng;WANG Chao;DENG Jia-lin(Institute of Advanced Design and Manufacturing,School of Mechanical Engineering,Southwest Jiaotong University,Sichu-an Chengdu 610031,China)

机构地区：[1]西南交通大学机械工程学院先进设计与制造研究所,四川成都610031

出　　处：《机械设计与制造》2020年第3期18-23,共6页Machinery Design & Manufacture

基　　金：国家高新技术研究发展计划(863计划)—高速铁路动车组全生命周期数据集成管理与综合(2015AA043701-02)。

摘　　要：经典聚类算法在机车异常轴温诊断应用中存在判别阈值参经验化设定与漏判率、误判率较高的问题。利用机车轴温测点的关联性、异常温升特征分布特点,提出一种基于Kmeans-DBSCAN融合聚类的轴承异常温升诊断模型。首先将轴承异常温升的诊断转化为多组关联序列中少数持续离群子序列的检测问题,再根据温度序列特征空间分布位置和局部密度差异性,通过Kmeans-BSCAN融合聚类分离出离群子序列,并实现了DBSCAN邻域判别阈值参数的自适应选取。基于某型机车履历数据的实例验证结果发现:该模型对异常温升诊断的准确率达100%,与Kmeans算法保持一致,比DBSCAN算法提高22.4%;误报率低至0.5%,比Kmeans算法降低18.5%,比DBSCAN算法降低12%。There are two problems in the application of classical clustering algorithm which were employed to diagnose the abnormal bearing temperature rise of locomotive vehicles. First,the threshold parameters for judging need to be pre-set by user experience,and the second is the high missing rate and false rate. An abnormal temperature rise diagnosis model was proposed according to relevance of locomotive axial temperature measure points,spatial location and local density difference of abnormal temperature rise feature distribution. Firstly,the judgment of the abnormal temperature rise was transformed into the detection of a few consecutive outliers in the sequence of multiple groups,and then Kmeans was used to detected the differences in feature distribution,the clustering radius of Kmeans is used as the clustering neighborhood radius of DBSCAN.The DBSCAN neighborhood discriminant threshold parameter is adaptively selected and the local density difference detection of feature is realized.Based on the example data of a certain locomotive vehicle,the result is verified:the model inherits the advantages of Kmeans and DBSCAN algorithms,the accuracy of fault diagnosis is 100%,consistent with the Kmeans algorithm,22.4% higher than DBSCAN algorithm;the false positive rate is as low as 0.5%,18.5% lower than the Kmeans algorithm,12% lower than the DBSCAN algorithm.

关 键 词：温度 异常检测 聚类 Kmeans-DBSCAN 诊断模型 机车车辆 

分 类 号：TH16[机械工程—机械制造及自动化] U270.7[机械工程—车辆工程]