• CN:11-2187/TH
  • ISSN:0577-6686

机械工程学报 ›› 2020, Vol. 56 ›› Issue (18): 7-14.doi: 10.3901/JME.2020.18.007

• 仪器科学与技术 • 上一篇    下一篇

扫码分享

数据驱动的聚类分析故障识别方法研究

王庆锋, 刘家赫, 卫炳坤, 张程   

  1. 北京化工大学高端机械装备健康监控及自愈化北京市重点实验室 北京 100029
  • 收稿日期:2019-12-05 修回日期:2020-05-11 出版日期:2020-09-20 发布日期:2020-11-17
  • 通讯作者: 刘家赫(通信作者),男,1995年出生。主要研究方向为故障诊断与健康管理。E-mail:liujiahebuct@163.com
  • 作者简介:王庆锋,男,1972年出生,博士,副研究员。主要研究方向为设备动态监测、诊断与维护;故障诊断与自愈;在役再制造;装置可靠性与风险评估。E-mail:wqf2422@163.com
  • 基金资助:
    国家自然科学基金(51775029)、中国石化“石化装置关键机组远程诊断故障智能预警技术研究”(CHG18064)和中央高校基本科研业务费(PYBZ1813)资助项目。

Research on Data-driven Clustering Analysis Fault Identification Method

WANG Qingfeng, LIU Jiahe, WEI Bingkun, ZHANG Cheng   

  1. Beijing Key Laboratory of Health Monitoring and Self-recovery of High-end Machinery Equipment, Beijing University of Chemical Technology, Beijing 100029
  • Received:2019-12-05 Revised:2020-05-11 Online:2020-09-20 Published:2020-11-17

摘要: 设备故障预测与健康管理已经进入了智能化时代,从设备状态监测数据中分析、提取故障特征信息,实现故障自动识别与分类是实现故障自愈调控的前提。以离心压缩机为研究对象,基于K-means聚类分析和数据驱动方法,构建K-means聚类故障模式识别模型,筛选故障识别灵敏度高的故障特征参数作为模型输入参数,基于监测数据训练得到不同故障的聚类中心,探索数据驱动的基于聚类中心距离判据的离心压缩机故障识别方法。应用中石油某公司离心压缩机正常和喘振、碰摩、不对中故障案例数据进行模型验证,结果表明:监测数据驱动的聚类分析故障识别方法能够实现离心压缩机多故障模式自动识别;与支持向量机故障识别方法相比,K-means聚类分析故障识别模型准确性更高;数据驱动方法计算实时监测数据聚类中心,利用直观距离判据实现故障模式自动识别,能够实现故障诊断和智能决策实现,为开展自愈调控提供技术支撑。

关键词: 数据驱动, 聚类分析, 故障识别, 自愈调控, 故障特征参数

Abstract: Equipment fault prediction and health management have entered the era of intelligence. It is the premise to realize fault self-recovery regulation by analysing and extracting fault characteristic information from equipment condition monitoring data to realize automatic fault identification and classification.The centrifugal compressor is taken as the research object. Based on K-means clustering analysis and data-driven method, the K-means clustering fault pattern recognition algorithm model is constructed. The fault characteristic parameters with high sensitivity of fault identification are selected as model input parameters, and the clustering centers of different faults are obtained by monitoring data training, and the data-driven centrifugal compressor fault identification method based on the cluster center distance criterion is explored. The model is validated by the normal and surge, rubbing and misalignment case data of a centrifugal compressor of CNPC. The results show that the data-driven clustering fault identification method can realize the automatic identification of multi-fault mode of centrifugal compressor; compared with the support vector machine fault identification method, the K-means clustering analysis fault identification method is more accurate; the data-driven method calculates the clustering center of real-time monitoring data, and uses the intuitive distance criterion to realize automatic identification of fault modes, which can realize fault diagnosis and intelligent decision-making, and provide technical support for self-healing regulation.

Key words: data-driven, clustering analysis, fault identification, self-recovery regulation, fault characteristic parameters

中图分类号: