基于层次聚类的状态监测数据衰退模式挖掘

故障预测与健康管理(prognostics and health management,简称PHM)技术,是在现代复杂设备的高可靠性和高安全性要求下,实现视情维修的一种新的技术理念。PHM技术的研究方向之一就是利用系统状态监测数据中包含的信息,对设备的健康情况和发展趋势进行评估、分析和预测。针对基于状态监测数据的衰退模式挖掘问题,提出了一种P-D-H聚类方法,以实现衰退模式的挖掘。首先,通过分段聚合近似(piecewise aggregate approximation,简称PAA)方法对由状态监测数据形成的退化轨迹时间序列进行模式表示;其次,采用动态时间弯曲距离(dynamic time warping,简称DTW)作为模式序列的相似性度量;最后,采用层次聚类的方法实现衰退模式聚类。用此方法对滚动轴承磨损状态监测数据进行了衰退模式挖掘,验证了方法的有效性。基于复杂系统状态监测数据的模式聚类方法能够有效实现系统健康衰退模式的挖掘,模式挖掘的结果可以为应用状态监测数据进行系统健康的预测奠定良好的基础。

Degradation Pattern Mining of Condition Monitoring Data Based on Hierarchical Clustering

Prognostics and health management (PHM) technology, in the modern complex equipment, high reliability, and high-security requirements, is a new technology concept to achieve condition-based maintenance. One of the research directions of PHM technology is to use the information contained in the system status monitoring data to evaluate, analyze and forecast the health and development of the equipment. Aiming at the problem of recessive pattern mining based on state monitoring data, a P-D-H clustering method is proposed to realize the mining of degradation pattern. First, the pattern of the degraded trajectory time series formed by the state monitoring data is represented by the piecewise aggregate approximation method. Then, the dynamic time warping distance is used as the similarity measure of the pattern sequence. Finally, the hierarchical clustering method is used to achieve the regression model clustering. In this way, the wear data of the rolling bearing wear condition is excavated and the effectiveness of the method is verified. The model clustering method based on the complex system state monitoring data can effectively realize the mining of the system health degradation pattern. The result of pattern mining can lay a good foundation for the system health forecasting.