IGMM结合区间统计的机械故障预警方法研究  

Study on Mechanical Fault Early Warning Based on IGMM and Interval Statistics

作　　者：苏方健 刘文才[2] 马波[1,3] SU Fang-jian;LIU Wen-cai;MA Bo(College of Mechanical and Electrical Engineering,Beijing University of Chemical Technology,Beijing 100029,China;CNPC Research Institute of Safety&Environment Technology,Beijing 102206,China;Beijing Key Laboratory for Health Monitoring and Self-Recovery of High-End Mechanical Equipment,Beijing University of Chemical Technol-ogy,Beijing 100029,China)

机构地区：[1]北京化工大学机电工程学院,北京100029 [2]中国石油天然气股份有限公司安全环保技术研究院,北京102206 [3]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

出　　处：《机械设计与制造》2024年第1期154-158,共5页Machinery Design & Manufacture

基　　金：国家重点研发计划项目(2018YFB1503103)。

摘　　要：针对机械工况恶劣、结构复杂,单特征门限报警的故障预警方法对其预警常出现误、漏报警事件的现状,提出一种无限高斯混合模型(IGMM,Infinite Gaussian Mixture Model)结合区间统计的机械故障预警方法。首先,将机械振动信号映射为高维特征空间,对其所在空间进行区间划分。然后,利用IGMM估计出机械健康状态下高维特征空间在各区间频数的分布;利用累计计数方法统计出机械在实时状态下高维特征空间在各区间频数的分布。最后,对以上两个频数分布计算距离并将其与自学习得出的预警阈值作比较,实现故障预警。验证结果表明,提出方法的预警准确率较高且时效性较好。Because complex machinery has bad working conditions and complex structure,the early fault warning method of the SF method often causes false and missing alarm events.A mechanical fault early warning method based on IGMM and interval statistics is proposed.Firstly,this method maps mechanical vibration signals to high-dimensional feature space and divides the interval.Then,the probability statistical model is established by using IGMM to estimate the frequency distribution of high-dimensional feature space in each interval under the condition of mechanical health;the frequency distribution of high-dimensional feature space in each interval under the real-time mechanical state is calculated.Finally,fault early warning is re-alized by calculating the distance between the two frequency distributions and comparing it with the warning threshold obtained by self-learning.The application results show that the proposed method has high accuracy and timeliness.

关 键 词：故障预警 无限高斯混合模型 机械设备 

分 类 号：TH16[机械工程—机械制造及自动化] TH17