航空学报 > 2015, Vol. 36 Issue (3): 898-906   doi: 10.7527/S1000-6893.2014.0124

基于数据关联性分析的飞轮异常检测

龚学兵, 王日新, 徐敏强   

  1. 哈尔滨工业大学 深空探测基础研究中心, 哈尔滨 150080
  • 收稿日期:2014-04-01 修回日期:2014-06-23 出版日期:2015-03-15 发布日期:2015-03-31
  • 通讯作者: 王日新 男,博士,副教授,硕士生导师。主要研究方向:船舶系统仿真、航天器故障诊断。Tel: 0451-86418320 E-mail: wangrx@hit.edu.cn E-mail:wangrx@hit.edu.cn
  • 作者简介:龚学兵 男,博士研究生。主要研究方向:人工智能算法、卫星系统故障诊断分析。Tel: 0451-86418320 E-mail: gongdanumber1@163.com
  • 基金资助:

    国家"973"计划 (2012CB720003)

Abnormality detection for flywheels based on data association analysis

GONG Xuebing, WANG Rixin, XU Minqiang   

  1. Deep Space Exploration Research Center, Harbin Institute of Technology, Harbin 150080, China
  • Received:2014-04-01 Revised:2014-06-23 Online:2015-03-15 Published:2015-03-31
  • Supported by:

    National Basic Research Program of China (2012CB720003)

摘要:

针对航天器早期故障在闭环系统下难以被检测、数学模型难以精确建立的问题,提出了一种基于数据关联性分析的归纳式系统异常监测(IMS)方法。该方法采用无监督学习的聚类算法,利用具有关联性的参数构建数据向量,通过聚类分析自动建立健康数据向量的族类阈值区间。关联关系的破坏将引起部分参数超出族类阈值区间,使系统的异常程度存在模糊性与随机性。引入云模型评价指标,将闭环系统异常程度的不确定性通过熵与超熵定量表示,从而更加准确地判断闭环系统的异常程度。仿真结果表明:该方法能够建立卫星飞轮闭环系统的族类知识库,并可以根据云模型提供的定性知识有效判断系统的异常程度。

关键词: 闭环系统, 飞轮, 关联性分析, 无监督学习, 异常检测, 云模型

Abstract:

In order to solve the problem that the early faults hardly to be discovered in the spacecraft under closed-loop system and the precise mathematical model barely to be established, inductive monitoring system (IMS) based on data association analysis is proposed to detect the abnormality in closedloop system. An unsupervised learning clustering algorithm is employed and can automatically characterize the threshold interval of each cluster by analyzing the nominal system operation data vectors with parameter correlation. Some parameters in the vectors may overflow their corresponding cluster intervals because of correlation damages during the abnormal operations. And there are fuzziness and randomness in measuring the degree of abnormality in the system. By introducing the cloud model index,the backward cloud can quantify the uncertainty of abnormal degrees in the closedloop system by the entropy and the hyper entropy indices, so that the abnormal degrees could be judgedmore accurately . Simulation results show that the method employed can build the cluster knowledge bases of satellite flywheels with closedloop systems, and the normal cloud model can offer the qualitative knowledge about damages in the simulink model. effectively judge the degree of system abnormality according to the qualitative knowledge of the normal cloud model.

Key words: closed loop systems, flywheels, association analysis, unsupervised learning, abnormality detection, cloud model

中图分类号: