中国空间科学技术

中国空间科学技术 ›› 2023, Vol. 43 ›› Issue (1): 88-99.doi: 10.16708/j.cnki.1000-758X.2023.0009

• 论文 • 上一篇    下一篇

一种改进的固液火箭发动机故障诊断方法

吴一凡，魏延明，杨博，于贺，刘超凡，魏翔   

  1. 1 北京航空航天大学 宇航学院，北京100191
    2 北京控制工程研究所，北京100190
  • 出版日期:2023-02-25 发布日期:2023-01-13

An improved fault diagnosis method for solid-liquid rocket engine

WU Yifan，WEI Yanming，YANG Bo，YU He，LIU Chaofan，WEI Xiang   

  1. 1 School of Aeronautic Science and Engineering,Beihang University，Beijing 100191,China
    2 Beijing Institute of Control Engineering，Beijing 100190,China
  • Published:2023-02-25 Online:2023-01-13

PDF (PC)

81

摘要： 针对固液火箭发动机的可靠性问题，设计了一种改进的贝叶斯网络故障诊断方法，可以通过网络化自主逻辑推理，对固液火箭发动机进行故障诊断。为了提取时序观测信号的故障特征，提出将步进法与核主成分分析（KPCA）相结合的分析方法，并根据模糊C均值聚类算法（FCM）建立模糊多态贝叶斯网络，实现对观测信号尺度的模糊处理，提高对不确定性故障的诊断能力。通过Matlab/Simulink建立改进的贝叶斯网络故障诊断系统。仿真结果表明，改进的算法能够实现对固液火箭发动机常见故障的有效诊断，并能够适应小样本集学习的情况。与传统贝叶斯诊断算法相比，故障诊断的平均准确率提高了20.9%。

关键词: 固液火箭发动机, 故障诊断, 贝叶斯网络, 模糊C均值聚类, 核主成分分析

Abstract: Aiming at the reliability problem of solid-liquid rocket engine,an improved Bayesian network fault diagnosis system was designed,which can diagnose multiple faults of solidliquid rocket engine through networking autonomous logic reasoning.In order to extract the fault features of time series observation signals,a scheme combining the marching method with the kernel principal component analysis (KPCA) was proposed.And then based on the fuzzy c-means clustering algorithm (FCM),a fuzzy polymorphic Bayesian network was established to realize the fuzzy processing of the scale of observation signals,which improves the diagnosis ability of uncertain faults.The improved Bayesian network fault diagnosis system was established by Matlab/Simulink.The simulation results show that the improved algorithm can effectively diagnose the common faults of solid-liquid rocket engine,and can adapt to the condition of small sample set.Compared with the traditional Bayesian diagnosis algorithm,the average accuracy of fault diagnosis is improved by 20.9%.

Key words: solid-liquid rocket engine, fault diagnosis, Bayesian network, fuzzy c-mean clustering, nuclear principal component analysis