北京航空航天大学学报 ›› 2019, Vol. 45 ›› Issue (7): 1303-1313.doi: 10.13700/j.bh.1001-5965.2018.0663

• 论文 • 上一篇    下一篇

面向复杂系统的三维Bayes网络测试性验证模型

史贤俊1, 王康1, 肖支才1, 龙玉峰1, 陈垚2   

  1. 1. 海军航空大学, 烟台 264001;
    2. 海军研究院, 北京 100161
  • 收稿日期:2018-11-19 出版日期:2019-07-20 发布日期:2019-07-25
  • 通讯作者: 史贤俊 E-mail:sxjaa@sina.com
  • 作者简介:史贤俊 男,博士,教授。主要研究方向:自动测试与故障诊断,测试性设计、验证与评估;王康 男,博士研究生。主要研究方向:测试性设计、验证与评估;肖支才 男,博士,副教授。主要研究方向:自动测试与故障诊断,测试性设计、验证与评估;龙玉峰 男,博士研究生。主要研究方向:测试性设计、验证与评估;陈垚 男,硕士,研究员。主要研究方向:测试与故障诊断。

Three-dimensional Bayes network testability verification model for complex systems

SHI Xianjun1, WANG Kang1, XIAO Zhicai1, LONG Yufeng1, CHEN Yao2   

  1. 1. Naval Aviation University, Yantai 264001, China;
    2. Naval Research Institute, Beijing 100161, China
  • Received:2018-11-19 Online:2019-07-20 Published:2019-07-25

摘要: 针对当前武器装备复杂的系统结构,现有基于装备整机系统测试性先验信息的测试性验证方法难以适用,基于分系统测试性先验信息的测试性验证方法不能系统有效地处理先验信息,导致测试性验证结果可信度不高的问题,提出一种面向复杂系统的三维Bayes网络测试性验证模型。该模型能充分运用装备各层级结构中所蕴含的条件独立性,有效降低构建Bayes网络模型的复杂度,同时能融合装备各层级单元的先验信息。通过给出的三维Bayes网络的条件概率学习方法及G/M-H算法,由底层单元数据通过模型逐步向上融合,得到顶层测试性指标的后验分布,进一步利用顶层后验分布求取故障样本量。结果表明:该模型能充分考虑复杂系统的系统结构及各层级单元先验信息,并能通过模型推理得到的指标后验分布达到有效减少测试性验证故障样本量的目的。

关键词: 先验信息, 测试性验证, 复杂系统, 三维Bayes网络, G/M-H算法, 故障样本量

Abstract: In view of the complicated system structure of current weapon equipment, the existing testability verification method based on the testability prior information of the equipment system is difficult to apply, and the testability verification method based on the testability prior information of the subsystem cannot process the prior information effectively, which lead to the low credibility of the testability verification results, so a three-dimensional Bayes network testability verification model for complex systems is proposed. The model can fully utilize the conditional independence contained in the various hierarchical structures of the equipment, effectively reduce the complexity of constructing the Bayesian network model, and at the same time integrate the prior information of each hierarchical unit. Through the given conditional probability learning method and G/M-H algorithm of the three-dimensional Bayes network, the underlying unit data can be integrated through the model to obtain the posterior distribution of the top-level testability indicators, and the top-level posterior distribution is further used to obtain the fault sample size. The results show that the model can fully consider the system structure of the complex system and the prior information of each hierarchical unit, and the posterior distribution of the testability indicators can be used to reduce the fault sample size of testability verification.

Key words: prior information, testability verification, complex system, three-dimensional Bayes network, G/M-H algorithm, fault sample size

中图分类号: 


版权所有 © 《北京航空航天大学学报》编辑部
通讯地址:北京市海淀区学院路37号 北京航空航天大学学报编辑部 邮编:100191 E-mail:jbuaa@buaa.edu.cn
本系统由北京玛格泰克科技发展有限公司设计开发