幂集辨识框架下置信规则库建模方法及在导弹武器故障预测中的应用

On the context of the key components of the new missile weapon system, the failure prognosis and optimal testing instant optimization problems of the complex engineering system are studied. The existing failure prognosis methods are mainly focused on the complex engineering systems in the continuous running state, which can not effectively deal with the failure prognosis problem and determine the optimal testing instant of those systems under conditions including state transition, numerous fault characteristics, strong coupling, serious noise interference and so on. Therefore, this project aims to solve the above problems. In theory: (1) Belief rule based modeling method under identification framework of power set (PBRB-R) is proposed for the first time, and the basic theoretical issues about the online modeling of the complex engineering system based on PBRB-R is tackled through solving the problems of PBRB-R model’s description, reasoning, and optimization of the structure and parameter. Moreover, a series of original research results can be obtained; (2) Based on the PBRB-R, a hidden failure prognosis model of complex engineering system, which takes into account the state transition, fault features’ correlation and reliability, is established, and then an online hidden failure prognosis method is further proposed; (3) Based on the results of the failure prognosis, an optimization model which takes the testing time as the decision variables and the future safety and the test loss as the decision target is developed. Meanwhile, an optimization algorithm based on covariance matrix adaption evolution strategy (CMA-ES) is presented to determine the optimal testing instant. In engineering, the failure prognosis method and the optimal testing instant optimization algorithm are verified by using the testing data of the new missile control system’s key components, which can improve the reliability and actual combat level of weapons.

以某新型导弹武器关键部件为背景，研究复杂工程系统的故障预测与最佳测试时机优化问题。现有的故障预测研究多针对处于连续运行状态的复杂工程系统，无法有效解决存在状态切换、故障特征众多等情况的故障预测模型建模和最佳测试时机确定问题。鉴于此，在理论上：（1）率先提出幂集辨识框架下考虑前提属性可靠性与相关性的BRB模型（PBRB-R模型），突破PBRB-R模型的描述、推理、结构与参数联合优化学习方法等难题，取得一系列原创性研究成果；（2）基于PBRB-R模型，提出考虑状态切换和故障特征可靠性与相关性的隐含故障在线预测方法；（3）基于故障预测结果，构建以测试时机为决策变量，以系统未来安全性和测试损耗为决策目标的优化模型，并提出基于CMA-ES的最佳测试时机优化算法。在工程上，利用某新型导弹控制系统关键部件测试信息对故障预测和最佳测试时机优化算法进行验证，为提高武器装备实战化水平服务。

对诸如导弹武器关键部件等复杂工程系统进行故障预测和最优测试时机优化具有重要意义。申请人在基金的资助下，系统深入地研究了幂集辨识框架下置信规则库（PBRB）建模方法，基于此提出了一种新的考虑属性可靠性的PBRB模型（PBRB-R）及其结构与参数联合优化学习算法，解决了存在状态切换、故障特征众多等情况下系统的故障预测建模和最佳测试时机优化问题。本项目在拓展BRB基本理论的基础上，建立了幂集辨识框架下考虑属性可靠性与相关性的BRB模型，提出考虑状态切换和故障特征可靠性与相关性的隐含故障在线预测方法，提出了基于协方差矩阵自适应优化策略（CMA-ES）的最佳测试时机优化算法，并将所提方法应用于某新型导弹控制系统关键部件。本项目取得的成果包括：出版专著和译著各1部；以第一作者或通讯作者录用和发表学术论文31篇，其中ESI高被引论文1篇；申请专利共13件，授权专利2件、软著作权6件。