在轨航天器诊断策略自动构建与学习调控方法研究

项目名称： 在轨航天器诊断策略自动构建与学习调控方法研究

项目编号： No.61503398

项目类型： 青年科学基金项目

立项/批准年度： 2016

项目学科： 其他

项目作者： 张士刚

作者单位： 中国人民解放军国防科技大学

项目金额： 21万元

中文摘要： 航天器状态监控和故障诊断是保证其安全运行的重要手段。当前，由于缺乏随遥测数据和运行知识积累而更新进化的指导理论和方法，其诊断策略一旦确定后很难改变，导致航天器运行管理系统普遍存在虚警率偏高、诊断精度不足的问题。为此，项目基于机器学习和贝叶斯网理论开展诊断策略自动构建和学习调控方法研究。首先，通过分析多信号模型和诊断贝叶斯网之间的映射关系，阐明多信号模型和贝叶斯网之间的转化机制，建立诊断贝叶斯网自动构建方法。然后，研究揭示诊断贝叶斯网模型缺陷对诊断精度的影响规律，构建模型缺陷识别和评价方法。在此基础上，建立学习进化模型，设计求解算法，实现诊断策略的学习调控。. 该项目提出的自动构建诊断模型并在应用中逐步学习熟化的方法为解决当前诊断系统中虚警率偏高且难以消除的普遍性问题提供了一条新的解决途径。所取得的贝叶斯网自动构建方法等成果对丰富贝叶斯网和机器学习理论具有重要意义。

中文关键词： 故障诊断；诊断策略；贝叶斯网；机器学习；航天器

英文摘要： Condition monitoring and fault diagnosis is an important method to guarantee the safe operation of a spacecraft. However, the diagnostic strategy can hardly be changed once it is designed because the lack of efficient approach that can update the diagnostic strategy using telemetry data and accumulated knowledge, which may lead to existence of false alarms and undetectable faults in the spacecraft operation management system. Consequently, this project studies diagnostic strategy auto-construction and regulation method for in-orbit spacecraft based on machine learning and Bayesian Networks theories. Firstly, by analyzing the relationship between multi-signal flow graph model and Diagnostic Bayesian Networks (DBN), the transformation mechanism of them is elucidated, and then a DBN auto-generation method is proposed. After that, the influence law between DBN defect and diagnostic accuracy is investigated, based on which a defect recognition and evaluation method of DBN is developed. Then, a diagnostic strategy evolution model is constructed. A solution algorithm is designed, by means of which diagnostic evolution and regulation is achieved..The proposed approach that a diagnostic model is automatically constructed and then matured during the application process provided a new solving way to the false alarm problem universally existing in the fault diagnosis systems. The achievement such as DBN auto-generation method has important significance to develop Bayesian Networks and machine learning theories.

英文关键词： fault diagnosis;diagnostic strategy;Bayesian Networks;machine learning;spacecraft