数据驱动与解析模型结合的民机飞控系统细粒度故障诊断

The combination of a multivariate statistical fault detection approach based on data-driven and a fault estimation method based on the adaptive nonlinear fault observer of analytical models with depth is applied to fault diagnosis of the civil aircraft flight control system, taking their respective advantages with mutual support and supplement. This research is important in realistic significance as well as theoretical value..An improved sliding window multiway kernel dynamic principal component analysis (SW-M-K-D-PCA) monitoring method will be put forward. Key issues of data-driven fault detection such as sub-stages' accurate segmentation and transition sections' fine extraction, synchronous alignment for unequal length data, active adaptive update of slowly time-varying statistical models can be solved, which realizes the system-level closed-loop fault detection and gives coarse-grained detection results of all component parts of the flight control system..A design approach for the novel adaptive nonlinear fault observer based on analytical models will be proposed. The fault estimation method is effective in relaxing the strictly positive real (SPR) condition of error dynamics, which expands the application scope of adaptive nonlinear observer and increases estimation rapidity. In addition, the fault observer can further give precise estimation results for core components of the flight control system based on coarse-grained closed-loop detection results, realizing fine-grained fault diagnosis eventually..A physical system will also be developed to serve in the civil aircraft maintenance based on above theoretical research.

把基于数据驱动的多元统计故障检测方法和基于解析模型的自适应非线性观测器故障估计方法深度结合，充分发挥各自优势并使两者相互支持、相互补充，应用于民机飞控系统的故障诊断。该研究具有重要现实意义和理论价值。.创立一种改进的滑动窗口多向核动态主元分析(SW-M-K-D-PCA)监控方法；解决数据驱动故障检测中的多时段数据精细划分与过渡时段数据准确提取、不等长数据同步对齐和慢时变统计模型主动自适应更新问题，实现系统级规模的闭环故障检测，给出飞控系统全部组成部件的粗粒度检测结果。.提出一种基于解析模型的新型自适应非线性故障观测器设计方法；有效地放松自适应观测器设计中存在的误差系统严格正实约束条件，扩展自适应非线性观测器的适用范围并提高故障估计快速性，在粗粒度闭环检测基础上进一步给出飞控系统核心部件的精确故障估计结果；最终实现细粒度故障诊断。.基于上述理论成果开发一套应用于民航机务维修的实物系统。

在充分尊重中国民航现有维修体系框架前提下，从数据驱动和解析模型相结合的角度出发，以一个新颖的视角来重新审视我们现有的海量运行数据，选择并开拓出一套富含中国民航特色的数据处理和飞控系统故障诊断方法，最终落脚于民航机务维修。本课题首先采用多元统计的数据驱动方法在不需精确已知解析模型的情况下完成大规模闭环系统的粗粒度故障检测；然后采用基于解析模型的故障估计方法在粗粒度故障检测的基础上进一步给出了飞控系统I类关键部件的“细粒度”故障估计值，使得故障描述更加精细、故障诊断结果更加明确。在研究过程中解决了过渡时段数据的细划提取、多时段背景下的数据不等长对齐、慢时变统计模型主动自适应更新和放松严格正实(SPR)条件的快速自适应非线性观测器设计四个理论难题；开发完成了一套应用于民航机务维修的故障诊断实物系统并通过了现场测试。课题资助正式发表学术论文19篇，其中SCI检索期刊论文15篇，EI检索期刊论文1篇，中文核心期刊论文2篇，EI会议论文1篇；出版图书1本；获得国家发明专利授权2项；项目负责人获得省部级荣誉称号1项，省部级人才称号2项，获得省部级研究成果一等奖1项；培养硕士生5名，其中1篇硕士论文被推荐申报天津市优秀硕士论文；课题研究期间，共采集并深入分析了约20G的飞控系统I类部件关键性能数据，目前飞控系统运行数据的规模达到30G，能够完全覆盖I类部件的全寿命周期。项目投入经费40万元，支出17.4393万元，剩余经费22.5607万元，剩余经费计划用于本项目研究后续支出。本课题研究所取得的理论和实践成果紧密围绕着民用飞机这类典型高端复杂航空装备综合保障的实际需求，对于推动中国民航维修理论创新和新型装备研发具有战略层面的重大意义。

内容获取失败，请点击重试