数据驱动的复杂结构动态系统故障预测与诊断

The on-orbit spacecraft is a tripcal dynamic system with complicated structure. Recently,there were a lot of faults and failures which took place in the international fields of spaceflighting engineering.It is very obligatory to explore some new techniques of faults prognostics/diagnostics by means of expansive approaches.In this proposal application, the researches aim at the three weakness aspects in the techniques fields of faults prognostics/ diagnosis for on-orbit spacecraft. Relying on the combination of the data-driven mode with the problems-driven mode, five topics of the main research contents will be explored and be solved, which include the following techniques of spaceflighting data mining and modality modlling, the analysis of associate relevancy in different changes, changes detection and faults diagnosis, tredency analysis and the long-term predicting algorithms of telemetry data, faults prognostics and health management of the on-orbit spacecraft,etc.In our research, three key scientific problems will be solved, which are the problem of modality classification and modelling telemetry non-stationary data series with outliers,strong noise and changes,the problem of location inference and forecast of faults by means of the changes information of telemetry in the case that the changes of telemetry data are mapped into faults of components in spacecraft with strong compuling relationship,and the problem of long-term preditcion and across-cycle change forecast of telemetry data series from the similar intermission process.At the end of our research, a technical frame and software platform will be built to integrate many kinds of functions, such as data mining, processes monitoring,faults diagnosis, faults prognostics and health management,etc. In our research,a series of technique innovation will be created in the following aspects: modality classification of telemetry data series, associate relevancy analysis of multidimensional telemetry data changes, multi-faults location and prognostics,etc. It is sure that these scientific researches are helpful for us to expand the fields of diagnosis & prognostics of faults in the dynamic system with complex structure and to increase the safety of spacecrafting engineering.

在轨航天器属于典型的复杂结构动态系统。近年来国际国内航天故障频繁，急需多途径探索航天故障预测与诊断技术。本项目以航天器在轨过程故障预测与诊断技术发展的3大薄弱环节为研究对象，采用数据驱动与问题驱动结合的方式，开展遥测数据挖掘与形态建模、异变过程关联分析、异变检测与故障诊断、趋势分析与长程预报、故障预测与健康管理等彼此衔接的5个方面技术研究，解决带杂质强干扰含异变非平稳数据的形态聚类与数学建模、遥测数据异变与系统部件故障"多对多"强耦合情形下的故障定位与故障预测、类间歇过程长程预报与跨周期异变预警等3方面关键科学问题，建立一套集数据挖掘、过程监控、故障诊断、故障预测和健康管理等多功能于一体的技术架构与软件平台,并在遥测数据形态聚类、多通道异变关联分析、多故障定位与故障预测等方面取得显著创新,有效拓展复杂结构动态系统故障预测与诊断技术的发展空间，并为航天器在轨安全运行提供有力的技术支持。

在轨航天器属于典型的复杂结构动态系统，急需多途径探索航天故障预测与诊断技术。本项目以在轨过程故障预测与诊断技术发展的3大薄弱环节为研究对象，采用数据驱动与问题驱动结合的方式，开展遥测数据挖掘与形态建模、异变过程关联分析、异变检测与故障诊断、趋势分析与长程预报、故障预测与健康管理等彼此衔接的5个方面研究。. 通过四年的探索和研究，项目已保质保量地完成了预定研究计划，在数据挖掘与建模、遥测数据关联分析、异常变化检测与趋势预测、状态趋势分析与长程容错预报、数据驱动的故障诊断与趋势预测、数据驱动的故障预测与健康管理，以及相关技术的应用等方面取得了系列性研究成果。在遥测数据的挖掘与知识发现方法、数据门限内异常变化的在线检测与状态监视、基于容错滤波的遥测数据“轴心+扰动+异常变化”三项分离、多维遥测数据的异常变化与故障关联分析、复杂结构过程故障定位、趋势预测与规则推理结合实现故障预测、自适应容错指数平滑预报等7个方面取得了显著的技术创新。. 研究过程中，发表论文49篇（EI收录11篇；Web of Science收录22篇）；获得省部级科技进步三等奖和二等奖各1项、通过部委级会议鉴定成果1项、著作2本、国家行业标准2项、国家软件著作权3项；结合项目研究，申报国家发明专利7项，培养硕士/博士研究生9人（已毕业4人），形成“可靠性增长与延寿技术”科研团队1个；组织国内学术技术交流会2次；在领域/行业会议上，受邀学术技术报告9次。. 在开展方法/技术研究的同时，关注研究成果应用转化，做到了针对工程技术难题边研究边转化应用，不仅形成了集多功能于一体的状态监视、趋势分析、异变预测、故障预警、故障诊断与远程诊断技术平台，而且部分技术成果已被成功地应用于多个不同平台异常状态分析、变化趋势预测与故障检测诊断,有效拓展了复杂结构动态系统故障预测与诊断技术的发展空间，为航天器在轨安全运行提供了有力的技术支持。