基于滑动时间窗主成分分析的液体火箭发动机传感器故障诊断方法

安装在发动机上的各种传感器是发动机状态监测的主要依据，由于工作环境恶劣，传感器失效时有发生。由于发动机运行过程中的性能蜕变和台次差异，现有基于主成分分析（PCA）的传感器故障隔离方法应用条件苛刻且诊断效果有限。针对这些问题，在对发动机数据分析的基础上，将滑动时间窗方法与PCA方法结合，提出双滑动时间窗的PCA方法用于故障传感器的隔离，并基于发动机试车数据进行了方法验证。结果表明：该方法能降低发动机性能蜕变和台次差异对发动机传感器故障诊断的影响，没有参数相关性的限制，可以实现对四种常见传感器故障的有效隔离，以及对两种发动机试验过程中故障的准确检测。研究证明了高速运转系统性能蜕变和强耦合复杂大系统台次差异对基于数据的故障诊断方法效果的影响，验证了在线学习/训练算法对这两种现象的鲁棒性。

Sliding Time Windows Principal Component Analysis Based Fault Diagnosis Method for Liquid Rocket Engine Sensors

Sensors installed on the engines are the main basis for monitoring the condition of the engines, due to the harsh operating environment, sensor failures occur frequently. Due to the metamorphosis of the engine performance during operation and differences among engines of the same model, existing methods for isolating sensor faults based on principal component analysis are applied under strict application conditions and have limited diagnostic effect. In response to these issues, the PCA method with a double sliding time window is proposed for fault sensor isolation based on the analysis of engine data, and validation was carried out based on engine test data. The results show that the method reduces the impact of the metamorphosis during operation and differences among the same type engines on engine sensor fault diagnosis, has no parameter correlation limitations and enables effective isolation of four common sensor faults as well as accurate detection of two types of engine faults. This study demonstrates that the metamorphosis of high-speed operating systems and the differences among strongly coupled complex large systems have significant impacts on the effectiveness of data-based fault diagnosis methods, the robustness of the online learning/training algorithm to these two phenomena is verified.