基于数据统计的液体火箭发动机地面试车故障检测算法

研制了一种用于液体火箭发动机地面试车的故障检测算法，既适于发动机启动瞬态过程又可用于稳态过程。算法能设置一些预先设置或在线计算的参数门限，检查参数测量值的平滑值，当连续多次检查到多个参数异常时就判定发动机异常。用正常试车数据和满足启动条件的异常试车数据进行测试，没有出现误报警和漏报警。对满足启动条件的异常试车，均在现有的红线关机系统关机之前报警。

A STATISTIC APPROACH TO FAULT DETECTIONOF LIQUID ROCKET ENGINE IN GROUND TEST

Presents a statistic approach to fault detection of liquid rocket engine to handle starttransient as well as steady-state operating condition in ground firing test. The detection approachfirstly checks the smoothed measurement signals using rectangular filter against safety limits equal topredetermined as well as on-line calculated mean values plus/minus the product of standard deviationvalues and multiplying factors, then if NE1 (the number of anomalous parameters required forconfirming a fault) or more anomalous parameters are simultaneously reported several times in a row,a fault is signaled. The predetermined mean values and standard deviation values are computed fromnormal test data while those on-line computed are from reaI-time measured test data' The multiplyingfactors are automatically selected through training with normal test data. Computer simulation on tealnormal and abnormaI test data indicated the effectiveness and reliability of this fault detectionapproach.