基于AR/CGARCH模型的液体火箭发动机自适应阈值故障检测算法

Liquid Rocket Engine Adaptive Threshold Fault Detection Algorithm Based on AR/Compact GARCH Models

关键词:  液体火箭发动机  时间序列分析  自回归模型  自适应阈值算法  故障检测

Key words:  Liquid rocket engine  Time series analysis  Auto-regressive model  Adaptive threshold algorithm  Fault detection

摘要:

为了解决传统自适应阈值算法对时间序列方差跟踪能力不足，以及故障阶段带宽自动放大的问题，提出了紧广义自回归条件异方差（Compact General Auto-Regressive Conditional Heteroskedasticity，CGARCH）模型。针对液体火箭发动机稳态试车数据的波动性特点，提出一种基于自回归（Auto-Regressive，AR）模型和CGARCH模型的自适应阈值故障检测算法。采用AR模型对稳态参数的均值进行估计，并采用CGARCH模型对稳态参数的方差进行估计，从而利用均值和方差的估计值自适应地构造检测阈值。用某氢氧火箭发动机的热试车数据进行验证，结果表明，该算法能够准确、快速、灵敏地检测液体火箭发动机故障，在正常工作阶段，能够有效跟踪数据波动性，在故障阶段，能够避免阈值变宽带来的漏检。

Abstract:

In order to solve the problem of the incompetence of traditional adaptive threshold algorithm in tracking the variance of time series， and the problem of automatic amplification of threshold in fault phase， the Compact General Auto-Regressive Conditional Heteroskedasticity （CGARCH） model was proposed. According to the volatility characteristic of static test data of liquid rocket engine， an adaptive threshold algorithm based on Auto-Regressive（AR） model and CGARCH model was presented. Using AR model in mean estimation of static parameters， and CGARCH model in variance estimation， the predicted values of mean and variance may construct the detection threshold adaptively. After validating the algorithm with the hot test data of a LH₂/LOX engine， the results show that the algorithm enables accurate， fast and sensitive fault detection of liquid rocket engine. In normal working phase， the algorithm is able to track the data volatility effectively， while in fault stage， it avoids missed detection caused by increased threshold bandwidth.