涡轮泵实时故障检测的改进自适应相关阈值算法

Improved adaptive correlation thresholds algorithm for turbopump real-time fault detection

关键词:  液体推进剂火箭发动机  涡轮泵  振动  实时故障检测+

Key words:  Liquid propellant rocket engine  Turbopump  Vibration  Real-time fault detection~+

摘要:

为了实时有效地检测液体火箭发动机涡轮泵的故障,避免分别计算各振动参数相同统计特征量的阈值和确定它们的故障检测逻辑,提出了改进自适应相关阈值算法(IACTA)。讨论了自适应相关算法(ACA)的原理,选用了线性相关性弱、能较好地反映涡轮泵故障特征的3路振动参数和3种统计特征量,并研究了这些统计特征量马氏距离平方阈值的自适应计算(自适应地计算均值和标准方差)和改进(双侧阈值改进为单侧阈值)方法,并剔除了故障数据对阈值的影响。在0.025的误检率下,通过8次转子碰摩试验的实时数据验证了IACTA故障检测的实时性;通过24次某型液体火箭发动机历史试车数据和196次转子碰摩数据验证了IACTA故障检测的有效性。结果表明,IACTA的平均故障检测延迟为226.625m s,故障检测率为96.0%。

Abstract:

In order to detect the Liquid Rocket Engine(LRE) turbopump faults in real time and effectively,and avoid estimating the thresholds of same statistical features for all vibration test parameters respectively and confirm their fault detection logics,the Improved Adaptive Correlation Thresholds Algorithm(IACTA) was presented in this paper.At first,the theory of Adaptive Correlation thresholds Algorithm(ACA) was discussed.And then three vibration parameters and three statistical features,which are weak linear correlation and well reflect the turbopump faults characteristic,were selected.At the same time,the thresholds adaptive estimation(the means and standard deviations adaptive estimation) and improvement(bilateral thresholds was improved to unilateral ones) methods for Mahalanobis distance square of these features were researched,and the thresholds refreshment without fault data were investigated.Under error probability 0.025,the IACTA fault detection real time capability was validated with 8 real time data from rotor rub-impact test;and its fault detection validity was validated with 24 historical data from LRE test and 196 data from rotor rub-impact test.In this case,it is shown that IACTA mean real-time fault detection time is 226.625ms,while its fault detection probability achieves 96.0%.