Research on an Adaptive Threshold Setting Method for Aero-Engine Fault Detection Based on KDE-EWMA

Abstract

It is challenging to set a precise threshold in fault detection and isolation, which helps in reducing false alarms and missed detection rates. In this paper, an adaptive threshold approach is developed for aero-engine fault detection. Based on kernel density estimation (KDE) and backward exponentially mean filtering method, the adaptive threshold setting result for a single steady-state point of aero-engine fault detection is obtained. The flight envelope is reasonably divided, and the fault detection threshold is obtained in each flight subarea. The exponentially weighted moving average (EWMA) method is used to obtain a threshold setting at different performance degradation levels throughout the life of the aero-engine. Then the proposed threshold setting method is utilized to compare the two traditional fixed threshold setting methods and a double threshold-based method. The results show that the proposed adaptive threshold setting method performs better in the fault detection under a single steady-state point. To be specific, the detection time was shortened by 0.44, 0.72, and 0.56 s, and the fault detection rate was increased by 0.46%, 6%, and 0.13%, respectively.