Towards Practical onboard PHM System for Aircraft Engines

Aircraft engines are essential for modern airplanes and monitoring their status is paramount for keeping our journey safe. Hence, various PHM systems for aircraft engines have been designed to supervise the engine operation and provide maintenance advice for engineers. The first step of those systems is to identify whether there exist faults from substantial amount of sensor data. Recent studies have shown the dominant capability of deep learning methods in discovering latent abnormal patterns, but the inherently high computational cost of deep learning models has made them impractical when deploying on embedded systems where onboard PHM systems reside. We proposed a more practical approach that only uses simple SVM model and achieved 86.7% overall accuracy for detecting gas path components degradation and 0.826 F1-Score for detecting rotor collision and abrasion on our engine dataset.