Anomaly detection of aircraft system using kernel-based learning algorithm

A real-time health monitoring framework is developed in this work to detect in-flight operational anomalies in aircraft subsystems. Relevant features with similar eigenvectors that characterizes dynamic flight behavior is extracted and used to train flight behavior and detect operational anomalies by comparison with statistical safety bounds. Additionally, the monitoring framework is implemented for real-time application by adopting kernel functions for computational acceleration. The accuracy and efficiency of the proposed algorithm is demonstrated with several case studies of operational anomalies in the aircraft engine systems.