Enhanced Isolation Forest Algorithm for Aero-Engine Fault Detection

Fault detection in aero-engines is critical for ensuring flight safety and operational efficiency. Detecting faults in aero-engines is challenging due to the complexity of the systems and the noisy nature of the sensor data. Traditional methods often fail to accurately capture the dynamic behavior of the engine over time. The isolation forest algorithm, known for its linear time complexity and low storage requirements, offers significant advantages in handling large-scale datasets and does not rely on distance or density measures. This study introduces an enhanced isolation forest algorithm incorporating a sliding window mechanism to optimize fault detection in streaming data environments, specifically for aero-engine applications. The proposed method includes a novel feature selection strategy based on the extreme difference of features, which mitigates errors caused by independent parameter introduction. Additionally, four strategies for updating anomaly scores are developed to improve detection accuracy. The algorithm's effectiveness is validated through simulations on three types of aero-engine failure data, demonstrating a high detection accuracy with response promptness indices F1 and F2 averaging 99.46% and 99.97%, respectively. This improved algorithm proves to be highly effective in detecting abrupt anomalies, showcasing superior performance in real-time aero-engine fault monitoring.