FG-DeformT: Grouped Deformable Temporal Modeling for Flight-Level Aircraft Anomaly Detection.

Area of Science:

• Aerospace Engineering
• Data Science
• Machine Learning

Background:

• Aircraft anomaly detection is vital for flight safety and predictive maintenance.
• Traditional methods may overestimate performance due to data leakage between training and testing sets.
• A stricter cross-flight generalization assessment is needed.

Purpose of the Study:

• To propose a robust method for aircraft time-series anomaly detection.
• To introduce a leakage-aware flight-level validation protocol.
• To develop the FG-DeformT model for improved cross-flight generalization.

Main Methods:

• Developed FG-DeformT, a grouped deformable Transformer.
• Integrated global temporal modeling with a grouped temporal offset mechanism.
• Utilized a leakage-aware flight-level validation protocol for evaluation.

Main Results:

• FG-DeformT achieved high performance (Precision: 0.925, Recall: 0.941, F1-score: 0.933) on the ALFA dataset.
• Conventional baselines showed significant performance degradation under flight-level evaluation.
• The grouped offset mechanism proved beneficial for anomaly detection.

Conclusions:

• Leakage-aware flight-level evaluation is crucial for assessing generalization.
• FG-DeformT demonstrates robustness and effectiveness in aircraft anomaly detection.
• Grouped temporal deformation is a practical approach for modeling heterogeneous temporal patterns in critical telemetry data.