Flight Parameter Exceedance Classification for Aircraft State Monitoring via Data-Augmented Self-Attention

To address classification challenges in flight parameter exceedance events during landing, specifically overlapping class boundaries and sensor noise, this study presents a framework that integrates tailored data augmentation with adaptive attention mechanisms. The method improves feature discrimination in overlapping regions, achieving 97% accuracy on open datasets and outperforming conventional architectures by 4%. Gradient-based attribution analysis identifies critical flight control variables, such as pitch angle, angle of attack, and altitude. This framework enhances operational transparency for aircraft deflection assessment and serves as a valuable tool for analyzing flight risk levels, identifying root causes of flight decisions, and evaluating flight quality.