PROTECTING PILOT COGNITION: WORKLOAD MANAGEMENT IN AI-AUGMENTED COCKPITS

Cognitive overload in commercial aviation crews has contributed to fatal loss-of-control accidents in which degraded decision-making, not mechanical failure, was the proximate cause. Existing cockpit monitoring systems based on threshold alert logic detect behavioral consequences of overload, producing alert latencies that consistently exceed the window available for effective intervention. This article presents and evaluates a dual-path neuroadaptive cognitive monitoring framework that fuses electroencephalographic spectral features, heart rate variability indices, and galvanic skin response through a combined LSTM and SVM classifier architecture to achieve real-time three-level pilot cognitive state discrimination. The framework couples the classifier output to a graded cockpit intervention module governed by per-pilot calibrated thresholds and a hysteresis-based authority recovery protocol. Tested on 24 licensed commercial pilots across 144 high-fidelity A320 simulator sessions, the dual-path system achieved a three-class weighted F1 of 0.851, a critical overload recall of 83.1%, and a median detection latency of 7.3 seconds, outperforming an EEG-only threshold classifier by 22.2% (weighted F1) and a rule-based EICAS-modeled system by 31.4 percentage points in critical event recall under identical test conditions.