Physics-Informed Machine Learning for Multi-Second Aircraft Stall Early Warning and Cross-Mechanism Generalization

This paper presents AeroGuard, a physics-informed machine learning approach for multi-second aircraft stall early warning and cross-mechanism generalization. The study uses a two-dimensional longitudinal point-mass flight-dynamics simulator with RK4 integration and an emergent lift-coefficient stall curve to generate synthetic aircraft trajectories. A temporal Random Forest classifier is trained to identify pre-stall conditions from sequential flight-state data. The research investigates the effect of flight-control profiles on available physical warning time, evaluates temporal stall prediction performance, and tests transfer between different stall-inducing mechanisms. The study also examines a regime-exclusion experiment to assess the limits of generalization. Results demonstrate multi-second early-warning capability in the tested simulation conditions, while highlighting important limitations in zero-shot generalization. The work is simulation-based and is not flight-tested, certified, or intended to represent a deployable aircraft safety system.