A mechanism-aided transformer may transform in-flight aviation safety

An accurate forecast on the in-flight parameters and aviation status is crucial for aviation safety, enabling timely responses to potential upcoming urgent risks during flights. Conventional physics-based method, the F-Factor, a wind shear hazard index associated with the total aircraft energy and its rate of change, face limitations such as sensitivity to physical assumptions, high false alarm rate, and high uncertainties. The advent of machine learning (ML) has significantly influenced various prediction tasks, offering new possibilities for aviation-related forecasting and hazard alarming. Here, we present a physical mechanism-aided TwinTransformer approach that integrates aviation dynamic constraints with deep learning to improve multivariate in-flight prediction, with targeted enhancement of windshear hazard alarming. Tested on real-world in-flight datasets, the proposed approach can consistently outperform cutting-edge benchmarks and provide pilots with reliable 15-s-ahead windshear hazard warnings. These results demonstrate a practical path toward intelligent decision support in aviation safety by AI and highlight how combining physical knowledge with ML can enhance in-flight aviation safety against windshear hazards.