Parameter hierarchy governing subgrade vertical strain in flexible airfield pavements from an ANN-based SHAP analysis

Predicting the structural performance of airfield pavements is challenging due to material variability and the complexity of aircraft loading. This study used trained Artificial Neural Networks (ANN) as instruments to examine how design parameters govern the vertical strain at the top of the subgrade in flexible airfield pavements. A database of 17,500 three-dimensional Finite Element Method (FEM) simulations was built in ABAQUS, covering three pavement types, five narrowbody aircraft, and varying layer properties. A SHAP decomposition, conditional on the aircraft, revealed that the parameter hierarchy shifts with typology, the subgrade resilient modulus governing conventional sections and the base layer governing stabilised ones. Layer thickness consistently outweighed layer stiffness by a factor close to three, as predicted by Odemark's equivalent thickness, and a stiff treated base attenuated the dependence on the foundation, an effect absent within the narrow modulus range permitted for granular materials.