Shallow Bearing Capacity of Unsealed Roads Using FELA and Machine Learning

Abstract

Abstract The bearing capacity of unsealed roads plays a central role in maintaining serviceability, especially because moisture variations and traffic-induced shear stresses govern the shallow surface instabilities that lead to corrugation. Traditional analytical approaches of bearing capacity estimations typically assume general shear failure involving the entire base–subgrade system and ignore horizontal tyre-induced shear stresses and moisture-driven strength gradients in unbound granular materials (UGM). Neglecting these factors leads to systematic overestimation of bearing capacity in the upper 20–100 mm, where shallow shear failure and corrugation are known to initiate. This study develops a hybrid modelling framework combining finite element limit analysis and machine learning to improve the predictions of unsealed road bearing capacity under realistic loading and moisture conditions. Numerical simulations were conducted using a two-layer base structure representing depth-dependent seasonal strength variation and incorporating both vertical and horizontal tyre interface stresses to capture shallow failure mechanisms. A parametric study examined key factors: top layer cohesion, internal friction angle, layer thickness, dynamic friction coefficient and tyre contact radius. Results show that surface cohesion and dynamic friction coefficient are the most influential parameters, with higher shear stress promoting localised surface failures and reducing mobilisation of shear strength of the deeper layer. A large dataset generated from numerical simulations was used to train two machine learning models: an artificial neural network (ANN) and a two-stage random forest (RF). The RF model achieved the highest accuracy (R 2 = 0.99, RMSE = 83.3 kPa). Sensitivity analysis and visual validation confirmed that corrugation initiation risk is strongly governed by surface shear stress and top-layer shear strength. This framework provides a physics-informed tool for evaluating corrugation initiation risk and offers practical guidance on tyre pressure control, surface material selection, and moisture management to enhance unsealed road performance.