Evaluation and prediction of interface fatigue performance between asphalt pavement layers: application of supervised machine learning techniques

Abstract

ABSTRACT Assessing interlayer fatigue performance is crucial for the durability and structural integrity of asphalt pavements. This study introduces supervised machine learning techniques to estimate interface fatigue life (IFL) and interface shear stiffness (ISS) between asphalt layers. Using 84 lab-prepared specimens tested under varying conditions-temperature, normal pressure, loading frequency and shear stress, with a constant tack coat type and application rate-predictive models were trained and validated. The influence of each factor on fatigue indices was evaluated before employing genetic expression programming (GEP) and artificial neural networks (ANNs) for prediction. Model efficacy was quantified using statistical metrics, highlighting the robustness of ANN models (R² = 0.980, RMSE = 0.826, MAE = 0.636 for ISS and R² = 0.972, RMSE = 0.498, MAE = 0.383 for IFL) over GEP models (R² = 0.921, RMSE = 1.402, MAE = 1.079 for ISS and R² = 0.913, RMSE = 0.623, MAE = 0.479 for IFL). Sensitivity analysis confirmed alignment with experimental data, identifying temperature as the most critical factor and frequency as the least influential on interface fatigue performance. These findings could lead to more reliable and efficient designs of interlayer bonding in asphalt pavement layers.