Asphalt concrete overlay thermal reflective cracking stress intensity factor prediction using machine learning

This work proposed an machine learning (ML)-based framework to predict stress intensity factor (SIF) in reflective cracking of asphalt concrete (AC) overlay under thermal loading, to predict overlay fatigue life. A database from 8750 simulations of 3D finite element pavement models was established to predict SIF. Considering the linear relationship between crack opening and SIF, two conditions were explored to predict SIF: (i) Crack opening was included in input variables along with AC layer thickness and modulus, Portland cement concrete (PCC) modulus, and crack propagation; and (ii) crack opening was excluded from input variables, maintaining other variables. Artificial Neural Network (ANN) outperformed other ML models in both conditions. The optimal input variables for SIF prediction should include AC layer thickness, AC modulus, PCC modulus, and crack propagation. In conclusion, ANN could be an accurate and robust method for predicting SIF in reflective cracking of asphalt overlay subjected to thermal loading.