Machine learning-based predicted stress intensity factor to estimate reflective cracking in airfield asphalt concrete overlay under aircraft loading

This study introduced a machine learning (ML)-based approach to predict the stress intensity factor (SIF) for reflective cracking in asphalt concrete (AC) overlay subjected to aircraft loading. A 3D Generalized Finite Element Method (GFEM) was developed to compute SIF for reflective cracking under aircraft loading. Subsequently, an extensive 3D GFEM pavement database comprising 3,101,679 datapoints was constructed, considering various influential variables. Four ML models were trained and evaluated for SIF prediction. GFEM results demonstrated that ratio and specific values of lateral and longitudinal loading locations exhibited different robustness and sensitivity to simulation results. Consequently, four cases were investigated to determine optimal input and output variables. Artificial Neural Network (ANN) achieved the best overall performance when using specific values of lateral and longitudinal loading locations for predicting all three SIF values. ANN demonstrated superior prediction capability for SIF with different longitudinal loading locations compared to SIF with different lateral loading locations. Sensitivity analysis revealed longitudinal loading location as the most important input variable for predicting all three SIF values, followed by lateral loading location. In conclusion, ANN proved to be an accurate and robust method for predicting SIF of AC reflective cracking due to aircraft loading.