Interface Bonding Strength between Asphalt Pavement Layers under Mixed Shear-Tensile Mode: Laboratory Evaluation and Modeling Predictions

Abstract

Highway traffic loads and environmental conditions, including temperature and moisture at the layer interface, could cause debonding or delamination between adjacent asphalt pavement layers in tensile, shear, or mixed-mode failures. Thus, studying the tensile and shear strengths of interface bonding is crucial to maintaining durable and functional pavement structures. The main objective of this research is to evaluate and estimate the interface bonding strength (IBS) between asphalt pavement layers under the mixed shear-tension loading mode. To this end, an experimental, statistical, and machine-learning (ML) approach was adopted. A total of 164 double-layered hot-mix asphalt specimens consisting of hot-mix asphalt AC-13 in the upper layer and AC-20 in the lower layer were tested via a direct tensile device with a supplementary shear fixture. The effects of test temperature, shear stress, and tack coat application rate on the IBS were considered. The results revealed that with increasing tack coat dosage, the IBS peaked at 0.8 kg/m2 and was subsequently decreased. It was also found that the IBS was very sensitive to temperature changes and heavily dependent on shear stress at elevated temperatures. On the other hand, with increasing shear stress from 0 to 0.20 MPa, the IBS at temperatures of 5°C, 20°C, and 35°C declined by 10.96%, 61.85%, and 83.16%, respectively. Two prediction models for the IBS based on the conventional statistical models of multiple linear regression (MLR) and nonlinear regression were successfully developed. However, the nonlinear model outperformed with a better prediction accuracy of 24.2% compared with linear regression (R2=71.8%). Finally, a highly accurate feed-forward back-propagation (FFBP) artificial neural network (ANN) model was developed to predict and form a relationship between the IBS and independent variables with an extremely low margin of error. It was revealed that the developed FFBP-ANN model could capture 99% of the measured data. Finally, a comparative analysis demonstrated that the developed FFBP-ANN model was superior to regression modeling in terms of predicting the IBS.