Modeling and prediction of surface free energy of pavement materials based on a single probe liquid using artificial neural network

Abstract

Surface free energy (SFE) is a vital material property employed for the assessment of various pavement performances such as asphalt-aggregate adhesion, moisture resistance, material compatibility, etc. However, the conventional models for estimating the SFE of materials require the contact angles (CA) of multiple probe liquids (PL) on the material's surface. In addition, the process requires further computational effort to arrive at the SFE components. In an attempt to offer an accurate and quicker alternative method of estimating the SFE of materials with lesser effort, this study has explored the use of artificial neural network (ANN) to predict the SFE of pavement materials. A simple but novel data structuring was employed to predict the SFE using inputs from just a single PL. Variables needed for the successful prediction of SFE components were identified and studied. Sensitivity analysis along with relative importance was used to rank the predicting variables. The Lifshitz-van der Waals (LW) component ANN-model showed the best accuracy (RMSE = 1.652 J/m2, R = 0.984). The base and acid components models showed relatively lower performances, (RMSE = 2.560 J/m2, R = 0.943) and (RMSE = 0.575 J/m2, R = 0.818) respectively. However, the SFE components estimated from the ANN-models yielded a total SFE with very good accuracy (RMSE = 2.049 J/m2, R = 0.982). Further insight into which variables such as substrate-type, CA test method, PL type, etc. facilitate model accuracy was provided.