Optimizing asphalt mix design using machine learning methods based on RIOCHTrack data

Abstract

Abstract Traditional mix design is a time-consuming and labor-intensive process ,requiring extensive testing and relying heavily on engineering experience. In order to enhance the speed and efficiency of asphalt concrete mix design process,this study investigated the use of machine learning techniques to predict key parameters of concrete mixture design,such as voids in the mineral aggregate (VMA), voids in the coarse aggregate(VCA), and dry density of the mixture(pd). Four machine learning methods, namely support vector regression, artificial neural network, random forest, and AdaBoost models were trained using data from RIOHTRack. Metircs releatde to asphalt mix design such as gradation, asphalt content, asphalt properties, compaction method, and compaction temperature were used as input variables. Various encoding methods were employed to encode classification variables, with the ordinal encoding method yielding the most favorable results. Through the calculation of different performance scoring metrics, such as coefficient of determination (R2), root mean square error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE), and by plotting the development curve of volume parameters and asphalt content, the most suitable prediction model was selected for each target variable. The analysis revealed that the random forest model (R2 = 0.8595 for pd, R2 = 0.9488 for VMA) demonstrated the best performance in predicting pd and VMA, while the Adaboost model (R2 = 0.9716) was chosen for predicting VCA. By calculating different performance scoring metrics, such as coefficient of determination (R2), root means square error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE) and plotting the development curve of volume parameters and asphalt content, the final prediction model was selected for each target variable. The analysis revealed that the random forest model (R2 = 0.8595 for pd, R2 = 0.9488 for VMA) demonstrated the best performance in predicting pd and VMA, while the Adaboost model (R2 = 0.9716) was chosen for predicting VCA.