Determination of asphalt layer temperature in FWD and TSD measurements using machine learning

The paper presents the application of machine learning techniques in estimating the temperature of asphalt layers during measurements using FWD (Falling Weight Deflectometer) and TSD devices (Traffic Speed Deflection). The problem of accurate determination of temperature is crucial for analysing the durability of road pavements. Traditional methods such as the BELLS3 model, although widely used, have limitations in forecast accuracy. The work presents the implementation of advanced algorithms such as multivariate adaptive regression spline (MARS), support vector machines (SVM), artificial neural networks (ANN), random forest (RF) and boosted trees (BT), among others, to optimise a model for estimating the temperature of asphalt layers Td. The BELLS3 model, used as the baseline in the optimisation process, was evaluated for prediction effectiveness. The results showed moderate effectiveness of this model (R2 = 82%, RMSE = 2.3°C), which triggered a need for further improvements. The use of machine learning techniques, particularly boosted gradient trees (BTs), has made it possible to significantly improve the precision of predictions. The BT model achieved the greatest fit for the dependent variable Td (R2 = 99% and RMSE = 0.61°C), indicating its clear advantage over other models, including the baseline BELLS3 model. Finally, the authors highlight the potential of integrating traditional approaches with advanced data analysis methods to further improve the accuracy of forecasting bituminous mixture layer temperature and effective management of road infrastructure.