Introduction. Rutting of flexible, super flexible and semi-rigid pavement structures is a typical and frequently decisive condition parameter, form of pavement deterioration. That is why, any research result in the field can have of high importance for the road engineers.
Problem Statement Rutting poses a significant challenge to asphalt pavements, causing permanent deformation under heavy loads, particularly in warm and wet conditions.
Purpose. This pavement distress type has — in addition to riding comfort challenges — important traffic safety consequences (e.g. aquaplaning), as well. The research work concentrates on the influence of the use of warm mix asphalt, reclaimed asphalt material and foamed bitumen binder on the rut depth of asphalt pavements.
Materials and Methods. In integration of machine learning techniques, a Feedforward Neural Network model was presented to analyse the relationship between pavement rut depth and Reclaimed Asphalt Pavement (RAP) content. The model, trained for RAP content ranging from 0 % to 100 %, showcased varying R-squared values, with the highest at 50 % RAP content. Additionally, a Gaussian Process Regression (GPR) model was employed, highlighting the significant effects of RAP content between 75 % and 100 %. Sensitivity analysis on the GPR model provided insights into parameter effects, while the significant influence of the number of wheel passes on pavement rut depth values emphasized the importance of optimal road maintenance timing.
Results The results of the machine learning model indicated a R-squared value of 0.476 for 0 % RAP content and higher values for mixtures containing RAP, with the highest value of 0.897 was found for 50 % RAP content in the asphalt mixture. A Gaussian Process Regression (GPR) model applied showed paradoxical effects between 75 % and 100 % RAP content. The derivative of the predicted mean rut depth as a function of RAP content revealed varying effects on Rut depth values in the case of different RAP content ranges. Sensitivity analysis on the GPR model was conducted by varying parameters such as Length Scale, Noise Level, and Amplitude. The results of this analysis provided insights into how changes in these parameters affected the mean squared error (MSE) for their various combinations. The influence of the number of wheel passes on rut depth values was examined, showing a significant increase in rut depth after 12,000 passes and reaching its maximum value after 20,000 passes.
