Abstract
This paper describes a new computational framework for characterising the effects of reclaimed asphalt pavement (RAP) parameters, sources and contents, on the structural performance of hot mix asphalt, in terms of dynamic modulus (E*), of using machine learning (ML) techniques. The ML models were developed for five RAP content percentages (0%, 10%, 20%, 30%, and 40%) for three different sources. The ML models showed promising results with Taylor diagrams and R2 values of 87% to 99% for all RAP sources and content levels. The results of pavement design using ML data showed the thicknesses of the asphalt mixture are slightly different depending on the RAP source. The same trend was found in the mixing and construction temperatures. Furthermore, analysis of sustainability showed that higher and stiffer RAP (higher E*) results in a more sustainable asphalt mixture because of lower thicknesses, and consequently lower aggregate requirements and CO2 emissions, for all ML methods and fuel types. Lastly, the incorporation of higher RAP content from any sources, the use of ML methods, and the adoption of cleaner fuel types results in massive material savings, higher energy efficiency, and improved sustainability in asphalt mix production for all RAP sources.