Machine learning techniques to estimate the degree of binder activity of reclaimed asphalt pavement
Ramón Botella Nieto, Davide Lo Presti, Kamilla Vasconcelos, Kinga Bernatowicz, Adriana H. Martínez, José Rodrigo Miró Recasens, Luciano Pivoto Specht, Edith Arámbula, Gustavo Pires, Emiliano Pasquini, Chibuike Ogbo, Francesco Preti, Marco Pasetto, Ana Jiménez del Barco Carrión, Antonio Roberto, Marko Оrešković, Kranthi Kumar Kuna, Gurunath Guduru, Amy Epps Martin, Alan Carter, Gaspare Giancontieri, Ahmed Abed, Eshan Dave, Gabrielle Tebaldi
Abstract
Abstract This paper describes the development of novel/state-of-art computational framework to accurately predict the degree of binder activity of a reclaimed asphalt pavement sample as a percentage of the indirect tensile strength (ITS) using a reduced number of input variables that are relatively easy to obtain, namely compaction temperature, air voids and ITS. Different machine learning (ML) techniques were applied to obtain the most accurate data representation model. Specifically, three ML techniques were applied: 6th-degree multivariate polynomial regression with regularization, artificial neural network and random forest regression. The three techniques produced models with very similar precision, reporting a mean absolute error ranging from 12.2 to 12.8% of maximum ITS on the test data set. The work presented in this paper is an evolution in terms of data analysis of the results obtained within the interlaboratory tests conducted by Task Group 5 of the RILEM Technical Committee 264 on Reclaimed Asphalt Pavement. Hence, despite it has strong bonds with this framework, this work was developed independently and can be considered as a natural follow-up.