Development of a model for the estimation of indirect tensile strength of RAP speciments using machine learning methods

Abstract

A model for prediction of Indirect Tensile Strength (ITS) of Reclaimed Asphalt Pavement (RAP) specimens is developed in this study using Machine Learning (ML) technique. Principal Component Analysis (PCA) was used to transform grading curves of RAP and obtain reduced amount of data for further analysis. Different Multivariate Polynomial Regression (MPR) models were developed considering properties of RAP (aged binder content and its penetration, grading curves before and after extraction (black and white curves)), manufacturing process (preheating temperature) and properties of testing samples (air void content). Analysis showed that PCA can be adequately used to reduce the number of variables required to describe grading curves (74% of variance was described with first two principal components). Additionally, it was concluded that the simplest (Linear) MPR Model was the most precise overall - coefficient of the determination was 0.59, which can be considered as quite high for such a data set (more than 40 RAPs from different sources were analyzed).