Mathematical Modeling and Optimization Methods for Characterizing the Properties of Recycled Asphalt Concrete Mixtures

Abstract

A huge amount of waste asphalt mixture (RAP) is generated in the process of highway development, and in order to rationally utilize the resources, the recycling technology should be born. The study focuses on a series of mathematical modeling and optimization methods for characterizing the performance of recycled asphalt concrete mixtures. Through gray correlation analysis, the degree of influence of variables such as RAP admixture on the performance of the mix is identified. A BP neural network was introduced to construct a recycled asphalt mixture performance prediction model, and further combined with genetic algorithm to globally optimize the neural network parameters. Experiments on the performance of recycled asphalt in terms of needle penetration, softening point, ductility and rotational viscosity at different RAP dosages were carried out. The results showed that with the increase of recycled asphalt dosing, the needle penetration of aged asphalt increased from 28.18 to 64.89, the ductility increased from 14.10 cm to 58.55 cm, and the softening point decreased from 63.91°C to 48.00°C, which indicated that its low-temperature deformation capacity and construction and ease of use were significantly improved. The R² of the combination of BP neural network-genetic algorithm in four performance indexes, including dynamic stability, residual stability, etc., exceeded 0.9, and the predicted Pearson correlation coefficient of maximum bending and tensile strain reached 0.980, which is superior to the machine learning methods such as multivariate linear regression, vector machine, and random forest.