Prediction of road properties of asphalt mixture subjected to three times aging-regeneration cycles based on a GA-BP neural network asphalt binder

Materials genome research has been rapidly evolved, aiming at the development of future pavement materials. It has been gradually applied to studying the properties of asphalt and asphalt mixtures. In this study, the prediction of the road properties of asphalt mixtures using asphalt binders subjected to multiple aging and regeneration cycles was systematically explored using various experimental tests. Additionally, various characterizations were carried out to analyse the variation law of road properties of the asphalt mixtures after three aging-regeneration cycles. Finally, a Genetic Algorithm-Back Propagation (GA-BP) neural network was adopted to establish a prediction model for the performance of asphalt mixtures based on asphalt binders subjected to multiple aging-regeneration cycles. Results showed that the penetration finally recovered to 80.7%, and the softening point ultimately reached 115% of that before aging. However, the road properties of the asphalt mixtures after the implementation of three aging-regeneration cycles presented a differentiated evolution. In terms of high-temperature performance, the dynamic stability reached 183.8% and the penetration strength rose to 150% with the increase times of regeneration. Regarding the low-temperature performance, although the flexural-tensile strength increased to 121%, the fracture energy and tensile strength gradually decreased, both remaining above 68% of those of unaged mixtures after the third regeneration. The material showed favorable water stability; specifically, its residual stability and freeze-thaw splitting strength ratio finally stabilized at over 90% and maintained this level. In terms of dynamic viscoelasticity, although three aging-regeneration cycles altered the viscoelastic balance of the mixture, the dynamic response characteristics similar to those of new mixtures were not eliminated. According to the grey correlation analysis of the performance of asphalt mixtures and asphalt, penetration, softening point, rotational viscosity, visco-toughness, and toughness, relatively high grey correlation degrees with the asphalt mixtures were shown. The established GA-BP neural network can effectively build a robust model for predicting the road properties of asphalt mixtures subjected to multiple aging-regeneration cycles, with small relative errors. Our work provides a valuable reference for systematically studying the materials genome of asphalt and asphalt mixtures.