Work

Optimizing Stone Mastic Asphalt mix design with TGAN-enhanced surrogate models

Year: 2024

Type: article

Abstract: The design procedure for asphalt mixes is still largely empirical, often requiring lengthy and resource-intensive trials in the laboratory. To streamline and accelerate the mix design process, this study investigates the use of a surrogate-assisted model to predict the volumetric characteristics of asphalt mixtures and automate the design process. Optimizing mixture design is essential for ensuring that the asphalt mixtures meet the required volumetric properties, involving complex decisions on the proportions of aggregates and binder content to satisfy multiple objectives. The mix design of Stone Mastic Asphalt (SMA) is often challenged by the intricate balance required to satisfy these multiple objectives simultaneously. Traditional optimization techniques struggle to cope with these challenges, especially under conditions of incomplete or scarce data, which is common in practice. In this paper, tabular generative adversarial networks (TGAN) have been specifically developed to address such problems. The paper further demonstrates the application of the developed framework in optimizing SMA mixture design, which proves to be a challenging problem in the realm of pavement engineering. Models trained on aggregated data – including those augmented with TGAN-generated samples – showed approximately a 50% improvement in R 2 values for maximum density and a 30%–40% improvement for bulk density at 120 and 350 cycles, respectively, compared to the non-TGAN models. Moreover, the results of experimental volumetric tests confirm that the TGAN-based models perform superior to the non-TGAN models. As such, the absolute error percentages of the candidate solutions obtained from the model trained on the TGAN-generated data were consistently below 10%, whereas the non-TGAN models exhibited higher variations, with errors exceeding 15%. The results of this study were further verified through experimental validation, as well as by analyzing parameters extracted from the Rosin-Rammler distribution function and the Bailey method. The qualitative and quantitative results demonstrate the effectiveness of the proposed framework in solving the optimization problem for Stone Mastic Asphalt mixture design.

Source: Construction and Building Materials

DOI: 10.1016/j.conbuildmat.2024.138863

Authors: Mahdi Zakerzadeh, Mohsen Mousavi, Babak Shahbodagh, James Ng, Nasser Khalili

Institution: UNSW Sydney

Language: English

Topic: Asphalt Pavement Performance Evaluation

Subfield: Civil and Structural Engineering

Field: Engineering

Domain: Physical Sciences

SDGs: Industry, innovation and infrastructure, Sustainable cities and communities

Open Access status: hybrid

Estimated APC paid: $4,080