ABSTRACT
The numerical simulation of concrete materials has gained widespread recognition, with mesoscopic modeling based on real materials playing a pivotal role. Traditional methods, generate aggregate sequences and arrange them incrementally, necessitating overlap detection during the process. This approach is computationally intensive and time-consuming, particularly for high aggregate volume fractions. In this study, a diffusion generative model is proposed to learn and reconstruct the distribution characteristics of 2D mesoscopic concrete. By leveraging the inherent randomness shared between actual aggregate distributions and the generative model, the proposed method achieves significantly improved generation efficiency compared to traditional techniques, offering a promising tool for efficient concrete modeling.
