New generation and separation method for the computational modelling of the mesostructure of concrete

Abstract

The computational modelling of concrete at the mesoscale depends on a realistic representation of the mesostructure, including random aggregate placement and wall effects. However, the generation of the geometry of the mesoscale usually requires significant computational effort, may have insufficient packing capability, and may not detect the overlap between aggregate particles. This paper proposes a novel generation and separation method based on analytical techniques that uses: (1) linear transformations to randomly position the particle inside the concrete specimen; (2) an efficient analytical method based on eigenvalues to deal with all overlap checks; (3) a regeneration strategy to treat the cases of failure in aggregate placing (due to overlap) or extensive iterations needed. Demonstrative applications, made for cylindrical and cubic specimens and for both the actual grading curve of a mix design and for an idealized Fuller grading curve, show that geometric models developed using the proposed method are computationally efficient and representative of realistic mix designs. A benchmarking shows the efficiency of the method, and an analysis of the regeneration strategy concluded that the representativity of the geometric model is ensured as long as a new particle is generated after 213 iterations of placement.