Segregation results in worse mechanical and durability performance of concrete. Therefore, an accurate modelling of segregation is required for reliable mesoscale modelling. In this context, this paper presents a method to develop the geometric mesoscale modelling of concrete taking into account segregation. In this method, coarse aggregate particles are generated as ellipsoids of random geometry and are randomly placed within concrete. The specimen is stratified in different numbers of layers ( \(NoL\) ) that represent different segregation conditions. The geometric models of four concrete mix designs are generated and the relevant key parameters, e.g. aggregate distribution are analysed and validated. In addition, the segregation level of generated models is classified based on the volumetric index and the correspondence between segregation level and the value of \(NoL\) is presented. The results show that 1) The generated geometric models with \(NoL\) ranging from 1 to 10 align well with real segregation in concrete and can accurately represent the majority of segregation cases resulting from various experimental factors; 2) The \(NoL\) reflects the segregation level: more layers indicate a more heterogeneous mesostructure; 3) The larger aggregates of the generated models tend to settle at the bottom, while smaller aggregates rise toward the top, leading to uneven vertical distribution, especially as the \(NoL\) increases; 4) the mix 1 with low aggregate content can represent the construction material with large segregation, while the mix 2 is inverse.
