A cellular automaton (CA)-finite element (FE) model was implemented for multi-scale modelling of micro-segregation, mesoscopic grain structure and macroscopic segregation during direct chill (DC) casting of industrial billets or slabs. The macroscopic transport of mass, momentum, energy and solutes is solved on an FE grid, while the mesoscopic grain structure governed by nucleation, growth kinetics and grain evolution was calculated on a CA grid. The solidification path was determined using a modified micro-segregation model for multi-component aluminium alloys. An Euler representation was used for pre-processing and post-processing, and a Lagrangian representation was used for expanding the calculation domain and for resolving the CAFE model. By simulating a DC casting experiment of the 2024 aluminium alloy, a typical grain structure was reproduced, and the composition map showed a reasonable deviation. This model was applied to an industrial-scale DC cast slab of an Al-3.5Cu-1.5Mg (wt. %) alloy, and three simulations with different nucleation undercoolings were performed for a grain-unrefined slab, a grain-refined slab and an equilibrium solidified slab, respectively. The slabs tended to solidify at equilibrium with the decreasing nucleation undercooling. The earlier release of latent heat yielded a smaller liquid undercooling region ahead of the solidification front, and a finer grain structure. A typical grain structure with coarse equiaxial grains at the centre and fine columnar grains near the bottom surface as well as sidewall was observed for the grain-unrefined slab. By contrast, the grain structure of the grain-refined slab was fully equiaxial. Furthermore, the grain structure, temperature field, melt flow and macro-segregation were quantitatively investigated.