Comminution processes are energy costly, representing 1% of the electrical energy consumption worldwide and have an overall efficiency rounding 1-3%. With a high potential for optimisation, multi-scale numerical modelling becomes critical. In the process of reducing material to a given size fraction, fragmentation details become important. High-resolved finite-element-based modelling has the advantage of giving accurate prediction of fracture and fragmentation using physical and measurable material properties. However, these models are computationally heavy, making it possible to study only a few particles and not complete industrial-scale systems. This study presents a computational framework for simulating the fracture of brittle materials on different scales aiming to enable AI-based optimisation of comminution processes. Industrial scale simulation of particle fracture is approached with dilated polyhedral discrete element models utilising graphical processing units (GPUs). This approach was implemented in the commercial software Demify® [1]. In the mesoscale, the KST-DFH [2] material model accounting for shear and tensile damage was used for detailed FE simulations of single particle fragmentation. Calibration and validation schemes were developed based on the experimental background of Brazilian disc experiments and single particle breakage of 3D scanned mineral material. A good correlation of force prediction and fracture behaviour was obtained. Finally, cone crusher simulations of granite accounting for high flows of material are presented, and the fragmentation of individual particles is studied in detail. In general, the framework presented here unlocks the possibility to accurately predict fragmentation processes of critical materials and optimize the overall system through AI algorithms within comminution processes.