Abstract
Compartmental models are widely used to reduce the computational burden of CFD-based reactor simulations, but their predictive capability often deteriorates during scale-up because the hydrodynamic history of the system is not explicitly retained in the reduced-order representation. In this work, we extend the CompArt framework, an artificial intelligence-powered compartmental modeling strategy, to the smart scale-up of multiphase reactors. The methodology combines CFD simulations, unsupervised learning, residence-time-based clustering, and compartment activation time estimation to construct reduced-order models that preserve both the spatial organization and the temporal ordering imposed by the underlying flow field. The framework is demonstrated on the precipitation of Mg(OH)2, used here as a representative case study, by considering two geometrically similar stirred tank reactors with working volumes of 785 mL and 10 m3. In contrast to conventional compartmental modeling strategies, the present approach does not rely exclusively on increasing the number of compartments to improve accuracy. Instead, it embeds the hydrodynamic history of the system through the age field for clustering and through tracer-based activation times for dynamic inter-compartment communication. This allows the model to reconstruct the transient development of the reference CFD-PBM solution in a physically meaningful manner. The results show that increasing the number of compartments mainly refines the reduced-order description, whereas the key ingredient for predictive accuracy is the preservation of the hydrodynamic activation sequence. Direct comparison with the CFD-PBM fields demonstrates that CompArt reproduces the evolution of the zeroth-order moment, m0, with high fidelity at both laboratory and industrial scale, while standard compartment models without activation produce overly diffusive and physically inconsistent dynamics. These findings demonstrate that CompArt provides a general and transferable framework for the scale-up of convectively dominated multiphase crystallization systems and, more broadly, for the simplification of complex CFD-PBM processes.