Compartment Models (CMs) provide a simplified yet effective approach for
modeling the hydrodynamics of chemical reactors, offering a computationally
efficient alternative to full-scale Computational Fluid Dynamics (CFD) simula-
tions. However, defining an appropriate compartmentalization that accurately
captures the underlying flow physics remains a key challenge. In this study, we
propose a physics-inspired optimization framework for defining CM compart-
ments in a batch reactor using a multi-objective genetic algorithm (NSGA-II).
The reactor is divided into compartments based on flow and turbulence char-
acteristics. Each compartment is designed to capture specific physical effects
influencing the crystallization process. For instance, zones with high turbulence
account for enhanced agglomeration. The optimization process aims to min-
imize the variance of both velocity magnitude (U) and turbulence dissipation
rate (ε) within the compartments, ensuring a physically meaningful partitioning
of the system.
Our methodology integrates CFD data from OpenFOAM with a machine-
learning-assisted optimization strategy. By extracting velocity and turbulent
energy dissipation fields, we define reactor regions using geometric and flow-
based criteria. A population-based evolutionary algorithm iteratively adjusts
compartment boundaries, optimizing key parameters such as impeller region
height, inner radii, and compartment limits. The Pareto front solutions reveal
trade-offs between velocity and dissipation uniformity, guiding the selection of
an optimal CM configuration.
The results highlight the ability of the multi-objective optimization to iden-
tify regions of high turbulence correctly. The proposed framework enhances
CM reliability and applicability for process optimization, reactor design, and
scale-up in chemical engineering.
