A machine learning-assisted multi-criteria decision making framework for chemical reactor channel geometry selection

Additive manufacturing enables the realization of diverse and complex reactor channel geometries, substantially increasing the number of feasible design variants. While this design freedom offers opportunities for improved performance, it also complicates the systematic selection of geometries suitable for a given application. This paper presents a data-driven framework to support this selection process in the early reactor concept design phase. The framework employs surrogate models to predict key physical characteristics such as heat transfer, radial temperature difference, pressure drop, and flow profile, and integrates these into a multi-criteria decision making method to guide process engineers in selecting appropriate channel geometries. By focusing on fundamental physical characteristics rather than reaction-specific results, the framework ensures that the surrogate models remain transferable across different reaction systems. Incorporating probabilistic dimensioning further narrows the solution space and reduces the need for iterative experimental campaigns or computationally demanding simulations in reactor design. The applicability of the proposed framework is demonstrated through two case studies in methanol synthesis, highlighting its potential to streamline early-stage reactor development.