Machine Learning-Driven Multi-Objective Optimization of Microchannel Reactors for CO₂ Conversion

Abstract

Recently, the power-to-gas (PtG) concept, specifically thermocatalytic CO₂ conversion via the Sabatier process, emerges as a promising route for mitigating greenhouse gas emissions. The process transforms CO₂ and H₂ into methane and water under low-temperature methanation conditions. This study suggests a new way to improve the performance of a microchannel reactor by combining computational fluid dynamics (CFD), response surface methodology (RSM), machine learning (ML), and multi-objective optimization. Key design variables include inlet velocity, temperature, and channel length ratios. The RSM approach is for generating datasets for simulation; while, data augmentation assists ML model training. Six ML models—linear, ensemble, tree, Gaussian, support vector machine (SVM), and neural networks are evaluated for regression accuracy against RSM-based correlation. The Gaussian process model is found superior and integrated with a multi-objective optimization algorithm. A decision-making score (DMS) levels and normalizes performance indicators. It finds the best reactor designs with CO₂ conversion rates of ≈78.6% and CH₄ selectivity close to 99.9%. These results demonstrate an advanced approach for significantly reducing computational demand (24 h to 1.471 ms) against CFD simulations; while, maintaining accuracy, thereby enabling cost-effective, efficient solutions for reactor design optimization across various engineering applications in real-world PtG applications.