In the present study, we investigate an operating strategy for membrane-based direct air capture (m-DAC) processes that improves energy efficiency by partitioning the overall pressure ratio between feed-side compression and permeate-side vacuum. To th...
In the present study, we investigate an operating strategy for membrane-based direct air capture (m-DAC) processes that improves energy efficiency by partitioning the overall pressure ratio between feed-side compression and permeate-side vacuum. To this end, a steady-state membrane model was integrated with an energy accounting framework that captures the efficiency characteristics of compressors and vacuum pumps. However, due to the extremely low carbon dioxide (CO2) concentrations and high nonlinearity of the model, the physics-based model exhibited numerical convergence failures (BVP instability) in approximately 30% of the operating domain, acting as a constraint for stable optimization analysis. To overcome these numerical limitations and perform analysis across the entire operating domain, a machine learning based surrogate model was developed. In particular, predictive precision was maximized by applying log transformations to flow rate data, which exhibit wide magnitude variations. A Random Forest model, achieving a coefficient of determination (𝑅2) of over 0.99, served as the foundation for the optimization analysis. The developed surrogate model predicts permeate and retentate flow rates under constraints of initial feed pressure, flow rate, and membrane design. Based on these predictions, key performance indicators including CO2 purity, recovery, and specific energy consumption (SEC) were calculated to systematically analyze the results. Furthermore, this study proposes a multi-stage membrane process for high-purity CO2 capture and derives feasible operating regions and quantitative pressure setting guidelines that maximize energy efficiency. This research is expected to serve as a core tool for the industrial application of m-DAC-based CO2 capture process design in the future.