(510g) Accelerating Direct Air Carbon Capture Design Optimization with Fourier Neural Operators

The urgent need to scale direct air carbon capture (DACC) systems demands rapid design optimization to maximize CO₂ capture while minimizing energy expenditure and operational costs. Simulation time remains a critical bottleneck in the process, slowing the design and deployment of new DACC systems. In this work, we present a deep-learning approach for modeling the parametric PDEs governing the behavior of DACC devices. We demonstrate that Fourier neural operator (FNO) methods can reduce the computational overhead required to simulate DACC system dynamics by orders of magnitude whilst maintaining high fidelity. We model the mass transfer between gas and liquid phases in a wetted wall column (WWC) configuration, a canonical setting for absorption-based DACC research. We validate the accuracy of the workflow against both high resolution computational fluid dynamics (CFD) simulations taken from the existing literature and against experimental data from a lab-scale prototype. We systematically explore the impact of key operational parameters, such as inlet gas velocity, oscillating fluid injection and film thickness for a range of CO2 concentrations. We demonstrate that a neural operator framework can serve as a surrogate model capable of accurately resolving changes in mass-transfer dynamics when design parameters are modified, accelerating design optimization for a WWC DACC device while achieving close agreement with both traditional CFD models and experimental benchmarks.