Industrial carbon dioxide emissions are a major driver of climate change, and

structured packed columns are widely employed in post-combustion carbon cap

ture processes due to their established advantages of low pressure drop and high

mass transfer efficiency. Despite decades of use, the geometric optimisation of

corrugated-sheet structured packings to further reduce frictional pressure drop re

mains an open engineering problem, since the gas–gas interaction between flows in

adjacent triangular channels is not yet systematically addressed by existing designs.

The primary contribution of this thesis is a novel cosine-curved structured packing

in which the corrugated sheets are shaped by cosine curves rather than straight folds,

creating an additional inter-sheet gap that directly reduces gas–gas friction whilst

preserving the surface area and column volume of the corresponding conventional

packing. To provide a computationally efficient and physically accurate tool for

evaluating this design, a GPU-accelerated Multiple-Relaxation-Time Lattice Boltz

mann Method (MRT-LBM) coupled with the Wall-Adapting Local Eddy-viscosity

(WALE) subgrid-scale model is developed as the enabling numerical framework.

A comprehensive theoretical foundation is first established, covering the funda

mentals of the Lattice Boltzmann Method, collision models including BGK, MRT,

and TRT, and Large Eddy Simulation turbulence modelling. A systematic compar

ison of subgrid-scale models—including the standard Smagorinsky, Smagorinsky–

Van Driest, Vreman, WALE, and Sigma models—is presented. The WALE model

is selected for its superior near-wall behaviour, exhibiting the correct cubic asymp

totic scaling (νt ∝ y 3 ) without requiring explicit damping functions or wall distance

calculations, making it particularly suitable for the complex corrugated geometry of

structured packings.

The GPU-accelerated MRT-LBM framework is implemented and validated through

turbulent flow past a circular cylinder at Re = 3900. A comparative study between

the Smagorinsky and WALE models demonstrates that the WALE model achieves

superior predictive accuracy across all tested grid resolutions (ND = 16, 20, and 24),

capturing sharper shear layers, more organised vortex shedding, and better agree

ment with experimental PIV data. The WALE model exhibits reduced grid depen

dency, achieving reasonable accuracy at coarse resolutions where the Smagorinsky

2model requires significantly finer meshes. Computational efficiency is further op

timised by comparing strain rate tensor calculation methods, with Yu’s direct ap

proach achieving approximately 10× speedup over Chai’s method, and by identifying

the 4-4-4 GPU thread block configuration as optimal for overall throughput.

The validated solver is applied to systematically evaluate the novel cosine-curved

packing across 12 geometric configurations spanning three channel heights and four

opening angles. Simulations conducted within three-REU periodic domains demon

strate that the novel packings reduce dry pressure drop relative to conventional

packings by an amount that increases with opening angle, reaching reductions of

up to approximately 50% at α = 120◦ . Gap-isolation experiments confirm that the

inter-sheet gap is the primary mechanism responsible for this reduction. Empirical

correlations for dry pressure drop prediction are developed using a genetic algo

rithm. These results provide quantitative guidance for the geometric optimisation

of corrugated structured packings targeting reduced compression energy in carbon

capture applications.