Alkaline water electrolysers are the primary technology for industrial-scale production of low-carbon hydrogen. Large current densities lead to increased hydrogen production but decreased efficiency due to ohmic, activation, and concentration losses. The two-phase flow plays a significant role, for example, through bubble coverage of the electrode and reduced electrolyte conductivity. At low current densities, product crossover through the diaphragm reduces gas purity, causing production shutdowns to avoid explosive gas mixtures. Understanding the behaviour of the bubbly flow is crucial in improving the design and operation of electrolysers. Direct numerical simulations (DNS) can simulate many fundamental processes in electrolysis, such as bubble growth, detachment, and coalescence. However, their high computational cost prevents the study of larger-scale configurations, leaving gaps in our understanding of industrial electrolysers. Approximations used to simplify the problem introduce significant sources of error and uncertainty, leading existing computational approaches in the literature to employ a range of assumptions of often unknown validity.
This thesis introduces a multi-physics model developed with the OpenFOAM libraries, linking multiphase flow, turbulence, population balance modelling, heat transfer and electrochemistry. We validate our computational approach with experimental measurements and examine common assumptions to define a robust numerical configuration. We leverage OpenFOAM's open-source nature to integrate our solver with machine learning and statistical analysis libraries. We first focus on a coarse industrial-scale configuration, conducting a global sensitivity analysis to assess the impact of uncertain parameters, followed by Bayesian optimization to adjust unknown parameters. We then explore physics-informed neural networks to accelerate Poisson equation solvers, applying this to our electrolyser model and the pressure-velocity coupling in incompressible flow solvers. To address uncertainty in two-phase turbulence uncovered in the sensitivity analysis, we perform large-eddy simulations, bridging the gap with the RANS simulations typically used in industry and providing new insights into the flow dynamics within electrolysers.