(87l) Multi-Physics Simulations and Bayesian Optimisation of Flow in Alkaline Water Electrolyzers

Hydrogen will play a critical role in the energy transition, with the 2021 UK hydrogen strategy aiming to replace 20% of the UK’s natural gas usage with low carbon hydrogen by 2030 [1]. With the potential to replace fossil fuels and decarbonise sectors such as energy, transport, and heating, reducing the cost of green hydrogen is key to achieving net zero by 2050. Alkaline water electrolysers produce hydrogen and oxygen through electrolysis of water and are commonly employed for industrial-scale production of low carbon hydrogen. Operating at high current densities leads to enhanced hydrogen production but reduced cell efficiency, partly due to ohmic losses from the gas-liquid flow. Bubble coverage of the electrode surface and reduced effective electrolyte conductivity from increased void fraction are major contributors to these losses.

To gain further insights into the flow in electrochemical cells and optimise the design and operation of AWEs, we perform three-dimensional transient multi-physics simulations with the OpenFOAM libraries. We simulate the bubbly flow with a multi-fluid Eulerian model and consider bubble coalescence and polydispersity through population balance modelling (PBM). The electrical current distribution in the electrolyte governs the gas production rate at each electrode through Faraday’s law. We solve the electrical potential and employ an iterative method to impose boundary conditions that satisfy the non-linear relationship between current density and Butler-Volmer kinetics at the electrodes [2]. We model heat transfer arising from Joule heating in the electrolyte and activation overpotentials at the electrodes. We compare our results with experimental work, and study free convection [3] and forced convection configurations with diaphragms [4] over a range of current densities and operating conditions. We examine the assumptions commonly employed in multiphase modelling of AWEs, including constant bubble diameter, no coalescence, isothermal conditions and uniform current density distribution, and analyse the contributions to flow and cell performance predictions from flow, electrochemistry, and heat transfer models.

Simulations results suggest a high sensitivity to various modelling parameters, particularly within PBM and momentum-coupling terms. Several of these parameters, including bubble size, distribution and turbulent dispersion coefficients are obtained through experimental methods. However, significant uncertainty is present in some of these measurements, with large variations in published bubble diameters, and unknown coalescence rates and turbulent dispersion coefficients for bubbles in electrolyte solutions, potentially resulting in significant errors in simulation results and inaccurate predictions. To avoid large scale parametric studies which are computationally expensive for 3D high resolution simulations, we employ a Bayesian optimisation and surrogate modelling approach, allowing us to search the parameter space and optimise our model in a significantly lower number of iterations and at lower computational costs. Our OpenFOAM solver is coupled to the scikit-optimise library and runs autonomously to determine optimal input parameters by minimising an objective function based on error between predicted and experimental gas distributions.