Integrative CFD and AI/ML-based modeling for enhanced alkaline water electrolysis cell performance for hydrogen production

Abstract

A comprehensive model based on CFD modelling as well as AI/ML based modelling for an alkaline water electrolysis (AWE) cell is presented. A single cell 2D multiphase CFD model is solved in COMSOL Multiphysics 6.1® and is successfully validated with the experimental results for different operating conditions. The CFD model accurately computes the concentration and flow profiles of produced oxygen and hydrogen gases, the movement of the bubbles, and turbulence within the cell as well as the impact of current density, electrolyte flow rate and electrode-diaphragm distance. Further, integrating the CFD model with a neural network model enhances its potential for better cell design and performance. Multiple inputs and single output (MISO) artificial neural network (ANN) models are developed to predict the performance of the AWE cell. The ANN models are trained using the Levenberg-Marquardt algorithm, which operates as a feed-forward back-propagation network. The trained ANN models accurately predicts the complex relationships between input parameters (temperature, initial current density, and electrolyte weight concentration) and output parameters (actual current density and cell voltage) with an R2 value of 0.999 for both the outputs. This integrative CFD and ANN approach provides a comprehensive understanding of the AWE cell's behavior, further optimizing its design for efficient hydrogen production, offering a robust process that minimizes both computational resources and time as well as contributes for the scale up of the process.