Leveraging Machine Learning and Data Mining for Enhanced PEM-Electrolysis Using Standardized Data Management

Abstract

Listen

Translate article

Green Hydrogen production by water electrolysis is paramount in the transition from fossil fuels towards a carbon neutral industry. Hydrogen is expected to function as energy carrier in mobility, energy storage medium in electricity grids and as regenerative fuel in industrial processes such as steel production. Proton exchange membrane water electrolysis (PEMWE) stands out as a promising technology for hydrogen production due to high current densities with high efficiencies and highly dynamic operation. While the number of publications on the production of PEM electrolysers grows steadily, a lack of standardized methodologies and experimental guidelines persists. Many studies focus on specific components and their manufacturing steps, hindering direct comparison and complicating data interpretation. Consequently, standardization of research data and specification of parameters is the objective of the current research landscape.In this work, we systematically analyzed literature and gathered meta data regarding experimental procedures for lab-scale production of PEM electrolysers. Thereby, special attention is on the identification of potential improvement in data management. The production of electrolysers can be divided into the various tasks as the fabrication of membrane electrode assemblies (MEAs), which includes catalyst ink production, coating and assembly, electrolyser design and characterization. All areas are in turn subdivided into metadata, for example the ionomer to catalyst ratio and catalyst content for the catalyst ink, the coating technology such as spray or slot coating for the coating. In total, over 60 different parameters were considered and a database with over 600 different samples from publications around the world was created. Incomplete datasets were determined by classification using various methods, such as k-nearest-neighbors.The extensive database was implemented in various machine learning algorithms with the goal of identifying parameters with a significant influence on the target variable based on the polarization curve. We then identify correlations between different parameters and formulate guidelines for individual fabrication steps and materials for electrolyzers. An optimization is then applied to determine the most favorable procedure for MEA fabrication and assembling electrolysers, incorporating all the parameters from the database.This work lays the foundation for future publications regarding the standardization of research data and shows how correlations between different parameters can be identified and optimized with the help of data management.