UTILE-Oxy - Deep Learning to Automate Video Analysis of Bubble Dynamics in Proton Exchange Membrane Electrolyzers
Creator: Andre Colliard
Hydrogen production through polymer electrolyte membrane water electrolyzers (PEMWEs) is crucial for efficient energy systems. However, analyzing the dynamic behavior of oxygen bubbles in these systems has been challenging. A recent study, "Deep Learning-Enhanced Characterization of Bubble Dynamics in Proton Exchange Membrane Water Electrolyzers," published in PCCP, introduces a groundbreaking deep learning tool to tackle this challenge.
The study leverages a deep learning framework to automate the analysis of video recordings of oxygen bubbles in PEM electrolyzers. Using just 35 annotated images, various U-Net models were trained to achieve high precision (up to 95%) and F1-scores (86%). This method enables rapid extraction of key parameters like bubble area, size distribution, and shape, providing comprehensive insights into bubble dynamics.
The tool benchmarks three U-Net models, with the U-Net model featuring the ResNeXt101 backbone slightly outperforming the others. This model accurately segments and analyzes bubbles, offering detailed visualizations of bubble behavior over time. Such insights are crucial for optimizing PEMWE performance and enhancing diagnostic techniques.
Why experimentalists are interested in this tool?
This automated framework can process thousands of frames in minutes, enabling extensive experimental analysis previously unfeasible due to time constraints. It opens new avenues for research and development in electrochemical energy systems, aiding in the optimization of cell designs and material performance.
Are you interested?
This deep learning tool marks a significant advancement in analyzing bubble dynamics in PEM water electrolyzers. By providing detailed and rapid analysis, it paves the way for future innovations in hydrogen production technologies. The full study is available in PCCP as open access, and the UTILE-Oxy repository on GitHub offers access to the implementation and datasets.
Publications
Deep learning-enhanced characterization of bubble dynamics in proton exchange membrane water electrolyzers
Colliard-Granero A, Gompou K, Rodenbücher C, Malek K, Eikerling M, Eslamibidgoli M - Physical Chemistry Chemical Physics - 2024
![HI logo](/staticfiles/img/connect/HI_logo_only_Black.png)