Our century is defined by a race against time to limit global warming to 1.5 °C above pre-industrial levels, as agreed in the Paris Agreement by 192 Parties in December 2015. In this context, the massive deployment of intermittent renewable energies (RE) is a national and European priority, particularly in non-interconnected island territories such as La Réunion, where grid stability relies on energy storage and decarbonized hydrogen production.
Proton Exchange Membrane Water Electrolysis (PEMWE) is a key technology for converting surplus renewable electricity into hydrogen. However, the accumulation of oxygen bubbles at the anode significantly reduces system performance and durability. These complex two-phase phenomena (bubbly, slug, and stagnated regimes) lead to partial coverage of the active surface, increased overpotentials, and accelerated component degradation.
To overcome these bottlenecks, artificial intelligence represents a breakthrough innovation. Pioneering work carried out by Idriss Sinapan led to the development of deep learning-based bubble detection and recognition tools (YOLO), initially in single-class and then in multi-class configurations, combined with a transparent PEMWE cell and a high-resolution video acquisition system on the SysPacRevers test bench. These approaches have already enabled precise quantification of coverage rates, bubble counts, and flow dynamics, as well as the identification of counter-intuitive phenomena related to water flow rate and current density.
The H2-DurabilitAI project – Improving H2 system durability through AI, led by Idriss Sinapan as principal investigator, builds on this body of work. It aims to establish a comprehensive experimental database, develop an advanced AI pipeline (multi-class detection/segmentation, non-uniformity heat maps, bubble residence time estimation), and propose new component topologies and optimizations, in partnership with Fraunhofer ISE for experimental validation.
The H2-DurabilitAI project is funded by the European Union in the amount of €167,924.67 under the ERDF-ESF+ Réunion 2021–2027 programme, for which the Région Réunion serves as the Managing Authority. Europe is committed to La Réunion through ERDF funding. The Région Réunion supplements this funding with a national counterpart contribution.
This project strengthens local expertise in artificial intelligence applied to hydrogen and contributes to the decarbonization and energy resilience of the Réunion territory, while promoting open innovation through the public release of the database and the AI model on GitHub.
This project builds on the pioneering work of Idriss Sinapan, which led to the development of deep learning-based oxygen bubble detection and recognition tools in PEMWE electrolyzers. This work enabled, for the first time, a multi-class analysis (bubbly, slug, and stagnated) of bubble dynamics at the anode, using a transparent PEMWE cell coupled with a high-resolution video acquisition system on the SysPacRevers test bench.
The H2-DurabilitAI project continues and expands upon this work through four specific objectives:
1. Large-scale data acquisition and establishment of an experimental database
The objective is to generate a rich, multi-condition database on the transparent PEMWE. Several tasks are carried out:
Recommissioning and calibration of the SysPacRevers test bench;
High-resolution video acquisition synchronized with operating signals (current density, water flow rate, temperature, pressure, channel and porous medium topologies);
Structuring, organization, and open sharing of data via a NAS system.
2. Development of an artificial intelligence pipeline for bubble analysis
The objective is to create an advanced AI tool dedicated to the detailed analysis of two-phase phenomena. Several tasks are carried out:
Development of a multi-class bubble detection/segmentation model (bubbly, slug, stagnated);
Generation of coverage non-uniformity heat maps;
Implementation of bubble tracking to compute dynamic indicators such as residence time and mean evacuation time;
Post-processing and performance evaluation (mAP, IoU, etc.).
3. Analysis of results and proposal of new topologies and optimizations
The objective is to leverage AI-derived indicators to improve PEMWE performance and durability. Several tasks are carried out:
Comparative analysis of bubble regimes according to operating conditions and topologies;
Proposal of component modifications (channels, porous transport layer, MEA assembly);
Validation in partnership with Fraunhofer ISE.
4. Dissemination, outreach, and preparation of future projects
The objective is to ensure the dissemination of results and the strengthening of local expertise. Tasks include:
Publication of results in a Category A scientific journal;
Public release of the AI model and database on GitHub;
Preparation of analysis reports and recommendations;
Prefiguration of a Horizon Europe consortium.
The H2-DurabilitAI project aims to produce concrete and reusable results, both scientifically and technologically, with a strong territorial impact.
The main expected outcomes are:
A rich and open experimental database: Collection and structuring of a large volume of high-resolution videos synchronized with operating conditions (current density, flow rates, temperature, pressure, topologies). This database will constitute a valuable resource for the international scientific community working on two-phase phenomena in PEMWE electrolyzers.
An advanced artificial intelligence tool: Development of a high-performance AI pipeline for multi-class bubble detection (bubbly, slug, stagnated), generation of non-uniformity heat maps, and computation of dynamic indicators (residence time, coverage rate, etc.). The model and source code will be made available as open source on GitHub to promote reproducibility and collaborative innovation.
Scientific and technological advances: Publication of results in a Category A scientific journal, including a comparative analysis of bubble regimes and component optimization recommendations (flow channels, porous transport layer, MEA assembly). This work will significantly improve the performance and durability of PEM electrolyzers.
Strengthening of local expertise and international outreach: Development of computer vision expertise applied to hydrogen within ENERGY-Lab, strengthening of the partnership with Fraunhofer ISE, and preparation of a consortium for Horizon Europe projects.
Through these deliverables, the H2-DurabilitAI project will contribute to reducing green hydrogen production costs and to the energy resilience of island territories, concretely supporting the ecological transition of La Réunion.
Financial partners
The BECOME project is funded by the European Union under the ERDF-ESF+ Réunion programme, for which the Réunion Region is the Managing Authority. Europe is committed to Réunion through the ERDF.
Academic partners
Fraunhofer ISE (Fraunhofer Institute for Solar Energy Systems) – Germany
Discussions are underway with Fraunhofer ISE regarding a scientific partnership, particularly on the use of AI for the segmentation and analysis of porous media from 3D images acquired by laser microscopy. This collaboration would extend the scope of the H2-DurabilitAI project towards the detailed characterization of internal components of PEMWE electrolyzers.
