Hydrogen Generation – Optimal Locations using AI/ML

October 11th, 2023

R&D Focus Areas:
Natural hydrogen

Lead Organisation:

Edith Cowan University (ECU)


Start date:
February 2023

Completion date:
June 2025

Key contacts:
Marina Pervukhina – marina.pervukhina@csiro.au


Project total cost:

Project summary description:
Hydrogen is expected to play a major role in global decarbonisation, including a large part of heavy industry and transport system that together are responsible for 52% of carbon dioxide emissions. According to International Energy Agency estimates, hydrogen production may reach 400 million tonnes by 2050. This may require alternative clean hydrogen options other than green hydrogen (that uses renewable energy to split water).

One such alternative might be underground hydrogen generation. Developing underground hydrogen production faces a number of challenges that require fundamental research. These challenges can be grouped into following categories:

  • Potential host rock description, quantification and mapping,
  • Understanding of optimal generation and production conditions and specifically the rock physics of coupled processes in porous media, and
  • Field scale optimisation of hydrogen generation and production to minimise the ecological impacts and maximise production rate.

This project will concentrate to the two first challenges, namely, potential host rock description, quantification and mapping and specifically the rock physics characteristics required for successful development of underground hydrogen generation.

To this end, the project proposes to use the Australian National cuttings collection to search for Hydrogen Generation Optimal Locations (HGOLs). As geological lithotypes are critical for screening for HGOLs and subsurface energy storage, a rapid spectroscopic logging and imaging scanner (HyLogger) will be used to digitise rock chips/cuttings, the only continuous record of lithotypes extracted from the subsurface.

Recently developed AI/ML and computer vision algorithms, including but not limited to Mask Region-based Convolutional Neural Networks (2015), Vision Transformers (2017), and ConvNeXt models (2020), will be trained, validated, and tested for automatically extracting lithological information from the digitised database. The project will concentrate but not limit itself to the datasets from the onshore wells in Australia, including those where geogenic hydrogen has been detected.

Related publications and key links:

  1. Wang, H. and Pervukhina, M. and Shulakova, V. and Beloborodov, R. and Kempton, R. and Piane, C. Delle and Clennell, M.B. and Nanjo, T. and Miyoshi, K., Ai/Ml Approach to Lithology Quantification from Rock Chips Analysis, European Association of Geoscientists & Engineers, 2022, 1, 1-5, https://doi.org/10.3997/2214-4609.202210316,
  2. Takashi Nanjo, Akira Ebitani, Kazuaki Ishikawa, Yusaku Konishi, Keisuke Miyoshi, Valeriya Shulakova, Roman Beloborodov, Richard Kempton, Claudio Delle Piane, Michael Benedict Clennell, Arun Sagotra, Marina Pervukhina, Yuta Mizutani, Takuya Harada, Automatic Lithology Classification of Cuttings with Deep Learning, https://www.earthdoc.org/content/papers/10.3997/2214-4609.202210316 , ISSN 2214-4609.

Higher degree studies supported:
One PhD student at ECU is supported by this project.


October 2023