Pope, Tom;
(2024)
From Pixels to Porosity: Computer Vision and Geometric Deep Learning for MOF-Based Separations.
Doctoral thesis (Ph.D), UCL (University College London).
Preview |
Text
TOM_POPE_THESIS_FINAL.pdf - Accepted Version Download (24MB) | Preview |
Abstract
This thesis explores the application of advanced deep learning architectures to predict and characterise the performance of Metal-Organic Frameworks (MOFs) in various gas separation processes. Leveraging and parameterising an extensive dataset, this research seeks to identify and optimise frameworks for the efficient separation of critical gases, including methane (CH4), carbon dioxide (CO2), sulphur dioxide (SO2), and various mercaptans. The research begins by utilising high-accuracy vision transformer networks trained on a comprehensive MOF database to predict methyl mercaptan (MeSH) adsorption properties within binary gas mixtures. These models facilitate a transfer learning approach, extending their predictive capabilities to complex biogas mixtures. Attention mechanisms within these networks are transformed and visualised to establish a novel linkage between adsorption mechanisms and predictive insights derived from machine learning. Moreover, this thesis investigates the use of various graph neural networks (GNNs) for semi-supervised prediction of SO2 adsorption in MOFs. This section underscores the potential of utilising limited labelled data for materials prediction, introducing advanced graph construction techniques to enhance both the predictive performance and interpretability of GNNs. The final chapter focuses on methane capture from dilute sources such as coal mine ventilation air, evaluating MOFs for their efficacy in low-concentration CH4 and CO2 environments. Utilising multitask convolutional neural network architectures alongside chemically intuitive data augmentation strategies, the study enhances predictive accuracy and explores innovative methods for dataset expansion. Overall, this work not only advances the field of deep learning-driven materials discovery but provides a robust, scalable framework for future studies aimed at identifying and evaluating novel materials for environmental protection and sustainability.
| Type: | Thesis (Doctoral) |
|---|---|
| Qualification: | Ph.D |
| Title: | From Pixels to Porosity: Computer Vision and Geometric Deep Learning for MOF-Based Separations |
| Open access status: | An open access version is available from UCL Discovery |
| Language: | English |
| Additional information: | Copyright © The Author 2024. Original content in this thesis is licensed under the terms of the Creative Commons Attribution-NonCommercial 4.0 International (CC BY-NC 4.0) Licence (https://creativecommons.org/licenses/by-nc/4.0/). Any third-party copyright material present remains the property of its respective owner(s) and is licensed under its existing terms. Access may initially be restricted at the author’s request. |
| UCL classification: | UCL UCL > Provost and Vice Provost Offices > UCL BEAMS UCL > Provost and Vice Provost Offices > UCL BEAMS > Faculty of Maths and Physical Sciences UCL > Provost and Vice Provost Offices > UCL BEAMS > Faculty of Maths and Physical Sciences > Dept of Chemistry |
| URI: | https://discovery-pp.ucl.ac.uk/id/eprint/10202532 |
Archive Staff Only
 |
View Item |
