Wakerley, James Kenneth Malcolm
(2023)
Material discovery and modelling for solid-state hydrogen storage and fuel cell applications.
PhD thesis, University of Nottingham.
Abstract
This thesis covers an attempt to construct a supervised machine learning model, for use in prediction of formation enthalpy values for novel metal hydride compositions. Further work, making use of static density functional theory calculations as well as ab initio and machine learning force field molecular dynamics simulations, to model oxygen transport in a La-Mg co-doped barium titanate system is also reported.
Utilising open-source, readily available repositories of previously calculated results, two gradient boosting regression models are developed; separately trained to qualitatively predict formation enthalpy data for metal hydrides, and for intermetallic alloys. Once developed, such predictions are compared to enthalpy values, calculated from first principles, for held-out samples from the original database, and known experimental values for select materials. A process is outlined for generating new ternary hydride compositions, previously unseen to the model, from which a select sample of promising predictions are subjected to crystal structure prediction processes. By introducing structural information, first principles calculations are used to determine formation enthalpies for comparison to predictions.
Intentionally trained using descriptors derived solely from chemical composition, without any dependence on crystal structure, the resultant model ultimately struggles to generalise prediction of formation enthalpies across the diverse geometry space of hydride materials. The decades-long quest for reliable crystal structure prediction simply from chemical composition proves to be a challenge for effective model validation by calculation, given the range of hydride classes. Oxygen transport through the prototypical perovskite system of barium titanate is studied to investigate the methodology of characterising oxide ion diffusion through the bulk of such a material. Inspired by unpublished experimental results, this system is then co-doped with lanthanum and magnesium, thus introducing titanium and oxygen vacancies, allowing for investigation of oxygen diffusion by means of dynamic simulation methods. This is performed using the relatively new method of on-the-fly machine learning force field molecular dynamics, an approach to the modelling of dynamical systems which, in theory, drastically reduces the time and computational cost of traditional methods based solely on ab initio molecular dynamics.
Approximations of low-energy transition paths for oxygen movement in the local vicinity of point defects suggest energetically favourable diffusion pathways introduced by lanthanum doping. Molecular dynamics simulation methods are used to construct oxygen self-diffusion trajectories, from which diffusion mechanics can be determined. These results suggest higher rates of diffusion events in magnesium-doped barium titanate than when co-doped with magnesium and lanthanum. Additionally, mobility in the co-doped system is shown to be influenced by geometry of lanthanum dopants.
| Item Type: |
Thesis (University of Nottingham only)
(PhD)
|
| Supervisors: |
Ling, Sanliang
Walker, Gavin S
Grant, David M
Li, Ming |
| Keywords: |
Hydrogen as fuel; Fuel cells; Machine learning; Heat of formation; Enthalpy; Hydrides; Titanates |
| Subjects: |
T Technology > TP Chemical technology |
| Faculties/Schools: |
UK Campuses > Faculty of Engineering |
| Item ID: |
73082 |
| Depositing User: |
Wakerley, James
|
| Date Deposited: |
21 Jul 2023 04:40 |
| Last Modified: |
01 Sep 2024 04:30 |
| URI: |
https://eprints.nottingham.ac.uk/id/eprint/73082 |
Actions (Archive Staff Only)
 |
Edit View |
Tools
Tools
