E-Locus - Institutional Repository of the University of Crete

Home Search

Results - Details

Search command : Author="Τρικαλίτης" And Author="Παντελής"

Current Record: 2 of 70

[Add to Basket]

Identifier

000463155

Title

From potential energy surface to gas adsorption via deep learning

Alternative Title

Από τη δυναμική ενεργειακή επιφάνεια στην προσρόφηση αερίων μέσω βαθιάς μάθησης

Author

Σαρικάς, Αντώνιος Π.

Thesis advisor

Φρουδάκης, Γεώργιος

Reviewer

Τρικαλίτης, Παντελής
Νεοχωρίτης, Κωνσταντίνος

Abstract

Μetal-organic frameworks, or in short MOFs, thanks to their ultra high poros- ity and surface area, are deemed as prominent candidates for applications involving gas adsorption. However, their intrinsic combinatorial nature translates to a practically infinite material space, rendering the identification of novel materials with traditional methods cumbersome. Over the last years, machine learning approaches based on pre- dictive models have been developed, allowing researchers to rapidly screen large databases of MOFs. The quality of these models is highly dependent on the mathematical representation of a material, thus necessitating the use of informative inputs. In this thesis, we propose a generalized framework for pre- dicting gas adsorption properties, using as one and only input the potential energy surface. We treat the latter as a ₃D energy image and then pass it through a ₃D convolutional neural network, known for its ability to process image-like data. The proposed pipeline is applied in MOFs for predicting CO2 uptake. The resulting model outperforms both in terms of accuracy and data efficiency a conventional one built upon textual properties. Additionally, we demonstrate the transferability of the approach to other host-guest systems, by examining CH₄ uptake in covalent organic frameworks. The perfor- mance and generality of the proposed approach along with the fast input calculation thanks to paral- lelization, renders it suitable for large scale screening. Finally, discussion for improving and extending the suggested scheme is provided.

Language

English

Subject

Machine learning

Porous materials

Μηχανική μάθηση

Πορώδη υλικά

Issue date

2024-03-21

Collection

School/Department--School of Sciences and Engineering--Department of Chemistry--Post-graduate theses

Type of Work--Post-graduate theses