Back to search

University of Crete (UOC)

Χρήση και ανάπτυξη μεθόδων μηχανικής μάθησης για μελέτη αποθήκευσης μορίων σε νανο-υλικά

Abstract

dc:description

The transition to clean energy and the response to global water scarcity are among the most pressing challenges of our time. Hydrogen, as a clean energy carrier, and Atmospheric Water Harvesting (AWH), as a decentralized solution to water scarcity, both rely on porous materials with optimized sorption performance. Metal-Organic Frameworks (MOFs), owing to their near-unlimited structural and chemical diversity, are prime candidates for both applications, yet their vast combinatorial space makes purely experimental exploration intractable. At the same time, computational screening studies typically focus exclusively on predicting functional properties, without assessing whether the proposed materials are experimentally synthesizable, a gap that this thesis places at the center of its investigation. To address these challenges, an integrated computational framework of four complementary tools was developed, each covering a successive stage of computationally guided MOF discovery. MOF360*automates the construction, geometric optimization, and structural/chemical descriptor calculation of hypothetical MOFs, generating homogeneous databases suitable for training Machine Learning (ML) models; it was applied to predict the working hydrogen storage capacity, yielding concrete design rules (framework density 0.2–0.4 g·cm⁻³, PLD 10–13 Å, LCD 13–20 Å) and identifying the structure edq_N19_N1000_E14_E14 as the top candidate, with a predicted gravimetric capacity of 13.11 wt% and volumetric capacity of 44.12 g·L⁻¹, both exceeding the 10 wt% technological target. MOFSynth, and its advanced version MOFSynth-ADV, introduce synthesizability assessment as an integral screening stage, quantifying the energetic strain and geometric deformation of the organic linker upon incorporation into the framework relative to its free, optimized conformation. Applied to more than 40,000 structures, the tool discriminates experimental from hypothetical databases while the MOFSynth-ADV version, based on semi-empirical quantum-mechanical methods (GFN2-xTB) and machine-learned interatomic potentials, proved both faster (−13% runtime) and substantially more numerically stable (−73% non-converging cases) than the classical version, while being fully open source (ASE-based architecture). Finally, AquaMOF implements a multi-target ML framework for the simultaneous prediction of water uptake, selectivity, and water stability of MOFs, using gradient-boosted tree algorithms and explainable AI (SHAP) analysis that surfaces design rules (e.g., void fraction > 0.1, KHW > 0.016 mol·kg⁻¹·Pa⁻¹); the sequential multi-criteria screening identified 3,733 MOFs with a predicted water-stability probability above 90% and ranked the top 100 candidates for AWH applications, with the model successfully validated against independent, experimentally documented structures (MOF-303, CaSyr-2, Cr-mos-MOF-1, UoC-13(1F)).All tools are released as open-source software, via GitHub repositories and web applications, making the developed methodologies directly accessible to the wider research community. This thesis thus contributes a comprehensive, extensible, and experimentally validated methodology for computationally guided MOF discovery, bridging the gap between theoretical prediction and experimental reality, with direct application to hydrogen storage and atmospheric water harvesting.

Degree

thesis:*
Grantor dc:publisher
University of Crete (UOC)
Year dc:date
2026

Author and committee

dc:creator, dc:contributor.*
Authors dc:creator
  • Livas, Charalampos
  • Λίβας, Χαράλαμπος

Subjects

dc:subject × 16

Rights

Language dc:language
gre

Identifiers

dc:identifier.*
Identifier
10.12681/eadd/62865

Chain of custody

source
Harvested from
Greek National Archive of PhD Theses
Base URL
phdtheses.ekt.gr/eadd_oai/request
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
citation

Livas, Charalampos; Λίβας, Χαράλαμπος. Χρήση και ανάπτυξη μεθόδων μηχανικής μάθησης για μελέτη αποθήκευσης μορίων σε νανο-υλικά. University of Crete (UOC), 2026. http://hdl.handle.net/10442/hedi/62865