Back to search

University of Thessaly (UTH)

Τεχνητή νοημοσύνη και μοριακή δυναμική για τον υπολογισμό των ιδιοτήτων των ρευστών σε ατομική κλίμακα: μέθοδοι και εφαρμογές

Abstract

dc:description

Computational methods in fluid research have progressed significantly in recent years, driven by the integration of massive amounts of data, in either textual or graphical form, generated from multiscale simulations, laboratory experiments, and real-field data. Multiscale simulation approaches, fuelled by high-performance computing architectures, are now a reality, bridging distinct research fields into common platforms. Physics-based descriptions are essential for understanding fluid behavior, as electronic and atomic properties most often affect the macroscale behavior of a substance. Advances in materials science have also enabled the conduction of laboratory experiments close to the atomic scale with increased accuracy, while industrial and large-scale research continuously provides real-case data to guide ongoing investigations. Despite its widespread success in studying atomic-scale phenomena, traditional Molecular Dynamics (MD) is constrained by high computational demands, especially for long-timescale simulations. Force-field accuracy constitutes another major limitation; while empirical potentials are computationally efficient, they often lack generalizability across diverse chemical environments. These bottlenecks have driven research efforts toward enhancing MD with the aid of Machine Learning (ML). Consequently, this dissertation focuses primarily on the current impact of these methods on fluid research, highlighting advances and opportunities, identifying proposed techniques, and tabulating as well as testing some of the most popular algorithms that have demonstrated significant accuracy and performance in fluid applications. Furthermore, it investigates algorithmic accuracy across various fluid datasets. Non-linear, decision tree-based methods have shown remarkable performance in reproducing fluid properties. In addition, challenges associated with the interconnection of MD and ML are discussed. The movement of ions in water nanochannels under the effect of an electric field is shown to be multifactorial. Potential energy regions of varying intensity are formed inside the nanochannel, causing ions to either drift toward the walls and thus be rejected from the solution or form clusters trapped within low-potential energy regions. To address this, a twofold computational approach is proposed, constituting a fast and accurate solution that could be adapted to similar ion separation models for property extraction. Viscosity and thermal conductivity values for four basic elements, namely argon, krypton, nitrogen, and oxygen, were gathered from experimental and simulation literature data, forming a primary database for further investigation. The data span a wide pressure temperature (P-T) phase space, covering fluid states from gaseous to liquid and supercritical. This database is enriched with new simulation data extracted from equilibrium molecular dynamics (MD) simulations. In terms of algorithmic performance, stacked and tree-based ML models yielded the most accurate results for all elements, proving to be excellent choices for small to medium sized datasets. In this manner, a twofold computational scheme is constructed, serving as a computationally inexpensive route that achieves high accuracy, aiming to replace costly experiments and simulations whenever feasible.

Degree

thesis:*
Grantor dc:publisher
University of Thessaly (UTH)
Year dc:date
2026

Author and committee

dc:creator, dc:contributor.*
Authors dc:creator
  • Stavrogiannis, Christos
  • Σταυρογιάννης, Χρήστος

Subjects

dc:subject × 28

Rights

Language dc:language
gre

Identifiers

dc:identifier.*
Identifier
10.12681/eadd/62935

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

Stavrogiannis, Christos; Σταυρογιάννης, Χρήστος. Τεχνητή νοημοσύνη και μοριακή δυναμική για τον υπολογισμό των ιδιοτήτων των ρευστών σε ατομική κλίμακα: μέθοδοι και εφαρμογές. University of Thessaly (UTH), 2026. http://hdl.handle.net/10442/hedi/62935