Application of physics-Informed neural networks for the solution of non-Newtonian fluid flows

Ζαχαρόπουλος, Χρήστος; Zacharopoulos, Christos

dc.contributor.author	Ζαχαρόπουλος, Χρήστος	el
dc.contributor.author	Zacharopoulos, Christos	en
dc.date.accessioned	2025-04-02T06:18:25Z
dc.date.available	2025-04-02T06:18:25Z
dc.identifier.uri	https://dspace.lib.ntua.gr/xmlui/handle/123456789/61557
dc.identifier.uri	http://dx.doi.org/10.26240/heal.ntua.29253
dc.rights	Αναφορά Δημιουργού - Μη Εμπορική Χρήση - Παρόμοια Διανομή 3.0 Ελλάδα	*
dc.rights.uri	http://creativecommons.org/licenses/by-nc-sa/3.0/gr/	*
dc.subject	Physics-Informed Neural Networks	en
dc.subject	Non-Newtonian Fluids	en
dc.subject	Forward Problem Solution	en
dc.subject	Inverse Problem Solution	en
dc.subject	Power-law Fluid Flow	en
dc.subject	Physics-Informed Νευρωνικά Δίκτυα	en
dc.subject	Μη-Νευτωνικά Ρευστά	el
dc.subject	Επίλυση Forward Προβλήματος	el
dc.subject	Επίλυση Inverse Προβλήματος	el
dc.subject	Ροή Power-law Ρευστού	el
dc.title	Application of physics-Informed neural networks for the solution of non-Newtonian fluid flows	en
heal.type	masterThesis
heal.classification	Computational Fluid Mechanics	en
heal.classification	Machine Learning	en
heal.classification	Neural Networks	en
heal.language	en
heal.access	free
heal.recordProvider	ntua	el
heal.publicationDate	2024-10
heal.abstract	In this Master’s Thesis, Physics-Informed Neural Networks (PINNs) models are developed to approximate the solutions of non-Newtonian fluid flows. PINNs are deep learning models incorporating a specially designed loss function, which allows for respecting the physical laws. This loss function consists of several terms, representing the residuals of the partial differential equations (PDEs), the boundary conditions and the observation data, if available. Dynamic balancing of these terms during training is utilized, enhancing the robustness and the effectiveness of the PINN model. The purpose of this Thesis is to solve both forward and inverse problems involving PDEs that describe the flow of power-law fluids. In the forward problem, the PINN model produces accurate results for a range of power-law fluids. For the inverse problem, the PINN model effectively infers unknown power-law parameters utilizing a limited set of observation data, derived from a CFD simulation of the examined flow. It is important to note that said parameter inference is achieved with minimal modifications to the code developed for the forward problem and only a small increase in the computational cost, indicating an advantage over classical numerical methods for inverse problems. The outcomes show great promise for expanding this approach to higher dimensions, more complex geometries, and more advanced models of non-Newtonian fluids.	en
heal.advisorName	Kavousanakis, Mihalis	en
heal.committeeMemberName	Boudouvis, Andreas G.	en
heal.committeeMemberName	Sarimveis, Haralambos	en
heal.academicPublisher	Εθνικό Μετσόβιο Πολυτεχνείο. Σχολή Χημικών Μηχανικών	el
heal.academicPublisherID	ntua
heal.numberOfPages	62 σ.	el
heal.fullTextAvailability	false