Technical University of Crete (TUC)
Contribution to the numerical solution of hyperbolic conservation laws with applications in Computational Fluid Dynamics
Abstract
dc:descriptionHigh-order Discontinuous Galerkin (DG) numerical methods offer increased accuracy for the numerical solution of hyperbolic conservation laws; however, in the presence of discontinuities they may exhibit non-physical oscillations and numerical instabilities, making the use of limiting mechanisms essential. This doctoral dissertation focuses on the development and integration of a novel troubled-cell indicator (TCI) for high-order (DG) methods, with particular emphasis on the ADER-DG framework on unstructured triangular meshes. In modern a posteriori limiting strategies, the troubled-cell indicator plays a central role by deciding whether an element retains the unlimited high-order solution or is recomputed using a more robust scheme. Classical TCI approaches are often based on problem-dependent heuristics, adjustable thresholds, or geometric assumptions, which can reduce the robustness and generalization capability of a computational method. This limitation becomes particularly pronounced on unstructured meshes, where the same local solution pattern may appear under different element orientations. To address these shortcomings, this dissertation introduces a novel machine-learning-based troubled-cell indicator that enforces geometric symmetry directly at the architectural level. The proposed indicator, termed SCNN-TCI, is based on a symmetry-aware Siamese Convolutional Neural Network designed to be invariant with respect to rotations and reflections of the reference triangle. This design ensures that physically equivalent local configurations lead to consistent decisions, independently of the local mesh orientation. The SCNN-TCI is embedded into an ADER-DG solver within an a posteriori limiting loop, in combination with a conservative flux-correction mechanism based on the mortar method and a subcell finite-volume scheme applied to troubled elements. This approach introduces additional numerical dissipation only where necessary, while preserving the high-order accuracy of the ADER-DG method in smooth regions. Numerical experiments on analytical test functions, as well as benchmark problems for the two-dimensional Burgers and Euler equations, demonstrate the effectiveness, stability, and geometric consistency of the proposed indicator. Overall, this dissertation presents a robust and interpretable approach for the detection of discontinuities on unstructured meshes, enhancing the reliability and generalization of high-order computations in complex flow problems such as those described by hyperbolic conservation laws.
Degree
thesis:*- Grantor dc:publisher
- Technical University of Crete (TUC)
- Year dc:date
- 2026
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Kokkinakis, Georgios
- Κοκκινάκης, Γεώργιος
Subjects
dc:subject × 32- Αριθμητική ανάλυση
- Αριθμητική επίλυση διαφορικών εξισώσεων
- Αριθμητικές μέθοδοι επίλυσης ΜΔΕ
- Υψηλής τάξης ακρίβειας ασυνεχείς μέθοδοι Galerkin
- Μονοβηματικές μέθοδοι ADER-DG
- Υπολογιστική ρευστοδυναμική
- Υπερβολικοί νόμοι διατήρησης
- Μηχανική μάθηση
- Συνελικτικά νευρωνικά δίκτυα
- Επιστημονικοί υπολογισμοί
- Numerical analysis
- Numerical solution of differential equations
- Numerical solution of PDEs
- High-order discontinuous Galerkin methods
- One-step ADER-DG methods
- Computational fluid dynamics
- Hyperbolic conservation laws
- Machine learning
- Convolutional neural networks
- Scientific computing
- Μαθηματικά
- Φυσικές Επιστήμες
- Υπολογιστικά μαθηματικά
- Εφαρμοσμένα μαθηματικά
- Επιστήμη Ηλεκτρονικών Υπολογιστών και Πληροφορική
- Τεχνητή νοημοσύνη
- Mathematics
- Natural Sciences
- Computational Mathematics
- Applied Mathematics
- Computer and Information Sciences
- Artificial Intelligence
Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/60981