National Technical University of Athens (NTUA)
Massively parallel implementation of finite element, meshless and isogeometric analysis methods in computational mechanics
Abstract
dc:descriptionThe primary purpose of engineering analysis is to provide a numerical simulation of a physicalphenomenon in a way that is accurate but also computationally feasible. The need to accuratelysimulate various physical processes in complex geometries is important, and has perplexed scientistsfor many years. The up-to-date simulation methods can accurately model the physical domain but oftenrequire high computational effort. A simulation needs to be performed within a reasonable time-framewith the given computational resources in order to be affordable in real-world applications. Thus, animportant aspect in terms of feasibility is the efficient implementation of a simulation method thatenables its application in large-scale problems. While the prevailing cost in traditional finite elementanalysis (FEA) is in the solution phase, the main drawback of meshless methods (MMs) andisogeometric analysis (IGA) when addressing real-world problems is the high cost for the formulation ofthe characteristic matrices. Therefore, in order to make them efficient in large-scale simulations, thesemethods require massively parallel algorithms not only for the solution phase but also for the assemblyphase.The aim of this work is to accelerate computationally expensive parts of simulation methods in amanner that is both efficient and scalable. Algorithms in the context of this aim are explored andimplemented in this work. For the solution phase, domain decomposition is an attractive option as itsplits the domain into several subdomains and allows their concurrent solution. As for the formulationphase, assembling the matrix by non-zero allows different parts of the matrix to be calculated inparallel. The calculations involved in a particular algorithm must be performed efficiently. Hence,calculations like matrix operations, which are omnipresent in the simulation, should be handledappropriately. Each matrix format has its own strengths and weaknesses and should be used for thetask it is most suitable for. All of the above are combined with GPUs (graphics processing units), whichhave been attracting a lot of attention in recent years due to their remarkable performance features.GPU implementations are developed, for the solution phase in FEA and the formulation phase of MMsand IGA, which led to a great reduction of the time required for the simulation of a particular model.
Degree
thesis:*- Grantor dc:publisher
- National Technical University of Athens (NTUA)
- Year dc:date
- 2014
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Karatarakis, Alexander
- Καραταράκης, Αλέξανδρος
Subjects
dc:subject × 20- Ολοκλήρωση gauss
- Παράλληλη επεξεργασία
- Επιτάχυνση με κάρτες γραφικών
- Κατασκευή μητρώου
- Ισογεωμετρική ανάλυση
- Μη-πλεγματικές μέθοδοι
- Gauss quadrature
- Matrix assembly
- Parallel computing
- GPU acceleration
- Isogeometric analysis
- Meshless methods
- Φυσικές Επιστήμες
- Επιστήμη Ηλεκτρονικών Υπολογιστών και Πληροφορική
- Επιστήμες Μηχανικού και Τεχνολογία
- Επιστήμη Πολιτικού Μηχανικού
- Natural Sciences
- Computer and Information Sciences
- Engineering and Technology
- Civil Engineering
Rights
dc:rights- Statement dc:rights
-
- BY
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/38801