Back to search

National Technical University of Athens (NTUA)

Ανάλυση φορέων με συμβατικά, προσαρμοστικά και στοχαστικά πεπερασμένα στοιχεία σε διαδικτυωμένους ηλεκτρονικούς υπολογιστές

Abstract

dc:description

Parallel programming is widely accepted by engineers and scientists as the today’s and future’s tool to overcome the computational barriers imposed by CPU speed and memory limitations in sequential processing. A platform for parallel computing that is continuously gaining popularity over the last years is the cluster of networked computers. Such clusters can be used as a single distributed computational resource for high performance computing, which is characterized by its low installation, interconnection and maintenance cost and thus its increasing availability. The research conducted in this thesis has been directed towards the exploitation of such computing environments, since even small engineering enterprises can afford today their own networks of workstations and/or Pentium PCs executing distributed programs. The engineering approach towards parallelism for the Finite Element (FE) analysis of a structural model involves the partitioning of the FE mesh into submeshes called subdomains. This process, termed as domain decomposition, has served as the basis for the development of efficient matrix handling techniques and solution algorithms for the parallel FE analysis of large-scale problems encountered in computational structural mechanics. Domain decomposition methods allow the concurrent handling of the subdomains of the structure on different processors accelerating this way the overall FE analysis. The research work in this thesis focuses on a method called FETI (Finite Element Tearing and Interconnecting), which is an iterative domain decomposition solution algorithm. According to the FETI method the domain is partitioned into a set of totally disconnected subdomains and the global problem is replaced by subdomain equations of equilibrium subject to the compatibility of the subdomain displacements across the subdomain interfaces. The resulting interface problem is in general indefinite due to the presence of floating subdomains, which do not have enough prescribed displacements to eliminate the local rigid body modes. The indefinite problem is handled by an iterative Preconditioned Conjugate Projected Gradient (PCPG) algorithm. The original contributions presented in this thesis involve both theoretical and practical aspects of the FETI method. A theoretical issue of the FETI method concerning the determination of the subdomains’ rigid body modes has been treated with a hybrid approach, which combines analytical and numerical computations. The development of new preconditioners for the iterative PCPG algorithm constitutes another innovation of theoretical nature. A practical contribution of the thesis is the developed FETI implementation, which has been programmed according to the master/slave model for distributed computing, in order to minimize the effect of the network’s low communication speed on the performance of the network-distributed FE application. Another practical issue addressed is the generation of subdomain clusters, which emerges due to the fact that the FETI method is used in practice with the number of generated subdomains being generally larger than the number of available processors. The task of handling the resulting matrices and solving the governing system of algebraic equations becomes even more demanding and challenging in parallel adaptive FE applications. The domain decomposition-based procedure adopted in this thesis to treat large-scale h-, p- and hp-version FE problems is implemented as an extension to a conventional parallel FE program, in which the powerful parallel FETI solver is employed to handle the FE system for each adaptive step. According to this approach, each adaptive step is treated as a stand-alone FE problem, avoiding this way the complexity of a dynamic load balancing strategy with the migration of elements from one processor to another. Another field of structural engineering involving computationally intensive applications is stochastic FE analysis. The Monte Carlo Simulation (MCS) technique, which is the most effective and widely applicable method for handling large-scale stochastic FE problems with complicated structural response, involves an expensive computational procedure requiring successive FE analyses. The techniques implemented in this thesis accelerate the solution of each simulation’s FE system by taking advantage of the fact that the simulations’ stiffness matrices have relatively small differences. This set of nearby FE problems is solved in parallel with the hybrid iterative domain decomposition technique PCG-FETI, which effectively combines the conventional Preconditioned Conjugate Gradient (PCG) method and the FETI method. The presented numerical results demonstrate the efficiency of the parallel PCG-FETI method for solving nearby problems arising in MCS-based FE analysis of large-scale shell structures with stochastic material properties.

Degree

thesis:*
Grantor dc:publisher
National Technical University of Athens (NTUA)
Year dc:date
2002

Author and committee

dc:creator, dc:contributor.*
Authors dc:creator
  • Χαρμπής, Δήμος
  • Charmpis, Dimos

Subjects

dc:subject × 16

Rights

Language dc:language
gre

Identifiers

dc:identifier.*
Identifier
10.12681/eadd/16724

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

Χαρμπής, Δήμος; Charmpis, Dimos. Ανάλυση φορέων με συμβατικά, προσαρμοστικά και στοχαστικά πεπερασμένα στοιχεία σε διαδικτυωμένους ηλεκτρονικούς υπολογιστές. National Technical University of Athens (NTUA), 2002. http://hdl.handle.net/10442/hedi/16724