Reduced numerical solution times for combined boundary‐initial value problems using parallel computing
Autor: | Behnam Salimi, David R Hayhurst |
---|---|
Rok vydání: | 2010 |
Předmět: |
Speedup
Computer science General Engineering Boundary (topology) Parallel computing Degrees of freedom (mechanics) Solver Finite element method Computer Science Applications Computational science Matrix (mathematics) Computational Theory and Mathematics Conjugate gradient method Initial value problem Software |
Zdroj: | Engineering Computations. 27:746-772 |
ISSN: | 0264-4401 |
DOI: | 10.1108/02644401011062126 |
Popis: | Purpose — The purpose of this paper is to seek improved solution techniques for combined boundary‐initial value problems (IVPs) associated with the time‐dependent creep deformation and rupture of engineering structures at high temperatures and hence to reconfigure a parallel iterative preconditioned conjugate gradient (PCG) solver and the DAMAGE XXX software, for 3‐D finite element creep continuum damage mechanics (CDM) analysis.Design/methodology/approach — The potential to speed up the computer numerical solution of the combined BV‐IVPs is addressed using parallel computers. Since the computational bottleneck is associated with the matrix solver, the parallelisation of a direct and an iterative solver has been studied. The creep deformation and rupture of a tension bar has been computed for a range of the number of degrees of freedom (ndf), and the performance of the two solvers is compared and assessed.Findings — The results show the superior scalability of the iterative solver compared to the direct solver, with larger speed‐ups gained by the PCG solver for higher degrees of freedom. Also, a new algorithm for the first trial solution of the PCG solver provides additional speed‐ups.Research limitations/implications — The results show that the ideal parallel speed‐up of the iterative solver of 16, relative to two processors, is achieved when using 32 processors for a mesh of ndf = 153,238. Originality/value — Techniques have been established in this paper for the parallelisation of CDM creep analysis software using an iterative equation solver. The significant computational speed‐ups achieved will enable the analysis of failures in weldments of industrial significance. |
Databáze: | OpenAIRE |
Externí odkaz: |