The goal of this type 2 project is the design, analysis and implementation of numerical algorithms on parallel computers for predicting nonlinear material and structural behaviour.
New computational methodologies have been introduced for some of the target applications.
New solution methods have been implemented and tested in the framework of parallel processing. Domain decomposition has been applied whenever possible, and many new developments have been made for its application to nonlinear problems. Geometric decomposition algorithms have been proposed and implemented, new solution strategies (mostly based on the Newton method) have been developed and tested, load balancing strategies have been proposed in the context of some applications. In particular, comparisons has been made between direct and iterative solvers for the implementation of the domain decomposition method on MIMD machines, and new iteration techniques such as GMRES have been tested on SIMD architecture.
Prototype softwares have been developed for the different application cases, and tests have been performed on computer platforms with different architectures: SIMD, MIMD with shared memory and distributed memory architectures, clusters of workstations.
During the conduct of the research, the partners have demonstrated a high capacity to adapt to the evolution of the computer technology and market. Indeed the technology of parallel processing is a rapidly evolving technology, and it turned out necessary to readapt the technical choices in terms of programming paradigms and environment during the conduct of the research. The expertise acquired by the partners is thus in itself an important result of the project.
Most programs were written in Fortran and under PVM parallel environment.
The demonstration has been made on selected benchmark problems.
Publications and technical reports have been issues which present in length the results of the research, and presentations have been made to specialized meetings.
The physical problems under concern, namely : thermo-elasto-viscoelastic material behaviour, geometrically nonlinear behaviour of composite structures, polymeric materials (viscoelastic fluids), steel-elastomer composite structures, powder materials, have been selected according of the partners' experience and the expressed industrial needs. They raise computational problems of a generic nature which justify an integrated approach.
Optimum implementation of simulation methods on parallel computers requires the development of an appropriate methodology and the use of adapted numerical algorithms. It will allow the full potential of the simulation methods existing nowadays to be tapped.
The fulfilment of this objective implies close collaboration between research teams specialized in structural implies close collaboration between research teams specialized in structural and material behaviour, on the one hand, and in parallel computing, on the other.
The research will address following issues : mathematical models, numerical discretization methods, nonlinear parallel algorithms, implementation on parallel computers, solution of demonstrative application cases.
Funding Schemeundefined - undefined