With Moore’s Law being verified, all we had to do was wait a few years to get more computing power. However, this increase in power is no longer exclusively based on the performance of the processor, but more and more on its ability to execute several tasks simultaneously. In the context of an internship at Dassault Aviation, various methods have been studied to take advantage of parallel computing, applied to a structural finite element code. This report presents the use of several processor cores in parallel with OpenMP. Then, two methods of decomposition into sub-domains allowing, thanks to MPI, the use of several computers to solve the same problem are presented. These two techniques are complementary and together allow to take thebest advantage of highly parallel computing servers. / När Moores lag hade bekräftats var det tillräckligt att vänta några år på ökad datorkraft. Denna effektökning baseras dock inte längre enbart på processorns prestanda, utan i allt högre grad på dess förmåga att utföra flera uppgifter samtidigt. Inom ramen för en praktikperiod på Dassault Aviation studerades olika metoder för att dra nytta av parallella beräkningar, tillämpade på en strukturell finita elementberäkningskod. I den här rapporten presenteras användningen av flera processorkärnor parallellt med OpenMP. Därefter presenteras två metoder för dekomponering av delområden som gör det möjligt, tack vare MPI, att använda flera datorer för att lösa samma problem. Dessa två tekniker kompletterar varandra och tillsammans gör de det möjligt att dra full nytta av högt parallella datorservrar.

When doing finite element analysis upon the structure of Saab’s aeroplanes a coarse global model of mainly shell elements is used to determine the load distribution for sizing the structure. At some parts of the aeroplane it is however desirable to implement a more detailed analysis. These areas are usually modelled with solid elements; the problem of connecting the fine local solid elements to the coarse global model will shell elements then arises.   This master thesis is preformed to investigate possible Global-Local methods to use for the structural analysis on Gripen. First a literature study of current methods on the market is made, thereafter a few methods are implemented on a generic test structure and later on also tested on a real detail of Gripen VU. The methods tested in this thesis are Mesh refinement in HyperWorks, RBE3 in HyperWorks, Glue in MSC Patran/Nastran and DMIG in MSC Nastran. The software is however not evaluated in this thesis, and a further investigation is recommended to find the most fitting software for this purpose. All analysis are performed with linear assumptions.   Mesh refinement is an integrated technique where the elements are gradually decreasing in size. Per definition, this technique cannot handle gaps, but it has almost identical results to the fine reference model.   RBE3 is a type of rigid body elements with zero stiffness, and is used as an interface element. RBE3 is possible to use to connect both Shell-To-Shell and Shell-To-Solid, and can handle offsets and gaps in the boundary between the global and local model.   Glue is a contact definition and is also available in other software under other names. The global respectively the local model is defined as contact bodies and a contact table is used to control the coupling. Glue works for both Shell-To-Shell and Shell-To-Solid couplings, but has problem dealing with offsets and gaps in the boundary between the global and local model.   DMIG is a superelement technique where the global model is divided into smaller sub-models which are mathematically connected. DMIG is only possible to use when the nodes on the boundary on the local model have the same position as the nodes at the boundary of the global model. Thus, it is not possible to only use DMIG as a Global-Local method, but can advantageously be combined with other methods.   The results indicate that the preferable method to use for Global-Local analysis is RBE3. To decrease the size of the files and demand of computational power, RBE3 can be combined with a superelement technique, for example DMIG.   Finally, it is important to consider the size of the local model. There will inevitably be boundary effect when performing a Global-Local analysis of the suggested type, and it is therefore important to make the local model big enough so that the boundary effects have faded before reaching the area of interest.

No description available.