Resumen de A high-performace computing tool for partitioned multi-physics applications
The simulation and modelling of complex applications involving the interaction of processes governed by different physical principles is addressed in this thesis. The interaction of a fluid with a deformable body, or the exchange of thermal energy between fluid and solid are examples of these multi-physics applications. In these two cases, the modelling strategy proposed here combines the solution of separated physical systems to account for the interactions taking place through the entire domain. As a consequence, the simulation process resulting from the use of separated systems considers independent codes to find the solution of each system, while the entire system is reconstructed through an iterative approach combining these solutions.
One of the main advantages of this partitioned approach is that each parallel code can use the most appropriate model and algorithm which allow achieving an accurate solution for the complete physical system. Nevertheless, several challenges must be considered when using this approach. For instance, from a physical point of view, the most of variables involved in the modelling of a multi-physical application must be continuous across the entire domain. From a computational point of view, efficient data transference between parallel codes is required to model the physical interactions taking place through the entire system. In addition, the simulation of multi-physics applications must be robust and maintain scalability not only for each parallel code, but also for the coupling problem.
The present work describes the development, validation and use of a high performance computing coupling tool designed for solving efficiently partitioned multiphysics applications. The emphasis has been placed to the development of strategies to make efficient use of large-scale computing architectures, but always keeping the robustness and accuracy of the solutions. The coupling tool developed controls the data transference between the parallel codes establishing peer-to-peer communication layouts between the processors, the dynamic localization of regions where physical interactions take place, and the possible interpolations required between the different meshes composing large-scale multi-physics application. In this work, these features are applied to solve two multi-physics applications: contact of deformable bodies, and conjugate heat transfer.
The contact problem involves the interaction of two or more solids which could deform. In this work, a parallel algorithm to deal with this problem is described. The continuity of the variables involved in the coupling problem is ensured using a domain decomposition method. The regions of the surface for each body where the contact takes place are identified using the localization process implemented in the coupling tool. The results show that the parallel algorithm used here for the solution of contact problems agrees well with those achieved by the elastic contact theory as well as those obtained by commercial codes.
The conjugate heat transfer problem referes to the thermal interaction between a fluid and a solid. In this case, the coupled process is similar to the contact problem. The results show the capability of the framework developed in this thesis to deal with practical engineering applications. In order to demonstrate the capability of the coupling tool to deal with large-scale applications, a parallel performance study of the partitioned approach is developed in this thesis. The study leads to a load balance strategy that allows estimating the optimal performance of a parallel multi-physics application. The parallel performance analysis of a conjugate heat transfer problem shows that the optimal efficiency of this application is well represented by the expressions derived in this study.
