COMPETENCE NETWORK FOR TECHNICAL, SCIENTIFIC HIGH PERFORMANCE COMPUTING IN BAVARIA
Field of work:KONWIHR Division South
Computational Steering of Complex Flow Simulations
Computational Science and Engineering faces a continuous increase of speed of computers and availability of very fast networks. Yet, it seems that some opportunities offered by these ongoing developments are only used to a fraction for numerical simulation. Moreover, despite new possibilities in computer visualisation, virtual or augmented reality, and collaboration models, most available engineering software still follows the classical way of a strict separation of pre-processing, computing, and post-processing. In the previous work of the applicants of this proposal, some of the major obstructions for an interactive computation for complex simulation tasks in engineering sciences have been identified and partially removed. These were especially found in traditional software structures, in the definition of geometric models and boundary conditions, and in the often still very tedious work of generating computational meshes. A generic approach for collaborative computational steering has been developed, where pre- and post-processing are integrated with high-performance computing and which supports cooperation of workgroups being connected via the internet. Suitable numerical methods are at the core of this approach such as the lattice Boltzmann method (LBM) for fluid flow simulation. The proposed project will extend this approach in different directions:
- The proposal focuses on interactive computational fluid dynamics simulations including particle transport. In order to tackle the huge computational effort of the underlying flow problems, parallelisation is inevitable. Hence, already existing codes (developed for the old HITACHI vector architecture) will be ported, extended, and optimised for the HLRB II dual-core architecture, and they will be prepared for future multi- and many-core architectures. This comprises parallelisation of particle transport, performance optimisation of LBM (exploiting the shared-memory concept of HLRB II), and load balancing techniques according to the changing geometry due to steering, e. g.
- The versatility of the underlying geometric models will be extended. Special emphasis will be laid on the development of grid generation techniques being robust against flaws in the geometric model (small gaps between or overlaps of surface elements). These flaws occur very frequently in practice and are one of the major reasons for too long engineering working cycles in computational engineering. Where these flaws cannot be "healed" automatically, methods will be developed to easier identify and interactively remove them. Automatic and robust mesh generation is an essential prerequisite for interactive computational steering.
- Simulation of particle transport will be included into the computational steering model. To this end, the KONWIHR-II-project proposed here will be associated to a recently started project group within the "TUM International Graduate School of Science and Engineering (IGSSE)" on Particle Dynamics in Turbulent Flows (see www.igsse.de for details). In this associated project, key research issues are: assessment of individual terms in the particle transport equation, modelling of mechanisms such as particle-fluid, particleparticle and particle-wall interaction, the transition from stochastic particle models to a deterministic description of second order moments, and multi-scale approaches for an enhanced numerical simulation. Based on the work in this IGSSE project, the research proposed here will integrate the interaction models in the lattice Boltzmann flow simulator and further develop the numerical methods for the transport simulation with respect to the demands of a computational steering environment.