ExaFSA - Exascale Simulation of Fluid-Structure-Acoustics Interactions

• Applicants: Professorin Dr. Sabine Roller; Professorin Dr. Miriam Schulte; Dr.-Ing. Dörte C. Sternel; Professor Dr. Alexander van Zuijlen
• Subject Area: Theoretical Computer Science ; Acoustics ; Mechanics ; Fluid Mechanics
• Term: from 2012 to 2020
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 230798225

Multi-physics simulations continue to be a very important type of simulation, if not the most relevant one since accuracy requirements enforce the consideration of more and more physical fields and phenomena in the underlying models. In many cases, the optimal model, the optimal type of solvers etc.~are still under investigation. This leads to a need for flexible modular simulation frameworks that allow for a fast and easy incorporation or exchange of existing stand-alone single-physics solvers. Since the accuracy requirements mentioned above can only be met with very high grid resolutions, using state-of-the art massively parallel high performance computing facilities is inevitable. However, the fast increasing degree of parallelism of such systems on the way to exascale performance poses severe challenges, not only but in particular for a simulation environment involving several independent software codes. Some of the respective open questions have already been answered in the first funding period for our chosen application type, i.e., fluid-structure-acoustics interactions: We have set up a full three-field simulation environment using a new, parallel and scalable type of numerics for the implicit fluid-structure coupling, advanced mapping and adaptivity strategies balancing accuracy and computational costs between the acoustic fluid in the near field and the pure acoustic far field, static load balancing over all involved codes, and a new, more efficient communication between the solvers at all interfaces. In addition, an efficient in-situ visualization is part of our environment, transforming results instantaneously into intuitively interpretable pictures and using gaps in the overall load balancing to minimize the visualization overhead. In the second period, the focus slightly changes from solving problems for the different forms of coupling involved to the overall picture of the fluid-structure-acoustics simulation environment. After many questions concerning numerical coupling scheme suitable for parallel simulations have been answered in the first funding period, this particularly means that our efforts in further algorithmic and technical optimizations are going to be enforced in the second funding period. In addition, considering the whole picture of fluid-structure-acoustics interaction also means that we have to validate our methods for further scenarios. This also includes the visualization that is going to become more focused on answering real application questions after general issues of integrating in-situ visualization into the overall simulation environment have been solved. Since fluid-structure-acoustics interactions combine a lot of typical challenges and characteristics of multi-physics applications, many of the methods and tools developed in our project are applicable also for the modular partitioned simulation of other multi-physics problems.

• DFG Programme: Priority Programmes
• Subproject of: SPP 1648: Software for Exascale Computing
• International Connection: Japan, Netherlands
• Partner Organisation: Japan Science and Technology Agency; JST
• Cooperation Partner: Hiroyuki Takizawa, Ph.D.