Modern engineering simulations consume a lot of resources in terms of computational capacity and energy which often results in workflows that last for days or weeks. Fortunately, most modern solvers are highly parallelizable and show excellent scaling when running on High Performance Computing (HPC) systems. The REGALE project is a European Union funded open-source initiative that aims to create an HPC architecture which integrates various heterogeneous computing systems to provide an extremely scalable solution for end-users, focused on performance, energy efficiency, reliability, and ease-of-use. As a demonstration of REGALE’s capabilities in handling massively parallel workloads, a simulation use case is presented by the National Technical University of Athens, Greece and TWT GmbH, Germany whose aim is to model the deformation of a carbon nanotube-reinforced polymeric car bumper under various loading conditions. A family of computational tools for all multiscale analysis stages has been customized and further integrated in the framework for this purpose. These involve: (i) Validation, Verification and Uncertainty Quantification (VVUQ) methods for the construction of a computational model of the bumper, tailored to HPC requirements, (ii) custom machine learning (ML) algorithms for accelerating multiscale analysis, (iii) Energy efficient, fault tolerant domain decomposition-based solvers that rely on the aforementioned ML algorithms. Call: H2020-JTI-EuroHPC-2019-1
Session:
Poster
|