ESMORGA · Exploiting Superconvergence in Meshes for Optimal Representations of the Geometry with high Accuracy
Horizon 2020 — Marie Skłodowska-Curie Actions
- Duration
- 2020-12-01 → 2022-11-30
- EU contribution
- €160,932
- Participants
- 1
- Scheme
- MSCA-IF
Lines connect the coordinator with its partners.
Results in brief
Exploiting Superconvergence in Meshes for Optimal Representations of the Geometry with high Accuracy
The issue addressed in this project is high-fidelity simulations featuring super-accurate geometric and numerical accuracy. The project investigated tools enhancing the accuracy of large-scale simulations using unstructured high-order methods. Super-accurate geometric and numerical accuracy are important for society because these capabilities are key to enabling high-fidelity simulation. Note that high-fidelity simulations using supercomputers play a major role in the production costs of the transport design industry: they reduce the number of full-scale tests. This reduction in the design process is key to accelerating the deployment of sustainable aircraft devices, a deployment that will be crucial to address the environmental challenge of air transportation. The overall objectives are: - Develop a parallel automatic mesh generation tool approximating general CAD data. This was achieved by minimizing a disparity measure that improves the accuracy compared to direct interpolation. The tool was implemented designing a Julia wrapper of the EGADS geometry kernel - Implement the SIAC filter (MSIAC) as a standalone tool for general use exploiting superconvergence: the filter can raise the convergence order from p+1 to 2p+1. The filtered data has increased smoothness and in general, reduces the numerical error. The main conclusions are: - The constrained disparity optimization preserves geometric super-convergence while significantly reduce the computational times and complexity - The solver was built on top of the EGADS kernel, enabling real-world applications such as aircraft design - The EGADS Julia port extended the applications of the EGADS kernel to other computing languages beyond C (C++) and it is now distributed within the ESP software - The code was parallelized and tested on distributed memory using the MareNostrum4 and revealing further computational efficiency - The MSIAC tool was successfully implemented as a standalone tool and handles handles 2D triangular and quad unstructured meshes, and 3D structured meshes - The filter was tested against turbulent flows from numerical solutions using FEM, DG (double Mach reflection) and including unstructured meshes (Trixie)
Data: CORDIS, © European Union
Project objective
High-fidelity simulations using supercomputers play a key role in aerospace and automotive design for reducing the number of ground and in-flight tests. The mesh lies at the heart of the simulation: it reproduces the geometry using elements that, for engineering applications, are required to be curved (high-order). This project exploits geometric superconvergence: extra accuracy using lower order elements. The main goal is to transform observed geometric superconvergence into theory. The originality of this work is to derive the mathematics behind geometric superconvergence including theoretical limits and to develop a parallel code for High Performance Computing (HPC). The action features a two-way transfer of theoretical and practical knowledge between the fellow and the host institution in the topics of superconvergence, high-fidelity geometry simulations and HPC. The fellow will be integrated in the internationally recognized Geometry and Meshing for Simulation group at the Barcelona Supercomputing Center (BSC). This action is devised to have significant impact on the fellow’s career, becoming an expert in superconvergence both for numerical approximations and geometry within the unstructured high-order simulations community focused on aircraft design. The proposal has the potential to establish the fellow as core scientist in the emerging research area of superconvergence and high-fidelity flow simulations, allowing the candidate to start and lead a new research group. The work packages and methodology have been devised to reach the goals on high-fidelity geometry for simulation including parallel implementation for large data clustering. A risk assessment plan has been designed ensuring that at least suboptimal superconvergence theory can be derived. The BSC infrastructure will be essential for the implementation of this high-fidelity geometry tool; it houses MareNostrum4, one of the most powerful supercomputers in Europe.
Original text from CORDIS.
Participants
- BARCELONA SUPERCOMPUTING CENTER CENTRO NACIONAL DE SUPERCOMPUTACION · BARCELONACoordinatorSpain
Links
Data: CORDIS, © European Union