High-fidelity large-scale CFD simulations on SuperMUC-NG: Improving accuracy and reproducibility of high-resolution schemes

Applicant

Prof. Dr. N. Adams
Institute of Aerodynamics and Fluid Mechanics
TU München

Project Overview

Improvements in numerical model development for multi-physics problems has enabled the research in fluid mechanics nowadays to consider very complex problems by high-performance computation. Such problems are characterized by nonlinear mechanisms that generate multiple temporal and spatial scales. Whereas turbulence is a broad-band phenomenon whose largest scales are determined by flow boundaries and exterior forcing, and whose smallest scales are determined by viscous dissipation length scales, singularities such as shocks and interfaces do not possess inherent length and time scales if considered in a continuum description. They generate small scales by instabilities, driven through their mutual interaction, and interact with broad-band flow structures, creating a scenario which is extremely complex for numerical flow modeling: high-resolution requirement of broad-band scales and instabilities, monotonic capturing of shocks and interface, tracking of interfaces without artificial diffusion and mass loss.
Over the years, efficient numerical models and advanced algorithms for sharp-interface representations of both phase interfaces and flow-field discontinuities have been developed that allow to enforce critical properties of the numerical discretization without compromising efficient computing strategies. As these numerical models are computationally expensive, best utilization of current and future high performance architectures is a necessity.
We have developed a multi-resolution simulation framework following a hybrid parallelization strategy to allow for large-scale simulations using HPC architectures. Application of massive MPI parallelization does not only allow for highly resolved meshes but also for higher-order methods. Recent publications provide so-called WENO (weighted essentially non-oscillatory) stencils up to 17th order. As shown in detail by different authors, these high-order stencils come along with a remarkable decrease in numerical dissipation of the simulation.

Aiming for accurate and high fidelity simulations of compressible flow dynamics for reliable research results, we want to improve our in-house simulation environment “ALPACA” to ensure reproducibility of simulations and eliminate numerical artifacts within this KONWIHR project.
The second key objective of this KONWIHR project will be improvement of our post-processing capabilities. Having a performant and scalable research code at hand, we want to perform highly resolved simulations of e.g. shock-bubble interactions in complex multi-bubble configurations. Visualization of such results from large-scale simulations using thousands of cores has proven to be a challenging task per se. In this project, we want to investigate different approaches for their visualization capabilities and performance.