Artificial Intelligence and Machine Learning
TU Ilmenau Researchers Use JUWELS, Machine Learning to Better Model Highly Turbulent Flows
Principal Investigator:
Prof. Dr. Jörg Schumacher
Affiliation:
TU Ilmenau
Local Project ID:
mesoc
HPC Platform used:
JUWELS at JSC
Date published:
Horizontal temperature field cuts through the convection layer taken close to the top plane inside the thermal boundary layer. The rotation about the vertical axis that points out of the plane increases from left to right. The aspect ratio is 60:60:1. The left case is non-rotating. The lower row is a magnification which displays individual convection cells that remain intact despite the large-scale rotation. Image credit:
A team of researchers led by TU Ilmenau Professor Jörg Schumacher have been using the JUWELS supercomputer at the Jülich Supercomputing Centre to run highly detailed direct numerical simulations (DNS) of turbulent flows at the so-called mesoscale—the intermediate range where both small-scale turbulent fluid interactions and large-scale fluid dynamics converge. As long-time users of GCS resources, the team has used state-of-the-art in high-performance computing to better understand turbulent flows at a fundamental level.
The team uses JUWELS to run computationally expensive DNS to chart the patterns and order that forms in turbulent flows at the mesoscale, then takes these valuable computational insights and uses it as reliable, detailed inputs for reduced-order models that can be run by more modest computational resources by taking advantage of machine learning models. With its proof of concept complete, the team uploaded its pre-publication results on arXiv.org. The team plans to use this method for studying compressible turbulent fluids moving forward.
For the full report, click here.
For the full journal article, click here.