SDL Energy Conversion

In the foreseeable future, combustion will still meet the major part of the world’s primary energy demand, but technologies will need to change for low-carbon and carbon-free energy conversion, using synthetic fuels as key contributors. System redesigns will have to heavily rely on high-performance reactive CFD (rCFD) with sophisticated numerics and accurate predictive physical models.

However, combustion applications are particularly challenging with respect to HPC due to the strong non-linearities and the large ranges of scales arising from the highly-coupled turbulence and chemistry interaction, the large number of partial differential equations for a detailed description of the multiscale phenomena, and the inherent multi-physics aspects.

The SDL Energy Conversion aims to enable computationally efficient simulations of real-scale combustion devices by developing HPC-ready rCFD software and methods by a co-design process.

Competencies

• Development of numerical methods for HPC architectures
  • e.g., improving efficiency through load-balanced mesh-refinement
  • e.g., porting the evaluation of chemical kinetics to GPUs
• Exploring the capabilities of AI-driven frameworks to enhance rCFD models
  • e.g., generative networks for SGS modeling
  • e.g., ANN for chemistry closure
• Leverage CPU- and GPU-based software
• Work centered around community-established open-source codes
• Expertise in conducting high-fidelity and large-scale numerical simulations

Current research topics

• Data-driven combustion modeling for laminar and turbulent flames
• Data-driven turbulence modeling for Large Eddy Simulations
• Evaluation of chemical kinetics and transport properties on heterogeneous architectures
• Development of Fortran-Python interface for machine learning (ML) inference on CPUs in collaboration with CSG Parallelism and Performance
• Inference of ML-based combustion models using GPUs in collaboration with CSG Parallelism and Performance

• Center of Excellence in Combustion (EuroHPC JU H2020)
• exaFOAM (EuroHPC JU H2020)
• CRC TRR 150
• CRC 129
• CRC 1194
• DFG FOR 2687

2025

• Parallel implementation and performance of super-resolution generative adversarial network turbulence models for large-eddy simulation (Ludovico Nista, Christoph D. K. Schuhmann, Peicho Petkov, Valentin Pavlov, Temistocle Grenga, Jonathan F. MacArt, Antonio Attili, Stoyan Markov, Heinz Pitsch, Computers & Fluids.
• Extrapolation Performance of Convolutional Neural Network-Based Combustion Models for Large-Eddy Simulation: Influence of Reynolds Number, Filter Kernel and Filter Size (Geveen Arumapperuma, Nicola Sorace, Matthew Jansen, Oliver Bladek, Ludovico Nista, Shreyans Sakhare, Lukas Berger, Heinz Pitsch, Temistocle Grenga, Antonio Attili), Flow, Turbulence and Combustion.

2024

• Influence of adversarial training for super-resolution models (Ludovico Nista, Heinz Pitsch, Christoph D. K. Schumann, Mathis Bode, Temistocle Grenga, Jonathan F. MacArt, Antonio Attili), Physics Review Fluids.
  
• Can flamelet manifolds capture the interactions of thermo-diffusive instabilities and turbulence in lean hydrogen flames? – An a-priori analysis (Hannes Böttler, Driss Kaddar, Jeremy Karpowski, Federica Ferraro, Arne Scholtissek, Hendrik Nicolai, Christian Hasse), International Journal of Hydrogen Energy.
• multiRegionFoam – A Unified Multiphysics Framework for Multi-Region Coupled Continuum-Physical Problems (Heba Alkafri, Constantin Habes, Mohammed Elwardi Fadeli, Steffen Hess, Steven B. Beale, Shidong Zhang, Hrvoje Jasak, Holger Marschall), arXiv preprint.
  
• Scalar mass conservation in LES of soot formation using mixturefraction-based combustion models (Marco Davidovic, Heinz Pitsch), Proceedings of the Combustion Institute.