ENGINEERING AND CFD

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.

Today, car manufacturers rely on CFD tools to design and optimise spark-ignition engines. However, current models of turbulent combustion—which are built based on the assumptions of the flamelet regime—lose their predictivity when used to simulate a highly diluted or ultra-lean combustion involving high turbulent intensities. Yet the combustion in a diluted boosted spark-ignition engine shifts from the flamelet to the thin reaction zone (TRZ) regime. This research project performed direct numerical simulations of premixed C₈H₁₈/air statistically flat flame interacting with a turbulent flow field. Results were analysed to develop a combustion model suitable for combustion in the TRZ regime based on the formalism of the coherent flame model.

Researchers of Koc University, Istanbul, performed extensive large-scale direct numerical simulations of turbulent bubbly channel flows to examine combined effects of surfactant and viscoelasticity by using a fully parallelized 3D finite-difference/front-tracking method. The insights achieved shed light for the first time on the intricate interactions of soluble surfactant and viscoelasticity in complex turbulent bubbly flows and reveal their effects on friction drag in channels. The results are expected to guide practitioners in engineering designs such as heat exchangers and pipelines. 

While for the design point operation of centrifugal pumps an essentially steady flow field is present, the flow field gets increasingly unsteady towards off-design operation. Particular pump types as e.g. single-blade or positive displace pumps show a high unsteadiness even in the design point operation. Simulation results for the highly unsteady and turbulent flow in a centrifugal pump are presented. For statistical turbulence models an a-priori averaged turbulence spectrum is assumed, and limitations of these state-of-the-art models are discussed. Since the computational effort of a scale-resolving Large-Eddy-Simulation is tremendous, the potential of scale-adaptive turbulence models is highlighted.

Flows over the curved surface of wings, cars, turbine blades in gas turbines and impeller blades in pumps have curved streamlines. The influence of streamline curvature on flows, drag and also heat transfer in flows is substantial to large. However, engineering models have difficulties in correctly predicting flows over curved surfaces and our knowledge on streamline curvature influences on flows is still limited. In this project, turbulent flows in moderately to strongly curved channels are studied by highly accurate, large-scale numerical simulations fully resolving the turbulent fluid motions. These give important insights into streamline curvature influences on flows, and produce data that form the basis for better engineering models.

Impressive progress has recently been made in machine learning where learning capabilities at (super-)human level can now be produced in non-spiking artificial neural networks. A critical challenge for machine learning is the large number of samples required for training. This project investigated new high-throughput methods across various domains for biologically based spiking neuronal networks. Sub-projects explored tools and learning algorithms to study and enhance learning performance in biological neural networks and to equip variants of data driven models with fast learning capabilities. Applications of these learning techniques in neuromorphic hardware and design for their future application in neurorobotics were also included.

Data sent over the internet relies on public key cryptographical systems to remain secure. A project under leadership of Dr. Paul Zimmermann of the French National Institute for computer science and applied mathematics (INRIA), run on HPC system JUWELS of the Jülich Supercomputing Centre, has been carrying out record computations of integer factorisation and the discrete logarithm problem, the results of which are used as a benchmark for setting the length of the keys needed to keep such systems secure.

The simulation of turbulent, partially premixed flames constitutes a challenge due to the complex interplay of the mixing process of fuel and oxidizer, chemical reactions and turbulent flow. Therefore, a detailed numerical simulation of an experimentally investigated flame of laboratory scale has been performed, which allows to study these fundamental interactions in great detail. The results have been compiled into a database which aids the improvement of future combustion simulations. The simulation has been performed with an in-house solver based on OpenFOAM, which includes several performance optimizations to maximize the hardware utilization on supercomputers.

Recent direct numerical simulations in closed slender Rayleigh-Bénard convection cells advanced to Rayleigh numbers of Ra = 10¹⁵ which were never obtained before and reveal a classical turbulent transport law for the heat transfer from the bottom to the top of the cell which is based on the concept of marginally stable boundary layers. Our simulations were able to resolve the complex dynamics inside the thin boundary layers at the top and bottom plates of the convection cell and to determine a steady increase of the turbulent fluctuations without an abrupt transition near the wall for a range of 8 orders of magnitude in Rayleigh number.