Abstract: With the term “high-fidelity CFD” we refer to the class of methods that resolve turbulent flow fluctuations directly, rather than representing their time-averaged effect through heuristic models. As the name implies, such methods give more reliable predictions than steady-state approaches. This benefit, however, comes at a cost of significantly increased computational effort.
The capability to accurately simulate turbulent fluid flow is an essential tool in the design of more energy-efficient, less noisy, and more competitive products in a wide range of fields such as transportation, renewable energy and many more. However, numerous barriers need to be overcome before the full potential of high-fidelity CFD methods can be exploited through widespread adoption. The talk will give an overview of the barriers to adoption, before presenting ongoing work at Upstream CFD to overcome these on three different fronts.
Firstly, limitations in hybrid RANS-LES turbulence models cause problems with significant practical impact. The “grey-area” problem denotes a delayed transition between fully-modelled turbulence in an attached boundary layer and the development of three-dimensional resolved turbulence following separation. An effective remedy has been developed in a European-funded research project, which will be presented. A second issue, whereby fine grids cause the collapse of modelled turbulent stresses in attached (RANS-mode) boundary layers is discussed. High quality numerical schemes are also key to unlocking the potential of high-fidelity CFD. These must reduce numerical dissipation sufficiently to resolve fine-grained turbulence in LES mode whilst retaining robustness for complex flows on unstructured grids.
Secondly, computational expense is a major barrier to adoption. The enabling role played by cloud HPC resources is highlighted, as well as efforts to improve the efficiency and scalability of the CFD methodologies, thereby reducing cost and turnaround time.
Finally, the unsteady nature of high-fidelity simulations introduces new questions compared to steady-state CFD. At what point has memory of arbitrary initial conditions been overcome and a statistically steady state reached? Once a steady state is reached, how long must a simulation be run for to obtain sufficiently accurate estimates of the mean flow and other statistical quantities? An algorithm to answer these two questions, named Meancalc, will be presented and demonstrated in a pilot SaaS implementation. The approach can be interfaced with CFD simulations at run-time to optimise their computational expense.
Bio: Dr. Charles Mockett (Charlie) is the Managing Director of Upstream CFD, which he founded together with four long-standing colleagues in 2019. Upstream CFD are experts in high-fidelity CFD for aerodynamics and aeroacoustics. He received his PhD on the subject of Detached-Eddy Simulation at the Technical University of Berlin in 2009. He moved to Berlin in 2003 after graduating with a Master’s in Aeronautical Engineering from the University of Bristol, UK, in 2002.