ENGINEERING AND CFD
Oscillation Control of Compressible Channel Flow
Principal Investigator:
Prof. Holger Foysi
Affiliation:
Chair of Fluid Dynamics, Universität Siegen, Siegen, Germany
Local Project ID:
osccompchannel
HPC Platform used:
JUWELS Booster of JSC
Date published:
Abstract
Reducing drag in engineering type flows is of paramount importance. Various approaches and configurations were tackled in the past, mostly, however, dealing with incompressible flow. In this project, researchers at University of Siegen were investigating a specific oscillation type control method for sub- and supersonic channel flow, a configuration where the fluid domain is restricted by cooled lower and upper walls. This canonical flow is well-known and usually the flow configuration of choice when investigating the efficacy of specific methods before proceeding to more complex wall bounded flows. While a laminar flow is preferred concerning the flow induced friction, technically relevant flows are turbulent, though. Due to high momentum transfer through vortices between the outer and the near-wall region resulting in large near-wall gradients, the flow inherently features an additional considerable amount of friction drag compared to laminar conditions. Many engineering applications in the aerospace industry additionally have to deal with supersonic flow, subject to large compressibility effects, which could enhance or attenuate the applied control methods. Here, we use an oscillatory flow control technique to influence the turbulence structure, studied already for incompressible flow [1], to investigate the influence of compressibility on its efficacy. The code PyFR [2] was run on the Juwels GPU cluster for this research project. A significant change in the flow structure with increasing Mach number was observed, requiring larger control wave lengths but also resulting in less drag compared to flow at lower Mach numbers.
Fig. 1: Vortical structures colored by the streamwise velocity in close vicinity to the wall in the natural flow (left) and with flow control (right).
Computational challenges
The temporal oscillation control procedure can be transformed into a spatial sinusoidal oscillation, increasing efficacy. This way, typical instabilites in the near-wall flow can be impeded, that are known to be primarily responsible for the generation of skin friction drag. An impression of the near-wall flow is given in Fig. 1. Vortical structures are depicted in the un-controlled flow (left), whereas the imprinted sinusoidal pattern is clearly visible in the controlled flow (right).
Turbulent flows contain a multitude of length- and time scales. The diameter of the largest eddies is of the order of the channel half width. They become progressively smaller when approaching the wall, before they are finally dissipated at the wall.
The majority of typical industry applications depend on models for the near-wall flow, since the numerical grid being relatively coarse to reduce computational time is not able to resolve these small length scales sufficiently. By using the time-averaged equations of motion for fluid flow the global character of the flow can be reconstructed with sufficient accuracy for these cases of application.
In turbulence research, however, we are particularly interested in the dynamics of the near-wall flow. This necessitates a much finer numerical grid to resolve the corresponding length scales directly. As a result, the computational cost grows massively and high performance computing is hence necessary even for simple geometries as is the case in channel flow.
Additional issues arise for the specific flow cases tackled in the present project. As mentioned above, supersonic flow is investigated as well, imposing further requirements on the stability of the simulations. With the computational capabilities nowadays, it is possible to simulate the flow in conditions that progressively approach those of technically relevant flows. Very large-scale structures appear in the outer layer of these flows affecting the near wall flow as well. While the interaction between the near-wall flow and the outer very large-scale structures is generally well represented even when cutting these structures by virtue of too small a domain, it is a research topic in the present project, too, in which way these large structures differ when subject to different flow conditions and flow control oscillation or wave lengths. Due to large flow domains required to incorporate the large vortical structures in the channel center, the computational cost increases additionally by a factor of twenty. More than one configuration is furthermore necessary to be able to judge the influence of various flow conditions. The JUWELS booster with its 936 compute nodes equipped with modern GPUs was especially suitable for the challenges in this project. Interesting results could be achieved, indicating a clear change in the flow structures with increasing Mach number, requiring larger control wave lengths compared to low Mach number flows, for example. For further results please refer to [3].
Computational details
The employed flow solver was PyFR [1], generating a platform specific code at runtime through its CUDA backend in case of the JUWELS booster. Selected simulations used for the characterization of the very large scales are exemplarily given in Table 1. 128 nodes each equipped with 4 A100 GPUs were used for the largest simulations. The file size of a single instantaneous snapshot amounts to 70 GB for the largest case. Parallel visualization tools were thus inevitable, too.
Table 1: Selected cases representing the largest resources required on JUWELS booster.
A volume representation of case M03R800 is given in Fig. 2. It highlights the very large scale structures in the core region. They appear in the form of elongated, meandering alternating low-speed / high-speed streaks. Their lengths amount to 2 − 5h, where h is the channel half-width. The employed computational domain is capable of resolving them properly. Results obtained so far indicate that there is no disparity between natural uncontrolled subsonic and supersonic flows regarding the dynamics and length scales of the very large structures. In the controlled flow the effect of these structures is relatively stressed all the way to the near-wall layer.
Fig. 2: Volume representation of the flow case M03R800 coloured by the streamwise velocity normalized with the bulk velocity (averaged over the whole domain).
References
[1] Viotti, C., Quadrio, M. & Luchini, P. 2009 Streamwise oscillation of spanwise velocity at the wall of a channel for turbulent drag reduction. Physics of Fluids 21 (11), 115109.
[2] Witherden, F.D., Farrington, A.M. and Vincent, P.E. 2014 PyFR: An open source framework for solving advection–diffusion type problems on streaming architectures using the flux reconstruction approach. Computer Physics Communications 185 (11), 3028 – 3040.
[3] Ruby, M. & Foysi, H. 2022 Active control of compressible channel flow up to Mab= 3 using direct numerical simulations with spanwise velocity modulation at the walls. GAMM-Mitteilungen 45.