WMLES

Large Eddy Simulation Workshop on Smooth-Body Separation at AIAA SciTech 2022

The AIAA Fluid Dynamics Technical Committee’s LES Discussion Group is organizing a workshop focused on the prediction of a separated flow using wall-resolved LES (WRLES) and wall-modeled LES (WMLES) at the SciTech 2022 meeting in San Diego, CA. The workshop will be held during the weekend immediately following the SciTech meeting, specifically on Sat/Sun January 8-9, 2022. A flyer for the workshop can be downloaded here.

Focus of the workshop

While there are many interesting and important aspects of LES, the organizing committee has decided to focus this specific workshop on the application of LES to wall-bounded flows at very high Reynolds numbers. As is well known, this requires the turbulence in the inner part of the boundary layer to be modeled rather than resolved. One of the most important challenges in such “wall-modeled LES” is the ability to accurately predict separated flows, specifically flows where the separation is not dictated by abrupt geometric changes (e.g., sharp corners). Therefore, the workshop will focus on a central test case that features an equilibrium incoming boundary layer, separation over a smoothly curved ramp, and reattachment. The test case will be run at Mach 0.2 in order to enable both compressible and incompressible codes to be used. The test case has further been chosen to strike a balance between physical realism and computational cost, where it is recognized that a limited computational cost implies both that a broader group of research groups can participate and (equally importantly) that participants can afford to compute on a sequence of refined grids in order to carefully assess grid convergence (or the lack thereof).
The purpose of the workshop is to provide a snapshot of the predictive capability of modern LES methods on smooth-body separation flows at high Reynolds numbers, and to build a comparative database for use in future method assessments.

Monthly videoconferences

Participants (and prospective participants, and interested observers) meet on the first Monday of every month at 12pm (noon) US Eastern Time. The Zoom link for these calls is https://umd.zoom.us/j/4487374796.

Notes and prior communications

• 02/07/2022 (telecon):
  • Post-workshop discussion about how to move forward
  • Will continue with the smooth ramp, but should we continue with exactly the same case specification (advantage: existing results don’t need to be re-run) or should we consider adjustments (advantage: if making adjustments creates a better case, then likelier to become a commonly used verification case in the future). Potential adjustments that were discussed:
    • Double the spanwise width. Reduces/removes the strong 2D vortex shedding dynamics which are unrealistic. Does not produce results that are independent of the spanwise width (much wider domains required for that). Should increase the total cost by 20-40% due to shorter averaging time required.
    • Change inflow turbulence to a steady inflow with a sharp step or similar to reduce variability due to synthetic turbulence methods.
    • Lower Reynolds number. Reduces the cost of WRLES or DNS. Current finest grids are close to WRLES limit, though, which would be a worse problem with lower Re.
  • Discussed whether to add the Boeing speed bump as an additional case. Three groups have (or are close to having) DNS results for Re=2M. Is likely a more expensive case than the smooth ramp due to the slightly wider domain (14 delta compared to 4 delta). Advantage: truth data exists. Disadvantage: quite low Re (Re_tau of 500) which makes good WMLES grids close to WRLES.
  • Current plan is do a 3-hour mini-workshop (for the LES case alone) online in July or September 2022, then another in early 2023, and then a full workshop in-person in summer 2023. Will continue telecons on the first Monday of every month.
  • Need to develop a better work flow for data processing and figure creation. Could involve more standardized data format, spreading the work among more people.
  • Agenda for next meeting:
    • Smooth ramp: keep the same or make adjustments?
    • Speed bump: should it be added, and if so, when do we add it?
    • Pick date for first mini-workshop.
• 12/06/2021 (telecon):
  • Reminder #1: register for the workshop ASAP!
  • Reminder #2: submit updated results by either (a) uploading directly to the Google Drive and e-mailing Johan, Ivan and Timofey; or (b) sending us the data directly.
  • The deadline for submitting data is fluid. We want to allow everyone as much time as possible to complete runs, but we also want to ensure that Ivan and Timofey can collect the data in a reasonable manner. If you submit data in exactly the same form as before, then we can accept submission updates up to a couple of days before the workshop. If you have never submitted data before, then we need time to understand your format, and thus we need your submission by Dec 15.
  • One updated request is that the mean pressure field should be included in the data (both 2D field and wall pressure profile).
  • The agenda for the workshop can be found at https://docs.google.com/document/d/1WqHZylASshGcWlTSoej5RGwpQHfEL-C9KHor-7WO6G8/edit?usp=sharing. The second page lists instructions for the participants’ presentation — please make sure to include all that information in your slides.
  • The participants should submit their slides to us a few days before the workshop. We will then place them on the Google Drive to give everyone the chance to familiarize themselves with the numerics, inflow techniques, domain sizes, etc from every participant before the actual workshop.
  • In addition to all the administrative things above, we also discussed some technical things:
    • Artem Korobenko showed their preliminary results with FEM.
    • The required time for the initial transient was discussed. Without knowing the correct answer, some groups found that they need 20-30 time units before averaging.
• 11/01/2021 (telecon):
  • Registration is now open and evidently works, see the link above.
  • Have received partial/preliminary results from 7 groups, all posted on the Google drive (should be accessible for everyone).
  • Johan will create a draft workshop agenda, this will then be edited during the Dec 6 telecon.
  • Ivan has volunteered to create all “spatial” plots (profiles, contours, etc — plots based on the 2D averaged fields).
  • Timofey Mukha has volunteered to analyse the time history data, both of separation/reattachment locations and the probes.
• 10/04/2021 (telecon):
  • Registration is now open, see the link above.
  • The format of the workshop will be as follows: we have one large room which will have a laptop running Zoom and projecting the Zoom output to the big screen in the room. Speakers (whether in-person or virtual) will connect to Zoom and share their screen. The room will have microphones and loud speakers to facilitate discussions to/from both in-person and virtual participants. We will probably be scheduled for 0.5-0.75 parts of a single day.
  • The extrude.f90 code for extruding the meshes was updated to correctly handle the finest mesh on Sept 20, 2021. Please download the latest version and report any problems.
  • Johan shared a preliminary analysis of the time histories of the separation and reattachment locations on grids 6-3 (the 4 coarsest grids). For the UMD/USC computations, the initial transient seems to be about 20 time units long. No sign of convergence of the reattachment location on grid 3. Johan has written a Matlab script that estimates the standard error in the sample mean, thus enabling computation of the 95% confidence interval on these quantities; this can be downloaded from larsson.umd.edu/code/ for interested participants.
  • Don Rizzetta discussed his and Dan Garmann’s WRLES results at both Re=300K and Re=1M. At both Reynolds numbers there is evidence of convergence on the finest grids. Don highlighted that they stretch the grid quite aggressively from x=2, which may affect the results.
  • Timofey Mukha discussed the variability he gets in the boundary layer thickness quantities depending on how he exactly defines the reference quantities. He will share the velocity profiles with Johan et al for additional checks making sure we are comparing apples to apples.
  • Homework for everyone: (1) register for the workshop, and (2) send preliminary results to Johan by Oct 25!
• 09/13/2021 (telecon):
  • David Gonzalez, Dan Garmann and Chris Rumsey informed us that AIAA is very committed to running Scitech 2022 as an in-person event. We agreed to move forward on that assumption until proven otherwise, meaning that we are expecting an in-person workshop in January 2022. AIAA is working to provide A/V access in the rooms to ensure that we can have virtual participants as well. More news on this front expected within the next few weeks.
  • We discussed the tentative agenda organized around 90-minute blocks for each of the main technical groupings (computational parameters, wall-models, numerical methods). Everybody thought the agenda was wonderful — or, at least, nobody objected to it.
  • Discussed possible ways to publish the workshop data/findings afterwards to make sure that it lives on. One option is an AIAA conference paper (Aviation 2022 or Scitech 2023), another is to place the material on the website. These are not mutually exclusive.
  • Mike Park asked to what extent we have a verification case for finding implementation issues. The argument is that the flat plate case will basically serve as such a case.
  • Timofey Mukha showed preliminary results for the flat plate case for grid levels 2-5. He raised the issue of how to compute the boundary layer thickness in a robust/accurate way, both for delta99 (what to take as the freestream velocity?) and the momentum thickness theta (really should use the wall-model solution below the matching point). He showed somewhat poor agreement of U(y/delta) between WMLES and DNS. Will plot U+ vs log(y+) and also U(y/delta) at different streamwise stations. He finally showed oscillations in the V component near the wall and wondered whether others have seen that.
  • The extrude.f90 script seems not to work for the finest mesh, most likely due to a Fortran issue with writing large records (actually, the script works, but the ASCII output format could not be read by some other software). Dan Garmann will fix it.
• 08/02/2021 (telecon):
  • Ivan Bermejo-Moreno showed preliminary results from runs on the community grids, levels 3-6 (i.e., the 4 coarsest levels). Despite going to 24 cells per boundary layer thickness, the results are clearly not converged. The message is that participants will likely require finer grids than anticipated.
  • Discussion about how long initial transient must be discarded and how long the results need to be averaged over. Participants should plot key quantities (e.g., reattachment location, velocity at some point) vs time from initialization to help judge these things.
  • Discussion about where to compare data along vertical profiles. Agreement that we should consider a few locations throughout the bubble, say x/L = 0.5, 0.75, 1.0, 1.25, 1.5 maybe.
  • Don Rizzetta spoke briefly about their WRLES runs. The finest grid has over 1 billion points and is showing convergence. Their runs are run for rather short times, but initialized from an interpolated instantaneous solution.
  • Participants were encouraged to start running as soon as possible. Preliminary data collection will start towards the end of October, although they will be able to submit revised data closer to the actual workshop.
• 07/05/2021 — CANCELLED DUE TO US HOLIDAY.
• 06/07/2021 (telecon):
  • In order to facilitate the formation of groups/connections with matching interests, please make an entry in this “interest map“. You can enter your group in multiple columns. In future monthly telecons we will create breakout rooms to facilitate discussions about specific interest areas.
  • We agreed to define the boundary layer thickness at the reference location as the place where the mean streamwise velocity is 99% of the upstream freestream velocity.
  • We agreed to postpone any decision on how participants should attempt to match the inflow boundary layer thickness until more groups have started running cases.
  • We current have 17 groups that have entered their details in the participant “survey”. Four groups introduced themselves during the telecon. Keith Walters and Wayne Strasser (Oklahoma/Liberty) will use Nek5000 to explore how their wall-models previously tested in nuclear-relevant flows perform here. Stephen Guzik and Xinfeng Gao (Colorade State) will use a high-order finite volume code with AMR, they are mainly interested in how numerics affect the predictions and how AMR can be used. Pinaki Pal (Argonne) will use Nek5000 and NekRS to explore data-driven wall-models. ZJ Wang (Kansas) will use their high-order flux reconstruction code; ZJ is also creating high-order grids and want to collaborate with other participants who use FEM-type codes.
• 05/03/2021 (telecon):
  • Clarified that the focus of the workshop is on the medium Reynolds number (Re=1M).
  • Don Rizzetta gave an update on the status of the reference WRLES: they have computed 3 grids for the low Reynolds number, and will then move to the medium Re. Don has tuned the inflow length for each grid but running on the flat plate, choosing the inflow length that produces the desired boundary layer thickness, and then using that length in the ramp domain; however, this has produced varying boundary layer thicknesses between the grids in the ramp domain.
  • We will collect flow field data using the Plot3D format. Data will include:
    • 2D fields of mean u, v, w (w can be used to judge averaging convergence)
    • 2D fields of all 6 Rij components (u’w’ and v’w’ can be used to judge averaging convergence)
    • Mean tau_wall and p_wall along the wall
    • Modeling interface height (exchange location for wall-stress models, or interface in LES/RANS) along the wall
    • Time series of some quantity at some locations. Most likely locations near the middle of the boundary layer, at some streamwise locations
    • HOMEWORK: we all should think about what quantities and what locations to extract time series from. Also, should the time series be at constant time steps or do we allow variable steps?
  • We current have 17 groups that have entered their details in the participant “survey”. Two of these groups introduced themselves during the telecon. Joao Azevedo from IAE (Brazil) are implementing wall-modeling into the high-order Flexi code from Stuttgart. Aniruddhe Pradhan and Karthik Duraisamy will use a DG-code with both a traditional and slip-wall wall-model, and will explore data-driven techniques for model improvement.
• 04/05/2021 (telecon):
  • ZJ Wang is working with GridPro to create high-order meshes. Will hopefully be done by next meeting.
  • It was agreed that the wall-resolved LES (WRLES) solutions will be kept blind from all participants until the workshop. At the moment only the AFRL group of Garmann/Rizzetta has committed to running WRLES; if other groups decide to do this, we will discuss whether the WRLES groups should compare their solutions or not prior to the workshop.
  • We discussed the viscosity-temperature relationship. The case description will be updated to specify that the viscosity should ideally be taken as constant, but participants who can’t run with constant viscosity should alert this group so somebody can test how important this factor is (the suspicion is that it will not be very important given the low Mach number).
  • We discussed whether to normalize flow quantities by quantities taken from the solution itself (e.g., scaling velocities by the velocity in the incoming freestream). We agreed to not do this and to plot “raw” data.
  • We discussed what the process for submitting/comparing results should look like. Specifically, do participants compute quantities or extract profiles themselves or do they submit “full” flow fields and the organizing committee performs the extraction? The current plan is for participants to supply raw data fields and for the organizing committee to extract the relevant quantities/profiles. The file format for the flow fields will be somewhat strict. The Python scripts for extracting and plotting quantities will be posted online so participants can try themselves. We will need to be careful about derived quantities, including some boundary layer thickness definitions that require the wall-model velocity field, and vorticity and other fields that depend on the numerics of the code.
  • We current have 16 groups that have entered their details in the participant “survey”. Three of these groups introduced themselves during the telecon. The team of Adrian Lozano-Duran, Jane Bae and Konrad Goc (MIT/Caltech/Stanford) will focus on how errors behave with the grid (topology and size) and will explore how machine learning can improve wall-models. Johan Larsson and Ivan Bermejo-Moreno (UMD/USC) will focus on the effect of the grid and will compare structured adapted grids with the provided ones. Dan Garmann and Don Rizzetta (AFRL) will perform wall-resolved (WRLES) of the low and medium Reynolds numbers, thus provided reference data. They also hope to explore the effect of the domain size in both span and the downstream direction.
• 03/01/2021 (telecon):
  • The second version of the grids have been uploaded to the Google drive. The README file should explain everything. The extrude.f90 Fortran code should be used to extrude the grids towards the chosen inlet location, the chosen outlet location, and in the spanwise direction.
  • Rob Baurle will update the extrude.f90 code to also produce grids suitable to the flat plate case, by simply removing all grid nodes for x > 0.
  • It was discussed that the grids will not be suitable for everyone. For example, high-order finite-difference codes may require smoother transitions between the different stretching regions, and high-order DG-type codes will require curved elements. Some participants have tools to produce curved meshes, and we encourage participants to collaborate to share such meshes. Participants can also apply smoothing to the meshes if needed. Note that the grids are supplied in a very easy-to-read format, so smoothing could be done in Matlab or Python prior to the 3D extrusion.
  • The Mach number should be 0.2 based on conditions in the freestream of the inflow.
  • It was suggested that we include two-point correlations in the spanwise direction in the list of quantities to compare; this has been added to the flowQuantities document on the Google drive.
  • We currently have 14 groups that have entered their details in the participant “survey”. Six of these groups introduced themselves during the telecon: Jan Nordstrom from LiU (Sweden; interested in the numerical stability of WMLES), Timofey Mukha from KTH (Sweden; Nek5000, effect of high-order code), Marilyn Smith from Georgia Tech, Christoph Brehm from Maryland (non-equilibrium effects; immersed boundary methods), Rozie Zangeneh from LTU (DES/IDDES; spanwise domain size), Cetin Kiris from the LAVA group at NASA Ames.
• 02/01/2021 (telecon and follow-up e-mail):
  • Participants are asked to fill out the “survey” at https://docs.google.com/document/d/1GUK5tBuqZXdGEbPe8NmFl56fS8gGVfPAmnYczWE8GnU/edit?usp=sharing. This information will be used to help plan the workshop (room size, agenda, …) and also to help participants team up to investigate issues of particular interest to them.
  • We will discuss what flow quantities to extract and compare during the next telecon in March.
  • The preliminary deadline for submitting results (in a format TBD) is October 2021.
  • Michael Adler and Johan Larsson shared some preliminary experiences from their preliminary simulations of the problem.
  • Rob Baurle has created smoother grids, will be finalized and uploaded to the website soon.

Google drive and case description

Grids and a description of the workshop cases can be found at https://drive.google.com/drive/folders/1Vi4UlbLemPHxCRq8_hFN0-ZjFLlJJLqm?usp=sharing.
Note that there are both regular and high-order grids, in two different directories.

The case description file is updated occasionally, with the full history given here:

• Version 5, updated 08/18/2021: data submission information
• Version 4, updated 05/19/2021: updated fluid properties, boundary layer edge definition, data format
• Version 3, updated 02/24/2021: version 2 of the grids
• Version 2, updated 02/16/2020: version 1 of the grids
• Version 1, created 09/10/2019: original version