SimScale CAE Forum

Optimizing mesh quality and Y+ layer formation on an FSAE car

A way to use nut/nu to look at boundary layer encapsulation (I still haven’t gotten my nut/nu plots to look anything like theirs :frowning: ). But since nu is a constant, even the nut plots should look like their Eddy Viscosity Ratio (nut/nu) plots with only a scaling factor difference…

Thanks for the article! I have uploaded a full car model and am now running the mesh using your methods. Here is the link. I have used 1 BL cell for all geometries at 3.4mm and level 9 surface refinements for wings / suspension parts and level 8 for the mono and tires. Hopefully the region refinement areas will not have too many cells.

To combat a potential memory overload problem, if the total cell count is close to 30 million, perhaps your ORSI can help determine the exact iteration zone the sim should run to. Then, I can reduce the iterations from lets say 1000 this run to whatever the ORSI says (hopefully less then 1000).

Like you said it would be great if Simscale could automatically stop the simulation when everything is in the 1% fluctuation range. Would it also work, If I am making successive runs of the same geometry, and know when the simulation should be at 1% to just set the number if iterations to this level so that I dont have to monitor it?

Yes, you have a good understanding now…

In effect once you have determined a 1% ORSI iteration for a geometry (or very close geometry) and a mesh and a sim setup, I have not found the 1% iteration to change much… Of course, any final runs should be manually stopped at 1% indicated (not the educated guess of the 1% iteration)…

Don’t worry too much about your core hours depleting, I will make sure you have enough for this amazing project… I am also pretty sure, if you ask @jousefm nicely, that you could get some more core hours, heck I think we NEED you as a PowerUser… :slight_smile:

2 Likes

Haha those are kind words. Im not sure im at the Power user level but i already have an agreement with Jousef about core hours because im technically sponsored with Simscale for my Masters thesis. I dont plan on needing an excessive amount because my plan is to have 3 simulations to match the 3 skid pad sizes. Then maybe a bit of optimization to change / verify / improve the center of pressure for the car. The entire goal of this project is to improve our cornering speeds through controlling the center of pressure location, which is basically done through increasing or decreasing the down force levels of front or rear wing. This allows the force exerted on the tires to be even on both axles ( if this is what we actually need - only the experimental results will confirm that)

2 Likes

Great!

I just updated my turbulence plots, click here to have a look :sunglasses:

@dschroeder, the offer is active! Good that Dale mentions it as I wanted to contact you anyway :wink: If you have the time and like to tinker around (like Dale & Andrzej do) we would be happy to have you as a PowerUser!

More information here: The PowerUser Program

All the best!

Jousef

Yes lets take a look at this Art.

Fist of all i notice two re-circulatory zones in the 3 layer mesh. This begs the questions.

  1. Is this here because if the 3 layer Approach?
  2. Is this here because the iteration shown could be stopped at a different spot of the fluctuation curve? (upper limit of the percent away form ideal ORSI values vs lower limit)

The second observation is that the 3 layer mesh has the BL cells inside of these re-circulatory regions (shown by arrow) Again questions:

  1. Does BL cells in this area give inaccurate results?
  2. Would BL cells in this area give better results. I ask this because this larger BL encapsulates more area transitioning from the log-law turbulent region into the free stream velocity, especially in the more important Low pressure side (shown in green arrows)

1 Layer

image

3 Layer

@jousefm I would be very interested in becoming a PowerUser! I definitely plan on continuing using Simscale for my other projects after my thesis is completed as i am very interested in learning more about CFD. As you may have seen I am currently asking more questions then answering, but over time i hope to change that.

1 Like

Even in steady state simulation the flow is ‘alive’. Re-circulation will change shape and transit from two to one seed and then back. On my simulations I usually set a series of snapshots and when playing with E423 airfoil and it’s modifications I’ve seen such a transition systematically. I believe it should be considered normal with strongly cambered shapes… So not a BL 1 or 3 layers are the culprit, it is a nature of flow.

Cheers,

Retsam

Ah that is extremely useful to know! Its interesting how such data can be misconstrued as evidence for something that is or isnt actually there. Seeing these differences could potentially alter the design flow if these results were accurate. I guess this is why experiments take data from a minimum of three trials, so that results are validated.

Now would i be correct in assuming that if we did a couple of sim runs for each, and they were still different, that we could use this as evidence. A better method might be to take data from time steps at 925, 950, 975 and 1000 iterations and compare those results together.

How are you getting full datasets over 1 period of oscillations (say 50 iterations worth of fulls sets at iteration 1100) at the 1% stable ORSI stage of a sim run :question: I did not think we can do that in SimScale, I think we may be able to do it using OpenFoam, which I am investigating right now…

My gut experience agrees with Retsam on this one, #2 is my guess…

I hope you have noticed that I have not filled in my conclusions yet :wink:

1 Like

The surface LIC integrator in Paraview is amazing at what it does, I LOVE it…
But we have to understand that it is interpolating those nice fine flow lines…

  1. I don’t think the solvers use different formulas for prism cells, they are just better suited in BL layers than Hex, so I do not think the question is fair…
  2. Good question, in turbulent/direct hit surface areas between eddies I find it hard to believe that the BL cells even matter… and then in your low pressure ‘lower side of the vortex’ the flow is generating ‘thrust’ by the shear drag :wink:

In any case until we have full sets of data over one or two periods of the oscillating results at the 1% stable ORSI stage of the sim run, it is all just educated conjecture :wink:

EDIT: Maybe I should try the 1 vs 3 layer comparison on the fully attached flow on NACA0012 at 10AOA :rofl: (very small results oscillations in that case)

Actually, I see 4 recirculating zones in the 1 layer mesh and 5 in the 3 layer mesh but I do not think this distinction matters at this point…

And for completeness, here are the 1 and 3 layer P plots:
.


.

BINGO ! :rainbow::sunny::beers::sunglasses:

Look what I have, 1 and 3 layer plots where the solution sets are shown at their final iteration AND the solver log result values at the final iteration MATCH the ORSI value at that iteration (well at least within a measly 0.2% to 0.8 % of the ORSI value).

This means that the LIC flow patterns and all of the various parameter color plots are representative of the ORSI values, this is a huge advance :slight_smile:

In fact, if there were ever an automatic simulation ‘Stop at a specified ORSI’ feature developed for the SimScale UI, the code for that can easily be written to stop when the solution set results match the requested ORSI :slight_smile:
.


.

2 Likes

amazing, i will look deeper into this after work!

Here is my finished mesh of the full car. feel free to run your own sim, but as soon as im free i will be doing one. I noticed some small CAD changes i want to make but overall the 1 layer method looks solid!!! Im so pumped to see the Y+ Histogram results and i am also looking forward to trying your ORSI program for the first time. This sim run can be a benchmark for when to have the final iteration.
Also more good things

11 illegal cells!!!
14.6 million cells - this can be easily reworked to about 10 by reducing the high quality mesh box behind the rear wing!
Mesh looks great!

All credit for this goes to you Dale!

1 Like

I am afraid my latest plots are bad news for 1 layer meshing though…

If we look at the ORSI CD difference between 1 and 3 layer meshes, the ORSI CD of the 1 layer mesh is only 0.8% lower than the 3 layer ORSI CD, which is sortof OK…

BUT, the ORSI CL difference shows that the 1 Layer mesh ORSI CL is a larger magnitude by 7.1%, this is NOT OK…

And in the U plot we can postulate that the higher speed air on the high pressure side of the airfoil is the cause of the significantly higher ORSI CL magnitude for the 1 layer mesh…

We can further postulate that this increased speed of the air, which is significantly inside the outer 2 layers of the 3 layer mesh, was ‘allowed’ to speed up because the solvers were not treating those HEX cells with same accuracy with regard to wall function calculations as the 3 layer prisms cells in that area were treated…

In effect, the air in the 1 layer mesh was allowed to speed up there because the air did not know it was so close to a surface that should have slowed it down by shear forces…

I can test these postulations by running a few more sims, with 0 layers to see if the air speeds up more in that area, and with 5 layers and 7 layers to see if we reach a point of diminishing returns (perhaps at the same expansion ratio as the 3 layers)…

I have some work to do :wink:

Very impressive geometry file… Thanks for using STEP file…

That is a LOT of wings :slight_smile:

I am not sure the 1 layer mesh is your best option but it is for sure worth a sim run or two :wink:

I think maybe you should investigate splitting your wing surfaces where they get too close together to use 3 ‘wall function’ layers… On these faces that are now only near the tight areas, simply layer them to y+ 1 and then use full resolution wall functions on those faces when the sim is run… At 15M cells you can add quite a few at y+1 in those areas and still stay under or near 20M max cells… (I have never tried using wall and full res layering like that but it may work :wink: )

I have a few other jobs for today and if I get time I will likely work on the 0,1,3,5,7 layer study…

I assume that you are taking these results from the numerical data of the Cd and Cl plots where both sims are in the 1% zone using the ORSI. I am not entirely convinced that the difference is not ok due to the lack of a “control” sim. We do not know what the Cd or Cl should be for this profile and we are comparing results based off of a lot of variables. Building on this I have different views on what you say here:

If we are basing assumptions on visual data, then i would like to point out the following. You have made some nice pictures of each wing comparison between the 1 and 3 layer sims where both wing profiles are in the exact same place. This allows us to compare everything much easier. Taking the U magnitude scale at the top of the picture and using it as an unofficial marker, I have placed arrows at the point in each sim where the air separates from the low pressure side of the wing. As you can see the point of separation for the 1 layer sim is at 14, while the 3 layer sim is at 13. From what i know about aerodynamics, is that the low pressure side contributes more to overall lift (in our case downforce) then the high pressure side. This difference in separation points is what i would attribute most to the increase in downforce from the 1 layer sim. HOWEVER!! based on what Retsam mentioned about flow being “alive” these assumptions from the visual data are not valid and everything we have thought could be wrong

I do not know much at all about how the solvers treat HEX vs BL cells. I would agree that this could be partially if not entirely the reason why we have a difference in CL. I hope that your new method of validating this will work and will help to clarify how accurate each result is.

1 Like