CFD analysis of Lightning McQueen

[Richard]

I guess the intention is to show scenario x is bigger/smaller than scenario y. Admittedly there is a lot of trust in black box systems and they could be wildly wrong, however sometimes things have to be taken on trust. Just like when I ask someone the time or for directions I trust they are reasonably competent to answer the question.

If I used CFD for my work I'd ask the questions you raise because my professional reputation relies on it. That's we employ doctorate level analysts. However that'd be excessive for amateurs following a hobby, the level of rigour has to be proportionate.

IMHO its great that amateurs can play with CFD, while also bearing in mind it is just playing compared to professional application of the tools. A bit like amateur astronomers peering into the sky are nowhere near the level of NASA or Stephen Hawkins, but we wouldn't say they shouldn't peer at the stars.

[KeithYoung]

I don't see the reason for what seems to be to be bordering on bashing.

Would I ride in a plane designed using just this service? No.

Would I use this service if I was deciding on a design for a splitter for my Miata I want to take to autocross? Yes.

How did we go from a cool video of a guy taking a Pixar car, scanning it to get a 3D image, and running CFD on it, to this? My inclination is to hope the guy gets an old Tyrell from Exoto, or mail him a clay/foam model I made, not bash what he's doing.

Did he use the right Near Wall Modeling for Lighting McQueen? Who cares. It's obvious this service has its unique advantages as well as its limitations. Stop making this thread negative.

[silente]

i agree with KeithYoung.

as en ex racer, i would not trust the absolute numbers out of this sim, but still i think it could be somehow useful, above all if the website owner replies to customer's question regarding method and boundaries...

[Spirits of Senna]

Hi Julien. I am going to take a Rapid Prototype class soon and the course has subjects on 3D scanning, I am playing to bring my own model cars to see if I am allow to scan and then play around in CFD.

Once you've scanned your model, did you use any CAD on the scanned model to modify? Can you share your experience on which CAD you use if you have done so?

[Greg Locock]

In a closed thread g-force addict asked "Can the average Joe get his model aero tested for drag and lift?
....
For wind tunnel do you just send the scale model and they test it for you or Do wind tunnels get rented by the hour and you have to test it yourself? I guess they provide the scales and instrumentation.
What wind tunnels are available for hire and How much do they typically charge?..."

A full size wind tunnel will cost hundreds or thousands of dollars per hour. Even a relatively small one suitable for 1:5 scale models will run a couple of hundred bucks an hour. A full size tunnel with a rolling road will have a staff of a dozen or so. You are paying for all of them.

Typically they'run the tunnel for you to do the experiments, but will give as much help as you need, up to and including running standard tests. Very few people would do it that way as the tunnel costs so much and there is nothing like witnessing the test as it is done, in general.

[julien.decharentenay]

Hi,

First, I would like to apologise for not replying earlier. I falsely assumed that I would be notified of response (but did not) - which is my mistake for not ticking the box.

Andy & Keith, thanks a lot for stepping in to "defend" my pet project.

I agree with the principle Garbage-In-Garbage-Out. The portal as is does not allow one access the inside of the black-box. It's aim is to minimise the risk of Garbage-In. Some may still get the speed and yaw angle wrong, but I can't do much about that? If the black-box does the right thing (big "if", isn't it?), then minimising the number of possible "garbage" input would minimise the number of "garbage" outputs. Please note: the portal is placing constraints on the geometry (orientation, units, location) - the portal try to check what it can but it is difficult to cover all situations, and you do get to download the openFoam results for viewing on ParaView.

3D scanning wise - as this was primarily the focus of the video:

As explained it was done using a set of photographs (about 30 to 40 photographs for the top and 30-40 photographs for the bottom). The photographs were used to generate a 3D shell model using 123D Catch (by Autodesk, it is free to download and use). Note: they have a iphone/ipad app, but prefer to use a proper camera and the desktop app. It is a great system, but:
- practice is important (my first 5 to 10 models were really bad);
- use a good camera and practice how you place the subject in the view. I found it worked best when the view is composed mainly of the object to scan;
- if scanning small object: place newspaper under the object. I don't fully know why, but it seems to work better;
- Last, but not least: lighting, lighting, lighting. This is probably one of the most important aspect.

Also: it does not scan transparent surfaces! I am thinking of trying some kind of powder to minimise reflection, but haven't had time to do anything about.

Accuracy of the scan: you get an overall shape, but don't raise your hopes too high. You get what you pay for (the same applies to my portal, as well: don't expect F1 CFD modelling accuracy for the price of a t-shirt).

Overall: it is a great toy, with good potential. If you are ok to powder your model and really work on the lighting, you would probably get something reasonably accurate. But it will take time to get the right setup.

From a practical perspective, I used the following list of software:

1) Take pictures;
2) Upload pictures to 123D Catch and download model (STL);
3) Load STL and texture in MeshLab to export in DAE format;
4) Load DAE top and bottom shells in SketchUp. Clean top and bottom shells - I removed the parts that were not needed but did not attempt to change the geometry itself) and align them. Alignment was done manually (ie manual rotation and translation and scaling), but I am considering trying to simplify the alignment aspect using a plugin. Let me know if interested as I would make it available for free.
5) Export the SketchUp model in DAE format;
6) Load the DAE file in MeshLab and apply a remesh filter (surface reconstruction: poisson);
7) And voila: a closed shell.

What are my next steps on the subject: I have been looking at other techniques (Microsoft Kinect based) to use in combination with the above. I have to buy a software license, and will apply it to another one of my kid toy, before trying it on R/C cars. If this goes well, I will then try to expand to a motorbike and then a kart. I will be producing videos and be more than happy to share experience - I have no desire to become a 3D scanning expert, but would love to see its availability more widespread. I think that the low-cost 3D scanning landscape is likely to change a bit in 2014 with the release of an upgrade Kinect sensor as well as low-cost (~ $1000) high resolution 3D camera.

Meshing, boundary conditions, etc: The detailed option is +/- what was done for the KVRC challenge. Happy to discuss specific questions. I think that most of the information is available on the website, but would be happy to clarify further. I should have (and will be) looking at running standard cases for validation purposes.

Julien

[julien.decharentenay]

That was a long post....

To a certain extend a commercial CFD software is also a black-box. It does take training and practice to get confidence with the results. I am happy to give reduction/free simulation in return of the following:

- Public comparison with other simulation tools on the same geometry - be prepared to share the results;
- Public feedback on the experience (good or bad) with the portal.

If you want to take up the offer, send me an email describing what you intend to do and a bit of background on yourself.

[gorich761]

I would see a potential to do that on real cars models, maybe accurate and big scale ones, something like what Suzuki has done with his Scale Wind Tunnel but in CFD...Could be interesting.

[silente]

i am totally qith Gorich...

This is what i meant, a virtual wind tunnel based on model scanning....would be at least curious to see the results

[julien.decharentenay]

To keep you updated. I just started working with a couple of guys doing remote control car racing. The plan is to :

- do a 3D scan of their car using a Xbox Kinect - it is fast and cheap but the scan is very noisy;
- run time test on two aero configuration on their track with (let's hope) section time measurement and wool tufting;
- correlate results (and driver impression) with CFD modelling.

The above will be qualitative more than quantitative. If anyone has idea on how to make it more quantitative, I would really appreciate.

[silente]

Julien,

how are you planning to correlate your results?

what i mean is, with track testing you are only measuring lap times, there is potentially very little connection with a pure downforce/drag number. Lap performance depends on a lot of other things that could also have an influence on car handling and driver impressions.

I guess the only way would be to somehow measure some loads, but i guess it is pretty impossible to do it precisely enough on such a scale model.

It could anyway be a starting point...

I don´t want to be negative, don't take me wrong. Simply from my experience i know that track test correlation is always very difficult, also when you have a lot of sensors measuring every small detail of car motion...

I would try to do something more basic somehow, but at the moment i don't have a better idea on how to measure loads without a wind tunnel. The only thing you could maybe do is a coastdown test, but still you would have only info about drag and with a lot of undesired things playing a role in your results (see mechanical friction that could potentially be very big in R/C cars).

[julien.decharentenay]

Silente,

Thanks for the feedback.

The testing protocol is likely to be an alteration of the aerodynamic of the same car (such as removing the back wing) during one testing session. Only one aspect will be changed (obviously you have other factors such as weather - it is an open-air track, tire, etc). The aim would be to run 5 to 10 laps to reduce variability of lap-time. There will be three runs: normal configuration, modified configuration, normal configuration. As a first step, we can assess if the modification has any effect on lap-time/driver impression.

Correlation between track and modelling: I agree with you. On my side this is an opportunity to learn through practical experience - the conclusion may well be that this is too hard. There are other aspects of interest, the R/C body shells are not very rigid and likely to be subject to deformation - but this is an unknown variable.

Wool-tufting will be easier to correlate - but obviously of limited information. It would be great to have a pressure sensitive paint - but I am unsure if they are available and cost...

Coast down: great idea. I will give it a try, but do not have access to velocity measurement (so will be difficult).

Wind tunnel: would be great, but I don't have one...

[silente]

i don't know about data logging systems for these cars. Depending on how big they are, i am pretty sure something exists (i have a colleague racing with small airplanes models that have much less space than an R/C car, even a 1/10 one and they have some data acquisition) but i don´t have any idea about costs.

But i guess, for a basic speed measurement, you could even use a smartphone with a gps system. I had a friend who once even turned a smartphone into an onboard camera!

Of course the measurement will not be 100% accurate, but better than nothing. If you have the possibility also to test on a very smooths surface with a long straight it would be even better (the first thing that comes to my mind is the athletic tracks that normally surround stadiums...)

Anyway it's a very interesting project! I wish you all the best!

[KeithYoung]

For speed, you can use a laser system with known distances between the sensors and time the duration. Simple, cheap, can be done with an Arduino.

Another option is put an optical sensor passing through a hubbed wheel in an upright (old mouse ball sensors), or use a Hall Effect Sensor. All possible on Arduino or Parallax Propeller.

Currently on site for a client so I can't look up the proper terminology for you at this time.

[ace37]

Hi,
...
Meshing, boundary conditions, etc: The detailed option is +/- what was done for the KVRC challenge. Happy to discuss specific questions. I think that most of the information is available on the website, but would be happy to clarify further. I should have (and will be) looking at running standard cases for validation purposes.

Julien

I think the GIGO commentary is reflective of just a few fundamental things that you could, to some extend, address with rather minor effort. Plus, while you may be sharing a fun idea you're excited about, you have a conflict of interests if you're also promoting your own product. Sometimes that makes folks irritable, right or not.

1. How do you define "good enough" for mesh and solution convergence criteria. Based on your coarse model generation technique (showing surface gaps at the underbody for instance), your answer may be simply that you are not rigorous at all in defining either. And if that's stated, that's fine, but your lift and drag predictions would be expected to be off by more than 100% compared to the real case, and it's not going to be a consistent error that can be corrected, so your numbers don't mean much. Also, the serious aero people will stop being interested at this point, and there's nothing wrong with that either. We know this is a $10-$40 answer, and that's not enough to buy an hour of anybody's time, so selling a nice CFD-based picture for $10-$40 is a fine enough business. But if that's the product, consider calling it something that implies a CFD-based picture rather than a CFD analysis.
2. If you have a reasonable criteria for (1), how do you model boundary conditions and so forth. Do you have a moving floor, how far does your mesh extend outward in all directions, etc. At a conceptual level, what is your black box assuming besides the specified parameters.
3. Show some honest typical results compared to public domain wind tunnel tests or validated CFD runs, and help people understand what they will be buying based on what they send you. Even NACA airfoils would be a start. Run a bunch of cases, and ideally, show us that you're always within a factor of 3. Because hey, that's an order of magnitude answer, and for $10-$40, that's something.

When you call this a CFD analysis, that generally implies it is validated, converged, etc. On the surface, this process doesn't appear to be robust enough to demonstrate that. That's why the comments sound negative - missed expectations. And if on the other hand this product provides an order of magnitude answer with real consistency - which it may do - put together some simple data to demonstrate that fact and then take full credit for everything it is and everything it isn't - the price is so low that it would be an interesting product to the right audience! After you've done this, you'll get a warmer reception, and you should still expect additional constructive criticism to help you further improve the product.

And, just in case this had a negative tone, the results you produced do look cool. Thanks for sharing.