[wouterremmerie]

Dear CFD users,

I'm looking into the possibility to create an online wind tunnel, where users can upload their 3D models and have the flow around the object and the respective forces calculated. No CFD knowledge required, a tool to get a quick first feedback on airflow.

There are already a number of downloadable software packages available for a virtual wind tunnel, but my assumption is that an online version of such a tool could have a number of advantages:
- The hardware of the customer is not used (calculations performed in the background on the server)
- No threshold to install and buy software: simple pay-per-case/upload system
- Platform independant (windows, linux, ...), providing access for any designer, architect, engineer, ...
- Significant potential to grow an online community, as all calculations are on the same server and experience can be exchanged.
- ...?


With this post, I am hoping to get your feedback on whether you agree or disagree and if you have other aspects that I could or should consider.
I'm looking forward to a vivid discussion!


Thanks,

Wouter.

[wyldckat]

Greetings Wouter,

You might be interested in contacting julien.decharentenay, because he's already done some work in this and is still working on it:

• http://www.cfd-online.com/Forums/mai...challenge.html
• http://www.khamsinvirtualracecarchal.../one-click-cfd - read this page in detail and you'll see why I'm referring you to this page

There are already some other services similar to what you're proposing, but they probably aren't as simple to use as you're proposing.

Best regards,
Bruno

[Artur]

Hi Wouter and Bruno,

As I said to W. in our private correspondence, I really like your idea and I think it could really make a big difference.

I've been helping a student recently who's working on using OpenFOAM on Microsoft Cloud (http://www.microsoft.com/en-gb/enter...id=nJzu8Zyzo5H). He's been getting very promising results in terms of performance, data transfer rates, etc. I imagine that for very big meshes with lots of communications this will most likely become inefficient more quickly than a typical cluster but on the other hand it's much more easily accessible and could be paid for by the users themselves. Many other companies offer similar services, like Google, for instance.

I imagine that while the BCs, schemes and solver settings may be "relatively" easily set for most engineering applications of the sort I imagine will be most frequently used, the meshing process may be a significant issue. I do have some experience in automated grid generation for airfoils which I'd be happy to contribute but that's probably just the tip of the iceberg.

I'll do some more reading and get back to you with more thoughts, would be happy to hear some of yours w.r.t. what I've written above.

All the best,

A

[wouterremmerie]

Hi Wyldckat & Artur,

thanks for your replies!
Recently, I found a program developed by Autodesk, called Flow Design:
http://www.autodesk.com/products/flow-design/overview
It is impressive, it calculates the flow nearly instantly (check the video's!).
I can't figure out how it does so (which solver does it use? how accurate is it?).

I pitched the online wind tunnel idea to "startup support" community in Belgium and they approved the idea. I guess now the goal is to find out how I can differentiate it from the Flow Design software, mainly through the online aspect (I can still see quite some benefits to that aspect).

Can you think of any unique points for an online wind tunnel? Online platform with result postings, higher accuracy/calculating power, things like that? Any input would be great!

I'm looking forward to your response! (and apolagies for my own late reply!)

Thanks again

Wouter.

[Artur]

Hi,

I remember thinking about this thread sometime last week, I was wondering if your initiative was still persisting, glad to see it is

I also remember having an identical discussion about the SolidWorks-related flow solver. I guess if it would be so easy to make simulations like they show in such a short time scale then probably 3/4 of people involved in this community would be out of work, StarCCM, Fluent, etc. developers included, but that's just my opinion. Point being, I wouldn't trust the results from that solver unless I would be presented some in-depth validation.

Regarding the potential advantages of what you are proposing, here are some of my thoughts in random order
- easy access from all over the world, making long-distance collaboration easier
- access to HPC to people and companies otherwise unable to afford it
- pay for the resources only when you need them, no unnecesary upkeep during times when CFD is not needed in the design cycle (I guess this only applies in some cases but I did an internship at a company where this was quite an issue)
- OpenFOAM as a baseline tool is quite well validated and mature so (most) people will be confident in the results produced
- if done cleverly there could be plenty of room for customisation of the cases, post-processing, etc. building on the already existing code
- the meshing tools are reasonably mature now so should be OK in most applications, eabling novice users to get decent results
- with cloud computing the timescale from case submission to post-processed results could be very short
- even more importantly, with distributed computing one could get results for LOTS of cases reasonably quickly, which on many tranditional clusters won't be possible due to queue limitations (I'm looking at you, Iridis); this would make it an ideal tool for quick design evaluation, allowing that info to be fed into the design process during the same day it was realised it is needed

I suppose the last point is really the key, but maybe I am a bit biased on the design optimisation side of things. I'm sure there are plenty more advantages to this, I guess there will be some disadvantages as well. I'll spend some time thinking about those since nothing relevant comes to mind at the moment.

Peace,

A

[wouterremmerie]

Hi Artur & Wyldcat & other forum members,

it's been a while!
A first prototype has been built, have a look at www.airshaper.com
You should be able to create an account and upload an STL file.
Some stl files cause problems, still not sure what it is, but the more cases we have, the better!

Mesh is very coarse for now, but can be scaled up with a single command, once we move to a decent server.

I would very much appreciate your comments on all possible aspects:
- concept
- look & feel
- information that you get (limited for now, will improve a lot)
- tips for improvement
- ...

Thanks a lot in advance!

Best regards,

Wouter.

[wyldckat]

Originally Posted by wouterremmerie

Some stl files cause problems, still not sure what it is, but the more cases we have, the better!

Quick suggestion: Public issue/bug tracker, so that people can help with that

edit: It will also give you a platform for the suggestions and so on

[Artur]

Hi,

Sorry it took me so long to answer, been pretty preoccupied recently.

Overall I'm very impressed, I think this is definitely going in what I'd say is the right way. Here are some more specific comments:
- I like the overall design of the website and the fact that one may browse through similar things by category
- It's pretty neat how the basic results are easily accessible
- I think it would be good to have some measure of mesh quality, like the percentage of cells extruded in the layers phase of snappy, etc.
- maybe add a slider or something which would allow background mesh size to be adjusted by the user with finest levels accessible to premium users (at the moment you seem to have only two levels)
- Not sure about how difficult it'd be to achieve, but perhaps allowing a vtk file of the results to be downloaded would be a nice feature? (I suppose you could make that a premium to cover storage costs etc.)
- I suppose having an option to run a 2D simulation would also be beneficial; making 2D grids with snappy is a bit of an overkill but can be done with the extrudePatch utility
- It's worrying how the automobile wheel from the "Hello" case gives 2e6 N drag at 11 ms-1...

All the best and please do keep me posted on the progress, fingers crossed!

A

[wouterremmerie]

Hi Artur,

thanks a lot for your feedback!
I'll definately integrate something that will give an indication of the applied mesh quality.

As for the high forces on the tire: likely the scale factor was wrong (meters versus milimeters).

again, thanks for the feedback

Regards

Wouter.

[rupole1185]

There are already a few on line actually. I know conself.com, simulationhub, simscale.com. they all are CFD on cloud solutions so not really what you were asking