Those designing their own high MPGe cars.

[smilingcat]

Are you using any CFD software? and have you tested your model in a windtunnel at even at 1/10 or even at 1/12 scale.

My partner is trying to talk me into designing a respectable electric assisted velomobile. The speed is limited by law but aerodynamic is still a significant portion of such vehicles.

I've been meaning to replace my aging computer with something more up to date. Loaded it with Nvidia graphics card with Fermi GPUs. I wanted two or three of the GPU in my PC so that I could run CUDA software and run CFD programs and other simulation programs. Even though my background is in electrical engineering, I took classes in fluid mechanics so things like Reynolds number, shear forces, laminar flow, boundary layer, turbulent flow, and how to reduce wake turbulence are things I do understand.

I told her that if I want to do it seriously, I would need a reasonable wind tunnel with cross section of about 2'x2' in the test area. 30' long open ended system... Maybe this would shut her up. But seriously, have you thought of running CFD on a PC loaded with graphics GPU with CUDA and or run a model in a wind tunnel. You can run a very precise CFD with two or three GPU processors in a PC with about 12G of main memory and with a Terabyte HD. And the simulation would only take few hours.

[PatQ562]

I too am interested in this question. And, what are the rules for scale-model testing (ie, how do you scale the air speed)?

Pat Q

[smilingcat]

simple answer is 1:1 between air speed for the model and the real thing. The air velocity for yours and my wind tunnel is too low to make much of a difference so it is 1 to 1. At higher speed where you start to worry about compressibility of gas with adiabatic heating (transonic speed) there are corrections required. But you won't even begin to see those effect till you reach several hundred miles per hour.

[NeilBlanchard]

I looked into CFD, and it requires a solid model, so even though my SketchUp model could be converted, it was still tessellated. I have now seen *much* smoother models out of SketchUp: http://www.youtube.com/watch?v=LjtplWKJfK0

I also looked into putting a 1/4 scale model in the MIT Wright Brothers wind tunnel, which is a 10' x 7' oval 2 atmosphere facility. It would have cost $1k plus set up fee for 4 hours. They need a testing program proposal, though I'm not sure whether this applied to me.

I tried using 5 (!!!) leaf blowers which can get the needed velocity (200+ MPH) for a quarter scale model, but the air flow is way too uneven to be very useful.

[smilingcat]

Common wind tunnel configuration today has a large inlet which tapers down to the cross sectional area of the test area. Then it has a whole bunch of slats/tubes to calm the turbulence so that you get laminar flow. It then leads into the test area then a expansion tube/diffuser and lastly the fan and an exit "cone" to match the "impedance" to the open air. This help with expelling the air into the still outside atmosphere.

The reason you place the fan behind the test area is just as what you noticed, is that the fan in front creates too much turbulence.

Hope this helps. 200MPH wind out of leaf blowers??

At $1,000, you should be able to build a reasonable wind tunnel. Not sure about the size of the test area. And do you really want to test something that large?

The thing that skews with automobile wind testing is the ground effect. I think most automobile wind testing now uses a conveyer belt underneath the body of the test vehicle to improve the test result from the ground effect. Air is moving, the ground is moving car stationary is equivalent to car moving in still air and stationary road way.

If you want some help with wind tunnel feel free to ask though I'm no engineering expert on the subject. I know lot about computer architecture and chip design but not mechanical engineering.

[palmer_md]

just buy one...

Originally Posted by old

http://hgpauction.auctionflex.com/sh...enum=1&lang=En

http://hgpauction.auctionflex.com/sh...enum=1&lang=En

[smilingcat]

My partner talked me into R&D for ultra light vehicle. Electric assisted velomobile with electric assist.

To this end, I am replacing my current Linux based computer with a new computer. Type of computer is called HPC.
Host computer is dual XEON E5649 six-core, with 48G DDR3 memory and RAID disk array. Another Linux machine.

GPU computer for now is Geforce 560Ti. It will be used to verify the applicability of CUDA software, CFD, magnetic simulation software. If it is verified, then Geforce 560 Ti will be replaced by Tesla 2050 "supercomputer."

Will be busy getting the new computer up and running with all the needed software.

one step at a time.

[dfbvt]

Hi Smiling Cat,

What are your goals for this investment And effort? Are you doing this just for your own use or are you intending on doing contractual work?

Dave Bowles

[smilingcat]

Hi Dave,

It will take time to get the software ported over to run on Tesla aka nVIDIA's Fermi GPU. C code needs considerable extension to take advantage of the massive parallelism afforded by the Fermi GPU. The extension is known as CUDA.

I have not really given much thought about doing contract work since it will be some time until everything I want is up and running cleanly. There is also a legal question regarding Gnu open source licensing. I'm not sure if I'm allowed to use it for contract work.

The good news for everyone is that when I do port it over, I will have the ported software available under the Gnu open source license. It will be free.

One step at a time

The porting of C code can be easy, not do much, or can be extensive overhaul. I need to see how bad it is going to be.

For now, I should be able to run the needed software on the host computer without the acceleration of supercomputer. It will run pretty fast even without the supercomputer. Host computer is very powerful in its own right. I can help you set up a high end enterprise grade workstation. Cost of such computers are around $4,000 to $5,000. My setup is costing in that range.

I did look at EC2 at amazon cloud computing. First year of use is free? but one needs to port software over then you are stuck with Amazon cloud computing. And when they start charging real money well. So I decided not to go that route.

if you don't know the terms check out on wiki: Gnu open license, CUDA, EC2

planned work flow.
1. get the computer up and running under Linux OS (Mandriva is my choice)
2. Get FEMM (FEA magnetic simulation) up and running without GPU supercomputer.
3. Get FreeCad or equivalent running
4. Get Salome mesher running
5. Get Elmer CFD simulator running.
6. Get CUDA installed and running.

when step 6 is completed I can start porting in the background.

First physical builds are the electric motor. Magnet design, coil design, rotor builds and stator builds. Current motor could use lot of improvement. To give you an example, GE has an electric motor 12 inches in diam. x 12 inches long and produce 250KW (340hp). Another reference point is an electric motor the size of a regular house brick produces 20hp.

It's a good thing I lost my job as a program manager/engineer two years ago and forced me into early unplanned retirement. It affords me the time to work on this kind of project full time.

my background:
I was a Research Staff Engineer on Strategic Defense Initiative program aka star wars program under Reagan

then I went to commercial work designing things called ASICs...

I will let you know when if and when I can do contract work.

[dfbvt]

smilingcat

Hi,

Well, it sounds like you got a lot of work and $$ to get it all working. I use Solid Works and I could always purchase their CFD add on which they recently ported into Solid Works. Desalt Systems bought FloXpress flow simulation software and integrated it with Solid Works and I guess Autodesk recently purchased Blue Ridge CFD which they integrated with Inventor (Blue Ridge quoted me 20K+ a few years ago) so I'm sure its not cheap either. So its not cheap no matter which way you cut it plus the fairly steep learning curve.

I guess for now I'll stick with playing around with the watered down CFD sim built into the basic Solid Works program and just use good rules of thumb. Maybe I'll build a 1/6 scale model and run it on top of my PU to see if I am on the right track.

Good luck with your project.

Thanks,

Dave Bowles