Duke University Motorsports is a student group that designs and builds open wheel, single seat race cars to compete in the Formula SAE competition sponsored by the Society of Automotive Engineers. The team consists of Duke students from both Pratt and Trinity, in all classes. The purpose of the team is to provide students with a way to gain practical design and manufacturing experience in a fun and challenging setting.

Wednesday, July 27, 2011

Aero and CFD

Having spent a good bit of time in the GM Aero Lab this summer, I understand how important a wind tunnel is to developing a good aerodynamic package. But, as a FSAE team on a budget, that's not something that we have available to us at this stage of the vehicle's development. So, our aero package is designed with the tools we do have available, namely computational fluid dynamics (CFD).

CFD is extremely useful in that you can easily visualize flow, something that is a bit more difficult in the wind tunnel, and you can get numbers without having a physical model.  However, it takes a long time to run (and thus take a long time to see the effects of each change) and it's not very good for absolute numbers, especially with the limited computing power I have on hand.  But where CFD can be extremely useful is relative results, in a sense optimization- as long as your CFD setup doesn't change.

RP model in the wind tunnel
I don't have a physical model or a rolling road wind tunnel (which I think is a must for an open wheel vehicle with 1" of full scale ground clearance), and so we're not going to talk absolute numbers at this point.  The aero senior design team last year did do some scale model tunnel testing last year (15% scale) in the Duke wind tunnel, and the numbers actually matched up pretty well (within 5%!) to what we predicted in CFD.  We didn't apply a tare correction to the wind tunnel data for the sting, but the full setup was modeled in CFD, including the sting, so it should be a pretty good representation of what's actually happening in the tunnel.  Of course, they missed the whole point of "ground" vehicle aerodynamics when they decided not to have a ground plane, but at least it's nice to know that our CFD isn't spitting complete BS back at us, but that was with a simplified model.  I'm not sure it would do so well with a high-res model, especially when it comes to drag.

CAD representation of the wind tunnel model
We use Solidworks Flow Simulation for CFD.  Why?  It integrates well into Solidworks (which is our CAD package), it's extremely easy to use, and I think it's accurate enough for our purposes.  The downsides?  Because it's so easy to use, it's really quite limited.  You have no way of viewing the surface mesh (or solid mesh, which I think is the case with SW Flow), the volume mesh, or any local meshes, so you really do have to trust that the program did it correctly.  If something goes wrong with the mesh, there's no way to see what went wrong, and more importantly, no way to fix it.  I do like the adaptive refinement (that is, if you trust that it's actually refining the right cells...), since it lets you start off with a lot fewer cells in the initial mesh.  Compared to Fluent, it's a lot less process oriented and a lot simpler to get started.  It still doesn't change the fact that you have to understand what the program is doing to get good results, which I think is a bit worrisome since SW Flow Simulation hides all these elements from you.  Garbage in, garbage out...

Maybe we'll move to another package someday, but for this year, I think I'm more or less done with aero design.

One issue I'm having is model size.  On the simplified full car, I can't really run more than a few million cells (and that's after refinement) without running out of RAM.  I need a dedicated CFD machine...  at one point this summer I had 5 computers cranking out CFD simulations 24/7 just to get through the simulations faster.  One thing I will probably end up doing is a simulating a half car instead of a full car.

The way I model the car is on a moving ground plane with rotating wheels (at least a rotating reference boundary condition on the wheels).  I've done my tests at a range of speeds up to 60mph and found the results to be more or less Reynolds number insensitive.

Eventually we'll have to translate these CFD results to the real world.  I tried setting aero balance in CFD, but we'll have to see how well that translates to real world. Luckily, that's something that's relatively easily measured through suspension travel, so I'm not too worried about it.  I have a little more faith in the downforce numbers from CFD since the body itself isn't going to be a huge generator of lift/downforce - the wings and undertray will dominate the total downforce, and those aren't too geometrically complicated for CFD to give us reasonable results.

No comments:

Post a Comment