4th gen aero

[Flinkly]

Holy cow. 1 day in and so many posts. Keep up the momentum!

 

I do design work and have access to NX and AUtoCAD. But this group already seems too fast for the time i could donate.

 

I do need to find the subaru testing around MPG and the outback's rear angle...

[boxkita]

Holy cow. 1 day in and so many posts. Keep up the momentum!

 

I do design work and have access to NX and AUtoCAD. But this group already seems too fast for the time i could donate.

 

I do need to find the subaru testing around MPG and the outback's rear angle...

 

so far, it's awfulwaffle doing all the work using a couple drawings off the web. Getting components modeled will take time. If you have skills, we can use them. I have a 2d autocaf class from 30 years ago, but going to give it try.

[awfulwaffle]

^^^yup, i think we'll need as many hands on deck as we can get for 3d stuff, it'll slow down real fast. Maybe we can divvy up components for everyone to model? Would just have to define interfaces between all the parts so they go together right.

 

...and don't expect me to be that productive all the time. I happened to be 2nd shift, on a Friday night, babysitting the startup of a bigger analysis at work si it could run all weekend, so I was able to justify the time

 

I'm good with trying to learn FreeCAD if everyone else is

Edited May 16, 2020 by awfulwaffle

[boxkita]

post #2 has a list of components. Getting them allows for swapping components in/out for testing.

[awfulwaffle]

Think while we're spooling up it makes sense to do a little bit of exploration in 2D. I'm thinking along the lines of:

 

-Model sedan and compare flowfield, downforce and drag vs the wagon @ the same conditions

 

-Add in a 2D representation of the hood scoop to one (or both) of the two 2D models. I'm thinking set a flow corresponding to utc_pyro's measurements, and have it vent underneath the vehicle roughly where it would in the actual car.

 

-Add a channel at the nose representing the grille and radiator. Where would be the best place to route this flow out?

 

- Add a representative cavity to the aft of the car to approximate what goes on between the rear subframe and bumper, on center plane.

 

All of this should be pretty low effort since the 2D model is just modifying a sketch.

 

Also gave it some thought, and does anyone think it would make sense to do some parametric study in 2D? Reason I ask is that I can slave the CFD solver to an optimizer code, and give the computer control of whatever parameters we think are relevant to adjust. Ride height, splitter length, spoiler height, etc, that sort of thing. The optimizer can then automatically come to a configuration that optimizes for downforce increase, drag reduction, or a combination of the two. This would be a big ask in the 3D model since each run would take a bit of time, and it takes a lot of work to build a model robust enough that it doesn't blow up as soon as the optimizer tries to change something. Pretty easy in 2D, though.

[boxkita]

parameter on 2d, yes.

 

the same pdf that had the wagon profile should have sedan.

[awfulwaffle]

Alright folks, here's a set of plots like before, this time for an N/A sedan.

 

It's looking like the lack of underbody geometry is having a pretty significant impact on the downforce predictions, and likely on the absolute value of the drag force value. I'm going to try something before I head out for the night, so standby on those values. Either way, the qualitative nature of the flowfield going over the top of the car should be representative.

 

Apples to apples, these two analyses seem to indicate that the sedan is the winner when it comes to drag, though.

[awfulwaffle]

Alright, after some refinement and turbulence model adjustments, the following ranges in drag force apply at 100 mph:

 

Sedan: 90 - 153 lb

Wagon: 170 - 191 lb

 

The significant increase in the range of predicted values for the sedan stem from the fact that its drag calculation is much more sensitive to the turbulence model's impact on prediction of the point where flow separates from the rear windshield. The 90 lb figure matches the flowfield above. However, adjusting how the transport and dissipation of turbulent kinetic energy is modeled in the region near the wall can shift the location where the flow separates as far as the roof. The wagon is less susceptible to this issue as the flow is pretty much guaranteed to separate somewhere on the rear 'wing'.

 

I don't usually work with such low speed flows, so I did a little bit of research in literature to see what's usually done in the automotive industry. It appears that the turbulence model used for the wagon and for the set of plots for the sedan provides robust predictions of the drag coefficient in these cases. I touched base with a coworker who used to do aero analysis at Chrysler and he confirmed that the same turbulence model was the workhorse during his time there.

 

It's quite mathy, but the following is a good reference that covers common turbulence modeling approaches as well as some results from comparing them on the case of a Hyundai Veloster, in case anyone is interested.

 

https://www.mdpi.com/2311-5521/4/3/148/pdf-vor

 

 

 

I think we oughtta hold off on lift predictions until we get some better resolution on the underbody of the car. Drag predictions should shift at the same time, but I suspect it will be to a lesser degree. Currently seeing lift predictions varying from about 100-1000 lbs.

Edited May 19, 2020 by awfulwaffle

[awfulwaffle]

Having some wool tuft data down the rear windshield of the sedan would be pretty useful.

[boxkita]

what speeds? just rear window or across trunk lid too?

[awfulwaffle]

Let me get back to you on that. I'll run a few different speeds tomorrow to see where the most conservative turbulence model stops predicting seperation right off of the roof. Hopefully, this will be somewhere below the legal speed limit.

 

Trunk lid probably wouldn't hurt to have either. If it turns out that the flow does follow the rear window most of the way down, we could get some idea where that separated zone near the window to trunk lid transition in the above pictures starts and ends.

 

I also have some other separation prediction tools that I can try and run on the predicted pressures on the rear windshield just for kicks, that would provide a much more confident answer as to where the separation occurs. We'll see how motivated I end up being after my shift tomorrow

Edited May 19, 2020 by awfulwaffle

[amptramp]

I am patiently waiting for someone to try dimpling the bodywork like a golf ball to achieve that last bit of drag reduction. Might be messy on a roof after a rain but someone has to do it. There have been experiments with dimpling the inside of intake manifolds and intake runners on heads that show about a 1% increase in flow.

[Chocoholic005]

There have been experiments with dimpling the inside of intake manifolds and intake runners on heads that show about a 1% increase in flow.

 

This is why high end factory manifolds aren't a smooth surface. If you were to go in and "polish" the surface, you'd decrease flow. One of those typically misunderstood "modifications" people make.

 

I had an engine airflow class in college; but i'm wholly useless when it comes to exterior flow characteristics. That being said, I might be able to pass some questions by the wind tunnel guys if we get that far.

Edited May 19, 2020 by Chocoholic005

[awfulwaffle]

^^ what he said. Some localized turbulence ends up being a good thing because it helps pipe some energy into the 'lazy' portion of the boundary layer, and keeps it from popping off of the surface for a while longer than it would otherwise.

 

As a corollary - if we do confirm that the sedan has the flow separate somewhere near the top of the roof, it would mean that vortex generators ala Evo 9 would likely be effective.

 

 

Also, just remembered that I have a 1D tool stashed away that can reliably predict separation from the pressure field along simple surfaces like the rear window. I'll see what that says as well.

Edited May 19, 2020 by awfulwaffle

[boxkita]

still reading the veloster paper. The 3000 core-hour vs 50,000 core-hour comparison really puts a damper on doing a full 3d model test.

 

My original plan was to get a 3d printer and make a wind tunnel with an ardiuno load cell setup to test lift/drag.

 

re:golf ball dimples, the spec miata racers collected wrinkled body parts as they more easily matched their aero needs than the oem bits. As a beneficiary of this, my miata had ram air from a crunched left corner. The "fix" created an airbox with cold air intake from the damaged headlight cover flap.

 

Dimples in the intake runners? Thought got disproven with newer versions of Nascar. Certainly the 3d printed/cf air intakes in F1 are smooth.

[awfulwaffle]

I wouldn't count the 3D run out. Our company policy at work states that use of company resources for personal projects is OK, as long as it doesn't interfere with regular business activities. While. LES/DES sim is questionable, there's no reason we couldn't do a full 3D RANS run in a day on a weekend when the cluster isn't too loaded up. I'll triple check to be sure.

 

On the subject of rough vs smooth surfaces and their aero benefits - a smoother surface will flow better than the rough surface as long as there isn't a sufficient static pressure gradient against the direction of flow anywhere in the passage to cause the flow to separate. So, think diverging walls in a diffuser or a turn in the pipe. Where dimpling or other passive boundary layer control schemes will work better than just a bare smooth wall is when you need to do a lot of diffusion or turning in a really tight space. For example, vortex generators wouldn't do a thing on the roof of a car if the slope of the rear windshield was gradual enough that the static pressure along it rose very slowly as the flow decelerated.

 

There are also active BL control schemes that can be used with a smooth wall passage to give them more margin before the flow separates. I bet the trick to the NASCAR/F1 findings you mentioned is clever design to control the adverse pressure gradients and turbulence in the intakes.

Edited May 19, 2020 by awfulwaffle

[awfulwaffle]

Here's a really good review paper that I found a while back on the subject of passive boundary layer control/vortex generators. It has some application examples towards the end too.

 

http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.462.2962&rep=rep1&type=pdf

[awfulwaffle]

Spent a bit of time today dusting off an old Fortran code that is used to accurately predict flow separation off of surfaces without too much curvature, such as the rear windshield of the LGT. Tested on a very coarse sampling of the pressures from the CFD run from a few posts ago, and the result supports what's reported there - the flow follows the rear windshield most of the way down. I'll get a finer sampling and try again tomorrow to confirm, but looks like the NA sedan is the clear winner in terms of drag (at least in a simplified 2D sense).

[awfulwaffle]

Took a first cut at getting something sort of representative for the engine bay and the cavity under the rear of the wagon.

 

The red boundary at the radiator will be defined as porous w/ a certain restriction, so it will naturally pull in flow. The boundary going into the scoop I'm thinking will use a mass flow value derived from utc_pyro's data, and the same amount of flow will be piped into the bayvia the boundary directly beneath it.

 

Thoughts?

[boxkita]

does the shape of the trailing edge of the splitter stay as a square? or just the roughness of the drawing due to low pixel count?

 

The cavity at the rear ought to produce a nice tail wave. looking forward to seeing that graph.

[awfulwaffle]

Not sure I see the area you mean. Maybe mark up the image? I can reshape whatever.

[boxkita]

talk about killing a thread. Ignore me. Im an idiot. Keep doing whatever you were doing.

[awfulwaffle]

Nah, I just had to do some actual work at the ends of the last few shifts instead of babysitting an automated process

[awfulwaffle]

Womp womp, 2D solve says the flow just skips right over the aft cavity. I don' think this is actually quite the case, we'll have to see when we get better resolution underneath the car.

 

For this run, I backed out a scoop mass flow from utc_pyro's data extrapolated out to 100 mph, and defined the radiator as a porous interface with 50% pressure drop to start. Encountered a bug where I wasn't able to select some of the surfaces along the body of the car, so no useful lift or drag values to report. I'll see if I can fix that next time I work on this.

 

I think it's safe to say that the scoop has a detrimental aero effect, which makes sense. I did only rough in the shape, though, so will take a second look to make sure it's not overstated/protruding too far.

[awfulwaffle]

Thoughts on improvements that can be made to the 2D analysis? Anything else that we want to explore?

 

If someone wants to get me a shot of some of Madrig's bits in profile, could add those in and see if they make a meaningful impact to lift/drag in 2D-land.