F1technical.net

kilcoo316 wrote:

miqi23 wrote:CFDesign can be a good tool for you as a starter. CFDesign is like other CFD tools however designed for engineers and not CFD specialists. Basically it captures the flow domain in your geometry and meshes it automatically and later on solves for a solution. Post-processing is also done in CFDesign using its built in post-processor.

Hmmm.... dunno if I like the sound of that, I'd be weary of the accuracies of a mesh I had no control over.

Thats why I said the automatic mesher concerns me a lot when it comes to large models. I have tried it on simple models and the accuracy is quite good

What are the objectives of the CFD calculations people are running?

Without knowing the objective in performing a CFD calculation it is difficult to recommend one CFD package or technique over another.

For example if the objective is for an understanding of say diffusersor wings then a panel method is viable.

syguy wrote:What are the objectives of the CFD calculations people are running?

Without knowing the objective in performing a CFD calculation it is difficult to recommend one CFD package or technique over another.

For example if the objective is for an understanding of say diffusersor wings then a panel method is viable.

Not in a highly turbulent free-stream its not.

Nor in a situation where boundary seperation from the wing/diffuser is a distinct possibility.


There is no such thing as a free lunch when it comes to aerodynamics - what you gain on the swings, you lose on the roundabouts.

kilcoo316 wrote:Nor in a situation where boundary seperation from the wing/diffuser is a distinct possibility.

There is no such thing as a free lunch when it comes to aerodynamics - what you gain on the swings, you lose on the roundabouts.

Of course you are right. A panel method can't predict viscous effects, unless using a coupled boundary layer method which assumes full attached flow.

However, the likely hood that a RANS CFD field solver (such as Fluent, CFX, STAR-CD) can accurately predict separation from smooth surfaces (not sharp corners) on airfoils/diffusers within a full car simulation is low. In order to get close we'd have to use an extremely fine mesh normal to walls, coupled with a non-wall function based turbulence model. Such a mesh (of optimally placed cells) for a full car will likely approach over 10,000,000 cells which in Fluent will require upwards of 10GB RAM (most likely run in parallel). Also to really nail airfoil flows at such relatively low speeds (compared to aircraft) we'll need to model laminar to turbulent boundary layer transition, another notoriously difficult task.

Next up in our simulation toolbox is the wind tunnel, which has no trouble simulating full cars and boundary layer transitions. However, using a scale model means the Reynolds number won't match that of the full size car. Without matching the Reynolds number there is uncertainty as to whether the laminar/turbulent regions on the model will be equivalent to those on the full size car, which relates directly to downforce and drag predictions. Also wind tunnels have issues with blockage effects and stationary/rolling roads.

My point (and I think your point is too) is that no single form of simulation is perfect - each has strengths and weaknesses. F1 teams know this (the best ones do anyway) and use a variety of techniques from hand calculations, through panel methods to RANS CFD and wind tunnels. If we know our objectives (e.g. better understanding, comparison with other designs, optimization), then we are in a good position to select the most appropriate technique, in terms of turnaround time, accuracy and cost.

As you say there are no free lunches, but there are cheap and expensive lunches, knowing which to choose and when, is what separates engineers from apprentices.

10 million cells is a very conservative estimate. I'd reckon you'd be looking at well over 30 mil to do the job right (unless the solver is quite robust).



Well, Menter's K-W SST model has been shown to be pretty decent at capturing most detachments. (as for the laminar to turbulent transition - forget about it, in a highly turbulent freestream it trips as near to the LE as makes no difference).


You don't have to do low-Re modelling on all surfaces, just the wing suction surfaces and the diffuser suction surface. If your solver has an automatic trip (Fluent doesn't for KE, does for SA, and not sure about K-W) between wall-functs and low-Re, then you can mesh appropriately saving considerably on densities.


Bottom line - I wouldn't recommend panel methods for F1 car flow. There is not one part of the car design suited to it IMO.

Front wing has clean inflow, but the wake interaction with the suspension is crucial.

Rear wing has dirty airflow and also complex endplate vortex interactions, panel methods are uhmm.... yeah...

Diffuser... well, maybe, the flow should be decently clean underneath it (probably pretty crap above it though), and entrainment won't be captured.

kilcoo316 wrote:10 million cells is a very conservative estimate. I'd reckon you'd be looking at well over 30 mil to do the job right (unless the solver is quite robust).



Well, Menter's K-W SST model has been shown to be pretty decent at capturing most detachments. (as for the laminar to turbulent transition - forget about it, in a highly turbulent freestream it trips as near to the LE as makes no difference).


You don't have to do low-Re modelling on all surfaces, just the wing suction surfaces and the diffuser suction surface. If your solver has an automatic trip (Fluent doesn't for KE, does for SA, and not sure about K-W) between wall-functs and low-Re, then you can mesh appropriately saving considerably on densities.


Bottom line - I wouldn't recommend panel methods for F1 car flow. There is not one part of the car design suited to it IMO.

Front wing has clean inflow, but the wake interaction with the suspension is crucial.

Rear wing has dirty airflow and also complex endplate vortex interactions, panel methods are uhmm.... yeah...

Diffuser... well, maybe, the flow should be decently clean underneath it (probably pretty crap above it though), and entrainment won't be captured.

I agree with everything you've said. I think panel methods are very ill suited to open wheel Aero. Viscous effects are just too dominant on an F1 car.

There are some panel methods that capture wake by introducing point vortices and thus enthalpy through crocco's theorem, but they are really an academic excursion as far as I know.

I think there are a good number of things CFD doesn't capture, but ten times as many that panel codes do not. You can use integral BL and treffetz plane to capture drag, but its going to be very ballpark, IMO.

That said I still love the NASA panel codes I've used

So to let everyone know, I have virtually unlimited access to a brand new 68 node cluster. My department only has 40 licenses of Fluent, so I can't run all cores, but it's still a good resource if anyone wants to get something going.

Since I can't contribute to anything as far as meshing/CAD goes (I have a big gradient based optimization project I'm trying to get going in Fluent/ICEM), I'll run cases for people who would like them run.

Hi,

I'm very happy that I found a community like this, it's a pleasure to be member of it. I'm a 16 years old student and CFD and CAD are my hobbys since I was 11 Years old. Now, I'm working on a research project about the turbuelnces behind F1 cars and i take part with this project at the german high schools scien competition "Jugend forscht"

I'm sponsored by Microsoft Compute Cluster 2003, ANSYS Inc. ans transtec AG (German clustervendor). I present my results at the EACC and the EACC ans I'm still working. I use Fluent 6.3, ICEM CFD, TGRID and SolidEdge (maybe I get CATIA). I mesh on a Workstation with 16G, 2 AMD Opteron and a Quaddro FX. I'm going to mesh a complete car, do you have any tips for me.

Here are some of my Screenshots.

[IMG:1669:1313]http://img530.imageshack.us/img530/6268 ... g22xd0.jpg[/img]
Shot at 2007-07-09

Your 16, I am well impressed mate. Nice job there for a 16 year old. Here in the UK, the only thing a 16 year old can do is to get drunk and fool around!

Good luck with your project and you sound like you know what you are doing. One thing though, why Tgrid if you are using ICEM?

Hi,

in my opinion Tgrid ist the best volume mesher. Unfortuanelly, you can't mesh surfaces in Tgrid and so I need a surface meshing application.
TGrid is very important for generating the prism at the boundary layer.
If anybody have got CAD geometries, please send them to me, I'm also interested in msh or cas files. I think we're a small community using cfd for our hobby and for personal education, we should organize in a community for improving our chances.

Send the files to miladmafi (at) gmx.de

Hi everybody,

this is my first time posting on here. I have recently started using Fluent with some lab work at school, but would like to know if anybody has any manuals or instructional guides with CFD Fluent? I have started working with 2D grids, and would obviously like to work my way up to becoming familiar with 3D grid modelling and 3D aerodynamic analysis.

Any help would be much appreciated!

Welcome-ccs2003 and mkhry...I've posted several Fluent tutorials that might be helpful on the Engineering Student Thread...when you finds odds and ends add to the thread. There is also a Software, Books, Links Thread that was started by the provocteur/graphics artist Manchild, F1Technical's most prolific poster. Both threads are posted temporarily on my signature line. When you have time you might want to read through the aero threads. Welcome

Is the offer to run the CFD on our .dwg files still available? I have a drawing of a rear wing assembly that I know isnt perfect, but would at least give me the ability to see where I need to clean up the design.

Anyone still willing to do this for me?

Email me at [email protected] and I will send the .dwg file.

Thanks!

Chris

On the topic of panel methods:

http://www.flowsol.co.uk/products/newpan/

I would have thought several if not most teams use panel methods at some point in the design process, particularly for streamlined surfaces such as wings.

I haven't read much of the preceding posts, so forgive me if I say something that's already been said, but panel methods are rubbish.
It completely ignores the viscous forces, so there goes your calculation of boundary layers and proper wake flows. The problems are only compounded with complex geometry and ground effect aerodynamics complicates things ever further .
Despite what that site says, I'd be amazing if an F1 team would bother to use a panel method even as a preliminary tool - it's more likely to send you down the wrong path.
That's my opinion anyway