CFD Online Discussion Forums

CFD Online Discussion Forums (http://www.cfd-online.com/Forums/)
-   Main CFD Forum (http://www.cfd-online.com/Forums/main/)
-   -   Scaling a model for aerodynamic analysis via cfd simulation (http://www.cfd-online.com/Forums/main/114653-scaling-model-aerodynamic-analysis-via-cfd-simulation.html)

addy01 March 15, 2013 02:04

Scaling a model for aerodynamic analysis via cfd simulation
 
Hello Sir,
I want to calculate co efficient of drag on heavy vehicle with the help of CFD simulation as my undergraduate project. Please tell me how to scale down my model for better meshing or i should not scale down model?? i mean scaling down of model effect Cd???

Please also tell me optimize setting for analysis or some technical journal where i can study that...plz reply..

Thanx & Regards

julien.decharentenay March 17, 2013 19:35

If you were doing a wind-tunnel experiment, spatial constraint (as well as other constraints) would require you to scale down the experiment.

I do not quite understand the benefits you would gain from scaling down in the CFD simulation (apart from comparing with a wind-tunnel experiment, or investigating the effects of scaling down). Can you clarify the benefits you are expecting from scaling down?

addy01 March 18, 2013 01:41

I jst want to reduce computation time. and the reason behind this is that i hav a laptop with 4 gb ram & 1 gb nvidia with i5 processor. it is very hard and time consuming approx 11 hours...plz suggest me some beneficial ways...

RodriguezFatz March 18, 2013 05:00

This does not work. There is no benefit. If you want a representative simulation, you must keep the Reynolds number. So if you scale down the simulation you must also keep the Reynoldsnumber, which means you have to increase your inlet velocity (for example). Thus, even for smaller domains, you have the same need for grid points.

flotus1 March 18, 2013 05:28

The only thing you can do to speed up the simulation is to make use of symmetries.
If you are using a RANS approach, your model and your boundary conditions have the same symmetry planes, then you can simulate only a part of the model.

In your case I guess the vehicle has one symmetry plane which allows you to halve the simulation time.

julien.decharentenay March 18, 2013 06:47

Alex option is a very good option.

Otherwise there are a number of options to speed-up analysis, but usually with trade-offs or increased complexity:
- Use a coarser mesh, at a trade-off with accuracy;
- Numerical methods: some methods are typically faster than others and using appropriate tolerance can speed-up results;
- Hardware: faster processor, more memory, GPU or parallel processing can help speed things up;
- Software: the software you use may have tricks that can speed calculations (such as memory access).

11hours per run is not bad (from my perspective). Doesn't your uni give you access to some computer resources to allow you to run 3-4 cases overnight?

Good luck.

addy01 March 18, 2013 09:45

Thanx a lot julien.decharentenay..i will surely mention ur name as a guide in my research report...u helped me a lot..at every stage of my project...:)

Thnx Alex & RodriguezFatz..thnx 4 ur kind reply..

please just help me once more...please can u tell me what is appropriate (best suits to my project "aerodynamic improvement of truck and trailer" as per ur knowledge and experiance) setting of k-e model such as y+ value nd turbulence intensity and other factor....by which i can get most appropriate answer...
Waitng 4 a nyc reply...Thnx & Regards

ADDY

atmcfd March 18, 2013 14:41

Addy,

Keep in mind that y+ is dependent on the mesh you build, which is in turn dependent on the quality/accuracy of the simulation you want. If you are more concerned with the boundary layer etc. on the surface of the vehice, its good to have a well resolved mesh, with low y+ values.

It also depends on the type of turbulence modeling tool you choose. For RANS, any decent y+ will be okay, but if you are looking at LES, you need very low y+, and for DNS, even lesser. Just look up about y+ in cfd-online wiki page in detail and you will have a better idea on how to play with it. Bottomline is : design a mesh based on the accuracy you want. Make sure that though you may want to reduce computation time, you shouldn't lose focus on the physical validity too much. Any decent turbulence simulation WILL require lots of run time. As Julien said, 11 hours is nothing for this.

If you are into RANS (which is more likely) you might want to try Spalart Allmaras/k-eps approaches. This is something you need to do literature survey about, since many papers are available online which do automotive aerodynamics on heavy vehicles. Shouldn't be too difficult. Read the papers and you will figure out most of what you want.


All times are GMT -4. The time now is 11:07.