CFD Online Discussion Forums

CFD Online Discussion Forums (http://www.cfd-online.com/Forums/)
-   Main CFD Forum (http://www.cfd-online.com/Forums/main/)
-   -   Validation with experiments - without error (http://www.cfd-online.com/Forums/main/86806-validation-experiments-without-error.html)

Connor April 2, 2011 14:48

Validation with experiments - without error
 
Hi,

when CFD users want to show the results of their validation, they often plot diagrams comparing CFD results and experimental data.

However, why does nobody use error bars on the experimental data?

I am looking forward for you answer!


best regards
Connor

Florent April 2, 2011 16:13

that's a very good and interesting question ... !

I hope their will be many responses to this post. And I think this will change: as soon as UQ will be more and more used (thanks to big computers), errors bars both on experiments and cfd results will b given

flo

Lysistrata April 2, 2011 17:47

Quote:

Originally Posted by Connor (Post 301967)
Hi,

when CFD users want to show the results of their validation, they often plot diagrams comparing CFD results and experimental data.

However, why does nobody use error bars on the experimental data?

I am looking forward for you answer!


best regards
Connor

Maybe because the errors are so large it makes the experimenters look incompetent.

Martin Hegedus April 3, 2011 12:45

The answer to this depends on what parameters will have uncertainty measurements. For example, uncertainty values can exist for position (i.e. angle of attack), load cells, geometry, and upstream conditions. Those are usually given in the report. However, uncertainties also exist for the physics. To determine that can be very difficult and expensive. This can require that the experiment be run repeatedly in a single wind tunnel and that the experiment be run in several different tunnels. For example, determining lift and drag for transonic flow over a 2D airfoil. The uncertainty for this is large and not clear to people. It is what it is.

Florent April 3, 2011 13:47

Experimentalists give also the uncertainty of their measures without repeating several times the experiments on different wind tunnel ...

Martin Hegedus April 3, 2011 14:23

This is where the art comes in. In general the authors of the WT report will explain what went into their error/uncertainty bars. The error bands should capture the errors that the experimentalist are aware of, load balance, angle of attack, Mach number, geometry, etc. Also, the experimentalist should document any WT induced flow features that he or she was aware of. However, those uncertainties may not represent all the uncertainties that are important to you. For example, if one could run that same experiment without wind tunnel walls and a sting you may get a value that falls outside of the error bands given in a report.

sazabi2001 April 5, 2011 09:12

Most of the CFD validation experiment would be far too expensive (large scale wind tunnel ) or almost impossible to do (large buildings, weather or so).

Therefore, the way of full simulation => full experiment => improving simulation may never work out in most companies.. except in those military or big government funded projects.

===========

But there is still a way of doing so when such validation is a very simple type of experiment, such as the very classical 'back step turbulent flow' thing.

So the process now becomes like this: simple simulation => simple experiment => extended to complex object simulation. This might have less theoretical accuracy, but still good enough for most engineers.

===========

However, those classical tests like back step model was completed / published in the mid last century and has been repeat doing by every mechanics studnets in every university every year.. So now people just skip this and use the model directly.

Also, like we said, wind tunnel may has many uncertainties as well; and because of the non-linearly of N-S function and the fluid dyanmic properties, we may never get a model has error less than 30%. So in some point people just looking for more fundamental things like flow property in CFD rather than a series of accurate data.


All times are GMT -4. The time now is 04:22.