Verification and Validation needed for LES model?

fish500 · September 6, 2017, 19:15

Hi all,

I have created a 3D plume with turbulent water flow triggered by staggered blocks along the bottom and an injected additional variable in the middle of the plume using LES in CFX. After reading a CFD textbook and other articles, I know most experiments require some level of verification and validation, but am not sure exactly what steps to take. The most basic verification in convergence by residuals lower than 1E-4 has been met and courant numbers are close to 1 already. I've read papers that have compared horizontal turbulent intensity (u'u') vertical profiles of the model and experimental data, so that is one method I think would be helpful too. What are the standards that must be met to verify and validate a LES model? This may be a very simple question, but I have had trouble finding a clear answer in ANSYS resources or online.

ghorrocks · September 6, 2017, 20:18

I am glad to hear you ask this question - too many times on the forum there are questions about why simulation results are weird, and 95% of the time the answer is because they did not do the basic checks that your simulation is accurate.

Validation and verification of CFD (and numerical methods in general) is a complex topic in itself, and any works have been written on it. We have a FAQ which discusses some basics: https://www.cfd-online.com/Wiki/Ansy..._inaccurate.3F

The core of it is to check your convergence tolerance, mesh resolution and time step size (if transient) is OK. You do this using a sensitivity analysis. Simple sensitivity analysis on convergence tolerance would be:
* Do a simulation with a certain convergence tolerance (say 1e-4)
* Extract from the simulation the results which are important to you. It could be pressure drop, temperature of something, flow rate; what ever is important for your work.
* Do a significant change to the convergence tolerance. I would recommend going to a convergence tolerance of 1e-5, so x10 tighter. Repeat the simulation and converge to the tighter tolerance.
* Extract the parameters of interest to you. Are they the same as the looser tolerance, within an error you are happy accept? If yes then you can use the 1e-4 convergence tolerance as a good convergence tolerance for your work. If not, then repeat the process with a tighter tolerance again (1e-6 would be the obvious next step) and continue until the results do converge.

You do this process on the mesh (halving the edge length each time) and time step size (halving the time step size each time), and then you have established the convergence tolerance, mesh and time step size required for the accuracy you want.

This is a very simple, crude way of doing it. The link in this FAQ: https://www.cfd-online.com/Wiki/Ansy...publishable.3F starts describing some more sophisticated methods such as Richardson extrapolation. Also the book "Computational Fluid Dynamics" by Roache is the seminal textbook in this field. He describes methods of speeding the process up and quantitative methods of establishing simulation accuracy.

September 6, 2017, 19:15	Verification and Validation needed for LES model?	#1
fish500 New Member Kyle Leathers Join Date: Feb 2017 Posts: 1 Rep Power: 0	Hi all, I have created a 3D plume with turbulent water flow triggered by staggered blocks along the bottom and an injected additional variable in the middle of the plume using LES in CFX. After reading a CFD textbook and other articles, I know most experiments require some level of verification and validation, but am not sure exactly what steps to take. The most basic verification in convergence by residuals lower than 1E-4 has been met and courant numbers are close to 1 already. I've read papers that have compared horizontal turbulent intensity (u'u') vertical profiles of the model and experimental data, so that is one method I think would be helpful too. What are the standards that must be met to verify and validate a LES model? This may be a very simple question, but I have had trouble finding a clear answer in ANSYS resources or online.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Verification and Validation databases for CFD	Reeb	Main CFD Forum	3	December 18, 2016 16:20
Request for Lagrangian Particle Tracking Validation or Verification Paper	Mojtaba.a	OpenFOAM Verification & Validation	6	May 23, 2016 02:47
verification and validation	karthikpaulraj	Main CFD Forum	2	February 22, 2015 09:04
Verification and validation	mashinsazi	Main CFD Forum	0	November 28, 2014 03:36
New OpenFOAM Verification & Validation Forum Opened	jola	OpenFOAM Announcements from Other Sources	2	October 1, 2011 18:21

September 6, 2017, 20:18		#2
ghorrocks Super Moderator Glenn Horrocks Join Date: Mar 2009 Location: Sydney, Australia Posts: 17,854 Rep Power: 144	I am glad to hear you ask this question - too many times on the forum there are questions about why simulation results are weird, and 95% of the time the answer is because they did not do the basic checks that your simulation is accurate. Validation and verification of CFD (and numerical methods in general) is a complex topic in itself, and any works have been written on it. We have a FAQ which discusses some basics: https://www.cfd-online.com/Wiki/Ansy..._inaccurate.3F The core of it is to check your convergence tolerance, mesh resolution and time step size (if transient) is OK. You do this using a sensitivity analysis. Simple sensitivity analysis on convergence tolerance would be: * Do a simulation with a certain convergence tolerance (say 1e-4) * Extract from the simulation the results which are important to you. It could be pressure drop, temperature of something, flow rate; what ever is important for your work. * Do a significant change to the convergence tolerance. I would recommend going to a convergence tolerance of 1e-5, so x10 tighter. Repeat the simulation and converge to the tighter tolerance. * Extract the parameters of interest to you. Are they the same as the looser tolerance, within an error you are happy accept? If yes then you can use the 1e-4 convergence tolerance as a good convergence tolerance for your work. If not, then repeat the process with a tighter tolerance again (1e-6 would be the obvious next step) and continue until the results do converge. You do this process on the mesh (halving the edge length each time) and time step size (halving the time step size each time), and then you have established the convergence tolerance, mesh and time step size required for the accuracy you want. This is a very simple, crude way of doing it. The link in this FAQ: https://www.cfd-online.com/Wiki/Ansy...publishable.3F starts describing some more sophisticated methods such as Richardson extrapolation. Also the book "Computational Fluid Dynamics" by Roache is the seminal textbook in this field. He describes methods of speeding the process up and quantitative methods of establishing simulation accuracy.