Hello,
I used the turbulence generator plugin with the geometry and settings already discussed in https://pyfr.discourse.group/t/turbulence-plugin-setting-input-plane/988 and I get the following time-averaged “Numerical schlieren” (density gradient magnitude)
Best
How long a period did you time average for to make the image?
More than 15 flow trough, where flow trough =Lx/u_inlet where Lx is axial length of the domain
Thanks. How exactly are you obtaining the 15 flow through time average? Are you dumping averages of smaller time windows and then averaging the averages? Also, what does average of u,v, w, p etc look like?
I’m currently running a full case to better understand the relative importance of those w.r.t. to main flow events. I’ll update on that. From preliminary results seems that running longer solved the issue (25 passages within the box at least with a TI = 8%)
OK, keep me updated. I have a Masters student looking at properties of the turbulence that is generated too. Will look to share report once she finishes the project.
OK, keep me updated
Yes sure, it will take a bit for the simulation though. One difference I noticed is the number of eddies used in PyFR, namely:
self.nvorts = int(ydim*zdim / (4*ls**2))
while a commercial software I found it was using:
“The number of eddies, N, is obtained from the ratio of this
volume (the box that generate the eddies) to the average eddy volume, computed as sigma^3 , where sigma is the average eddy length scale”
For the case here reported this would give:
ILS = sigma = 0.002
y-dim = 0.396 + 2 ILS
z-dim = 0.0108 + 2 ILS
xdim = 2 ILS
PyFR Number Of Eddies : 370
Commercial Software: 2960
Do you think this can play a role in a longer runtime needed for streaks to disappear? Also by comparing the Q criterion, visually in PyFR we tend to generate “bigger” structures (although the Q Criterion has not been normalized for a fair comparison with the same value)
I have a Masters student looking at properties of the turbulence that is generated too. Will look to share report once she finishes the project
I did a similar study as well for 1% and 8%, if you want we can get in touch to compare results, you can email me whenever you want, let me know
Best
For me I get the same number?
PyFR -> ydim*zdim / (4 * ls**2)
Commercial -> 2*ls*ydim*zdim / (8 * ls*3) = ydim*zdim / (4 * ls**2)
But maybe I have made a mistake?
Peter
Professor Peter Vincent MSci ARCS DIC PhD FRAeS
Professor of Computational Fluid Dynamics
Department of Aeronautics
Imperial College London
South Kensington
London
SW7 2AZ
UK
Shouldn’t commercial be:
(2 ILS) * ( 0.396 + 2 ILS) * (0.0108 + 2 ILS) / (ILS^3)
In the expressions I provided ls was the eddy radius - is ILS for you the eddy diameter?
It is just mentioned:
sigma is the mean eddy length scale in coordinate direction .