Heat Flux Pattern

Jeff P. · August 19, 2016, 15:53

Viewing a steady state, compressible, supersonic, SST, tetra-based, simulation I have come across an abnormality in the calculated heat flux into the leading face of a blunt geometry. Variations are double the base rate in a radial pattern. The regularity is leading me to question the validity of such results. Other properties such as velocity, temperature, and density each have similar irregularities. I suspect this may be a discretion / mesh issue.

Any thoughts as to what is causing such a pattern?

Heat Flux.jpg
Mesh.jpg

ghorrocks · August 20, 2016, 08:03

Your mesh is not fine enough. You need to do a mesh sensitivity study to find out how fine your mesh needs to be for accurate results.

Jeff P. · August 20, 2016, 10:48

Ah, makes sense. Thank you very much!

ghorrocks · August 21, 2016, 06:39

While you are at it you should check your convergence tolerance and time step size (if transient).

These are the standard checks you should do for any new simulation.

Jeff P. · August 21, 2016, 12:22

My normalized domain imbalances are in the order of 0.000 2% but looking at the actual imbalance report, all seems within reason at a glance. Furthermore, my max residuals are within the shear layer of the wake as I would be lead to expect but still below 0.000 1 RMS on all accounts. Decent for a first run I suppose. (I achieved 0.000 01 RMS overall to end the run)

I guess now I'll need to do some refinement around the shock, leading face, and the wake.

Let me also ask, is there anything inherently wrong with my residual convergence in terms of accuracy. I read somewhere in the manual that monotonic convergence is best for accuracy.....why? As it was a hard start, I began with a local time scale, let momentum converge then switch to progressively larger physical time-steps as stability increased.

Residuals.jpg

ghorrocks · August 21, 2016, 20:47

Loose convergence tolerances are OK for debugging the run and getting started. But you need to do a sensitivity check on your configuration to see how tight you need to go for the accuracy you require. The documentation has a guide for how tight to aim for, but this is only a guide. You should check on your case.

Monotonic convergence indicates the solver is progressing well and proceeding to a converged solution. The measure of a converged solution is the residuals tolerance (and/or imbalances and/or other convergence measure) achieved.

What the monotonic solution does tell you is that you may be able to run with a larger physical timestep and converge quicker.

Also your approach of local time scale for a while and switching to physical time scale to finish the run is a good approach, commonly used for shock modelling. As long as you have a good number of physical time scale iterations at the end to get to convergence this is a good approach.

Jeff P. · August 22, 2016, 13:32

Makes sense. I'll try some refinements and get going with sensitivity checks. Thanks for the input!

August 19, 2016, 15:53	Heat Flux Pattern	#1
Jeff P. New Member Jeff Join Date: Jun 2016 Posts: 20 Rep Power: 10	Viewing a steady state, compressible, supersonic, SST, tetra-based, simulation I have come across an abnormality in the calculated heat flux into the leading face of a blunt geometry. Variations are double the base rate in a radial pattern. The regularity is leading me to question the validity of such results. Other properties such as velocity, temperature, and density each have similar irregularities. I suspect this may be a discretion / mesh issue. Any thoughts as to what is causing such a pattern? Heat Flux.jpg Mesh.jpg

August 21, 2016, 12:22		#5
Jeff P. New Member Jeff Join Date: Jun 2016 Posts: 20 Rep Power: 10	My normalized domain imbalances are in the order of 0.000 2% but looking at the actual imbalance report, all seems within reason at a glance. Furthermore, my max residuals are within the shear layer of the wake as I would be lead to expect but still below 0.000 1 RMS on all accounts. Decent for a first run I suppose. (I achieved 0.000 01 RMS overall to end the run) I guess now I'll need to do some refinement around the shock, leading face, and the wake. Let me also ask, is there anything inherently wrong with my residual convergence in terms of accuracy. I read somewhere in the manual that monotonic convergence is best for accuracy.....why? As it was a hard start, I began with a local time scale, let momentum converge then switch to progressively larger physical time-steps as stability increased. Residuals.jpg Last edited by Jeff P.; August 21, 2016 at 12:23. Reason: Addition

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Total heat transf. rate vs Total surface heat flux	Renato Sousa	FLUENT	1	April 14, 2020 04:27
CEL expression for calculating heat flux to be applied as surface boundary condition	shadnia	CFX	14	August 20, 2018 19:32
Basic Nozzle-Expander Design	karmavatar	CFX	20	March 20, 2016 09:44
Forced convective heat transfere with fixed heat flux?	BenFranklinIII	OpenFOAM Running, Solving & CFD	0	July 30, 2015 06:31
Enforce bounds error with heat loss boundary condition at solid walls	Chander	CFX	2	May 1, 2012 21:11

August 20, 2016, 08:03		#2
ghorrocks Super Moderator Glenn Horrocks Join Date: Mar 2009 Location: Sydney, Australia Posts: 17,870 Rep Power: 144	Your mesh is not fine enough. You need to do a mesh sensitivity study to find out how fine your mesh needs to be for accurate results.

August 20, 2016, 10:48		#3
Jeff P. New Member Jeff Join Date: Jun 2016 Posts: 20 Rep Power: 10	Ah, makes sense. Thank you very much!

August 21, 2016, 06:39		#4
ghorrocks Super Moderator Glenn Horrocks Join Date: Mar 2009 Location: Sydney, Australia Posts: 17,870 Rep Power: 144	While you are at it you should check your convergence tolerance and time step size (if transient). These are the standard checks you should do for any new simulation.

August 21, 2016, 20:47		#6
ghorrocks Super Moderator Glenn Horrocks Join Date: Mar 2009 Location: Sydney, Australia Posts: 17,870 Rep Power: 144	Loose convergence tolerances are OK for debugging the run and getting started. But you need to do a sensitivity check on your configuration to see how tight you need to go for the accuracy you require. The documentation has a guide for how tight to aim for, but this is only a guide. You should check on your case. Monotonic convergence indicates the solver is progressing well and proceeding to a converged solution. The measure of a converged solution is the residuals tolerance (and/or imbalances and/or other convergence measure) achieved. What the monotonic solution does tell you is that you may be able to run with a larger physical timestep and converge quicker. Also your approach of local time scale for a while and switching to physical time scale to finish the run is a good approach, commonly used for shock modelling. As long as you have a good number of physical time scale iterations at the end to get to convergence this is a good approach.

August 22, 2016, 13:32		#7
Jeff P. New Member Jeff Join Date: Jun 2016 Posts: 20 Rep Power: 10	Makes sense. I'll try some refinements and get going with sensitivity checks. Thanks for the input!