Interpreting residuals, in ramped pressure jet

NotOsborneReynolds · January 11, 2022, 10:56

We use residuals as a guide to how well simulation is working. For example when they're above 1e-3 we're not so sure but if they're below 1e-4 we're happy about them. But are they so simply interpreted when we are ramping up pressure?

I'm studying impinging air jet heat transfer with choked flow. If I start with maximum inlet air pressure, the sim diverges and usually crashes. So, I'm ramping up the pressure from a very small value (and harvesting numerous intermediate values along the way).

The problem with residuals in this case is that what is reported is normalized residuals. Using STAR-CCM+ defaults, they're normalized to the highest values seen during the first 5 iterations -- but I think regardless of the exact normalization method, this problem will still exist.

Namely, the residuals are normalized to a situation where the pressures, mass flows, velocities, fluxes, turbulences, and everything else is at small values. In an absolute sense, the sim is accurate there because there's hardly any signal.

As the pressure ramps up, the residuals should grow even if everything is working as I hope it does. So, I should no longer hope to see small residuals. How do I use residuals as one of the guides to simulation quality?

Isn't there some way to use the absolute values of the residuals (not normalized ones), perhaps in conjunction with cell statistics and parameter scales, to judge how well the simulation is meeting my true accuracy goals?

Thanks!!!

LuckyTran · January 11, 2022, 12:13

Only the continuity (and a few other very niche) residuals are normalized by the worst in the first 5 iterations if you use the defaults. The rest are normalized by the mean fluxes of those fields. Your momentum residual for example is always with respect to the mean momentum flux across cells. If you ramp slowly enough, those residuals should still decrease as you increase the pressure. If you see them increase, then it is because you're ramping faster than it converges. Which isn't necessarily wrong, you just keep iterating until it converges.

You rarely want to use absolute residuals ever, they are dimensional quantities! They scale with your flow and domain size and you have to re-calibrate your feel for what they mean with each and every case (which is effectively re-normalizing them in your mind, every time).

January 11, 2022, 10:56	Interpreting residuals, in ramped pressure jet	#1
NotOsborneReynolds New Member Chris Bailey Join Date: Dec 2019 Posts: 16 Rep Power: 7	We use residuals as a guide to how well simulation is working. For example when they're above 1e-3 we're not so sure but if they're below 1e-4 we're happy about them. But are they so simply interpreted when we are ramping up pressure? I'm studying impinging air jet heat transfer with choked flow. If I start with maximum inlet air pressure, the sim diverges and usually crashes. So, I'm ramping up the pressure from a very small value (and harvesting numerous intermediate values along the way). The problem with residuals in this case is that what is reported is normalized residuals. Using STAR-CCM+ defaults, they're normalized to the highest values seen during the first 5 iterations -- but I think regardless of the exact normalization method, this problem will still exist. Namely, the residuals are normalized to a situation where the pressures, mass flows, velocities, fluxes, turbulences, and everything else is at small values. In an absolute sense, the sim is accurate there because there's hardly any signal. As the pressure ramps up, the residuals should grow even if everything is working as I hope it does. So, I should no longer hope to see small residuals. How do I use residuals as one of the guides to simulation quality? Isn't there some way to use the absolute values of the residuals (not normalized ones), perhaps in conjunction with cell statistics and parameter scales, to judge how well the simulation is meeting my true accuracy goals? Thanks!!!

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Generate convective pressure fluctuation	Bananenflanke	CFX	10	May 12, 2021 19:33
Fail to converge when solving with a fabricated solution	zizhou	FLUENT	0	March 22, 2021 07:33
Pressure Outlet Guage pressure	Mohsin	FLUENT	36	April 29, 2016 18:16
Setting up the pressure variation due to tornado in a duct(UDF)+animation	guillaume1990	FLUENT	0	March 3, 2014 12:59
Hydrostatic pressure in 2-phase flow modeling (long)	DS & HB	Main CFD Forum	0	January 8, 2000 16:00

January 11, 2022, 12:13		#2
LuckyTran Senior Member Lucky Join Date: Apr 2011 Location: Orlando, FL USA Posts: 5,761 Rep Power: 66	Only the continuity (and a few other very niche) residuals are normalized by the worst in the first 5 iterations if you use the defaults. The rest are normalized by the mean fluxes of those fields. Your momentum residual for example is always with respect to the mean momentum flux across cells. If you ramp slowly enough, those residuals should still decrease as you increase the pressure. If you see them increase, then it is because you're ramping faster than it converges. Which isn't necessarily wrong, you just keep iterating until it converges. You rarely want to use absolute residuals ever, they are dimensional quantities! They scale with your flow and domain size and you have to re-calibrate your feel for what they mean with each and every case (which is effectively re-normalizing them in your mind, every time).