Momentum imbalances

AlexRonto · September 19, 2016, 05:16

Hello to everyone,

I make a transient analysis in CFX of an electric generator. It is actually two rotating disks and a stationary rim between them. The gap between the rotors and the stator is very small (about 0.6% of the rotors diameter), so there are many small cells in my mesh. My BCs are Pstat= 0atm (relative) at the external boundaries of my domain and rotating walls with a certain rotational velocity at the "interface" of my domain with the rotors.

My problem is that my residuals reach values of 10^(-14), but the U and V momentum equations' imbalances do not fall under 10% for V and 100% for U (the other imbalances are under 1%). So my questions are:

1) How do I decide whether I need the imbalances to judge my simulation's convergence? Could I consider my results as OK if ALL the imbalances have not reached the 1% value?

2)If we accept that a further reduce of my imbalances is needed, what could I do for it? I have tried different timesteps but the problem remains.

3) Could it be a problem of inappropriate BCs? Truth is I need pressure boundaries, as I want CFX to calculate the mass flow and therefore, I don't have any velocities to specify.

4) Could it be a problem of inappropriate turbulence model? I have SST and I tried k-ε too.

Thnx and sorry for the long post.

Alex

ghorrocks · September 19, 2016, 07:19

Imbalances are not appropriate measures of convergence in some cases. For instance in domains with no inlets or outlets tiny amounts of numerical noise causes imbalances of 100%. This is why imbalances are an optional convergence tolerance.

If your domain has no inlets or outlets it is likely this is what is happening. In this case imbalances are not a useful measure of convergence and should not be used.

This is caused by imbalances being normalised against the total flow rate (or momentum or heat flux). If there is no total flow then the imbalances calculation is meaningless.

However, looking at your convergence graph: Changes of residual of 10E-6 to 10E-12 in one iteration is not good. This is too fast. It means something weird has happened. Did you change to a microscopic time step size? Or change some other convergence parameter at this time?

AlexRonto · September 19, 2016, 08:07

Thank you Glenn! Really really helpful. I had spent hours trying to find what was going on.

Yes, my domain has no total flow applied at any boundaries so what you said is probably what happens here.

As for the residuals, at the beginning I ran a steady-state simulation in order to get some more proper initial conditions for the transient simulation that I ran next. So I suppose this is the reason for changing so steeply. Beside, the timesteps I used at the transient analysis is smaller than the one at the steady state.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Momentum source coefficient and convergence	meh	CFX	9	October 12, 2020 08:37
Simulation of Axial Fan Flow using A Momentum Source Subdomain	Liam	CFX	28	July 16, 2013 08:24
Spray total droplet momentum calculation help	tarnsharma	AVL FIRE	0	July 19, 2012 08:17
Derivation of Momentum Equation in Integral Form	Demonwolf	Main CFD Forum	2	October 29, 2009 19:53
Nonstaggered Grids and Momentum Interpolations	S. Wang	Main CFD Forum	3	March 2, 2001 13:44

September 19, 2016, 07:19		#2
ghorrocks Super Moderator Glenn Horrocks Join Date: Mar 2009 Location: Sydney, Australia Posts: 17,703 Rep Power: 143	Imbalances are not appropriate measures of convergence in some cases. For instance in domains with no inlets or outlets tiny amounts of numerical noise causes imbalances of 100%. This is why imbalances are an optional convergence tolerance. If your domain has no inlets or outlets it is likely this is what is happening. In this case imbalances are not a useful measure of convergence and should not be used. This is caused by imbalances being normalised against the total flow rate (or momentum or heat flux). If there is no total flow then the imbalances calculation is meaningless. However, looking at your convergence graph: Changes of residual of 10E-6 to 10E-12 in one iteration is not good. This is too fast. It means something weird has happened. Did you change to a microscopic time step size? Or change some other convergence parameter at this time?

September 19, 2016, 08:07		#3
AlexRonto New Member Alexandros Join Date: Jun 2016 Posts: 28 Rep Power: 9	Thank you Glenn! Really really helpful. I had spent hours trying to find what was going on. Yes, my domain has no total flow applied at any boundaries so what you said is probably what happens here. As for the residuals, at the beginning I ran a steady-state simulation in order to get some more proper initial conditions for the transient simulation that I ran next. So I suppose this is the reason for changing so steeply. Beside, the timesteps I used at the transient analysis is smaller than the one at the steady state.