CFD Online Discussion Forums

CFD Online Discussion Forums (http://www.cfd-online.com/Forums/)
-   CFX (http://www.cfd-online.com/Forums/cfx/)
-   -   multiphase flow ,mass and volume fraction imbalance (http://www.cfd-online.com/Forums/cfx/104645-multiphase-flow-mass-volume-fraction-imbalance.html)

sope111 July 12, 2012 06:05

multiphase flow ,mass and volume fraction imbalance
 
Hi,everyone!
I'm puzzled a problem for a long time.:confused::confused: Could anyone help me?

I try to calculate a multiphase flow model containing oil and air. There are two inlet, air and oil separately, one outlet,and a high speed rotar.
The BCs are as follow:
1.oil inlet with a mass flow rate
2.air leak to the cavtiy , so quantity is unknown, i set the boundary as opening with direction and press.
3.outlet is opening with direction and press.
4.the rotar runs at high speed(12000_14000rev/min)

the multiphase flow pattern changes with different work conditions. I choose steady model to calculate when the work condition is constant .

So,I change multiphase model with different work conditions.
when the flow pattern is homogeneous, i choose homogeneous model and set the timescale is antomatic and timescale factor is 2. The outfile shows mass and volume fraction imbalance are about 98%. Then i set timescale factor as 100 for mass and volume fraction equations , mass and volume fraction imbalance are down to 1%:confused::confused:


Then i set the work conditions differently and choose imhomogeneous and free surface model ,The outfile shows mass and volume fraction imbalance are still about 98%. :eek::eek:And i changes the timescale seems useless.



+--------------------------------------------------------------------+
| P-Vol |
+--------------------------------------------------------------------+
Boundary : AIRIN 4.6508E-03
Boundary : OILIN 2.6500E-02
Boundary : OUT -5.7588E-03
-----------
Domain Imbalance : 2.5392E-02
Domain Imbalance, in %: 95.8191 %
+--------------------------------------------------------------------+
| Mass-oil |
+--------------------------------------------------------------------+
Boundary : AIRIN 1.4396E-15
Boundary : OILIN 2.6500E-02
Boundary : OUT -9.4913E-04
-----------
Domain Imbalance : 2.5551E-02
Domain Imbalance, in %: 96.4184 %

The mesh is produced by ICEM, the Jacobian is 0.55 above and angle is 27 above . So i think there is no problems with mesh.

ghorrocks July 12, 2012 08:15

Sounds like an FAQ: http://www.cfd-online.com/Wiki/Ansys...gence_criteria

sope111 July 12, 2012 08:33

Quote:

Originally Posted by ghorrocks (Post 371107)


Hi ,ghorrocks.
I'v confused by this problems for a months,and have tried many times.
But the imbalance can achieve the target only for homogeneous.
Could you give me some advice ?
Thanks a lot.

oj.bulmer July 12, 2012 16:19

I would recommend first, if you are not already doing it, to monitor the imbalances along with the residuals when the run is in progress.

It has been my experience that with multiphase, you have to be a real gentleman. I had similar problems, when I used physical timescale though I used a fraction of residence time (Going for higher values resulted in divergence).

Then I switched to Auto timescale with factor of (factor=1) with conservative option. Yet, my imbalances kept oscillating. I tried factor of 0.1 and 0.01 and surprisingly, the balances started reducing to much bearable extent.

Also, to get a usable results (not perfect), you can also enforce the steady state by using first order discretization scheme, as it will diffuse the turbulent effects to some extent.

sope111 July 12, 2012 22:49

Quote:

Originally Posted by oj.bulmer (Post 371188)
I would recommend first, if you are not already doing it, to monitor the imbalances along with the residuals when the run is in progress.


It has been my experience that with multiphase, you have to be a real gentleman. I had similar problems, when I used physical timescale though I used a fraction of residence time (Going for higher values resulted in divergence).

Then I switched to Auto timescale with factor of (factor=1) with conservative option. Yet, my imbalances kept oscillating. I tried factor of 0.1 and 0.01 and surprisingly, the balances started reducing to much bearable extent.

Also, to get a usable results (not perfect), you can also enforce the steady state by using first order discretization scheme, as it will diffuse the turbulent effects to some extent.


Thanks for your reply . This problem bored me for one month .

First ,i've try to moniter the imbalances along with the residuals ,but the mass and volume imbalances are always oscillate around 90%.

I even calculate it using transient . At first , the imbalance is up to 1, but it reach to 1% after some timesteps. But transient simulation must employ a steady result as initial valve. However ,I didn't get a convergence value,
so i used this unconvergence value.

Now ,I wonder whether the transient results are usable( unconvergence value as initial value,but the imbalance and residual seems right)?

oj.bulmer July 13, 2012 08:44

If I am right, in the event where Steady State solution is not at all possible, it may be beneficial to use first order upwind (for stability owing to numerical diffusion) to get some sensible solution with SS and initialize a transient solution with it using High resolution. Though, more resourceful CFD guys here may have better suggestions.

In transient solution, if you make sure that the convergence is obtained for every time step by choosing appropriate timestep (using adaptive timestepping/ timestep sensetivity); the time averaged results can be fairly representative of physics you wish to see.

sope111 July 13, 2012 23:12

Quote:

Originally Posted by oj.bulmer (Post 371305)
If I am right, in the event where Steady State solution is not at all possible, it may be beneficial to use first order upwind (for stability owing to numerical diffusion) to get some sensible solution with SS and initialize a transient solution with it using High resolution. Though, more resourceful CFD guys here may have better suggestions.

In transient solution, if you make sure that the convergence is obtained for every time step by choosing appropriate timestep (using adaptive timestepping/ timestep sensetivity); the time averaged results can be fairly representative of physics you wish to see.


Hi,oj.bulmer.
Thanks for your attention.

But i still don't know whether the result is usable with
the unconvergence result as initial value.
I'm not sure whther the unconvergence result is credible.

And another question , as i mentioned above ,the rotar speed is 12000 rev/min ,
so the timestep must be set to 10E-4(1/w). But the initial value is unconvergence, so i must set much time(i set the value as 16s) to get a usable value . However ,the amount of computation is heavy.

Besides, i don't concern about the transient process, so i still want to get a steady value as soon as possible.
I reduce the timescale factor to 0.1 and 0.01 with automatic timescale,
but the imbalance stay in the same level(90%) 。

So , what can i do for steady simulation??

Thanks.

ghorrocks July 14, 2012 05:29

If all you care about is the final steady state solution (or final pseudo-steady if transient and it is flapping about) then a non-converged initial condition will not matter. As long as it is close enough that the solver converges OK in the transient run, that is the main issue.

There is an extensive discussion on setting time steps int he CFX documentation. For tricky steady state simulations you want to start with a small time step to get it going, but once it is starting to converge OK then you increase the time step size. You can increase it massively if it is converging well).

sope111 July 14, 2012 06:00

Quote:

Originally Posted by ghorrocks (Post 371416)
If all you care about is the final steady state solution (or final pseudo-steady if transient and it is flapping about) then a non-converged initial condition will not matter. As long as it is close enough that the solver converges OK in the transient run, that is the main issue.

There is an extensive discussion on setting time steps int he CFX documentation. For tricky steady state simulations you want to start with a small time step to get it going, but once it is starting to converge OK then you increase the time step size. You can increase it massively if it is converging well).


Dear ghorrocks,

Thanks for your help .

Another question, multiphase model are homogeneous and imhomogeneous, i want to know what's the relationship between vof in many textbooks and FLUENT?

ghorrocks July 14, 2012 06:05

CFX has a lot more multiphase options than homogeneous and inhomogeneous. That merely refers to whether different phases share a variable field or not.

I do not understand your question to VOF and textbooks and Fluent.

sope111 July 14, 2012 10:36

Quote:

Originally Posted by ghorrocks (Post 371425)
CFX has a lot more multiphase options than homogeneous and inhomogeneous. That merely refers to whether different phases share a variable field or not.

I do not understand your question to VOF and textbooks and Fluent.

Hi,ghorrocks

There are two main multiphase phase model:homogeneous and inhomogeneous(free surface,mixture,particle).

But Vof appears in many papers and textbooks to track the interface of different phase.

I don't know the way to track the interface in cfx, by solve the volume fraction equtions??

Thanks

ghorrocks July 15, 2012 08:20

There are a few methods to track free surfaces and VOF is one of them. CFX does not actually implement VOF as it is usually defined, CFX uses a volume fraction equation with an advection scheme designed to keep the interface sharp.


All times are GMT -4. The time now is 10:48.