CFD Online Discussion Forums

CFD Online Discussion Forums (http://www.cfd-online.com/Forums/)
-   FLUENT (http://www.cfd-online.com/Forums/fluent/)
-   -   Explicit vs Implicit VOF in Fluent (http://www.cfd-online.com/Forums/fluent/115272-explicit-vs-implicit-vof-fluent.html)

MachZero March 27, 2013 08:38

Explicit vs Implicit VOF in Fluent
 
I was wondering if anyone can shed some light on explicit vs implicit VOF methods in fluent. My understanding is that explicit is simply an interface tracking scheme. It "sees" the interface at time t, obtains the momentum at that point, then extrapolates to get a continuous interface at time t+1. My understanding of the implicit formulation is that it actually calculates the interface location, therefore inverting a matrix with the volume fraction equation. This method is significantly more stable because of it, and much larger timesteps can be used. In theory, since the interface tracking is done by calculation and not tracking, it should be more accurate, no?

Where this stems from:
Whenever you talk to one of fluent's service engineers and ask anything about the methods (whether it be "how do I interpret this huge color band, or what is most appropriate for this particular problem) they always say that the method that gives you the clearest and cleanest interface is the explicit with geo-reconstruct. While this makes perfect sense after reading through the guide... is this actually more accurate or just "prettier". Sure the interface looks better, but is the solution any better? Or is it actually less accurate.

Any insight and experince would be much appreciated!

Ford Prefect March 27, 2013 18:37

Quote:

Originally Posted by MachZero (Post 416719)
I was wondering if anyone can shed some light on explicit vs implicit VOF methods in fluent. My understanding is that explicit is simply an interface tracking scheme. It "sees" the interface at time t, obtains the momentum at that point, then extrapolates to get a continuous interface at time t+1. My understanding of the implicit formulation is that it actually calculates the interface location, therefore inverting a matrix with the volume fraction equation. This method is significantly more stable because of it, and much larger timesteps can be used. In theory, since the interface tracking is done by calculation and not tracking, it should be more accurate, no?

Where this stems from:
Whenever you talk to one of fluent's service engineers and ask anything about the methods (whether it be "how do I interpret this huge color band, or what is most appropriate for this particular problem) they always say that the method that gives you the clearest and cleanest interface is the explicit with geo-reconstruct. While this makes perfect sense after reading through the guide... is this actually more accurate or just "prettier". Sure the interface looks better, but is the solution any better? Or is it actually less accurate.

Any insight and experince would be much appreciated!

The original method was explicit. You can find it here.

I don't understand your statement about accuracy. Advecting a volume fraction more than one cell per time-step makes no sense and the solution you obtain will not be time accurate. Most free surface flows are not steady. I also believe that you will suffer from diffusion of the interphase if you use an implicit scheme for the VOF equation.

So, yes the solution is most likely better with the explicit scheme.

MachZero April 10, 2013 06:55

examples
 
1 Attachment(s)
FP,

Thanks for your response! Sorry for my delay, I have been on travel for some time. I had trouble accessing the link you sent. I understand what you are saying regarding not being able to step larger than one cell in either implicit or explicit without it becoming time inaccurate.

That being said, regardless of the original method, what would be the cause of explicit being more accurate than implicit using the same timestep? I imagine (again, just my opinion) if you are doing a surface tracking that involved looking at where the surface is, looking at the momentum term, and then making the interface shift accordingly, you could get a lot of instability waves forming (in the simulation, not numerically). Whereas if you were inverting a full matrix (as I believe is done in implicit), you are calculating where the interface will be a given timestep later (yes, again, can lead to diffusion, but with sufficiently small timestep...) I turn to the fluent advanced training manual for a better explanation...

Please take a look at the attached photo. It is taken from the Fluent advanced training in VOF methods. These images represent the pure advection of shapes (I like to think of it as a very, very viscous fluid being advected in water, where no distortion of the interface is expected). The only one I am showing is the hollow square here. These are the images after they have been advected with explicit(left) and implicit (right) with several VF disc. What one might expect to see is, at a given time down the road, the same shape. Note for the explicit methods, the interface is wavy, almost like it has a kelvin-helmholtz instability forming. BUT... as everyone says, the interface is crisp! ;-)

With the implicit method, the hollow square shape is preserved far better (especially with compressive, but EVEN with HRIC), but the interface is mildly blurred, as might be expected from numerical diffusion.

Therefore, based on this (possibly incorrect) reasoning, I would estimate that the interface of the explicit method might be more inaccurate.

Side note: Thoughts on the new Ford (really Aston Martin) grilles?
Thanks for your feedback!

rmh26 April 10, 2013 10:05

I'm not sure what Fluent considers to be implicit VOF. But it would seem like it requires solving the momentum equation,pressure equation, vof advection, updates on density and viscosities and the surface tension forces. This would seem to be a monstrously large coupled nonlinear system. The only reasonable way to attack it would be to use a predictor corrector scheme which isn't really implicit at all. It would be interesting to see a precise definition of what they are considering implicit. I do remember seeing a paper once that did an implicit multiphase solution but i think it was for a very specialized case not general at all like VOF.

MachZero April 18, 2013 11:33

Quote:

Originally Posted by rmh26 (Post 419653)
I'm not sure what Fluent considers to be implicit VOF. But it would seem like it requires solving the momentum equation,pressure equation, vof advection, updates on density and viscosities and the surface tension forces. This would seem to be a monstrously large coupled nonlinear system. The only reasonable way to attack it would be to use a predictor corrector scheme which isn't really implicit at all. It would be interesting to see a precise definition of what they are considering implicit. I do remember seeing a paper once that did an implicit multiphase solution but i think it was for a very specialized case not general at all like VOF.

rmh,

Thanks for the reply. I saw a flow chart at one point that I have not been able to find recerntly that shows the way fluent does this. That being said, I know you have the option turning off some of the equations, such as surface tension. That being said, you can use a coupled solver for pressure and velocity, and then you have the option of coupling these with volume fraction. I have done some reading on this and the "coupled with volume fractions" option is one of those things that is new to v14, and is still being worked on. That being said, along with my confusion, some people at Fluent say that coupled solving is only beneficial when doing steady state calculations, while others say it is significantly more accurate. Either way, I am still curious to hear why explicit or implicit might be better than the other based on the image I uploaded.

ndabir August 7, 2013 18:17

If you look at the original picture you uploaded early, you can see that this figure is NOT meant to compare implicit versus explicit methods. But instead is to compare different methods (such as compressive and HRIC) for each implicit or explicit method. Even if you look at the figures caption, it says for the explicit one it is first order in time and for the implicit one is second order in time.
Moreover, if you take a look at a table which is provided in the same lecture file, you can see that both explicit and implicit methods have high and low accurate interface schemes. For example for explicit method, GEO reconstruct is very high in accuracy and for implicit method, BGM has very high accuracy.
It will be better to first use explicit method since it takes less memory and time to solve the equations of flow. If you saw instabilities and your solution was highly dependent on very small time steps, then change to implicit methods.


All times are GMT -4. The time now is 01:12.