[DoubleG]

Hi all,

I need to model mixing a large tank (a cylinder of r=20m and h=6m) with three or four mixers of two types (d=6m slowly rotating rotors and d=0.5m fast rotating impellers). the axis of the mixers horizontal (while the axis of the tank is vertical)

I wonder what kind of strategy is best to model this problem. I have non-newtonian behaviour and am interested in transient effects (homogenization). in a later stage I also would like to involve temperature effects, but initially it is not necessary to solve the energy equation. Ultimately I actually also need two phase (or even three phase) flow, but probably it is not realistic to involve these effects.

I have a cluster available to facilitate large computations, still I realize the problem is rather complicated, though I need to do it anyway as good as possible.

For now I have some troubles in determining a proper strategy to tackle this problem, starting with meshing. So here are some questions which might help me to start.

1. How can I determine what type of elements I should use? (hexa/tetra)
2. What defines MRF approach to be a proper method and when do I need sliding mesh?
3. Is it wise to use non-conformal mesh for the MRF interfaces or should I rather use conformal mesh
4. Is it a problem when the interface of MRF also is used as a boundary between hexa and tetra mesh?
5. The top of the tank is a free surface, is it really necessary to add a layer of air on top of the tank in order to model free surface flow? (It is quite important to have some realistic indication of the surface flow, because it is my only validation reference.)

I will probably come with some more questions, but if we could discuss this questions, I would be really grateful.

[energy382]

Originally Posted by DoubleG

Hi all,

I need to model mixing a large tank (a cylinder of r=20m and h=6m) with three or four mixers of two types (d=6m slowly rotating rotors and d=0.5m fast rotating impellers). the axis of the mixers horizontal (while the axis of the tank is vertical)

I wonder what kind of strategy is best to model this problem. I have non-newtonian behaviour and am interested in transient effects (homogenization). in a later stage I also would like to involve temperature effects, but initially it is not necessary to solve the energy equation. Ultimately I actually also need two phase (or even three phase) flow, but probably it is not realistic to involve these effects.

I have a cluster available to facilitate large computations, still I realize the problem is rather complicated, though I need to do it anyway as good as possible.

For now I have some troubles in determining a proper strategy to tackle this problem, starting with meshing. So here are some questions which might help me to start.

1. How can I determine what type of elements I should use? (hexa/tetra)
2. What defines MRF approach to be a proper method and when do I need sliding mesh?
3. Is it wise to use non-conformal mesh for the MRF interfaces or should I rather use conformal mesh
4. Is it a problem when the interface of MRF also is used as a boundary between hexa and tetra mesh?
5. The top of the tank is a free surface, is it really necessary to add a layer of air on top of the tank in order to model free surface flow? (It is quite important to have some realistic indication of the surface flow, because it is my only validation reference.)

I will probably come with some more questions, but if we could discuss this questions, I would be really grateful.

For this kind of simulation, you need the best mesh you can achieve! And I hope, you really have a very big cluster, as this simulation is very time consuming!!

1. hexa is always better in terms of computational effort (especially in huge models as yours). I would highly recommend to use hexa, as you also want to investigate heat effects. If it's not possible to achieve reasonable quality (cell angles etc.), I would suggest to use a hybrid mesh.

2. you alway use mrf. sliding mesh is for transient simulation, as frozen rotor and stage is for steady state simulation. Most often, you use a steady state solution as initialization for a transient simulation. But in your case, I'm not sure if it's possible to achieve a steady state solution (you've to provide more info).

3. depends on the kind of simulation. generally it's recommended to use 1:1 or fairly equal mesh sizes on both sides of the interface. in case of heat exchange etc. cfx always uses a ggi, no matter if it's a 1:1 connection.

4. sorry, i've no idea as i've never used this approach before

5. is it a closed volume (tank)? where are the inlet and outlet? please post some pics

[DoubleG]

Thank you for your reply!

Yes I realize this problem is rather complex. I just need to do the best possible, and we'll see how it works out. The amount of nodes available depends on my colleagues .

I know that structured is more efficient, though is the hexa also faster if I apply 12 to 1 hexa generation (and thus end up with an unstructured hexa mesh)? For the bulk, I probably will use hexa structured, but I wonder what to do for the refinements.

For homogenization I also need to go to transient analysis, but in this case the flow is steady, only the concentrations develop in time. I assume MRF is still better.

What is GGI exactly? I just found that it is general grid interface. Does it mean that it detaches anyway even when it is conformal?

Actually I use Fluent, but I think it doesn't make a difference.

Unfortunately I am not allowed to post my geometry. Though attached you find a very simple representation. The tank is open at the top. Not drawn is a roof which is just one meter higher. I think that one is not so important. There is a small impeller near the bottom and a big mill, let's say a rushton turbine of about 6 meters somewhere else. There are are even more impellers, and I have different geometries to do, though this is just the principle of the tank. The inlet is at sayone meter from the fluid surface, the outlet is near the bottom. Feed and outlet are not so important, this is often performed intermittent (say once a day, while I have retention times of tens of days), so I'd probably rather not consider them in my model.