Learning Forum

[shahriar_011]

Hi forum,

I would like to know how can I efficiently simulate a transient multiphase flow scenario of a vertical pipe having an L/D ratio in the order of 10^4 or so (5000 ft long with a 0.5 ft dia). I might need to simulate around 30 minutes flow. This is an example of an oil and gas well drilling scenario. I have read some papers saying the simulate may diverge due ot the high l/d value. Some even suggested my case may take a month to simulate even on an HPC.

One solution was suggested on this forum (link below) is to divide the pipe into multiple small segments and transfer profile data from segment to segment.

I plan to use a 2-D geometry, probably will go with axisymmetry along the vertical axis, use VOF model, no energy equation at this point. I will have High performance Computer access.

Please try to shed some light on the following points and beyond those if you can:

1) How can profile transfer technique be implemented for a transient case? It will be highly transient (bunch of gas bubbles being circulated), so no chance of SS

2) I can perform dimensional analysis to reduce my length but in that case, the diameter will be super small. What do you think?

3) I came across the idea of creating response surface for different smaller length pipe and try to extrapolate what the conditions could be for 5000 ft pipe. But I don't have good knowledge on that front, so any thoughts?

4) Any other technique you would suggest?

5) I know it is difficult to tell without knowing the configuration of HPC, but just to get a ball park as I have no knowledge using HPC, if I actually do the simulation full-scale with 2-D axisymmetry for 30 minute of flow, how long the simulation may take?

When drilling a well it can reach to lengths longer that 100000ft. The diameter however is less than a meter the whole way down. I am trying to mesh a 0.3m diameter, 30000m long pipe. I used dimensional analysis to scale the product down to 0.0003m diameter and 3m long. However, I am struggling to mesh such a long product.

Has anyone had this occur? Is there a scaling option in ANSYS to scale down the length or to use symmetry to create a long pipe?

[Rob]

If you're expecting multiphase flow regime change then breaking the domain up adds it's own problems. Namely getting the pressure right as the gas expands. This is why the papers are using HPC and modelling the whole length. Divergence isn't a feature of the domain, but more likely a result of the mesh and/or time step. nWhat are you wanting to see in the CFD model and can you reduce the length by using some calculations with a Baker Chart? n

[shahriar_011]

Rob, thanks for your comment. One of the objectives here is to observe how the contaminated mud column stretches (due to the gas expansion) while being circulated bottom up. So, essentially observing gas void fraction change with time at different lengths of the pipe. nCould you please also elaborate further how can I use a Baker chart in this scenario?n

[Rob]

The Baker Chart will tell you the likely flow regime, this can help guide which multiphase model to use and also whether you can ignore sections of the pipe. n

[shahriar_011]

Rob, while I understand the matter of identifying the flow regime, I don't understand how can I ignore some pipe sections. As the gas-liquid column will be traveling bottom up (pumping liquid ), I would like to see its evolution throughout the length and time. In addition my outlet boundary condition will have a varying back pressure. n

[Rob]

Yes, the Baker Chart will allow you to see what regimes you're working with. From there we can decide if we want VOF, Eulerian (multifluid or otherwise) as the multiphase model. We can also look at turbulence damping to pick up the regime change. Regime change will also influence back pressure, so you may well find you need to model longer and longer sections. nFor a 10-50m section I'd use CFD and have done. For 1500m or so I'd use a Baker Chart or OLGA and then use that as an input to a more detailed CFD model. We can model the whole system but would need a very good reason (and lots of computers) to do this. I have modelled sections for a mud displacement project but that didn't have a regime change and we used activate/deactivate mesh to keep the cell count at a manageable level. n