Learning Forum

[Zhongrui Liu]

I'm trying to simulate a microfluidic device in 3D Since it has unique wall materials on different inner surfaces. I looked up many tutorials but they are all in 2D. Every time I run the calculation y velocity and residue go sky high and end up in a floating point error. Any suggestions? Many thanks.

[NickFL]

Modeling wise, there is nothing different between 2D and 3D. I am not sure of which tutorial you are speaking of, but the traditional inkjet drop example (2D) is VERY good. 

Look at your results as the solution is progressing. Do the intermediate results make sense? Where are the problem areas in the domain? Are these accurately resolved?
Are you using the NITA (Non-iterative time advancement) approach? If you allow 30 iterations at the early timesteps, does it help it to converge better? 

Usually, it comes down to using a very small timestep and several iterations per timestep at the beginning. Later when we have a more established solution can we be more aggressive with the time step size and moving to NITA.

Good luck! These type of problems are never easy.

[Zhongrui Liu]

Thank you so much. The residual almost blowup immediately in very few timesteps. Could it be a mesh problem? I have a mms wide and ums thick channel and I've been sizing the faces. If I size the body it won't get small enough before creating too many nodes and crash (I'm using student version). 

[NickFL]

It could be a mesh problem. And keep in mind that we are also discretizing in the time space, so a smaller time step may help. What are you using for the time advancement? Allowing multiple iterations per time step is usually necessary.

If you are mesh size limited, I would try to reduce it to a 2d planar or an axisymmetric case. It may not give you the final results you want, but it will give you a way to quickly test the necessary mesh and time step sizes. From there you can estimate what the 3D mesh size will be, and the time step will be similar to what you used in the 2D case. From there you can judge if it is even possible to solve with the license.

[Zhongrui Liu]

I tried different timesteps from 0.1 to 0.0001. Iteration 20 for each step. Nothing seems to work yet. I'll be trying 2D setup and see if that works. Thank you so much for the suggestions!

[Rob]

Post a few images of the mesh - knowing what you're doing tends to help provide answers. 

[Zhongrui Liu]

I have a channel that is 50um thick, 3mm wide, and 30mm long with 1mm wide inlet and outlet vent on the sides. I initially want to simulate how liquid enters one of the inlets and seeps through the channel by capillary forces. However, the inlets are huge compared to the channel, and I think the mesh on the thin channel is messed up or too large. I moved to 2D, reduced channel length (so I need fewer nodes and can use finer mesh), and tried multiple methods to mesh. The only one that made it to completion without floating point error is the triangle method shown above but the results are pretty weird -- at least compared to what I see in the actual channel in the lab. I suspect the mesh is simply too large but I don't know any better ways to refine it without using too many nodes.

I appreciate your help! Hopefully, this conversation can help others trying to do similar things.

[NickFL]

You are going to want a more resolved near wall mesh. To get the proper angle of contact at the wall surface you will need several elements to capture this gradient. Six across the channel will likely not be enough.

 

The other part that could be leading to difficulty converging is the initial condition. To what area are you patching the solution?

[Zhongrui Liu]

I patched the inlet cylinder from the top to y=50um which is top of the channel. I also set inlet fluid volume fraction to 1.

 

Since I have a very wide inlet and relatively thin channel I'm having trouble creating enough nodes within the limit of student version 

[Rob]

As NickFL has noted you need a lot more mesh, and I tend to favour pure hex in these models where possible. It's also more efficient in terms of cell count. With capiliary driven flow you're also looking at resolving the wall adhesion, and the increased cell resolution (smaller cells) also tends to reduce the time step you can use. These models are fairly easy to set up, but significantly less easy to get good results from. 

Student is intended to learn how the software works and aid you in your studies, it's not there to replace the Campus licences so if you can't complete the work on Student you may need to speak to your supervisor about getting access to the Research licence as that doesn't have a cell count limit (hardware & parallel licences will give a sensible limit).