[y__tb]

Hello, guys. I'm a beginner of Ansys fluent and need some advice.

#I'm using Workbench 2021 R1#

I have several questions about simulating a high-pressure valve flow. The reservoir pressure is around 200 bar, so the flow becomes supersonic when the valve opens. The flow will be choked after the expansion, then I’m going to calculate the actual (or maybe approximated) mass flow rate along with the valve opening. Eventually, expressing the exhausted mass flow rate as a function of (or, a profile of) the valve opening area would be what I want to do.

So, for the first, I divided my geometry into about 30 bodies to use the sweep method, which is the most favorable method for CFD simulation as I know. Then I finally made a mesh whose cells are consisting of hex (except one body). It should be noted that I didn’t apply the ‘hex-dominant’ method (of course anybody would not be confused). But, I realized it is quite hard to resolve meshes near-wall when I use the sweep method despite considering a yplus value of 100 ~ 300, not 1. Applying bias on each edge also didn't be an adequate way. The inflation method made my work (w 30 bodies) all the way tetra, even the wall meshes were not resolved well. Layers were not accumulated adequately.

So, that's the situation so far, and here's my question.

1. Is it okay if meshes near a wall are not resolved ‘at all‘? (= or, what happens if I do not consider meshes near the wall at all ?)

As mentioned before, the main concern is calculating the exhausted mass flow rate under the condition of a certain valve opening area. So, if the quality of mesh near-wall does not affect the value of mass flow rate that much, well-defined near-wall meshes are not necessary to me. I know that it is way important to consider meshes near the wall to simulate the ‘actual viscous effect’, but I’m afraid that it will take countless time to get the exact mesh.

2. Then, how about using fully tetra–mesh with inflation?

Would making a whole mesh in the way that Ansys Meshing do automatically, all tetra, better ? Near-wall meshes were constructed far easier in tetra-mesh case by applying inflation method. However, the mesh quality criteria were not satisfied well of course.

For reference, the prior mesh (with 30 ‘swept’ bodies) satisfy mesh quality criteria such as max skewness ~ 0.75 and min orthogonality ~ 0.25 (I know that it could be satisfied because the wall mesh that requires harsh quality criteria conditions were not considered in this case).

3. Convergence issue : An additional question

I have conducted simulations with the mesh (w 30 bodies), and currently struggling with the oscillating residuals. It’s in the range of ‘ 1e-2 ~ 7e-2’. In some cases, The oscillating range goes occasionally around 1e-1. Though an acceptable mass flow rate is observed, I want to know if such oscillating phenomena occurred because of the poor quality of mesh around the wall. If not, what would be my next step ?? I can not ensure the results because of the oscillating residuals.

The issue was also not resolved in the case of ‘reducing relaxation factor of momentum and pressure to 0.1’, ‘applying first-order method on each scheme’. I'm using a pressure-based solver with k-epsilon turbulent model since solution with SST diverged. Note that the mesh quality is written in section 2.


I would be appreciated if someone gives me any advice. Have a nice day.
P.S. Some expressions could be weird since it's not my native language. Please understand.

[AlexanderZ]

you may try following:
1. use fluent mesher, which is more handy and rather powerful now.
2. instead of tetra–mesh you will generate polyhedral mesh, which will save you computational resources
3. as you have a huge pressure gradient you may try to simulate your case in transient, but use small enough time step. The flow suppose to pass your domain twice to consider the flow to behave as steady-state
4. you don't really need to resolve boundary layer for this case from my point of view, so y+ ~ 150 should be very reasonable

[y__tb]

Thanks for your reply, AlexanderZ.

I'm trying to do the same job in Fluent Meshing as you said, and it seems further easier to get a well-resolved near-wall mesh. I'll let you know if I get some reasonable results. And also, I'm observing more reasonable results in a transient simulation with the same mesh though more iterations seem to be needed yet. Thanks.

But, I still have further questions about your yplus suggestion, which was about 150. What if yplus value is 1000 or larger (for an example)? I mean just using "meshes similar with the internal meshes" as "near-wall meshes". Such as using meshes with an aspect ratio of close to 1. Will it give me horrible results? Or just results with some accepctable errors that occur in the process of interpolation near wall?