[Robert_B]

I am trying to simulate steady flow through an axisymmetric pulse combustor. I have constructed a mesh, constituting a 90 degree slice of the axisymmetric geometry. At the two faces where I have "sliced" through my combustor, I have defined cyclic boundary conditions. I am struggling to match these two cyclic boundary regions to each other.

My mesh is tetrahedral, with no extrusion layers on the two sliced faces (except at their edges where they meet adjacent non-sliced faces).

I am using arbitrary matching, as the boundary cells on each of the two faces do not explicitly match. One face has 65526 boundaries, the other has 64227 boundaries. The match type is regular.

I specify an offset of 90 degrees (under a cylindrical co-ordinate system) between the two regions to be matched. My tolerance is the default value of 0.01.

I start the matching process, and once finished StarCD tells me that "9027 MATCHES HAVE BEEN CREATED". I confirm that 9027 matches have been created by checking in Lists>Cyclic Sets.

When I try to validate my model (Analysis preparation > Model validation), I encounter problems. When I validate my boundaries, I am told "BOUNDARY CHECK PASSED SUCCESSFULLY".

But when I validate my cyclic sets I get a lot of error messages - almost one message for each cyclic set. (There are some cyclic sets that do not generate any error). A typical message reads "**ERROR - BOUNDARIES 309696 AND 373966 OF CYCLIC SET 9027 DO NOT OVERLAP".

I have tried raising the matching tolerance to 0.1 and then to 1.0, but this seems to have almost no effect. Almost the same number of matches are created (9043 with tolerance = 1.0), and I still get the error messages when I try to validate.


Can anyone offer any help with this? I have never used any CFD software before StarCD, and I have been using StarCD for about a month, so I am a novice to CFD.

Thanks
Rob

[Pauli]

Sounds like you are doing it correctly.

Here is what I'd do next:
1) Visually verify the boundary faces are complaining about. For example, plot boundary 309696 and 373966. Are they close to one another?
2) Verify you do not have duplicate boundary patches within the two regions. I'd just delete & recreate the two regions.
3) Send it to support.

[Robert_B]

Thanks Pauli. Unfortunately I had already tried steps 1 and 2, without success.

1) Yes, they are close to each other - in fact they should lie on top of one another when you project the first cyclic face onto the second one.

2) I've redone the regions several times. I have now also reconstructed my mesh (my supervisor told me that my original one was too coarse), and I am having the same matching problem with the new mesh.

I haven't sent it to support. Instead, as it is a 90deg slice of an axisymmetric problem, I have replaced the cyclic boundaries with symmetry planes. This seems to work fine thus far.

Cheers
Rob

[bjtuzhang]

Originally Posted by Robert_B

Thanks Pauli. Unfortunately I had already tried steps 1 and 2, without success.

1) Yes, they are close to each other - in fact they should lie on top of one another when you project the first cyclic face onto the second one.

2) I've redone the regions several times. I have now also reconstructed my mesh (my supervisor told me that my original one was too coarse), and I am having the same matching problem with the new mesh.

I haven't sent it to support. Instead, as it is a 90deg slice of an axisymmetric problem, I have replaced the cyclic boundaries with symmetry planes. This seems to work fine thus far.

Cheers
Rob

The cyclic boundaries is different with the symmetry plane. It will influence your results.
You would like to generate the mesh by ICEMCFD and then match the cyclic boundaries with integral. The tolerance is relative to the unit you have chosen. The larger tolerance is better for the uint(mm) and the smaller tolerance is better for the unit(m).

[Robert_B]

Originally Posted by bjtuzhang

The cyclic boundaries is different with the symmetry plane. It will influence your results.
You would like to generate the mesh by ICEMCFD and then match the cyclic boundaries with integral. The tolerance is relative to the unit you have chosen. The larger tolerance is better for the uint(mm) and the smaller tolerance is better for the unit(m).

How exactly is using cyclic boundaries any different to using symmetry planes in this case. I am modelling a 90deg slice of an axisymmeteric system with no swirl. Surely I should get the same results using symmetry planes or cyclic boudaries on my sliced faces?

[RobertB]

I would say that simple theory would say you were right and if you took your final symm plane solution and swapped boundaries then you might retain the same solution.

During startup it is very easy to generate a swirl component when the flow solution is getting going. Typically there is nothing to dissipate this swirl and it can become self reinforcing.

As a practical consideration it almost always works better if you minimize the domain to its most basic level, such as making what could theoretically be cyclics be symm planes.