[Jiricbeng]

Hi there,


I analyze a simple 2D model of a bearing to verify basic equations.
The issue is as follows:
The bearing domain consists of inner wall, outter wall and 2 symmetry planes. The inner wall rotates 80000 rev/min, the outer wall rotates 30000 rev/min.
1) I set stationary domain and wall velocity inner 80000 and outer 30000
2) I set rotating domain to 30000, inner wall velocity 50000 and outer is default.


When ploting the variable "Pressure" (which is scalar and invariant) around inner wall I obtain different curves in both cases. They are both perfectly converged. The Velocity in Stn Frame on inner wall is the same for both cases, which means that velocity field is the same. The question is why is the pressure different?


Btw, I modified the case 2) by swithing off rotating domain. So I kept the inner velocity 50000 and outer to be zero but domain was stationary. And I obtained the same results of the pressure around the inner wall. It looks as if the pressure was not affected whereas velocity were changed..

[ghorrocks]

Check your convergence in this comparison, and use double precision numerics. I would converge these simulations tighter than normal to reduce numerical errors caused by the large variations in velocities.

[Jiricbeng]

Thank you very much for suggestion. I tried it with double precision but I obtained the very same results. Imbalances are also again zero..
Do you think it is still matter of convergence (e.g. try out finer mesh) or is there some other misunderstanding?

[ghorrocks]

The residuals are more important that imbalances in this case, I suspect.

Making sure your results are properly converged is just the first thing to check. There are others, such as round off errors, simulation set up, mesh resolution and so on.

[Jiricbeng]

The residuals are below E-4 and monitored variables (e.g. force x,y) are very stable, several thousands iterations were done, when changing time scale no change of monitored variables occurs, this is valid for both cases. I found difference in velocity in Stn Frame across bearing thickness which likely is the reason of different "Pressure". I cannot converge more the cases, it seems there is another glitch...

[ghorrocks]

I would converge tighter than 1E-4 to check this. I would also try using double precision numerics to see if round off error affects it.

Can you post an image of what you are modelling and your output file?

[Jiricbeng]

I am sending in the attachment. I am modelling a bearing with eccentricity of several microns. It is a 2D model (I cell in axial direction). I tried also refined the mesh with double precision.

[ghorrocks]

Your domain is 9mm across, but you are trying to resolve the effect of a few microns displacement. That is a ratio of about 1000:1, which means round-off errors will make this simulation difficult, if not impossible by the approach you are using. So I am not surprised you are getting weird results, round-off errors are limiting the simulation.

I would recommend doing some background reading to understand round-off errors in CFD better, and then you will need to revise your simulation approach to reduce round off errors. This will probably mean you model this geometry in small segments rather than one big ring.

[Jiricbeng]

Thank your for advice. Do you have any documentation regrading this issue? I do not know what to improve basically, I refined the mesh, computed double precision, changed time step etc. and the results are the same.

[Jiricbeng]

Thank your for advice. Do you have any documentation regrading this issue? I do not know what to improve basically, I refined the mesh, computed double precision, changed time step etc. and the results are the same.¨


In addition, I analyzed the case in Fluent (double precision, converged below E-7) and obtained basically the same results. I think it is not a round-off error.

The question therefore is, is rotating domain approach fully analogous to rotating walls, in general? Must the results be the same?

[ghorrocks]

Please attach the output file of both runs.

[Jiricbeng]

Please find attached, case2 is rotating domain, case3 is wall rotation

[ghorrocks]

I can't see the problem. I do note that in the stationary domain case you have the inner wall rotating about axis 1.3 but the outer wall about axis 0.3. Are you sure this is correct? Also you should have a look at the results carefully and see if you can see the difference between the runs.

You appear to have a very coarse mesh on this and the residuals are very low compared to normal. This does not bode well. I recommend you try doing this with a small segment rather than the whole ring, as I said previously.

[evcelica]

"I recommend you try doing this with a small segment rather than the whole ring, as I said previously"


I agree. This is the perfect candidate for a periodic model.

[Jiricbeng]

Thank you both for the reply. However, there is eccentricity as I mentioned previously hence I think segment model cannot be considered.

Therefore the inner wall rotates around axis 1.3 and outer wall around 0.3. 1.3 is axis "z" shifted in x direction by several microns. Therefore, maybe, rotating domain approach cannot be used because there are two circles of different center (axis) and giving rotation to the whole domain around one axis may cause the problems.

[ghorrocks]

Looks like you worked it out. Your two simulations are modelling different things because of the eccentricity.

[Jiricbeng]

Perhaps, it might be main culprit