[funmaker]

Hi i try to simulate a rotating gear-wheel in a box.
The box is filled with oil so that the lower gear-tooth are dipped.

The wheel is rotating with 100rpm.
The grid consists of 3.5e6 elements.
For my simulation i used the following approaches:

Number of Elements: 5.5e6
Simulation Type: transient
Total Time: 1 s
Time Step: 1e-3 s
Turbulence Model: SST
Multiphase Model: Homogeneous
Surface Tension Coefficient: 0.028 N/m
Surface Tension Model: CSF
Interphase Transfer: Free Surfacce
Maximal Coefficient Loops: 15
Residual Type: RMS
Residual target: 0.003
Conservation Target: 0.01

if i use for volume fraction control the option "coupled" i got the following result for the imbalances:


if i use the option "Segregated" i got the result:


For the first case the maximum number of coefficient loops were used in 20% of all iteration steps.

From this point i got the following questions:
Why does the conservation target did not work for the oil-phase?
Could i say the the segregated option works better the the coupled?

Some guy told me that the Imbalances in a closed system without inlet and outlet are difficult to benchmark, because a real reference value is missing. Is this true?

I would be very happy if someone could help me with this topic.
Best regards
Mat

[ghorrocks]

For free surface simulations with surface tension you need to use VERY small time steps. Use adaptive time stepping, homing in on 3-5 coeff loops per iteration and let it find its own time step size.

And are you sure surface tension is significant? It is making your simulation much harder and slower. If it is not significant then do not model it.

[funmaker]

Hi ghorrocks,

thanks for your advice, i will try to use adaptive time steps.
What about the imbalance reslutls i got? Do you know if i can trust the results i got?

Best regards
Mat

[oj.bulmer]

Why does the conservation target did not work for the oil-phase?

Well, if I understand correctly, the imbalance plot for segregated option as you mentioned, doesn't contain the oil imbalance at all. So one can not understand how it performed.

As Glenn suggested, the adaptive timestepping will definitely help in arriving at a timestep for smooth oscillations with imbalances coming within the conservation target you specified. Not sure how you arrived at a value of 1e-3. A lot of aspects are involved in deciding the appropriate timestep that allows imbalances and residuals to reach their targets and one may need to find that out by trial and error. Adaptive timestepping is just easier. Make sure you monitor the progress to make sure timestep is not becoming unreasonably small, otherwise it will take a long time to have any sensible results. In which case, you may need to relax the residual/conservation target over an informed decision.

Could i say the the segregated option works better the the coupled?

With coupled option, the Solver uses implicitly coupled equations for velocity, pressure and volume fraction. This is beneficial for buoyancy dominating problems and (probably) for free surface as well. In many cases, one needs to reduce the timescales with these problems if they insist on segregated solver while coupled option can work with relatively larger timesteps.

What about the imbalance reslutls i got? Do you know if i can trust the results i got?

As long as the imbalances and residuals monotonically reduce to conservation/residual targets you are happy with, at the end of every timestep, you should be fine. Your plots don't exactly beam this very clearly!

OJ

[funmaker]

Hi,

thanks for your reply.
Right now i started a simulation with adaptive time step. If i get results i will let you know if it worked.

you may need to relax the residual/conservation target over an informed decision

What do you mean? Should I use increased target values?

In the end of a simulation i plot the wetting of the gear-wheel in order to compare it with other simulation results.
Is this quantity meaningful?
Which other properties could i monitor to get information about the convergence?

Best regards
Mat

[ghorrocks]

You seemed to have ignored an important comment I made previously - "And are you sure surface tension is significant? It is making your simulation much harder and slower. If it is not significant then do not model it."

Are you SURE you need a surface tension model? This model will be MUCH easier without it.

[funmaker]

Hi ghorrocks,

sorry i have not commented it.
A simulation without surface tension is already running.
It seems that this simulation is much faster than the same with surface tension. Thanks for your hint.

But actually the simulation is about oil-lubrication an wheels, so if i want to include wall adhesion i have to use surface tension. Am I right?

Best regards
Mat

[ghorrocks]

If you are looking at the adhesion of the oil onto the rotor then you will need surface tension.

If I was doing this simulation I would do it in Fluent. Fluent handles surface tension much better than CFX, so the simulation is likely to run much faster (like x10). But you will have to do a lot of tuning with Fluent to get this.

[funmaker]

I read about the advantage of fluent according this issue.
But i have to finish my thesis until the middle of april, so i don t know if i have the time to try this in fluent.

Are the imbalances importent to me in a closed system without inlet and outle?
Some guy of ansys cfx told me that the imbalnace values are useless due to a missing reference value.
Is he right?

Best regards
Mat

[ghorrocks]

The imbalances are always important. They are one of several measures of convergence. Large imbalances mean you are not converged. But you have to be careful of systems where there is only one contribution to the imbalance. Then any numerical noise becomes 100% imbalance.

"useless due to a missing reference value".... Hmmmm, as I said in the previous paragraph, if you only have one contribution to the balance then yes, it is useless. But almost always you have more than one and then they are important. And for CHT simulations they are critical.

[oj.bulmer]

This is an interesting point, Glenn. Never looked at it that way.
But what do you mean by only one contribution to the imbalance?

OJ

[funmaker]

I have the same question like OJ.
What is the meaning of one contribution and if under these circumstances numerical noise becomes 100% imbalance, how can i avoid it?

Mat

[ghorrocks]

If you have an imbalance which has only one contributor then yes, convergence on imbalances is a bad idea. Then converge on residuals only.

[funmaker]

Ah ok,
i my case discribed i have two domains with a transient rotor-stator interface. So i have two contributions to the Mass-Oil-imbalances. Am i right or is there only one contribution?

Domain 1

Neg Accumulation -5.18E-03
Domain Interface 2.475E-04
___________________________
Domain Imbalance -4.9325E-03
Domain Imbalance % -77.8861

Domain 2

Neg Accumulation 6.3329E-03
Domain Interface -2.4749E-04
___________________________
Domain Imbalance 6.0854E-03
Domain Imbalance % 96.092

Both Domain Imbalance values are normelized with the highst value (6.3329E-03) in this time step to get the percentage.

If i compare this absolut Domain Imbalances values to other simulations, like a pipe flow simulation with an inlet and outlet. I got almost the same absolut Domain Imbalance values, but the percentage is much smaller, caused by a higher value for the normalization.

Could I say that in the gear-wheel simulation the value for normalization is unsuitable and that i have to look at the convergenz on resuidual insteat of looking at the imbalances?

Best regards
Mat

[ghorrocks]

I should add that the single contributor thing only applies to steady state runs - for transient runs the change of the quantity integrated over the whole domain adds another term so you always has at least two. So if your run is transient and you are getting bad imbalances then you are not properly converged.