[h0rst]

Hello everybody,

I am considering a heated tube that is flowed though by water.

I made simulations in LES with serial and with parallel solvers and found out that there are slightly different results in temperature distribution.

Why is there a difference and can I avoid it? Can I trust the serial values more than I can trust the parallel ones?

Thank you, h0rst

[LuckyTran]

Are you using QUICK?

When you run in parallel the grid is chopped up into many partitions. Only the information at partition boundaries can be shared. So, some discretization schemes (such as QUICK) cannot be used on cells near partition boundaries. In the QUICK, Fluent switches to 2nd order upwind at boundaries.

[h0rst]

I don't know what QUICK is so I guess I don't use it

So as I understand parallel processing can make issues in some cases so I should better rely on the results from serial processing.

[Kummi]

Hello h0rst,
Could you please post your residual plots for serial and parallel processing. Might help to resolve your problem..

[h0rst]

Hello Kummi,

attached you find the plots of each residuals.

After 11 iterations, the results are more or less the same. But if I run 1000 iterations, than there is a bigger difference.

Maybe this has something to do with the fact that I have quite high values concerning convergation at the moment (0,1).


Best regards
h0rst

[Kummi]

Hello h0rst,
I have faced the same error long ago. Guess there is more than one inlet condition in your work.
However, try to initiate your problem with "Hybrid Initialization" and compare your residuals and final solutions..

Thank you..Good day ^^

[CeesH]

Since you are using LES, are the results truly due to the way of solving, and not just due to the finite runtime of the simulation (I assume you are looking at averages in distribution?)

[h0rst]

I've changed to hybrid initilalization and it looks like the difference between the results have decreased. I need to run more iterations to say for sure.

I am running a transient simulation with LES but my model converges to a stationary state, where the heat from the tube and the cooling of the flowed through water reaches an equilbrium.
I am considering the temperature distribution in the heated tube afterwards and compare at the moment the maximum temperature when running in parallel and in serial mode.

[h0rst]

Hello,

I've run many iterartions now and the temperature difference decreases very much to a value that can be easily neglected.

Can you or someone explain the reason for that and if this approach (hybrid initialization) is recommended?


Best regards
h0rst

[CeesH]

In the end, it shouldn't matter, but for simulations with in- and outlets the hybrid initialization method produces a better initial guess for the velocity and pressure fields, as it already solves an approximative equation rather than posing constant values as regular initialization would do. And a better initial guess typically means fewer iterations to converge the problem.

[rudipewe]

Originally Posted by h0rst

Hello everybody,

I am considering a heated tube that is flowed though by water.

I made simulations in LES with serial and with parallel solvers and found out that there are slightly different results in temperature distribution.

Why is there a difference and can I avoid it? Can I trust the serial values more than I can trust the parallel ones?

Thank you, h0rst

Hello Horst,

I have the same problem. I knew from the fluent in my case, when using parallel setting the node position is become counted from smallest y position to the largest. The average value of the pressure (for example) is same, but if we counted the pressure related to the node position, it will be different. So my conclusion, if we considering the value of pressure or something else considering the node position the serial setting is the best and the parallel (in my case) totally become wrong. Unfortunately, serial setting is very slow it takes 5 days for 1 point of data/case meanwhile I can did it only 1 day using parallel setting.

Thank you