[sam92]

Hi all,

I am trying to simulate the flow around the ahmed body(35 degree slant) using the k-w SST model with wall functions.I have used first order upwind for momentum,k and w for the first 100 iterations then changed to second order upwind.

pressure-velocity coupling- coupled scheme.

The URF's are as follows:

courant number-50
momentum - 0.25
pressure - 0.25
density-1
body force-1
k-0.8
w-0.8
turbulent viscosity-0.8

The residuals were going down fine until i got a error message "Divergence detected in AMG solver: pressure coupling".I have attached the error message and the residuals.
If anyone could help explain what this means and how to solve it, your help will be deeply appreciated.
Sam92

[IceWind980]

Sam,

I encountered issues on a past simulation using the coupled scheme. Typically I'd run analysis using realizable or standard k-epsilon. However, I experienced your AMG Solver failure when running realizable k-epsilon (standard wall functions). I swapped the solver to using standard, and it ran fine (I ran my simulation longer to allow it to converge, 5000 iterations from 2000 for a truck compressor intake duct analysis). One of the senior engineers also recommended to me to run sims using the simplest options (no frills). If you encounter issues, you can alter your set up from there. I am pretty new to CFD so that is just my two cents. I've run second order on mostly everything. Only had to alter relaxation factors for an extremely low flow problem because Fluent couldn't capture the data correctly.

Krystian

[sam92]

Thanks for your reply Krystian. It could be a problem with the coupled solver,I have read in the forums here that it is relatively more unstable compared to the segregated one.But I had done a simulation on ahmed body using the realisable k-e model(with wall functions) using the coupled solver and got no such error.Also in my sim, none of my residuals were going up which could have explained the error.I did some research about it and found that either improving mesh quality or reducing URF's could solve the problem.Regarding my mesh,I checked the skewness and quality and it was ok. I will try reducing the URF's and courant number and see how that goes.Also can I change the solver from coupled to standard in the middle of the iteration and start iterating from there?Will this impact the accuracy of the solution?

Regards,
sam92

[IceWind980]

Sam,

Sure! In regards to coupled, yeah I heard similar around here. One of the engineers said he never uses it because of its specific application and lack of stability. As for changing the solver mid-way, other senior engineers at my company have done this as well. Typically they'd run a simulation for a few hundred iterations using one solver, stop it, and start it again under a different solver locally. If there was a marked improvement, they could then start a "fresh run" using the better solver on a cluster. I've heard usage of journal files that alter the relaxation factors (or sweep them) to find the ideal numbers were also used for trial iterations.

Krystian

[DanielVe94]

Hello !

I´m facing a very similar problem to what you describe here. Could you find a solution for that ? any help is appreciate

Thanks a lot