[saugat]

Hello All!
I am using Cadence Autogrid/Fineturbo to work on a case of SCO2 compressor. I cannot share much details due to the sensitivity of the project but the issue I am having is: After running around 200 - 300 iterations, I get 'negative density' in multiple domains in run log and residuals shoot up and the flow solver crash abruptly.
The mesh is not as per the best criteria but the mesh generation window looks good enough enough overall and doesnot produce any warnings.

I am looking for a way to pinpoint exaclty what is causing the flow solver to crash. I have seen some suggestions in different forums to look at the residuals but I am not sure what is the best way to do this. Any suggestions would be very helpful.

Thanks,
Saugat

[colinda1]

Hi,

When a computation diverges in such way, I would relaunch it with a number iterations such that the simulation ends just before the divergence.
Then you can look in post processing where in the domain something unphysical happens: are there unphysical temperatures somewhere in the domain? Look at the minimum and maximum pressure and temperature ranges to see whether that could be the case. Then an iso-surface can help to locate these unphysical values. A velocity vector plot or streamlines throughout the domain may also help to identify the cause of trouble.

Best regards,
Colinda

[Hamidzoka]

Hi;
According to the attached picture, divergence takes place after termination of coarser grids iterations. Possible reasons can be: turbulent model, mesh quality and improper initial conditions. Try to locate the point at which divergence starts. This can normally be found through post-processing the case at 1 or 2 iterations prior to divergence. This may give an insight to origins of divergence. Also try more stable turbulence models like SA. More sophisticated turbulence models like SST are more sensitive to mesh quality.
One more thing that needs attention is that achieving a converged solution in case of multi-stage compressors can be an issue and needs a precise setting of initial solution. Improper setting of flow directions, temperature and pressure levels will definitely result in divergence which is not the case in turbines.
Maybe you need a mean-line quick calculation of flow directions as well as P and T, to find a proper initial solution prior to CFD simulation.
You also need to check your material properties.

Regards

[gsgt07]

But how is your mass flow rate is so much different at inlet and outlet

[Hamidzoka]

Originally Posted by gsgt07

But how is your mass flow rate is so much different at inlet and outlet

Inlet and outlet mass flow rates are quite different since the solution is diverging. Divergence is generally associated with unrealistic values in pressure, velocity, density and consequently the mass flow rate.

[gsgt07]

The flow basically is diverging because of implementation of wrong boundary condition or either may be you are using some bad quality mesh.
The selection of turbulence model I think so have no concerned with divergence of solution because in turbulence modeling we are modelling the turbulent viscosity that can be further used in NS equation if you are getting wrong velocity & pressure field that may be because of wrong implementation of BC's or IC's.
If you have any doubt on turbulence model in that case you can run your simulation by excluding the turbulence model means as DNS.

[Hamidzoka]

Originally Posted by gsgt07

The flow basically is diverging because of implementation of wrong boundary condition or either may be you are using some bad quality mesh.
The selection of turbulence model I think so have no concerned with divergence of solution because in turbulence modeling we are modelling the turbulent viscosity that can be further used in NS equation if you are getting wrong velocity & pressure field that may be because of wrong implementation of BC's or IC's.
If you have any doubt on turbulence model in that case you can run your simulation by excluding the turbulence model means as DNS.

From physical point of view you are right. However, note that solver tries to handle a system of equations with different levels of complexity and non-linearity. Different turbulence models, depending on their assumptions and formulations, have terms with different levels of non-linearity. It means that simpler models with less sophisticated source terms will have a better chance to converge while the more complex models engaging different combinations of flow parameters have issues in convergence.
Just as an example, there exist different vesrions for RSM turbulence models. some uses eddy viscosity to model the diffusion term while others use Reynolds stress components instead. The former definitely has less convergence issues, while the later models are prone to divergence. Thus, technically turbulence model has an impact on convergence

Regards

[gsgt07]

In such case you have to check that at which point (location) the maximum divergence has taken place but in Fidelity how to locate that point either may be in log file or even if you know that point then how to find that location in computational domain