[gemxx]

Hi all,

I simulate the flow around a submerged body for acoustic analysis. The analysis converges during steady state, but the residuals, especially x-y-z momentum residuals, are changing rapidly when the unsteady state analysis starts . I attach some pictures about residuals and physical models which I selected.

Please advise as to what the reason of this change on residuals. What can I do to reduce the residuals at the beginning of the unsteady state ?

[lcarasik]

Originally Posted by gemxx

Hi all,

I simulate the flow around a submerged body for acoustic analysis. The analysis converges during steady state, but the residuals, especially x-y-z momentum residuals, are changing rapidly when the unsteady state analysis starts . I attach some pictures about residuals and physical models which I selected.

Please advise as to what the reason of this change on residuals. What can I do to reduce the residuals at the beginning of the unsteady state ?

This is expected behavior. When running unsteady simulations, your residuals will observe a "saw-tooth" pattern. Just ensure the residuals are dropping significantly between each time step.

[gemxx]

Originally Posted by lcarasik

This is expected behavior. When running unsteady simulations, your residuals will observe a "saw-tooth" pattern. Just ensure the residuals are dropping significantly between each time step.

So, it is necessary more converged residuals at steady state to obtain at least 10^-4 convergence for unsteady state. Right ?
residuals are not dropping significantly between each time step. After 100-200 steps , they remains constant. I think that is the problem.

[fluid23]

It could also be a product of your time step. Either too large or too small of a time step and your residuals will fail to converge. You will still see the so called 'sawtooth' pattern but your residual magnitudes should be reducing at least 3 orders of magnitude as well. I suspect if you fix your time step you will get better convergence on residuals.

[Vincent_GDC]

You should not pay lots of attention to the residuals, I think your residuals looks fine. Its the relative error that is important.

[gemxx]

Originally Posted by MBdonCFD

It could also be a product of your time step. Either too large or too small of a time step and your residuals will fail to converge. You will still see the so called 'sawtooth' pattern but your residual magnitudes should be reducing at least 3 orders of magnitude as well. I suspect if you fix your time step you will get better convergence on residuals.

I tried 10^-6 and 6x10^-5 for acoustic analysis. Do you think that it is too small ? The total time of analysis is 0.12 second

[Vincent_GDC]

Don't focus so much on the residuals. It is better to create reports and monitors, from that you can determine if the results are reasonable. Also in the monitor settings set the setting from auto to off. That gives a better understanding of the residuals.

[gemxx]

Originally Posted by Vincent_GDC

Don't focus so much on the residuals. It is better to create reports and monitors, from that you can determine if the results are reasonable. Also in the monitor settings set the setting from auto to off. That gives a better understanding of the residuals.

Thank you for your advises. I'll set the setting from auto to off in the monitor setting. But still x-y-z momentum is high. I add the last residuals in attachment.

I also wonder the mesh structure is appropriate at the sharp bodies. I added two mesh picture and I marked the complex parts of the geometry. Do you think that the mesh structure is good especially at the marked parts ? How can I generate continuous prism layer layers along the geometry ?

[gemxx]

Originally Posted by gemxx

Thank you for your advises. I'll set the setting from auto to off in the monitor setting. But still x-y-z momentum is high. I add the last residuals in attachment.

I also wonder the mesh structure is appropriate at the sharp bodies. I added two mesh picture and I marked the complex parts of the geometry. Do you think that the mesh structure is good especially at the marked parts ? How can I generate continuous prism layer layers along the geometry ?

I also add the last residuals after I change the setting from auto to off.

[gemxx]

Originally Posted by MBdonCFD

It could also be a product of your time step. Either too large or too small of a time step and your residuals will fail to converge. You will still see the so called 'sawtooth' pattern but your residual magnitudes should be reducing at least 3 orders of magnitude as well. I suspect if you fix your time step you will get better convergence on residuals.

When I reduce the time step from 10^-5 to 10^-3, I get better convergence on residuals. The Prism layer near wall thickness is 1.9 mm. If I set the time step to 10^-3, the courant number is 1.5x10^-3 . Do you think that the value of CFL is good for density based implicit solver ? I perform an acoustic analyse. If the time step is 10^-3 , the maximum frequency resolution is 100 Hz. It is too small, right ?