[juan4815]

Hello,
I've been working with CFX on the ventilation of engine emissions in a maintenance workshop and I encountered a problem regarding the residuals during the simulation.

During the steady state solution (for the initial conditions of the transient one) everything is fine, but when working the transient sim the residuals (momentum, turbulence variables, imbalances) oscillate highly with periods of 0.05 sec to 0.1 sec approximately.

The solution in large timescales, over 30 seconds, it's fine and doesn't have any abnormalities, so apparently it's not a problem of convergence of the solution (topic that is covered in related threads, which I reviewed). I tried changing some parameters like min time step, incresing time step factor, max time step, without any appreciable effect.

I think it may have something to do with the numerical scheme being used, but I'm not entirely sure how to approach this problem as I'm fairly new to CFX (and to this forum too, I hope this is enough information as a starting point to a discussion, I didn't know what other info to give at this point).

I appreciate any information o suggestion you could give to understand and eventually fix this issue.
Regards

[Opaque]

A snapshot of the residual plot would help.

For transient simulations, the residuals nearly always go up every time step until it converges within the timestep. To see that you may need to activate monitoring within the coefficient loop.

If the flow is unsteady, the residual at which iteration is deemed converged can be different every timestep leaving something else to smooth out in the next timestep. Have you tried lowering the target residual a bit further (recall to increase the maximum number of coefficient loops as well) and see if the oscillation diminishes?

In theory the solution of a given timestep is achieved when the residual is 0; however, that is not practical and a compromise is used, i.e. target residual. Once there is a compromise, the question is how large can the target residual be and the transient solution still be representative of the physics being modeled and not become dependent on the selected value.

Hope the above helps,

[juan4815]

Here I have a picture of the behaviour that I mentioned.

Opaque, right, it may be what you mentioned I'll try changing those parameters. It's not happening for all cases (dimensions and boundary conditions change within a fairly narrow range), so maybe not all situations use many loops to achieve a certain residual target.

[Opaque]

For this particular plot, your steady state solution is far from converged and the transient simulation is smoothing out the incomplete solution and trying to reach some level of convergence.

Note the mean level of the transient residual oscillation is the same as the steady state before stopping that run.

[evcelica]

Looks to me like the time step is too large. How do you know what you chose it is appropriate?

[juan4815]

So I tried letting the steady state simulation run longer so it could achieve a better level of convergence, and that solved the issue finally. Initially I used only 100 iterations, but in reality around 500 were needed, this was not done before because of time restraints. Thanks for the assistance Opaque.
evcelica, the time step used followed some guidelines given past experiences on this kind of problem, and making it smaller (2 orders of magnitude) didn't solved the issue apparently.
I hope this info is useful for someone else in the future, I'll give more if needed eventually.
Regards