Hi,

I am doing a study around and inside a front rotating wheel of a car and have some thoughts about convergence.

I get velocity residuals (rms) below 4e10-3, pressure residuals below 10-5, k and e residuals below 10-3. I have ploted too the SCx (or SCd) and SCz (or SCl), the Surface Drag and Lift Coefficients. I was thinking of making a statement of convergence.

As it would take days for all my residuals (still rms) to go down 10-4, let's consider SCx and SCz. I have checked their value at every 50 iteration step.and I get variations below 0.05% in the end. For me it looks like converged. Maybe I could check the standard deviation? But with which criteria? arbitrary?

I am not focus on one particularity of my flow (separation, etc...) but mostly its behaviour, so its difficult to judge on a specific criteria.

Thanks a lot for a good advice!

What is your computation criteria? Steady? Unsteady? what else?

The best converged solution is when the residual is ZERO... but it is not going to happen. You set how accurate you want it, and make the trade-off... between computing costs and accuracy required. I like to use the residual of the density to at around 10e-5... The lower, the more accurate, the higher this residual the less accurate. But so are the computing costs...

Something looking converged at macro scale can look unconverged at micro scale... Looking at the percentage difference is a good practice for lift-drag values.

regards

Of course I know that the best converged solution is when residuals are zero...

I think what I have done for SCx and SCz is what you are talking about (percentage difference), I gave you the numbers. But for you what could be a 'good' percentage?

Besides I will check the density I you said.

Thanks!

Yes I forgot,

It's a steady simulation, using RNG k-e, constant property fluid (so check of density might be useless...), constant temperature, 'real' atmosphere boundary condition at inlet (not a low turbulence intensity level and not a low length scale as in a wind tunnel)

regards

Convergence has many properties...

One is the convergence of residuals... Another is when the changes of particular values of a particular steady computation does not change significantly with respect to the previous values... as you did...

But is the level still enough? If you are doing accurate research, we should know the validated values... that is easiest... if not... do a convergence study, and make sure the changes of, say grid refinements, do not change the values of lift-drag coefficients, for example... Compare this to the experimented value when available... Then you will have roughly the idea of your convergence criteria...

Basically, compare to experiments are best... at least one configuration, so that you roughly know the errors that the rest of your computations do...

I don't know whether this could help, but I hope it does.

regards Hafidz Rahmat