How to manually set max iterations on Design3D

The bad news: when the simulation stops at 218 iterations is because it diverged. You can see the residual for the continuity equation goes from -1 to +2, with the maximum going to +40. 

The file job_euranus stores the content of the *.std file produced by Fine Turbo. You can also open a monitor and look at the residuals of the simulation to get a better idea of what's happening. By looking at the files, it seems that the max residual is in block 5, which should correspond to the leading edge of the impeller (not betting any money on this).

The good news: there are a few tricks to make the optimization faster. My favorite option is to restart all simulations in the DoE from an already converged solution and "reset the convergence". Supposing that the geometry changes a little (a few degrees for the blade angles, something along these lines), the flow field won't be too dissimilar, therefore it represents a great initial solution for the other designs.

This means that the convergence level is not going down too much (maybe to -2), so better add a few iterations more to be sure to have a converged solution.

My favorite approach:

- run a geometry and get a converged solution

- create a _restart_ computation from the previous one: this is the computation that I'm going to use for the optimization

- in Fine Turbo, the convergence level is -10, the max number of iterations is to be chosen appropriately. The solver will run until reaching the max number of iterations, since the residuals won't go to -10.

- in Fine Design3D, go to Simulation settings, turn on expert mode, set the convergence level higher than -3. Fine Design3D will consider that all simulations with the job_euranus showing a last line with a residual higher than the number you have there.

The bad news: when the simulation stops at 218 iterations is because it diverged. You can see the residual for the continuity equation goes from -1 to +2, with the maximum going to +40. 

The file job_euranus stores the content of the *.std file produced by Fine Turbo. You can also open a monitor and look at the residuals of the simulation to get a better idea of what's happening. By looking at the files, it seems that the max residual is in block 5, which should correspond to the leading edge of the impeller (not betting any money on this).

The good news: there are a few tricks to make the optimization faster. My favorite option is to restart all simulations in the DoE from an already converged solution and "reset the convergence". Supposing that the geometry changes a little (a few degrees for the blade angles, something along these lines), the flow field won't be too dissimilar, therefore it represents a great initial solution for the other designs.

This means that the convergence level is not going down too much (maybe to -2), so better add a few iterations more to be sure to have a converged solution.

My favorite approach:

- run a geometry and get a converged solution

- create a _restart_ computation from the previous one: this is the computation that I'm going to use for the optimization

- in Fine Turbo, the convergence level is -10, the max number of iterations is to be chosen appropriately. The solver will run until reaching the max number of iterations, since the residuals won't go to -10.

- in Fine Design3D, go to Simulation settings, turn on expert mode, set the convergence level higher than -3. Fine Design3D will consider that all simulations with the job_euranus showing a last line with a residual higher than the number you have there.

Hi Domen,

Thanks for the advice and sharing your approach. I would like to clarify a few things regarding it:

1. for the ''create a _restart_computation from previous one'' - does this mean creating a new fine turbo project and using the run file from the first converged solution as the initial solution (in fine turbo, going to the Initial Solution> from file> selecting the run file of the first converged simulation)?. If that is the case, you then change the convergence to  -10 for this second simulation so the number of iterations becomes the stopping indicator for the simulation

2. For the design3D case, do you actually mean lower? like -4 etc. Besides more result fidelity, is there a reason to lower it in this scenario? 

1. yes, perfect!

2. No, I'd set it higher (+3). I let Fine Turbo converge (or not), and not Fine Design3D assess the convergence. 

Alright thanks, will try it out and let you know the outcome

Hi Domen,

I've encountered a few issues that i'd like a second opinion on if possible:

1. The Screening simulation's final residual at my imposed 6000 iterations was -2.4. Something seemed usual as the 3d associated computation I used reached residuals of -4 at this point. I'm still looking into how this discrepancy occurred.

2. Setting the requested convergence level to 3 and allowing 6000 iterations to be the stopping point has been resulting in more converged solutions in the database, thanks for that tip! However, upon checking the residuals via the Monitor module, I noticed that some residuals only reached a minimum value of -1.6 but was considered a successfully converged solution in the database:


Using this method of setting the iteration as the stopping point, my question is does this imply that any simulation in the database will be considered successfully converged as long as it reaches the 6000 iteration mark?

3. When using the Monitor to check the residuals progress with each database simulation, I noticed that there was a block button and I assumed the displayed residual corresponds to the selected block. If this is the case, then how would one assess convergence; would each the residuals in each block need to be checked? 

Sure, here my thoughts:

1) ok, let me know how it goes your investigation.

2) that's pretty normal. Fine scales the residuals to the ones at the first iteration, so the convergence depends on what happens at iteration 1. We rather look at the global data (efficiency etc) and add a control point if needed. When we restart, the scaling can even produce even residuals that are a few orders of magnitude bigger, (+1, +2).

3) you can see residuals for each group if you enable this option in the Control variables box: 

If the simulation doesn't converge well, then I'd look at where the residuals are higher in the Monitor. Otherwise, if the convergence curves for efficiency/pdiff are ok (flat, not oscillating, stable) I'd consider the run ok. Watch out: any oscillations in the p_diff can create _a lot of_ noise on the surrogate model.