Learning Forum

[cibi]

Hello everyone,

I am working on a CFD simulation of a design that I would like to optimise usig Fluent + optislang.

The process I have been following is: 1) sensitivity analysis; 2) MOP; 3) optimization wizard on the MOP.

The first two steps are okay but when I run the optimization my results are wrong, I obtain a pareto front like this:

but I know it's wrong because the quantity that I have on the y axis cannot be negative.

I have also solved some points and their outputs are very different from the predicted ones:

I think the only reason for this to happen is the bad quality of the MOP, but if I check my MOP most of them have good quality:

I tried to improve further the avg_pli model but the best I can get is 73%, which I don't think is that bad to give me such wrong values. Also, the avg_ss has good quality but the results are still wrong.

At this point I am not sure what to do.

Does anyone have any advice/ can explain why this is happening?

Thank you in advance.

[Markus]

 

Hi cibi,

in general you can conduct a a) optimization on the MOP or a  b) direct optimization.

You are working on a). Here I would concentrate on the quality of the Metamodels. You already named vol_avg_pli_tip_top with a CoP of 73%.
I would focus improving metamodel with a lower quality like these. Maybe you have outliers, maybe you need more designs, …
If you do not want negative values for a Metamodel of a response, you can set a min (also max) value for each response in the MOP settings, see here:

Then you need to run the MOP again to create new Metamodels.

If you can not further improve the MOP or you are finally still not satisfied with the results, you can  continue with b) direct optimization.
Then each design is evaluated with the Solver.

Best regards,
Markus