Hi, maybe a question for @Daniel.Zalkind .
If, in WEIS, I enable the analysis option driver:design_of_experiments:, is the constraints: section going to be ignored?
To give my example: I would like to perform a DOE analysis with the IEA 15MW UMaine ss, varying the tower external diameter and the thicknesss of each section, and I managed to do that.
Anyway, sometimes when generating these tower geometries, the slope of some of the sections can be negative (i.e., the bottom diameter is smaller than the upper diameter).
I tried to add design_variables:tower:slope:flag: True, and the WEIS run normally, but I suppose it simply ignores this setting, since I keep getting tower geometries with some sections having negative slopes.
Then I tried adding the slope as a constraint, but nothing happen: it is ignored.
So, my questions are:
- When running a DOE with WEIS, can we specify constraints?
- If not, do you know another way to force the DOE design_variable generator?
Many thanks for your help, and I keep being more and more impressed by WEIS and its huge potential
Hi Maurizio,
Thanks for the kind words! It’s good to see WEIS being exercised like this.
I don’t think the constraints are enforced when a DOE is run. The DOE only changes the design variables. Constraints are usually outputs that result from design variables. Adding the constraints will include them in output logging/plotting.
I’m not very familiar with this part of the code, but it may be possible to enforce that constraint with some code additions.
I’ll circulate this with some others who may have a more detailed answer.
Best, Dan
Hi Maurizio,
I’m a bit confused at what you would like to see happen during the study. The DOE driver runs a parametric sampling of the variables you specify, as if each were a stand-alone simulation. There is no sophisticated outer-loop driving things or determining if a run is “valid”. The OpenMDAO docs page may be helpful.
Constraints are only applied during an optimization run. However, the output variables that WISDEM flags as constraints during an optimization are always present. You can check the value of the various outputs on each run of the Design of Experiments if you are curious about buckling, geometry consistency, etc. You can check the source code here for an exhaustive listing.
Many thanks, @Garrett.Barter , and apologies for the confusion caused.
Ok, I think I understood, let me explain better what I meant.
I agree that, in general, constraints can be checked only after running some kind of analysis, so those cannot be considered when generating candidate geometries.
Nonetheless, some (few) of the constraints can be immediately checked after generation of design variable vectors (an example being checking if the generated diameters for the tower truncated cones always have the bottom diameter larger than the upper diameter, i.e. imposing a convex outer surface) prompting an immediate elimination of the generate candidate (and therefore preventing unnecessary analyses), and the generation of a new one.
But I understand that this can complicate things, since it is applicable only for few, simple constraints.
Hi Maurizio,
The capability you describe is not available. A complete WISDEM, and possibly WEIS, evaluation would have to be run to generate the constraint values anyway, so checking against the output parameters of interest is something you could do manually. There is no closed loop feedback to generate new input samples if a constraint is violated, sorry.
Many thanks, @Garrett.Barter, for taking the time to confirm this. Still a powerful tool, looking forward to applying it in my research.