Personal tools
 

Constrained space filling - Move option

In this part, we’ll look deeper into how the space filling point sampling algorithm has been improved in LS-OPT 4.2. Also, the elusive Move option for constraints will be explained.

It’s not uncommon to have optimization problems where simply providing individual bounds for every variable is insufficient, for example geometrical problems where a sum of thicknesses may not exceed a given value. With LS-OPT 4.2 such infeasible experiments can be avoided in the sampling (DOE), while maintaining a good space filling design.

Consider an LS-OPT project the following variable design:

And let’s say we constantly need the variable x to be larger than y for the experiment to be feasible. In order to do this, we need a Composite Expression to help us. These are created in the Response panel in LS-OPT. Here’s our expression, called VarConstr1:

Then, we move on to the Constraint panel and define a constraint on this composite as so:

Note the crucial Move option being active. This makes LS-OPT take the constraint into account while doing point sampling.

This is the point sampling we get after running LS-OPT 4.2 with these settings:

Here’s a more involved example with two constraints, sin(x*90)-y < 0 and
x*x-y > 0. (Did you know that LS-OPT uses degrees instead of radians for trig functions?)

 

Reference: http://www.erab.se/2011/05/30/ls-opt-4-2-part-1/