Info about the GRSM algorithm?

Hello,

I am doing a master thesis regarding structural optimization and I'm wondering if there are any proper information (in addition to the one in the help info) about how the GRSM algorithm in Hyperstudy works?

Things of interest that I'd like to know:

The DOE it creates after each iteration
The type of fit it does for its response surface
The actual search algorithm on how it finds the optima on the current response surface and how it handles constraints.

/Rhodel

Find more posts tagged with

English

HyperStudy

Accepted answers

All comments

Rahul Rajan_21763

Did you look in to online help.PFA GRSM description from online help.

Unable to find an attachment - read this blog

Altair Forum User

Did you look in to online help.PFA GRSM description from online help.

Yes, however, I did not think it was a sufficient description of the method.

Unable to find an attachment - read this blog

Rahul Rajan_21763

In addition of the information in the Help about usability and settings, we would like to share with you some suggestions and good practices you may want to use. A general suggestion in HyperStudy is to accept the default settings, which work optimally for most applications.

So, do run GRSM with 50 evaluations and then observe the system’s improvement:

If it is sufficient in your case (objective goals are already met or further improvement seems unlikely), then there is no reason to add more evaluations;
Otherwise, you can leverage in the existing 50 evaluations and ask GRSM to run for an additional set of evaluations. You can copy the first optimization and add the existing data as an Inclusion matrix; the included runs will be used to build the initial response surface and optimize on it. As a result, you can notice in the Iteration History tab that Iteration_1 provides the same optimum as the best solution from the 1^st optimization. It is like you start the 2^nd GRSM from the best solution of the 1^st one. This process could be repeated…

Another suggestion is to observe the Inputs vs Iterations. For instance, the plots below show that the bounds of the design space are reached and may be a better solution could be found by relaxing the bounds. If you want and can, do run a 2^nd optimization with new bounds. And don’t forget that if you change your bounds to still use the previously run data as an inclusion. The information from those points is still valid even if the optimization formulation has changed. This is why we sometimes refer to inclusion as recycling data – it is more than a restart!

<?xml version="1.0" encoding="UTF-8"?>

Mr.Alb

I have the same problem. Is not possible to know how the algorithm works in GRSM method? There are a lot of different algorithms, for instance:

Line-Search Approach
Trust Region Approach
Simplex Method
Newton’s Method
Quasi-Newton Methods
Conjugate Direction Methods
Levenberg–Marquardt Methods
Elimination Methods
Lagrangian Methods
Active Set Methods
Penalty and Barrier Function Methods
Sequential Quadratic Programming
Mixed Integer Programming
NLPQLP

Many Thanks

Diana_38240

Hi Rhodel,

Thank you for your interest in HyperStudy and GRSM.

I thought you may be interested in looking at this white paper dealing with HyperStudy optimization algorithms benchmark:

https://www.altair.com/resource/benchmark-of-hyperstudy-optimization-algorithms

Kind regards,

Diana