/opt module

[andreamordini]

I started using this module but I found out that the maximum number of parameters (design variables) is 60. It is a very small number to perform Finite Element Model Updating.
Can anyone explain why ? (from the theoretical point of view)
Is it possible to use more parameters using some numerical tricks ? (like damage functions for example)

andrea


-------------------------------------------
Andrea Mordini, Civil Engineer, Ph.D.

http://www.andreamordini.com

-------------------------------------------

 

[meshparts]

Hello,

I don't know why the maximum number of DV's is set to 60 in Ansys but this number is not small. In fact it unpractical to perform gradient based optimization with more then 10 variables due to local minimum problems and large computational time.

Think of it: With 10 DV's it will take 10 Solving steps to compute one gradient of just one iteration. Usually well posed optimization problems will converge in say 10 iterations. This means: 100 solving steps. In this calculations I have not included the line search iterations! With large models and 60 DV's it will take 600 solving iterations! And if the problem is not convex you can not be sure if the solution is a absolute minimum.

In fact, you should optimize a few times with different starting points in the design space. This means a huge computational effort it you have a large model an big number of variables. So you should think about using up to 10 DV's at a time, obtain a solution, then optimize with other 10 DV's and so on. This method can not assure an absolute minimum but it's better then optimizing with all DV's at a time.

Hope this helps.

Alex

 

[andreamordini]

Thanks for your feedback but I disagree. I have written a Newton-based FE updating code and I have performed analysis with 500 parameters with very good results (in publication to international journals). The computational time is acceptable. In order to improve it, however, I was thinking to use the /OPT module since an internal module of ANSYS should be faster than an external code where the data are exchanged by writing files.

andrea

-------------------------------------------
Andrea Mordini, Civil Engineer, Ph.D.

http://www.andreamordini.com

-------------------------------------------

 

[meshparts]

That's very interesting... Could you provide a link to your publication? I would be very interested in reading it.

I agree that there are some problems, how can be successfully solved with so much parameters but the Ansys code computes the gradient by solving as much times as variables defined. This means a huge computational time even if you defined "only" 60 DV's. Unless your model has only few DOF's it will take many days to finish the optimization task. I tried it!

I don't know how your code works, but I would be very interested to learn about it, if it can manage 500 parameters.

Regards,
Alex

 

[andreamordini]

I have uploaded one pdf to my website. You will find it in the page Publications, under Conferences:
"VCUPDATE, a numerical tool for Finite Element Model Updating", IOMAC.
In particular, check the last example of the beam using the Plane model.
Always open to discuss.

andrea


-------------------------------------------
Andrea Mordini, Civil Engineer, Ph.D.

http://www.andreamordini.com

-------------------------------------------