differences in optimization solutions from R12 and R14
Altair EmployeeHi,
I ran the same optimization problem with optistruct R12 and R14 and get 2 (very) different solutions.
indeed the optimization converges at the 44th iteration with R12 whereas it converges at the 15th itération with R14. visually both solutions are very different and the R12 one is much more acceptable than the R14 one which does not look actually converged...
I've read the release documents in order to understand this discrepancy and have found almost nothing.
can it be linked to the TOPDISC parameter which has been introduced in R14 (my problem includes a symetry constraint which is a manufacturing constraint)?
About this parameter I read 2 different things.
In the document 'OptiStruct_14.0_User_Guide.pdf' it is written 'Beginning in V14.0, an improved Topology Optimization formulation is used that aims at producing more discrete results for all topology runs including manufacturing constraints. This formulation is activated by default and you can revert to the previous formulation using OPTPRM,TOPDISC.'
In the document 'OptiStruct_14.0_Reference_Guide.pdf' it is written 'TOPDISC YES or NO Default = NO'
I attached both .out in order for you to have a look. It appears than from the 1st iteration both calculation part ways.
thank you for your help
ben
Answers
-
0
-
Hello @benf
The difference between two jobs is not just the version of OS. One is Linux and the other is Win. Also, v12 job is SPMD + SMP while v14 is SMP only. The optimization is very sensitive to any small numerical round-off and even any tiny difference in analysis caused by different version, different run options, machines etc…could lead to different path in optimization.
As far as the status of optimization in out file, both jobs seem to be converged. The objective function is even lower with v14.0.
If the concerned about the density results in topology, you can ask try to lower objective tolerance with DOPTRM,OBJTOL. turn TOPDISC on with v14.0, which could help produce more discrete results.
0 -
hello Prakash Pagadala ,
thank you for your answer.
indeed both calculations ran on different machines as we are experimenting issues on our Linux computation station...
I will make some experiments with the parameters you pointed out as soon as we solved these issues.
Ben
0 -
Hi Ben,
Sure. Let us know your findings
/emoticons/default_smile.png' srcset='/emoticons/smile@2x.png 2x' title=':)' width='20'>
0