optimization weight of a steel bridge, ansys

[19sbirro87]

Hey, guys.
I have to optimize the weight of a bridge with rectangular beams (link8 type elements) loaded with 1000 tons (equably distributed on each node) respecting the constraints on the maximum shift of each node[umax<10cm]and the maximum stress. I must also respect a constraint on the section of each beam: the sections must be between 1cm^2 and 400cm^2.
I built the bridge considering the starting structure constituted by elements with an initial section equal to 400cm^2, obtaining a weight of the structure equal to 968[tonnellate].
Now I have to check what happens if I change the starting section of the beams, so instead of 400cm^2, I have to consider 1.5,20,50,100,200,400[cm^2] and see what happens.
I tried to perform the analysis individually by creating a different project for each section and the weights of the structure vary according to the starting section, the lower the starting section, the more the weight decreases..... according to you the right thing?? ?
Shouldn't the results converge to the same weight?? ? ?

responses, thanks

 

[stefano.garbin]

I tried to perform the tests individually creating a different project for each section and the weights of the structure vary according to the starting section, the lower the starting section, the more the weight decreases....

more slender beams = lighter bridge... Amazing!!! :wink:
apart from the jokes I did not understand how you want to set the problem!
do the sections consider them all equal to each other or can each segment have its own section?

this would be the typical optimization problem that with a little apdl and using the "old" classic ansys optimization module (/opt) comes great!

 

[19sbirro87]

Hello stefano, thank you for the answer.. .I am at the first experience with this software so forgive me if I post "banal" questions. ...I try to be clearer:
I have to minimize the weight of a steel bridge composed of 7 elements of type link8, respecting the constraints in the construction:
Maximum nodal shift < 10[cm]maximum stress < 250[MPa]area of the sections of the plots between 1[cm^2] e 400[cm^2].
as you will well know in the definition of elements beyond the statement of the element
et,1,link8
I must insert the real constant characteristics of the element itself; In this case, being a link8 type beam, I only need to define the cross section area and therefore
r,1,s1
r,2,s2
.
.
(with I indicated the section area, i=1,2...7);
in the actual construction, before inserting the elements I must declare
type,
Real,
and proceed with assignment of constraints and loads (1000 [tonnellate]).
it is necessary to insert an initial value of the sections, which I will then optimize, I chose s1,s2....=400[cm^2].
using the following command lines I perform optimization

/opt
opanl,bridge

areamin=0.0001
areamax=0.04

opvar,s1,dv,areamin,areamax,0.0001
opvar,s2,dv,areamin,areamax,0.0001
opvar,s3,dv,areamin,areamax,0.0001
opvar,s4,dv,areamin,areamax,0.0001
opvar,s5,dv,areamin,areamax,0.0001
opvar,s6,dv,areamin,areamax,0.0001
opvar,s7,dv,areamin,areamax,0.0001

opvar,umax,sv,0,1

opvar,strmax,sv,0,1

Maximum stress and weight have been normalized compared to limit shift and stress limit in /post1, so the upper limit in the opvar is 1!!

opvar,peso,obj,,0.000001

optype,first
opfrst,20,50
op.

/out,results,txt
oplist,all
/out,term
finish

the following optimization brings my structure to a final weight of 968[tonnellate]. At this point I must check what happens if you change the initial value of the beam sections, leaving everything else unaltered (including optimization?!? );
instead of considering si=400[cm^2] I have to consider in order
Yes.[cm^2].

I did the 7 separate analyses, and the results lead me to get 7 different weights decreasing to decrease the value of the section up to a minimum of 92[tonnellate] in correspondence of sections 1[cm^2].
my question is: "Is it right to conduct the analyses separately and get the results mentioned above?"
Should the weight not converge at the same value in all 7 cases ??


p.s I tried to put 7 files of origin (each with a different starting section) in the same workbook and read them in cascade; typing /clear (under ansys) at the end of each optimization and reading from the next source file (other than the previous one), without being released from the program;

the program returns me the following warning(but proceeds with all calculations):
"modification of an optimization variable (dv,sv or obj) has forced a partial clear of your optimization date. design sets and global optimization settings are inherent, but all other data has been initialized"
at the end of the result file, the weight actually converges to the same value, after a tot of iterations, regardless of the starting section. Could it be a good resolution?

It forgives the excessive length of the answer but I did not know how to clarify my doubts :smile: thank you and I wait for answers!!! ! !

 

[stefano.garbin]

Hello stefano, thank you for the answer.. .I am at the first experience with this software so forgive me if I post "banal" questions. ...

Well as a first experience I compliment you because you're flirting with a certain not easy part of ansys!
now let's make some clarity: The /opt commands all seem correct, so I suspect that in the input file (I think you called it "bridge") you are making some confusion. Can you please post it?
I suspect that you overwrite the value of some variables for each iteration, for this reason the results are strange.

 

[MBT]

maximum stress < 250[MPa]

 

[19sbirro87]

for convenience you only place the final part. ..without the constructive one:
!!! Please note that in the initial phase I type the command *create,bridge and then poarto with the assignment of nodes and constraints

/solution
antype,static
Kbc,1
autots,on
nsubst,10
The
finish



/posts
etable,volu,volu
esel,all
SUMMARY
*get,vtot,ssum,item,volu
! !
weight=vtot*dens


*get,nelem,elem,num,max
*get,nodes,nodes,num,max

Umax=0.
Ulim=0.1
*do,inod,1,nodi
u=abs(ux))
v=abs(uy(inod))
w=abs(uz(inod))
*if,u,gt,umax,then
♪
*endif
*if,v,gt,umax,then
♪
*endif
*if,w,gt,umax,then
♪
*endif
*end
umax=umax/ulim


strlim=250. d+6
etable,normal,ls,1
strmax=0.
*do,iel,1,nelem
*get,strel,elem,iel,etab,normal
strel=abs(strel)
*if,strel,gt,strmax,then
♪
*endif
*end
strmax=strmax/strlim

finish
*
*use,bridge

to this then must be added the optimization part that I posted earlier... if also this is not suff. then placed directly the .txt of origin...thanks x the answer, aspect news.
soon

 

[stefano.garbin]

That's weird. have you taken as an example the list in the "ansys mechanical apdl" guide to chapter 1.7.1?
is it not that by chance you have "seminated" inside the "bridge" file some /clear type command that resets the variables?

 

[19sbirro87]

no the /clear does not appear...
the problem is that with different starting sections the results do not converge ....and I do not understand why....

 

[stefano.garbin]

You should "explore" the space of the problem with a "random design generation" to see what happens. You never know that there are any locals that mess up the optimizer. .
Maybe after the first random step on the best data set launches an "first order" optimization.
if you still have problems send me all the input file (maybe like private mex if you don't want others to see it).

 

[19sbirro87]

hi, I sent you an mp with the source file.... look news:smile: thank you 1000!!

 

[stefano.garbin]

the input file with some small correction is fine.
the question is that optimization with a range of variables so wide requires a particular care to "drive" the solutor towards the right convergence.
I sent you the modified inpt file in which optimization is divided into 3 parts, a first part with a zero order method to "find the right direction", and 2 other parts with order metid one and range of variables via reduced so as to help convergence.