Continue to Site

Eng-Tips is the largest engineering community on the Internet

Intelligent Work Forums for Engineering Professionals

  • Congratulations IDS on being selected by the Eng-Tips community for having the most helpful posts in the forums last week. Way to Go!

Submitting thousands of jobs

Status
Not open for further replies.

Calmitude

Computer
Nov 18, 2013
5
Hello!
I'm trying to implement a new optimization routine in Abaqus, and for this I'd need to submit thousands of jobs in as short a time as possible.
As a starter, is it possible to do something like this:
1) Start with a simple 2D structure (call it A)
2) Remove a fixed number of random mesh elements
3) Get max Mises on result
4) Repeat a few thousand times, each time starting from A

I've got a script that can do steps 1-3. Can I do this 1000s of times efficiently? Maybe with substructures?

A prod in the right direction would be really appreciated!
 
Replies continue below

Recommended for you

submodelling wont help, substructures might help depending on what your structure looks like and where you want to change it.
It's probably better to make your structure based on some parameters, and change those, instead of deleting random elements.
What some programs do is removing low stress areas, and adding to high stress areas, iteratively.
 
Thanks for your response!
What you're describing is Evolutionary Structural Optimization. I'm hoping to get better results using an approach based on Monte Carlo algorithms, hence the rather weird steps (which are a big oversimplification of what I'm trying to do).
I've found very little usable information about substructures (the official documentation doesn't seem to cover what I want). Essentially the models I which to test my approach on (as a proof of concept) are extremely simple. Imagine a 2D rectangle split up into a mesh. I remove an element, analyze the result, and use max Mises to give a 'rating' to the resulting structure. Since this is Monte Carlo, I need to do this kind of step as many times as possible. Yet Abaqus takes a lot of time submitting the job, pre-processing, saving the results in a file etc. which it evidently isn't doing when you choose the default structural optimization job. Is there any way I can have that kind of speed? Otherwise my whole project is unfeasible...
 
Is there no optimization function in existence? External processing is bound to be slow.

TTFN
faq731-376
7ofakss

Need help writing a question or understanding a reply? forum1529
 
Using very simple 2D models (20 elements, 80 nodes), making sure all the files Abaqus writes are kept only in the RAM (so as not to have slow hard disk access) and running with no GUI, every job takes roughly 3 seconds. This seems incredibly long since the calculations involved are relatively simple (and the only thing I want is the max Mises stress).
Over the last two weeks I've been increasingly impressed by Abaqus, but it seems to me it's not suited for what I want to do...

Does anybody know of something I'm missing? And if not, do you know of any other program that would be quicker?
 
Equivalently, does anybody know why the following:
"mdb.JobFromInputFile(...)
mdb.jobs[name].submit()"
causes pre.exe to be run? I thought pre.exe just made the input file for the next stage? Since this is the step that's taking the longest, removing it completely would be great.
 
Hi Calmitude,

I am also a new ABAQUS user and am curious why the pre.exe takes so long. I think it is because the input file can contain commands like "*NFILL" so the nodes are not defined in the input file and need to be generated by the pre.exe executable. I would be very interested if this step could be removed when a job is submitted.

I do not think that there is a quicker method if you want to still use abaqus.

if you find one, please post it to the forum.



 
Status
Not open for further replies.

Part and Inventory Search

Sponsor