Does scaling matter for analysis?

pberrypberry
Maybe a stupid question, but here it goes....

Is it ever computationally advantageous, or is it ever disadvantageous to scale the model below actual size for solving? Do size or units affect the result? A slightly different way to state this is, "will the solution for my small model be valid for one that is twice as big/thick"?

Primarily, I was thinking about solve time, but then I couldn't figure out if the analysis would even be good at all, if is not full size. I'm not referring to symmetry, where you can split it into half or smaller. Just from the standpoint of a full model.
I think that I can partially answer this, but I don't know what other circumstances would be bad, or if there are cases for under-scaling which might be beneficial.

And my own thoughts on answering this......

I think something like a spherical tank would have to be actual size if you were filling it with a liquid, and applying a gravity load, due to the square/cube scaling of the shell vs the contents. Is there anything else that is not going to scale non-linearly?

For a linearly scaling model, it would not be advantageous to shrink it, because you still need as many mesh elements, just smaller. But how about using 'easier' units. For (a bad) example, is it cheaper to use 1m as a dimension as opposed to 1000mm? Or maybe 1" opposed to 25.4mm?

Regards,
Paul

Comments

  • VictorVictor
    With Mecway, the solver converts everything to the same units for solving anyway so 1000 mm ends up numerically the same as 1 m.

    However, you could scale everything so that it's numerically smaller/bigger, like specifying 1 mm for something that's really 1 m. In that case, the easiest and safest way I've found is to specify m,kg,s units for everything and treat it like its dimensionless software. So, in your head, pretend they're any consistent unit system like mm,t,s, um,kt,s, etc. You still have to be careful to convert all the values correctly but you can do it fairly simply by writing every unit in SI base units then just swap mm for m, t for kg, etc. Converting units automatically takes care of any non-proportional scaling like squared-cubed.

    I don't think there's any performance difference but haven't measured it. You can run into numerical error in the extremes though.
  • pberrypberry
    edited January 3
    Something to think about.

    Thank you Victor!
  • DaveStuppleDaveStupple
    Victor, Paul,

    I've got a model that fails with CCX thermal static, as I had before, because of small elements. If I scale it by +1000x it solves. But, as previously, my conversions were way out.

    I persevered and after (too many) hours found the right conversion factors.

    A 1000x model (without the problematic elements, for testing purposes) solved identically to its little sibling but in < 5 mins rather than the 18 minutes that the other one took.

    CCX thermal static seems not to like very small elements.

    Cheers, Dave
Sign In or Register to comment.

Howdy, Stranger!

It looks like you're new here. If you want to get involved, click one of these buttons!

Sign In Register