karamba - parametric structural analysis

Comment

Comment by Robert Vier on May 14, 2012 at 1:52am

Hey Steven,

could you post your definition to have a closer look at it?

Thank you

Comment by Steven on May 2, 2012 at 9:29pm

Comment by Steven on May 2, 2012 at 9:28pm

Comment by Steven on May 2, 2012 at 9:26pm

Robert, thank you for your quick reply to my questions!

I started playing with the MinDist option and was getting quite confused by the results. I then realised that even with the same settings, the FFF can produce different results each time it is run. See images below, these were created with the same parameters, I just clicked on "recompute". I would have thought that if left iterating long enough, it would eventually converge on the same optimal solution every time. Regarding the symmetry, I would imagine that even taking away 3 elements at a time, given that it can add elements back in at any stage, it should still arrive at a stable, symmetrical solution...

It would be fantastic if you could discuss these issues with your colleagues next week and keep me posted!

I have also read the FFF section of the manual. It is a little hard to understand exactly what each of the inputs means. Would it be possible to illustrate them with a numerical example? eg "a truss with 100 members, and overdrive set to 1.4 will take away x members each iteration... etc etc" 

Thanks for you help,

Steven

Comment by Robert Vier on May 2, 2012 at 3:28am

Sorry Steven,

I didn't see you just set the 'WCompression' value different from zero. This indeed is kind of strange.

You can also set the weights to negative values, which means this section forces are 'even more avoided'.

Since I haven't written this latest version of FFF myself, I have to wait for my colleague to com e back next week for detailed discussion. But as soon as I have time I'll try to investigate the cause of this.

Best

Comment by Robert Vier on May 2, 2012 at 3:20am

Hello Steven!

Firstly, problem solving in general -be it with galapagos or our FFF-component - can be greatly eased by dividing a problem to sub-problems, e.g. using symmetries. If you divided you structure to four quarters, modeling just one of them with Rx,Ry,Rz and local-axe-Tx support conditions on the element-ends in the section planes (simulating the cut pieces of the structure), inherent symmetry is imposed. This also greatly increases performance with large structures.

Due to numeric inaccuracies, bot more due to the iterative incremental approach of FFF (where the number of elements to remove and elements to add in one iteration probably are not dividable by four in every iteration), symmetry is not always obtained. Either by splitting the model like described, or setting the parameters cleverly, you can avoid this.

Second, the karamba-manual http://www.food4rhino.com/project/karamba explaines the basic functionality of FFF-component and the meaning of the inputs.

The fact that it keeps working on the same element from approx. half of the process to the end in your example is explained as following: The 'Iter' -input specifies the number of iterations in which the structure is reduced to the specified 'Target'-mass-ratio, in your case 50. The remaining difference between 'MaxIter' and 'Iter', in your case 50, is used to 'smooth' the result by removing x elements (depending on 'Overdrive'-setting) and adding x elements again. The number of this smooting steps often can be reduced.

Your issue maybe also can be solved by increasing 'MinDist', which means that in one iteration only elements with a minimum distance of 'MinDist' between each other are processed.

'WLimit' specifies another 'smoothing'-value, which is a threshold-percentage of the average element weights. Elements weighted below this threshold will be removed.

Regarding the weights: In every step, each element is weighted by its average section forces:

W = SumOverAllSectionForces (Factor * AverageSectionForce)

The fact that tension elements remain with Tension-Factor set to zero probably is caused by specific structural situations where the element is e.g. much more needed for moments or shear and therefore remains in the system, though.

Hope I could help,

Best

Robert

Comment by Steven on May 2, 2012 at 2:23am

Hi there, I have been doing some structural optimisations using the Force Flow Finder component and I have a few questions about the results and the input parameters. A video of the optimisation I am working on can be found here:

https://vimeo.com/41407740

The script seems to generate very asymmetric results, which is unusual because the load is in the centre, and the supports are at the corners. After about 60 iterations it just keeps adding and removing the same member, but is clearly not at the most optimal solution.

I imagine it might be possible to tweak these results using the "Overdrive", "MinDist" and "WLimit" input parameters, but it is hard to understand exactly what these do and how they can be used to improve results. Could someone explain those to me?

Setting the tension input to 0 still yields members that are purely in tension, which also seems odd to me...

Thank you for your help!

Steven

Comment by David Shook on June 9, 2011 at 12:23am

There have been many great advancements.  Your work is very exciting.

Comment by Karamba3D on June 7, 2011 at 9:06am

karamba 0.9.06 is out.

These are some of its new features:

- bi-directional evolutionary structural optimization (BESO)

- axial force and resultant moment per element

- bending- and axial deformation energy per element

- nodal displacements: translations and rotations

Comment by Karamba3D on May 24, 2011 at 5:31am

There is no particular reason. I just thought that all the gha-files have to be in the 'Components'-folder of Grasshopper. Thank you for that tip!

Best,

Clemens

 

• ‹ Previous
• 1
• 2
• 3
• 4
• 5
• 6
• …
• 8
• Next ›
• Page