Test phase 4 - Genetic Optimisation of structure

Today and yesterday i've been randomly creating generation 0 of panels (see image of screen shot below) and breeding the 1st generation based on the best ranked members of generation 0.

Peter von Buelow ( http://www-personal.umich.edu/~pvbuelow/ ) has guided me intensively during this process. The constantly adjusted parameters of the members are being generated by his Genetic Algorithm linked to a customized website (image below). The website makes it possible to download a datafile, which can be read by GC. The dxf file of the panel is then loaded in the Finite Element software STAAD-Pro, which optimizes the construction applying different sizes of steel tubes. It calculates the total weight of the structure and the ratio of structural performance. This output is uploaded to the website again, which will visualize the new best ranked members.
This optimization loop can be repeated untill the results approach an optimum according to the previously defined criteria.

Link to the GC-GA Population website: http://www.frames.tcaup.umich.edu/radiolaria/


screen shot of breeding website white= gen 0, grey=gen 1

Test phase 3

Wow, it is a bumpy ride I must admit!
I just generated a population called generation 0.
I realized after a few hours of breeding, saving, pasting, generating, that the GC file was a bit expired.

I've tried to generate a population for an updated combined configuration (see image below) too, but this didnt work, because the GA has to change first to get that properly done (and i'm not able to do this myself yet)

After an inner discussion I've decided to keep these results and finishing the GA process with them soon. These structures, are all squeezed (see image below). So it wont be possible to construct a facade with only these types of panels. I decided to calculate a few types of the stretched out panels (image below), manually in STAAD-Pro. Because I'm interested in the different structural behaviour of these two types.


configuration of squeezed panel


configuration of stretched panel


Note: Together these two types of panels could form a continuous surface (facade).



configuration of combined panel

I considered to calculate these kind of combined panels before, but I guess that the option of either squeezed and streched panels will work better structurally because they are more symmetrical than this example.

Of course, a radial structure will be the best, taking in consideration the nature of most radiolarian skeletons, but for this course I stick to these panels. It is a more zoomed in approach. So not a complete radiolarian, but a part of it.

Below you see stretched and squeezed panels forming a continuous surface. I prefer the lower one, because the grid is in the same direction.

continuous surface sketch

Applications/design/types

This is a recording of the Spline Grid in GC. Playing with the variables and rotating the structure. This version will be structural analyzed and optimized this week. I recorded my laptop screen with a camera this time, while manipulating the variables.



I played with a radial gridsystem too in GC as you can see in the next videoclip. I will use the spline grid version to optimize within this course, but want to point out that there are more ways to manipulate the radiolarian structures.

Second test phase

Currently I'm working on setting up the genetic algorithm to optimize the structure and linking this to the structural performance software Staad-pro and to Generative Components aswell.
Peter von Buelow who designed the GA tool, helps me to set up this optimization loop.

• What is a Genetic Algorithm?

During the initialization phase a population is randomly created. This concerns the results of combining all possible values of the variables.

During the selection phase, a number of criteria have to be defined. The algorithm will test each member of the population on this criteria (fitness based process).

During the repoduction phase, the selected members of the solution population will be recombined (mutated) to breed a second generation.

The result of reproduction phase is repeated untill it is terminated mainly because 1: a solution is found that satisfies minimum criteria, 2: successive iterations no longer produce better results

What is the complete process of this course?

1. Analysing radiolarian skeleton
2. Extracting basic principles
3. Designing a simplified prototype structure
4. Defining the variables/criteria
5. Modelling the structure in Generative Components
6. Link data of variables via excel to Genetic Algorithm tool
7. Execute optimization loop as in next scheme:
8. 
9. Integrate the result of loop in design
10. Test hypothesis
11. Repeat cycle

Component Design - First test phase

For the first Diana (structural performance) calulation, I took the GC model as one complete slab. (as the "big component" variant in sketch below) supported in the 4 corners (two with a role and two pinned)

comments on Diana results:
I will have to adjust the sizes in GC. because now the model was randomly created. (about 12 mm height!)I adjusted some values in Diana in order to get a bit more realistic results without starting all over again.
The results are based on the following values in Diana for now:

1. N and meters (this is a bit unlogical, but a quick temporary solution, it should be N and mm next time)
2. the height of the structure in Diana is 12 meter (too big, should be about 3 meter next time)
3. The materialization of the structure are all equal solid round glass beams with a diameter of 0,1meter. to start with a reference. I would like to test it for example glass tube system of steel to compare it later on. I also have to look into glass properties, the get to know what the max stress levels are for solid glass without failure.

Next week, when I'm in Eindhoven for the WAAS workshop, I will try to get more input and some progress of my project there. The topic is related, so may be it is an opportunity to work ahead. while being there.. But I dont really know what to expect yet.

Because I won't be able to work with Diana for the next 9 days due to licence problems, I would like to try setting up the GC loop sytem in the evening hours, after the workshop. But I have to discuss the feasibility with my tutors.

===================================================================================

I'm focussing now on a simpified planar facade component to start the testing in Diana tomorrow.
For the first structural analysis, I'm planning to use hollow glass beams and standard joints.


I still have to decide on a scenario to chose

Radiolaria applet

I've found a java applet allowing us to manipulate a regular hexagonal grid which seems to react according to some basic principles of radiolarian skeletons!

Click on the image to play >


source: www.n-e-r-v-o-u-s.com

New GC model

I've decided to model one module in GC.
I want to use my basic idea, but the lines of the grid should all be splines, in order to get a fluent tessellation. I will script a relatively simple hexagon script to add the tesselation to the grid. (see image).
Tomorrow I will talk to Fred Veer (ir. TU Delft) about the possibilities of glass beams as structural elements.

Midterm presentation

Hereby a visual summary of the midterm presentation.

critics:

• Plans are far too ambitious to be able to finish it untill a final presentation within 3 weeks.

• I have to simpify my model to a module of the compete structural skin, in order to compete at least one compete loop of scripting, modelling, calculating and adjusting of the structure.

• Do you want to generate a radiolarian structure or manipulate a given structure?

• Start as soon as possible with testing a simple model in Diana (structural performance) while working on scripting more and modelling more in GC.

INTRO:


RESEARCH PROPOSAL:


TRANSLATION:


CLICK FOR G.C. ANIMATION >>


PLAN:

Possible future applications: Structural Skin

I expect that double curved surfaces are feasible to be built based on radiolarian surface structures.