I have continued on exploring the limitations and restrictions of the rigid-foldable tubes. The constructed prototype models showed that the sine curve would induce movement in two directions during contraction. In order to solve the problem within the limits of rationality, a decision was made to simplify into a curve, which implies movement in one direction. Additionally, I explored the positioning of the tubes and concluded that it is better to substitute the continuous repetition with dynamic shape-changing tubes as the slope descends. This gives a steeper curve that is smooth on the inside with increased structural integrity.
The units are positioned in a continuous chain around the defined areas, within limiting boundaries defined by the insolation conditions. The proposal demonstrates the foldability (contraction) of the shelters and their potential dismantling during winter.
During the final presentation, an observation was made that the shelters could remain attached during winter. Their structure would allow snow to fill the gaps (thermal bridges) and provide isolated volumes, which is favorable for winter conditions.
After the introduction of converging planes, I extracted one of the resulting curves and reintroduced convergence, but this time the limiting element was a point. A few tests and adjustments resulted in selection of two ‘visually optimal’ shapes. Thinking of small-scale structures I envisioned them as tents for one person. The next step was to explore different approaches for inside-outside movement with a system of foldable elements that allow opening/closing.
The first proposal is based on the idea of curved rigid-foldable tubes. Their structure would allow movement along tracks (rails) as they contract/expand to open/close. At this point, they can be observed as cells that group to form a small organism – camp on the island.
Trying to connect the proposed structures with the context, I reflected on the site-selection task. I concluded that the identified pixels with irradiation above a threshold value are to be used as important factor for arranging the elements. Thinking of the approaches for correlating them, I lost track of the true meaning of these pixels and took a ‘wrong direction’ in the process.
The proposal included two ways of manipulating the pixels into shapes with MATLAB and Rhino. First is the Point Spread Function (PSF), which generates Gaussian distribution of the pixels in a selected region (see graphs that visualize the algorithm). This gives a certain degree of blurring which is determined by the x- and y-axes and an intensity factor. The second proposal uses a Motion Function (MF) that is adjusted for different directions and intensities. The outcomes in both cases are treated as height maps that translate to lines which define their shape. Then, one of the results is chosen to try the principle of curved rigid-foldable tubes.
At this point the obtained product contradicts with the idea of utilizing irradiated pixels for daily activities. The structures on top of them do not justify their purpose, so I will iterate back to the previous step by further developing the contracting-cells proposal.
I tried to explore some of the possibilities from folding a piece of paper while constraining with various “levels of freedom”. The following diagrams explain the results from different limiting points and how they can be reproduced with one foldable element. By introducing the limiting plane the original constraining points are removed but the folding results are attained.
Quite often architects are challenged with the need to analyze large sets of data in the design process. However, there is a lack of defined approaches and therefore in most cases detailed analyses are avoided. Through this project I tried to observe a fraction of what could be used as a valuable assessment tool in various projects. Clustering approaches are not uncommon in different fields when in need to observe data, but in architecture these have been poorly investigated. I have used the MATLAB programming language developed by MathWorks to investigate the potential for architectural use.
This is a brief description of the steps in the process:
1. Looking at the height maps as a set of pixels containing height data and clustering with kmeans. (This is an unsupervised learning technique that does the analysis without predefined criteria.) The results give groups of heights that share similar ‘intensity’.
2. A region of interest is selected with clustering division at k=5.
3. Radiation maps of the site are generated using Ladybug. (Different relevant maps or parameters can be used for evaluation depending on the requirements.)
4. k-means clustering is applied on a radiation map. The points that satisfy values above a defined threshold are selected and interpolated to the previously defined region of interest. This gives points that satisfy certain irradiation criteria within the specified height region.
5. New clusters are created in order to define optimal building areas and building shapes are defined at locations where the points satisfy certain density.
The results from the steps are shown in the following isometric drawing.
An experiment with varying k-number values in the kmeans clustering was performed and the results can be observed in the short animation. For values above 200 it is difficult to read the information from the resulting images, but lower k-numbers can give different levels of detail depending on the requirements of the user.
The roman town Silchester is an interesting case if we analyze it at three scales. It can be seen through its irregular border containing strictly divided cells with ‘random’ clustered buildings in each. What fascinates me is the ‘logic randomness’ that is completely unexpected if we compare it to the rigorous meticulousness of roman military camps. Nevertheless, with some effort one can start to detect logical patterns and that is the first approach that I would consider in this project.
Another way of thinking about a camp is to have a predefined building element that will be repeated according to a certain parameter. Or we could take a predefined shape and fit the element in it. I have previously considered this approach in a similar studio program, and this is why other ideas are more compelling for me.
The approach I find the most fascinating is to observe clustering in every layer of ‘making’ architecture. What if we could define a path around which we would cluster buildings, object, elements and atoms? Or what if we could use clustering in a broader sense and try to cluster these predefined paths too? I would like to test the k-means clustering method on the given terrain to find areas satisfying certain parameters. Than it would be exiting to observe clustering in different directions and rates depending on the program and the needs. Although this might not lead to a successful solution, I am interested in exploring such potential.