#gh3d

2D Chladni Pattern

Here we are! Finally, we had the chance to move from bits to atoms … or better … to yarns. We had the opportunity to use a semi-automatic Jacquard loom for a two-day session that we decided to split between two tasks: 1 - to weave a physical catalog of 120 selected patterns to test the relationship between the clusterizing features we defined and thought were representative of the physical behavior of the textile … and the reality of it. 2 - to use the selected 120 patterns in a big composition:

to check how clusters of similar patterns work together.

to test the ‘computation to weave’ pipeline and all in-between steps, and possible bottlenecks.

to design and weave an artistic piece that reflects our approach, a sort of manifesto piece.

Generate patterns

We decided to work using an 8x8 pattern matrix (a set of 64 information), so we started writing an algorithm in Grasshopper to define each pattern and its relative visualization, and then to generate all the possible permutations of 0s and 1s. Obviously, we had to stop it at some point since all the possible permutations were a lot (1.8447e+19). In the end, we generated 3288 patterns, enough variety to choose 120 patterns among them.

8x8 generated pattern & its simulation

some generated patterns

all 3288 generated patterns

Defining and Extracting Features

The second step was about defining features. We want to understand which possible relationships between the information of the digital model and physical outcomes can be established. We identified 12 possible features (number of weavings, number of clusters, …) and created algorithmic sequences to automatically extract this information as values from all our 6888 patterns, normalize them, and save them in a database.

12 features value distributions and the values graph

Clusterization of Patterns based on Features

Clusters are a sort of families of similar patters grouped according to identified (by us) ‘features’ values. During the session it was evident that different features lead to different similarities grouping patterns differently but at the same time different similarities among groups were evident.

We used a K-means algorithm (commonly used for clustering high-dimensional data) to clusterize, by our features values, all our 6888 patterns into 8 clusters of 15 patterns.

Then we started selecting the candidate 15 patterns from each cluster 'manually', according to our experience and aesthetic criteria, but then we implemented an algorithmic method based on the measurements of ‘similarity’ between patterns (difference of the feature value) allowing us to navigate more easily the increasing complexity due to the vast amount of available patterns.

clusters of similar patterns

The Catalog

The aim of ‘the Catalog’ was to weave each selected pattern in a textile portion (160x160 pixels representing around 10x10 centimeters) to be able to compare our identified features with the physical behavior of the textile in terms of bending, softness, isotropic vs non-isotropic behavior, etc ... .

The catalog displays an 11 by 11 pattern matrix (the selected 120 patterns + 1 QR code), and its size is 141x141 cm. We choose to use black and white yarns to stay stick to our digital model (although they are inverted).

pattern to technical colors table

technical color drawing in Pointcarrè

Pointcarrè preview of the weaving patterns

The Weaving of the Catalog

During weaving, and because of the characteristics of certain patterns, the warp threads become tighter or looser. Then, because the tensions are not evenly distributed, this creates problems of imbalance between the different patterns that make up a line. That's why, before weaving, together with Mona Cara, we sorted patterns according to the characteristics I had intuited. To organize the lines, we classified them as ‘extremely tight’, ‘very tight’, ‘tight’, ‘not very tight’, and ‘very loose’. It wasn't easy to classify them, so we determined which category they belonged to by observing how similar they were to other patterns, for example, ‘extremely tight’ with plain weave.

What we see after weaving is that the bottom row, which includes the ‘extremely tight’ patterns, measures 12 cm x 10 cm (height x width), i.e. +20% in height. The patterns in the top row, the ‘very loosely packed’ ones, measure just 5.5 x 10 cm, i.e. -45% in height. This can be explained by the fact that, in the first case, because these patterns look like plain weave, there are a lot of crossings with warp threads, so the packing is less strong. In the second case, many warp threads do not cross weft threads – they are called ‘floats’ in weaving technical language - so all the threads are packed more tightly and lose 1 cm in height for every 16 weft thread crossings (because we have 16 pixels per cm).

weaving of the catalog

The Composition "Woven Cells"

‘Woven cells’ is a woven composition that represents a set of cells, using shades of grey. The weaving patterns are the structures used to create the fabric. When you look at it closely, in detail, and pay attention to the 120 patterns all mixed together, it can appear ‘chaotic’, very disorganised. But when you look at the fabric from a distance, it's this ‘disorganisation’ that gives shape to the design. Whether we look close up or at a greater distance, we will see a cellular whole: close up, it will be all the weaving patterns that give the impression of vibrating cells, and from a distance, it will be the drawing itself that represents a cellular whole.

How will 120 patterns work all together? Which is the visible and physical result of this level of complexity? Will we get or see some unplanned features? What are the possible feedback? These are some of the questions that drove us to work on a big composition showcasing the articulation of our 120 patterns, divided into 8 clusters of 15 similar patterns.

The size of the Jacquard loom was the first constraint framing our design opportunities. The maximum size of 141 cm (with 16 yarns per centimeter, so 16 pixels per centimeter) as width allows us to use 274 patterns (each pattern has 8 pieces of information ... 0s and 1s) to fill the width of the fabric. The fact that we want a ‘squared’ final piece then defines our canvas matrix of 274x274 patterns.

How to organize our 8 clusters (groups of 15 similar patterns) into that matrix was the next big question. As computational designers, we are used to working with complex data structures and organizing them into objects, embedding variations of similar elements to create gradients. Many different strategies can be adopted or created ex-novo. We started using an image to drive our first attempt of distribution (we used here the very famous painting “The Great Wave off Kanagawa” by Katsushika Hokusai) just to test and debug the workflow and then, refined a few code details, for the final piece we decided to use a Noise-4D algorithm to create a sort of organic/cellular distribution. We can see the result as an image, but this is just a way to convey the underlying set of information to the human eyes.

technical color drawing in Pointcarrè

Pointcarrè preview of the weaving patterns

274x274 patterns distribution

20x20 patterns "Woven Cells" close up

Image Gallery of the "Woven Cells" front

Image Gallery of the "Woven Cells" back

Final feedbacks

Having the opportunity to weave these two designed textiles allows us to understand a few things.

A) The transition from our developed digital tools to the Jacquard loom software (Pointcarré) had some bottlenecks.

Setting hundreds of patterns. If you use ‘common’ patterns, you can select them in the software database, but if you want to create your own or add a pattern that is not in the database, it’s a manual operation. Thus, if you have to define a few patterns, it sounds reasonable, but if you have to set hundreds or thousands of them, it becomes almost impossible. A better way to import pattern data into the software should be implemented to make it easier.

Colours table. The same issue can be found when assigning colors to patterns. The weaving software uses colors to map patterns and then to define which patterns has to be woven and where. Creating the image to drive the weaving using hundreds of specific colors is mandatory to achieve the result.

UI design. These considerations, among others, make evident the need for a better UI (user interface). In fact, the traditional software is not meant to manage such an amount of patterns & colors and to deal with that level of complexity. A better or different UI Design is a key point to allow users to explore more easily novel ways to organize complexity into textiles.

B)  A new approach leads to a new aesthetic.

Reconsider the sampling phase. Since each zone of the woven textile is made up of a mosaic of weaving patterns, rather than a single repeated pattern, the sampling phase has been reconsidered. Whereas traditionally textile designers sample all the weaving patterns contained in their weave, here the methodology is more like that of black and white silver photo development: the photographer doesn't test each shade of grey, he tests a strip containing a set of shades present in the photo. For us, it's not a question of sampling each weaving pattern one after the other, but rather of weaving ‘test stripes’. The designer then has to choose which areas to test: those that are the most uncertain or those that include a large number of variations. It makes evident the need for a specific loom, a small jacquard loom like the TC2 loom, for sampling this type of test strip.

An aesthetic based on the materiality’s relative control. Just as in silver photography, even after weaving a few test strips, there are still many unknowns when it’s time to materialize the weaving program. During the weaving phase, the design appears thanks to the threads that materialise it. By associating computational design and weaving, the textile designer has to accept that he/she cannot control everything and therefore has to collaborate with materiality, and above all, this has to be part of his/her design process. For example, it's during the weaving phase that we notice, as we watch the fabric being constructed, that some weaving patterns link up from one area to another. This means that the white and black pixels that make up the weaving patterns interact from one pattern to the next. This is particularly visible thanks to the ‘floats’, i.e. threads passing over or under the fabric. These floats create a network of threads and lines that visually give shape to a new design and therefore a new aesthetic.

Weaving patterns as cells. In the style of pointillist painters, our approach aims to create an overall sensation through a multitude of woven structures. The numerous weaving patterns, mixed to form more or less dark shades of grey, give a ‘vibrant’ effect to the textile. This ‘vibrant’ effect can be explained by the fact that in certain areas of the fabric, lighter or darker squares can be observed than others, and networks of threads can be seen. These squares are weaving patterns made up of a different proportion of black and/or white pixels that create different shades of grey. If we compare the aesthetics of ‘classic’ fabrics with those of Woven Cells, we can see that each weaving pattern is given greater prominence in the second. This is because in classic fabrics, the weaving patterns are repeated and therefore blend into the mass. In Woven Cells, each woven structure can be seen individually, while at the same time forming part of an overall aesthetic. Secondly, in Woven Cells, the zones of shading are not so radically separated: each of the zones draws a demarcation, a territory, but for all that, there is porosity between them. It is this porosity that visually creates an effect that could be described as ‘cellular’. Lastly, as this cellular effect is visually very present, the design or woven image must be sufficiently simple and large to be readable.