AdaCAD: Parametric Design as a New Form of Notation for Complex Weaving

We, and others, use the term “structure” to describe common patterns of weft interlacements and floats that give rise to mechanical characteristics in cloth [29] (though other texts refer to this as the “construction” [17] or “bindings” [22]). An interlacement describes a location in the weave where the weft (or warp) passes from the front to back side of the cloth, or vice versa. A float is a place where the weft is not interlaced and instead, floats over the surface of the cloth on the front or back face. Interlacements give cloth stability and shapes its mechanics. For example, a twill weave orients interlacements along a diagonal line, which results in the ability of the cloth to sheer in one direction. Double cloth (or double weave) orients interlacements on different groups or “systems” of yarns, allowing for pockets or multiple layers to form.

Because of the relationship between structure and function, a focus on structure can be a point of shared concern between weavers and HCI researchers. For example, woven structures have been used to create: many common visual interface elements [3, 31, 32]; multi-layered circuits that are unravelable and/or able to disassemble [30, 67, 73]; self-shaping fabrics [50]; methods of storing computer memory [63], and auxetic structures with applications in soft robotics [55]. The central role of structure is also emphasized by Buso et al., who argue that weave structure is an often overlooked element of textile design with huge potential for creating new and sustainable modes of innovation [23]. They argue that a shift from the term “smart” to “animated” textiles can crosscut concerns of electronics, bio-materials, and sustainability that all share a focus on innovating weave structure (as opposed to adding materials to existing structures or on top of pre-made cloth). McQuillan also uses the term “multimorphic” to describe the dynamic nature of woven structures [48]. In her book "Mastering Weave Structure", Alderman writes “when you understand the underlying principles that govern a particular kind of structure, then you can modify the basic draft [of the structure] to obtain the result you want” [18]. The AdaCAD software presents these structural principles and modifications as parametric design components which can be combined and linked together to generate new structures with emergent behaviors. Figure 2 aims to illustrate the relationship between structure, the effects of specific structural modifications on cloth, and how modifications are achieved in AdaCAD.

As interest in textile-based or otherwise soft electronics grows within HCI, so too do the number of techniques and systems to assist with their design. While this field is quite large, spanning work in knitting (e.g. [51]), embroidery (e.g. [36]), and paper-based electronics (e.g. [75]), we will focus this background very specifically on work in weaving. We do this because the procedures in different textile crafts may share some overlaps, but are largely distinct in their design processes and machinery. In their review of woven electronic systems in HCI, Pouta et al. provide a detailed summary of the state of weaving in HCI while noting the need for more design support tools that account for the new concerns that emerge with interactivity [61]. This implies a divide between the software weavers currently use and emerging systems situated towards the concerns of those in HCI or engineering, as well as an opportunity to more deeply integrate computational advances within the context of weaving. Currently, weavers use hand drafting with grid paper as well as many software tools to design woven structures. Since the weaver’s knowledge is the key source of innovation, these tools can and have been used to do innovative work.

Most of these tools follow a paradigm of direct-design, whereby a structure/draft is produced by manually clicking cells within an empty draft, and then filling regions of artwork with those structures to produce visual or mechanical effects. Some of these tools have “generators” for structures like twill, satin, and double cloth, but the primary mode of interaction remains to be directly clicking grid cells to define drafts and/or loom settings. The choice of software is often linked to the loom that one uses. WeavePoint [2] and FiberWorks [5] focus on floor and table looms, and advanced tools like PointCarre [10], ArahWeave [4], and ScotWeave [9] focus on jacquard weaving. Due to the high cost of the Jacquard tools, as well as the training required to use them effectively, many weavers opt to use Photoshop to make their drafts with the help of a book called “The Woven Pixel” [64] or training in Jacquard weaving courses.

Researchers in HCI and computer graphics have begun developing new systems for textile design which often focus on simulation and structural complexity, such as multilayered weaving. Two such tools include TexGen [21], an open source library for modeling textile structures using Python, and WeaveCraft [72] a tool that simplifies the process of making complex structures by focusing on the design of cross-sections of cloth. Other projects complement design with interactive fabrication with DIY desktop looms [16, 57], realtime design inputs [15, 16], and software to produce custom frames for hand looms [25]. Our original publication of AdaCAD in 2019 exists within this space by presenting designers with multiple linked views of their design: one emphasizing woven structure and the other emphasizing the path of electronics through the cloth [35]. In light of our findings from this original study suggesting that playfulness and documentation are key features weaving software ought to support, the present version of AdaCAD includes a different approach where a user develops a structure through a parametric design workflow. This approach shares similarities with approaches like "visual programming" or "procedural design", yet, we chose the term parametric design because we find our user interface to be slightly more abstracted from the code than in the other two cases. Nevertheless, model-based design tools have proven to be successful in many creative application spaces within HCI (e.g. [42, 43]) as well as in other successful products like Grasshopper [19], MaxMSP [12], and vvvv [11]. While we know anecdotally of some weaving teachers that have adapted Grasshopper to explain weaving, AdaCAD is the first parametric-based system designed specifically as a tool for weaving. While the underlying functions are specific to the concerns of weavers, we believe the parametric approach could generalize to other textile design spaces for instance, as a user interface to other code-based frameworks such as KnitOut [6] or Processing Based Embroidery Generators [8].

Consider the following situation faced by weavers (illustrated in Figure 5). Imagine a user wants to make cloth with a pleating structure, which can emerge when satin structures are arranged next to each-other with alternating faces (e.g. one weft-facing (showing the weft color), the other warp-facing (showing the warp color)). In AdaCAD, this weaver could begin by adding two satin operations, defining the satin for each region. A "satin" describes any weave structure that obeys a specific set of criteria (e.g. maximum spacing between interlacements). A weaver typically selects the parameters for their satin based on their loom density, how tightly they intend for their cloth to pack, etc. The weaver could then connect each satin draft to a rectangle operation which repeats the input satin across the region of the rectangle. The size of the rectangle would likely be determined by their loom density and the size they desire the block in the resulting draft. To place the rectangular warp- and weft-facing satins regions side-by-side, the user could choose the pattern across width function and define the repeat pattern with letters corresponding to each region (e.g. "a b a b"). The operation then creates inlets to use for "a" and "b" and the user would connect their rectangular stain regions to those inputs and immediately see the resulting pattern.

The role of parametric design becomes clearer when an initial design needs to be modified or updated. For instance, when placing satins side-by-side, ensuring the interlacement points are directly next to each other at the boundary where weft-and warp-facing regions meet is required to ensure a clean line between regions. In the example above, the weaver may notice that their envisioned rectangle width did not support this relationship between interlacement points. In AdaCAD, they can simply update the width until the outputs tile at the desired intervals (Figure 5 a and b). Next, they may test the file by weaving it and realize they want one region, let’s say the warp-facing regions, to pack more densely to more aggressively pleat on one side of the cloth. To do this in AdaCAD, they would go back to their original satin operation and change the repeat size to be larger (and thus produce a more dense satin). This change would ripple down, immediately showing the impact of the change on subsequent drafts in the design pipeline (Figure 5 b and c). When patterning two different sized blocks side-by-side, AdaCAD automatically knows to recalculate the draft height to ensure that as one repeats the pattern while weaving on their loom, all the varying sized satin intervals repeat at the same rate (Figure 5 c).

We approached the development of AdaCAD as first-person research [26, 52] and included collaborations with three weaving experts. Specifically, we studied weaving and drafting from a first-person perspective, developed the tool as a provocation or expression of “what if” for ourselves and others, and then used it to elicit conversations with the community members about their values and visions of weaves they’d like to make. This approach was motivated by an observation from Magrisso et al. that stated it can be difficult to evaluate hybrid-craft interfaces because the users whose practices require such tools do not yet exist [46]. Taking the approach of building software as an expression of a new possibility helped us explore a space and practice that we believe existed implicitly in the practices and mental models of weavers, but did not yet exist in this specific form.

We felt that evaluation required users to make use of the tool in their practice because relationships to tools in creative practice are highly personal. Yet, it can be difficult to ask someone to introduce an entirely new system into their practice (e.g. by analogy, we felt like it was asking a painter to start working with pencils). We addressed this by designing the tool for our own practices first; to present the tools we made to the broader weaving community through public talks, YouTube videos, zoom calls with individual weavers, providing a documentation site; and then working collaboratively with expert weavers who had showed interest in integrating AdaCAD into their practice. We will detail the modes of sampling, data collection, and analysis for each of these phases below.

As someone who studied computer science and was familiar with tools like Grasshopper and MaxMSP, I (Devendorf) always tended to approach weaving as a computer scientist, thinking of structures in terms of their algorithmic relationships. Before AdaCAD 1.0, I would program woven structures in Excel or Processing, yet, when developing AdaCAD, I defaulted to a direct design paradigm I had seen in other weaving tools. It was not until two expert weavers, Vibeke Vestby and Belinda Rose, showed me a tool called Proweave for playfully making drafts by dragging and dropping graphic and draft components on each other that I began to think of this “computational” approach as a method to center the playfulness and need for documentation we discovered in AdaCAD 1.0. The decision to push towards a parametric approach suited interdisciplinary weavers who were approaching weaving from the perspective of a fabrication system and may have used tools like Grasshopper in fabrication labs. Some weavers sent me messages about the software and what they made with it, such as one who suggested that “As I studied mechanical engineering before and started studying textile design, I had many frustrations [with textile design]...from your lecture I thought I could be a middle person who can improve communication between textile and engineering” and “AdaCAD is nice because even [a] beginner like me also can make patterns easily, after short instructions.” Despite self identifying as a beginner, this person shared the outcome of her AdaCAD file and weaving which is a quite geometrically double layered textile with openings across the edge (Figure 8). This communication was heartening as it suggested potential for the software to reach others negotiating their position between engineering and textiles, as well their ability to meaningfully adapt it into their process without my direct involvement.

Presenting AdaCAD to collaborators Kathryn, Marianne, and Etta revealed more challenges negotiating their backgrounds in weaving with AdaCAD 3.0. This emerged most notably with Etta, who found that AdaCAD did not support the drafting style for the loom she most commonly used, a direct-tie loom. Because I had developed the software for use on floor looms and jacquard looms (both looms to which I had access), I was unaware of the different conventions required by direct-tie looms. Where jacquard looms support complexity and non-repeating patterns by making every warp thread in a weave individually controllable, harness looms require different sets of warps (threaded into harnesses) to operate together. This creates an interesting "algebra" [38] whereby a weaver can create different patterns among threading harnesses and then use software to simulate all the different possible outcome patterns with that threading. Often this works in a back and forth, where a weaver desires a pattern of a particular type, uses that to create a threading, then plays within the bounds of the threading map, often moving between the loom and the draft in response to the material results. In this sense, harness loom weave drafting is inherently parametric and the process of hand computing the frames and possible outcomes can be quite joyful, or even, a way that one comes to understand the relationships between structure and draft on a deeper level.

The inherent generativity of harness loom drafting created a tension with the parametric framing of AdaCAD, which asks users to design structures first (and then can automatically generate the harness threading required to achieve them). This was most emphasized in the “blank space” presented to Etta by AdaCAD when it loaded: “I still felt intimidated and unsure of where to start within the blank workspace when confronted with it on my own. Since I do much of my drafting by hand (without much background in engineering), I found navigating the tools and parametric workflow to require some play and practice before it started to feel like a comfortable and generative tool.” Implementing software required to support Etta and drafting for direct-tie looms required a large rethinking and reworking of the underlying code base of AdaCAD. These changes were programatically complex, but when they were implemented, Etta found herself better able to make use of the tool. In this instance, our collaboration revealed a moment where the first-person design approach implicitly privileged the first-person’s equipment, resources, and personal "alchemy" in its design. Working collaborative with E, then, helped expand my knowledge of different looms and their associated drafting styles as more than just abstractions of woven cloth, but fundamental, generative, and already parametric systems of thinking and planning woven structure design.

A similar challenge emerged with Kathryn whose woven knowledge emerged through Scandinavian traditions, which are different than the American traditions in weaving I had learned. These differences are animated by the use of different terminology (e.g. binding instead of structure) and most notably, by a different orientation of the draft itself. Where I designed drafts from right to left, top to bottom, Kathryn worked from left to right, bottom to top. This orientation underpins her entire conceptual process for generating drafts and AdaCAD’s top right orientation presented a major hurdle to use, namely because she was unable to validate visually if the structures made sense, reducing her "trust" that the system would work. We were able to resolve this challenge in two ways: first, by having Kathryn produce the files generated by AdaCAD and verify it worked, and second, to add a workspace preferences setting that allowed the user to set their orientation point instead of assuming one orientation fits all.

AdaCAD opened up new ideas of possibility and inspiration to weavers, mostly in terms of adding complexity or otherwise computational modes of exploration into their process. Many weavers, often those who identify as “complex weavers” have expressed enthusiasm for the project. Many found the parametric approach unfamiliar but alluring, leading one weaver to say “I don’t think I’m going to be able to sleep tonight”, suggesting that it was introducing new puzzles that they found exciting. Many weavers requested features to map datasets to decisions in weave structures, or to even link AdaCAD with live data via Platforms like Open Sound Control to more closely connect real time interaction and weave structure similar to a concept explored initially by Albaugh et al [15]. Weaving-while-developing led me to wonder what I could draft programatically that would be impossible to able to accomplish otherwise and thus, I began to explore different methods of creating woven designs that evolved more deeply from code, like the creation of a random draft generator that would create bespoke designs with a given percentage of weft facing floats (thus allowing for the color shifting visible in Figure 7). In each case, a weaver is desiring AdaCAD to serve as a locus of connection between computational infrastructures and textile infrastructures, whether crossing between graphic design and cloth, data analysis and cloth, or real-time interaction and cloth design. While some features have been implemented, and others for future releases, the software does prompt new ideas, if not new puzzles for a weaver to solve.

Our collaboration with Marianne Fairbanks exemplified how AdaCAD, and the development of custom modules for AdaCAD, created a platform of new experimentation to take place: specifically experiments with combinatorics and artificial life. Marianne got involved in this project after watching a talk by Devendorf [47] and contacting her directly with an idea to implement in AdaCAD that she had struggled to find someone to help her with. Specifically, she wanted to implement an algorithm to find every possible structure for an N x N draft. The motivation was largely based on the observance that three structures are said to be the core of all weaving: satin, twill and tabby. Marianne wanted to know why these three? What other options were there? And what might be possible in these other options? She had been fascinated by historic weaving sample sets, such as Atlas de 4000 armures, that demonstrated and cataloged 1000’s of structures as both drafts and material samples for other practitioners to integrate into their practice [65].

With such a clear idea in mind, Devendorf quickly began implementing algorithms and functions in AdaCAD that could both generate the set of possible outcomes and explore them in a creative way. The very first, and fast, implementation simply asked for a numeric input value, converted that number to binary, and then used the bits to populate an NxN draft. Noting that many of the numbers made structurally invalid drafts we updated the algorithm to ensure that each N x N draft had at least one interlacement in every row and column (i.e. at least one cell of black and one cell of white). We updated the algorithm to include only valid drafts while ensuring that each draft maps to a unique index value (and then be consistently searchable across all instances of AdaCAD). This algorithm revealed that there are 102 possible 3x3 structures and 22874 possible 4x4 structures (though some structures are rotations, reflections, and shifts of others, leading us to questions of definition of equivalency in structures). We added a button on the operation to download the full set of options as an index sheet (an excerpt of which is shown in Figure 10). This outcome led to more ideas of both programmatic and playful nature: a set of stamps of each structure to play with color combinations; a material characterisation of each structure to be able to classify and group them by their functional specifications; a next step using AdaCAD to create a sample blanket with woven samples of every possible structure. She wrote in an email: “I find this inquiry somehow the weaver version of Sol LeWitt—trying to quantify angles, lines and shapes in new ways and combinations” which suggests that this particular systematic process of generating possibilities held both practical and metaphorical resonance with weaving and art history. Mariannes’s case also most centrally relates to particular next steps for exploration in woven structure design by engineers, specifically the systematic characterization and organization of woven structures by functional characteristics. Could it now be possible to highly customize the structure to function, rather than adapting the existing small set of structures? Here, AdaCAD leads to more openings for weaving and computation by focusing on how it can be used to generate corpora for analysis and exploration.

When using AdaCAD herself, Marianne described AdaCAD as “disorienting” in the way that it “makes me feel like I am working from the ground up - building from the micro to macro instead of how I think about designing in Photoshop or PointCarre which is the top down approach.” She suggested this disorientation could serve a practical effect, such as causing the user to rethink the processes and methods of drafting as she imagined with her structure generator. For Kathryn our collaborative development of AdaCAD invited her to explicate and document the thought process behind her "Dragon Scales" project (shown in 11). One of the key features of Kathryns’s practice is the use of multiple layers and weaving systems. Weaving multi-layer cloth is achieved by interlacing yarns in such a way that they can separate or intertwine, as desired, into distinct layers. These layer orders can switch across the cloth, creating pockets, or otherwise interesting folding structures. When designing, Kathryn maps out the layer order in different regions of the cloth, chooses which structures to apply on which layers (in which regions) to give her the intended structural or 3D effect and then develops the structures required to achieve those layers largely by hand through the direct clicking of pixels before translating them into a program called ScotWeave, a software she uses to design and communicate with her loom. One such project includes a challenge she took on to weave a flat textile that can be unfolded into a box (See Figure 9 B). This is the kind of challenge that captivates Kathryn and it involves a highly complex set of mental gymnastics in order to plan and execute, specifically as it relates to developing “suture” structures that could be used in the joints to support easy folding and unfolding. In our first meeting, we talked about this project and she walked me through her design process, including notes and images she used to aid the process. One image is a drawing of the cloth with codes in each region. Connected to this representation is a hand-written index code to structure in a notation that she had developed. In her words, “this is me working out in every region what layer needs to be doing what thing…”

Inspired by her ad hoc practice, we explored methods for implementing her notation directly into AdaCAD. After several exchanges where Kathryn and I (Devendorf) attempted to form a common understanding of her notation, we came up with a text-input operation that created inlets for each term in her notation (See Figure 11). She could then assign structures to the inlets to automatically generate the draft that assigned each structure to its correct position in the “layers stack.” At our next meeting, I walked her through the layer notation feature. She made a few samples to see if she could trust the interpretations and confirmed that it was both working and that she could effectively integrate the software into her process.

In terms of technological mediation, the parametric workflow brought two impulses to her practice: the ability to quickly identify and resolve errors and a prompt to rethink/reinterpret her process in terms of AdaCAD. Before AdaCAD, when errors emerged in her cloth, she would go back and make the drafts from scratch, in Photoshop and re-import them into ScotWeave – “Because the draft format (only indicating warp up or down) flattens the structure, it becomes impossible to distinguish between the warp movement in a specific layer and the warps lifted or lowered to separate the layers. Therefore redrawing becomes the only feasible option to fix errors.” She felt AdaCAD effectively “exposed the logic” to a degree that she could pinpoint and correct where errors occurred. Though, she also noted that translating process to AdaCAD to be a new challenge: “while I know (tacitly) the algorithms to generate a particular structure by pixel clicking, there was a lot of mental work going on to translate this into the parameterized operations of AdaCAD – I knew how to do things tacitly (in my hands!), but AdaCAD made me make them explicit.” The result of this process, though, was her “brain on a page” suggesting that while challenging, the work of explicating process to an alien system like AdaCAD was productive for documentation and sharing. She took an interest, then, in using AdaCAD to archive past projects.

In Kathryns’s case, AdaCAD became both a space where different processes and practices were discussed and communicated through both language, diagrams, and parametric operations in the software. The collaboration was aided by the fact that, implicitly, Kathryn’s thought process through weaving was very much systems oriented – evidenced by the fact that the notation system they had developed with a collaborator and used extensively could directly map into an algorithmic interpretation. This resonance between the models of AdaCAD and her thinking allowed her to go back and unpack her past processes through this frame, as, again, a way of accounting for and making her tacit practice explainable to others, but also herself (as weavers themselves can hardly understand their drafts). To some degree, the practice of going through past drafts and putting them into AdaCAD signals that the activity can be beneficial or enjoyable, as there is an inherent logic and “math” that comes into framing your practice through a new set of notations and systems: The work of documenting process was also seen by Kathryn as a process of revealing the complexity of weaving for further exploration to others: “there is so much potential in weaving that we have barely scratched the surface and so the more potential you build into things like AdaCAD, where you expose these opportunities, the more it reveals.” Like some others in our study (who also happen to have extensive experience in institutions of higher education), Kathryn shared an interest in not only weaving, but teaching about weaving, so the work of explicating weaves in AdaCAD offered multiple pathways to benefit. She wasn’t necessarily playful with the software, but used it as a more powerful and explainable tool to get the ideas in her head in conversation with the machine loom.

In their survey of weaving within HCI, Pouta et al. argue that to advance weaving research within HCI, “Authors should strive to make all relevant design files, schematics, layout drawings, and (especially) weaving patterns accessible” [61]. While we feel the spirit of this message to be correct, we see in our studies and experience that to reveal a draft or weaving pattern in its current form, even with notations of materials used, may not be sufficient to understand what it represents or the logic inherent in the construction. Weavers are rarely able to remember their own drafts and structures and the looms accessible to different weavers will also vary dramatically. The personal files used by artists to document their process, too, take on radically different conventions and styles.

We believe that AdaCAD offers a middle ground between the draft and woven cloth that can serve to advance textile innovation by providing a new means of notation for weaving. Notation systems, from musical notations to written language are “captured within the relations between the elements of a relevant ‘set or system,’ and this capturing creates a physical document that can be studied, analyzed, referenced, reproduced, and reinterpreted.” [7]. Throughout history, weavers have evolved through practices of developing and sharing different forms of notation, such as the bespoke notation Kathryn used in her practice. Prior to these notations, it is thought that knowledge was tacit, communicated locally through practice, or sold in small folios with bespoke notations [38]. To this day, weaving guilds share their process through textual descriptions, samples, and abstracted sets of rules that govern the structure. For instance, a copy of a 1949 weaving text by Ada Dietz relates color systems in weaving directly to bi- and tri-nomial expressions [74]. These weavers are not just sharing their drafts, they are sharing the logical units from which the draft emerged (and also the design space of other drafts that emerged though the application of such rules). Furthermore, the cloth serves as the ultimate record of the production process better than the draft in that “unpicking” the cloth by way of reverse drafting allows one to create the pattern file from the physical sample. This asserts the importance of physical sample sharing in a research-based textiles practice and lives on in the form of swatch exchanges [39, 58] but also suggests that AdaCAD provides a notation for a different part of the weaving process: the logic of the structural relationships as opposed to the resulting thing.

Here, the history of weaving notation reveals its heavy reliance on systems and rules in much the same way that parametric design explicates form in systems and rules. The components or “terms” of the parametric design landscape are malleable and multiple, allowing weavers to express individual style and interest in the combination of those components. At the same time, we observed that AdaCAD offers several different access points to this notation of weaving: first, through the user interface of packaged parametrized components, and second through the code base. Where the user interface offers itself to weavers, the code base offers itself to coders, and there is an open invitation to blur and work across those spaces as we did through our collaboration. The browser running AdaCAD becomes a translator between pixel based code operations and weaving draft. In our studies, this creates a (yet another e.g. [37]) spillover between weaving and computer science we we might come to understanding computational systems in terms of weaving and of weaving in terms of computational systems.