This action might not be possible to undo. Are you sure you want to continue?
Abstract. This paper puts forward the concept of Generative Fabrication as a conceptual framework for exploring the potential of material and device-specific design grammars to act both as machinereadable information and as platforms for creative design exploration. A series of experiments and prototypes are presented and discussed.
1. Aims The main aim of this research is to put forward the concept of Generative Fabrication as a conceptual framework for both generative design exploration and prototype fabrication; such framework is necessary as a seed for a new generation of design tools that provide a common, integrated platform for design and construction ideas. The role of bottom-up design generation is thus considered in addition to, and as an alternative to, topdown design methodologies where computer programs or manual labor are used to subdivide an initial shape into a set of distinct building components. Furthermore, a formalization of the trade-offs inherent to a generative grammar where design and construction logic overlap is attempted. 2. Significance Encoding design and construction information in a single digital entity has been object of many efforts in the field of design and computation (Eastman 1999). Rather than having to create separate descriptions for different fabricators or contractors, the collapse of design and construction data into a single information package promises to achieve greater simplicity and to add control in construction processes (Sass and Oxman 2006). Much of the research around such integrated descriptions, however, relates to the way such data structures can store the vast amount of information required for the design of complex projects, but few of them deal with the implications of this emerging type of description for design reasoning and
creativity. The division between design and construction remains thus largely untouched in the large specter of digital design culture, comprising software, hardware, pedagogy and practice. Our research presents a series of experiments and prototypes that explore ways for both design and construction logics to overlap and play. Single-material assemblies are used to reduce the conceptual complexity of the digital models and because of the possibility of fabricating them in the lab using a single machine 1 . This simplicity grants us the ability to focus on the design reasoning afforded by a given design language, and on formalizing the trade-offs inherent to it. 3. Methods
3.1. SINGLE MATERIAL ASSEMBLIES
Single-material assemblies for digital fabrication (SMAs) have been a topic of exploration in recent years (Kashiap 2001, Sass 2005, Griffith 2005, Botha 2006) The logic behind SMAs in these works is to avoid junctureonly elements. This is achieved by creating notches and slots along the pieces that constitute the design language in order to produce a frictionbased system of assemblies comparable to a LEGO kit, or a 3D puzzle. By relying solely on friction joints other materials like glue or nails are deemed unnecessary, and therefore the complexity of the fabrication and assembly processes is significantly reduced. The relevance of SMAs draws also on the assumption that in the close future digital fabrication devices will be cheaper and more accessible, having effects on the building industry of developed countries, enabling communities to engage in cheaper and quicker design-toconstruction processes. Under the light of these premises defining computational methods and procedures, as well as a solid theoretical framework, that enable creative design exploration using SMAs becomes a worthy arena of research. The following sections describe the computational methods used to develop a generative grammar for SMA fabrication under these ideas.
3.1. DIGITAL FABRICATION
The work presented in this paper builds on the knowledge gained in the experiences mentioned above, specifically those where the prototypes were constrained to 2D cutting techniques which include water-jet cutting, plasma cutting, CNC cutting, and laser-cutting. All the prototypes in this paper were fabricated using a Universal Laser Systems X-660 laser cutter (50 watt), which employs a laser of adjustable power and speed to cut planar sheet
Other reasons for this choice are explained in Section 3.
that sends instructions to the laser cutter to perform the cuts as specified in a loaded drawing file. the Sketchpad. is the ancestor of modern CAD systems (Sutherland 1963). the evolution of the interaction between the user and the machine as CAD systems developed can be described as a constant strive to mimic the interaction between the designer and his/her physical drafting table and drawing tools. The machine is physically connected to a computer equipped with a CAD software system (ACAD 2004). The notion of computer aided design as an automation of hand drawing and drafting methods was the guiding idea in the early development of these tools. When cutting. The prototypes presented in this document were built using 1/8” thick plywood and 1/16” thick cardboard. 2 Ivan Sutherland’s Ph. 4 and 5 axis that can vary the angle of the cut are available in the market at greater expense.2. which means that the arm that carries the cutting device (the laser) can only move in the X and Y axis. acrylic. and of variable thickness depending on the material employed. masonite. and Plexiglas.D. and until very recently. which evolved from the practical necessity of managing large amounts of information on designs in a faster and more efficient way than paper allows 2 . This characteristic constrains the laser cutter to make orthogonal cuts on the sheet of material. The process is similar to printing a document in an inkjet printer. Machines of 3. automated versions of the traditional drafting table of the architect or engineer. For the purposes of this document I will separate them into two categories. Interestingly the .1. This machine is capable of cutting sheets with a maximum size of 32x17in. as well as the age of the lens and other contingencies. 2. From that point on. 2. For detailed descriptions of different digital fabrication technologies refer to (Seely 2004). plywood. balsa wood. This burned section needs to be considered in the preparation of the file. CAD SYSTEMS There are several computer aided design systems in the market.2. consisted of a screen where the user could draw on using an optical pen. as the amount of material that gets burned by the laser depends on the type of material. A defining characteristic of this machine is that it is “two-axis”. and on the power and speed of the head. even though the line that divides them is not entirely clear. the machine is typically used to cut materials such as chipboard. thesis at the Massachusetts Institute of Technology. these packages are often regarded as “traditional” CAD systems because they can be seen as evolved. Solid Modeling and Drafting Systems Exemplified by software packages like AutoCAD and Micro Station. the laser beam burns part of the material.GENERATIVE FABRICATION materials.
The solid modeling and drafting package used for the first set of experiments is AutoDesk AutoCAD 2007. Two examples of CAD systems often referred to as parametric are CATIA. In his keynote speech in SIGRADI in November 2006. the flexibility afforded by parametric modeling makes these systems “the future of architecture” 4 .3. these allow users to define geometrical entities by establishing relationships of geometrical or mathematical dependency between different elements of the design. array. 3 CATIA was later adapted to aid in the architectural practice Gehry and Partners through a partnership between Dassault Systemes and what later became Gehry Technologies headed by Dennis Shelden. a set of elements that responds and changes in accordance to the dictates of the driving geometry. was called Digital Project. mirror.2. This hierarchy favors a top-down design manipulation (the changes in the general shape are propagated to the parts) over a bottom-up (the changes on the overall shape are given by the addition. also can be described as relational modeling systems. to a large extent a subset of CATIA. The power of a good parametric model is that once the constraints and chains of dependency are defined by the designer. described in the next section). and prevents almost completely the relational propagation of changes. etc. John Frazer also describes parametric . rotate. 2. This first CAD systems were implemented by the car industry in the sixties. and a driven geometry.. To the eyes of many professionals. and only decades later by the architectural industry. within the global and local constraints defined in the model. developed by Robert Aish at Bentley as a parametric plug-in for the solid modeling and drafting package Micro Station. developed by Dassault Systemes for the airplane industry 3 . The new software. the driving geometry can be changed in order to produce design variations that retain the logic of the constraints. solid modeling and drafting packages are primarily non-hierarchical and non-relational. and Generative Components. subtraction or interaction of elements). Parametric modeling is highly hierarchical. Parametric Modeling Systems The second category of CAD systems is parametric modeling systems. generally a small set of geometric elements and parameters that govern the behavior of the whole model. partly due to a distinction between a driving geometry.4 AUTHORS In contrast with parametric modeling packages (the second category. This favors a type of design moves based on design transformations and operations such as copy. 4 The expression was used by Peter Eisenman in a Spring 2007 lecture at the Massachusetts Institute of Technology.
The lexicon of the grammar is composed of a set of three compound types. A grammar composed of these elements can be used intuitively to compose spaces. modeling packages (specifically CATIA) as “the single most advanced piece of design software in the market today”. the geometric definition of these objects is stored in a 3-D DXF format. open to many choices. The sequences of assembly are studied so that there is only one assembly vector in each assembly move. The parametric modeling system used for exploring the possibilities of the grammar was the student version of CATIA V17 by Dassault Systemes. These sequences are meant to be deterministic derivations of grammar rules. Shape Grammars (Stiny 1978) provide a framework for the formalization of design languages. in a post-rational process a global shape is approximated by geometric strategies to infer the set of components needed for its construction. In contrast.1. Results 4. PRE-RATIONAL LOGIC The term “pre-rational” makes reference to a design strategy in which the components for the construction are defined beforehand. For a discussion on the origins and meanings of these two terms see (Loukissas 2003). . however. which in turn are composed of a total of 9 different basic types or morphemes. walls and endwalls returns a set of three basic lexical elements. 4.1.1. 4. The generation of a design with this set of building blocks is. The elements of the vocabulary are defined geometrically with the software tools. like a LEGO kit.GENERATIVE FABRICATION enthusiasm is notable since only very few firms in the world actually use parametric modeling packages due to their high prices and to the very high level of training required from both the designer and the fabricator: only expensive specialists can take full advantage of them (Wodzicki 2003). Lexicon A morphological subdivision of an abstract box in corners.
1. and defined to respond flexibly to many conditions. Morphemes It is important to note that the strategy explained above carries a great deal of design implications. Lexicon 4. Figure 2. and is a design decision in itself. Other abstract strategies will result in a different “universe” of designs that can be generated by using the grammar rules.6 AUTHORS Figure 1.1. Morphemes . The morphemes of the grammar were inferred from the abstract box morphology of the previous section. Some interesting alternatives were explored by the participants of the Digital Fabrication Workshop at the [INSTITUTION] in the spring of 2007.
1. rules of transformation provide the framework for generative design.GENERATIVE FABRICATION Figure 3. relationships of dependency between different values in the grammar can propagate changes parametrically. providing the framework for affording variation of the global shape. yielding a potentially infinite number of objects through operations such as addition and subtraction on the elements. In a grammar-based generative design process. and therefore can produce an infinite number of outcomes. Transformation rules In a generative process such as speech the application of transformation rules on the elements of a grammar is non-deterministic. Deterministic derivation of a corner 4.2. Furthermore. . This will be evidenced through a relational model built in CATIA and described in a later section.
4. these data must be defined in terms of mathematical relationships between different data fields. MATERIAL DIGITIZATION Transformation Rules Design entities such as lines. In order to make these flexible to global changes.2. curves and surfaces have specific data structures that allow the geometry to be constructed and reconstructed mathematically in modeling environments. or as a function of a global parameter that stores information about the material and structural issues.8 AUTHORS Figure 4. . points. or about the particular fabrication method of the element. This section shows how the understanding of material behavior affects the digital representations of the morphemes and the lexicon.
do not scale linearly. Coulomb experiment (Friction coefficient of plywood = 0. it is a function of the weight understood as normal force.01 in. The measurements and experiments resulted in an adequate value of 1/8 in for the H-joint piece. 1/8” Plexiglas.37) This study took into account vertical loads only. The experiment consists in finding the angle at which a surface is when an object placed on it starts sliding.2. 1/16 in. 4. 1/4 in and 3/4 in sizes.GENERATIVE FABRICATION The generality of the description will afford the versatility of the system to operate at different scales and materials (for example 1/32” chipboard. which in the model is determined by the pressure exerted by the braces caused by the tolerance of the joint. and a variable d. A key component of the grammar is de H-joint. as is shown in the following sections. Although the friction force is not a function of the area. Dealing with Friction The objective was to study the relationship between the length of the connecting piece (and therefore the friction area of the joint). and b) Confirm that the . and the strength of the connection. however. we chose a constant value for h (0.25 in). It was done using the Instron machine at the Lab. For the load measurements. The parameters taken into account in the grammar are a) Tolerance b) Thickness of material c) Size of the slot d) Distance between slot and end of the piece e) Depth of the slot f) Distance between slots. The goal is to find a joint size that a) is not loose but still b) can be assembled manually. As a result of these measurements and inquiries I was able to determine that a) H-Joints over 1/8 in behave almost the same. u = angle. and a tolerance of 0. in 1/8 in. or 1 inch plywood). and the basic behavior of the material used needs to be understood. Material properties. Figure 5. This machine measures accurately the strength required to pull apart the joints.1. The Coulomb experiment yielded the friction coefficient of plywood. In order to find an adequate dimension a series of test was conducted on several instances of the joint in different sizes.
Figure 6. I was able to determine that the right tolerance for the plywood I was using was 0. The tolerance value should be such that it a) provides firmness and b) can be assembled manually. displacement charts of two different joints 4. As a result of the trial and error process. and not by the area. Dealing with tolerance and non-perpendicularities The correct tolerance value needs to be assigned to the joints. Strength vs. Schematic diagram of friction force distribution along H-piece Figure 7.10 AUTHORS strength of the joints is to be determined by the tolerance.2. as this is the critical value that will provide rigidity to the assemblies. and of the geometrical solution of the angled slot problem. The assessment of the tolerance is to a large extent a result of a trial and error process.01“).254 mm (0. .2.
and it was to play an important role in the goal creating a flexible parametric model. Figure.2. and tol the value of the tolerance.GENERATIVE FABRICATION Figure 8.3. And that the variation of the slot size is given by the formula: W=[d-(2*tol)]/tan(o) + [L-(tol)]/sin(o) (1) Where w is the width. Finding this value is a precision aide in the modeling process. Variation of the slot relative to the angle 4. Dealing with structural optimization concerns . Finding the right tolerance in different materials A related problem is the assessment of the variation in the size of the slot relative to the angle of the crossing elements. L the width of the horizontal part. d is the width of the vertical part. Alpha (o) is the angle between the vertical and the horizontal elements. Figure 9.
thickness of materials. Any design generated by the sequential application of the grammar rules is constructible as the parameters for tolerance. however. (b) Parameters of the joint: The size and depth of the slots. These substitutions provide increased structural rigidity by breaking continuity of joint nodes along the structure (rule 1) or by conflating sheathing panels (rule 2). etc. between the end of a structural element and the intersection with another element. The structural behavior changes between scales even in objects of the same material. However in this chapter I provide three open strategies for structural adjustments to take place in an object derived with the grammar. An informed use of the strategies in this chapter can provide non-linear scalability to the system. (a) Modulation of the structure: By changing the distances between different structural elements.12 AUTHORS Actual examples of structural adjustment for specific designs made with the grammar are outside the scope of this thesis. It is worth noticing that material properties do not scale linearly. The optimal structural solution for a 1/8” thick plywood box at 1” = 1’ is not the same to that for a 1” thick plywood at full-scale. The “building blocks” of the grammar result in an inconvenient continuity of the connection nodes that increase the bending moments transversally to the continuity lines. The height of the structural element is also a defining factor of the structural performance of the object. which will affect specifically the resistance to certain lateral loads. Structural modulation parameters The structural performance of the objects that result of this sequential process of addition and subtraction is not. (c) Substitution rules: A set of rules of substitution specific for certain elements is defined. Figure 10. . structural modulation. optimal. and between the intersection of two structural elements and an end. are built in the members of the vocabulary. Figure.
and to modify this design flexibly by adjusting the global geometry of the object. holes can be opened. designs can change. 4.3. GROWTH AND FLEXIBILITY In this chapter it will be shown how the grammar can be used generatively to derive a specific design.GENERATIVE FABRICATION Figure 11. There is an infinite number of different sequences of rules that can generate the exact same closed box. As rules of transformation are applied nondeterministically to the components.3. Growth It is the ability of having open-ended design processes what makes this particular grammar generative. and there is an infinite number of design variations that can be performed on the object. Examples of Substitution Rules 4. . and new sections may appear. grow. The image shows the derivation of a very simple object. a closed box. It is worthy of noting that a pre-rational process with clear rules can be described computationally very easily by the sequence of rules that was used to derive the design.1.
or 1 inch plywood).3. The parameters taken into account in the grammar are a) Tolerance b) Thickness of material c) Size of the slot d) Distance between slot and end of the piece e) Depth of the slot f) Distance between slots. . CATIA software provides a framework to encode the dynamic relationships of dependency between the parameters described in section X. Using the grammar to derive a box 4.2 Flexibility The general description of the language will afford the system to operate with versatility at different scales and materials (for example 1/32” chipboard. A model was built of the closed box derived with the generative grammar.14 AUTHORS Figure 12. 1/8” Plexiglas. In the model all the faces except the floor can be changed and moved.
adapt easily to different materials and shapes. Due to its very hierarchical nature (everything gets created as a branch of a large data tree and is either “children” or “father” of its neighbors) there is no simple way to use go back to the “generative level” and operate changes in the topology of the object. however. Figure. The model can. Image of the CATIA model on screen . size and shape of the object derived with the grammar. Using the parametric model to flexibly modify materials. Figure 14.GENERATIVE FABRICATION Figure 13.
There is room for future development in computational design systems towards finding active ways to involve such constraints in design environments and dynamically explore material trade-offs. Model comparison table .design process. TABLE 1. The trade-off lies in that the highly hierarchical and constrained nature of a parametric model makes it virtually impossible to organically “add” or “subtract” elements to the composition (a characteristic of a generative process): a box remains a box. the same changes can take a tenth of the time (See Table 1). others are found along the process of design exploration. Flexibility Making small changes in the global shape of the design in a closed box is extremely time consuming using a solid modeling and drafting system. Architects who are familiar with parametric modeling often refer to this condition saying that the design ideas must be completely clear before starting a to model. some of which are imposed by the problem given. Growth vs.1. In a way “there’s no way back”. We’ve focused in discussing the role that material constraints play in computational design environments. this is the price of a certain amount of global to local design flexibility. keeping in mind that these constraints are a narrow subset of a much larger universe of issues to consider in any –even a very simple. In contrast. using a relational. Conclusions 5. 5.1. DESIGN TRADE-OFFS One of the many ways in which design can be seen is as a continuous bargain between the designer and the constraints. even if we can change its angles or the properties of its elements.1.16 AUTHORS 5.
Solution Space The prototypes show how a material and device dependent physical grammar can provide 1) a certain amount of design flexibility and 2) an optimal connection to fabrication technologies. and (b) The height of the box is at least 300mm. the non-hierarchical nature of solid modeling allows for addition and subtraction of elements (a typically generative or constructive approach). These boundaries can change substantially by manipulating the global parameters that control the structural and physical properties of the grammar. We were able to determine that for the material most used in the prototypes (1/8“ thick plywood). They also show how the inherent geometrical and physical properties of the design language affect and characterize the language by limiting it. 5. It is clear that the boundaries of the language define a constrained world of shapes and variation.GENERATIVE FABRICATION In contrast. but the trade off lies in that there are no ways to propagate changes across the composition. the universe of “safe” designs that the grammar is defined by the set of volumes that comply with the following two conditions: (a) The angles described by the walls and the horizontal plane are greater than 80deg and smaller than 120deg. As mentioned in the previous section.2. Two similar structures built from different materials using the same file. .1. Ease of Manufacturing vs. The parametric boxes show how this flexibility can be increased by defining mathematical relationships between the different dimensions of the elements and global parameters that relate the elements to specificities of the materials and other dimensions. Figure 15.
in contrast. and therefore are very impossible to fully predict in a digital 5 See (Loukissas 2003:33) . is a process where the construction technique is defined early in the design process. Figure 16.1. How flexible can building blocks be? How generative can fluid shapes be? We hope that the experiments presented in the technical sections of in this paper provided useful insights into this question in the light of the current landscape of digital design and fabrication technologies. often a post-rational process involves rational subdivision of the “fluid” shape into components. temperature.18 AUTHORS This is the inherent trade-off of facing a design problem with a design language fixed beforehand. becoming a major constraint in the form-finding 5 . PROBLEMS 5. These distortions are caused by different factors such as humidity.2. A post-rational process typically starts with a shape devoid of material considerations and then infers methods for it’s construction. Warping Sheet materials distort from their original plane in a phenomenon known as warping. A pre-rational process.2. Two corners where the components are generated automatically by the model when the shape and/or material of the box changes. 5. This problem is usually described by the opposition between two mutually exclusive processes: post-rational and prerational.
CONCLUSIONS: TOWARDS GENERATIVE FABRICATION This paper exposes the lack of an integrated software platform that simultaneously supports both growth and flexibility as design operations. b) allow for both local and global dynamic design variation c) have a detailed and extensible construction knowledge-base. Figure 17. 5. 5. CONTRIBUTIONS - A generative grammar for manufacturing 3D objects from sheet materials that: o Is based solely on friction. The explorations conducted in this research suggest that in order to achieve this. Possible solutions can be .3.A system where the panels instead of being fixed by friction can be “slidein”. acknowledging it and providing a margin of movement to the panels. A novel kind of design tool can be outlined on the basis of these observations: a tool that is able to involve material knowledge in the early conceptual stages of design and keeps the design process constrained to the requirements of specific material and device concerns.GENERATIVE FABRICATION model.Additional joints that connect perpendicularly between panels . Warping Although the prototypes show that the structural elements of the objects derived with the grammar are able to self-correct the errors produced by the warping of the plywood sheets.4. this is not the case for the paneling. . Warping falls perfectly in the category of “sticky and inconvenient contingencies of dealing with physical matter”. the design tool needs to a) allow a designer to switch between different levels of representation (considering fabrication-ready information as one of these levels). and easy to implement material constraints. Instead of fighting the warping with more material. and warping has a negative impact on the sheathing that needs to be addressed in order to advance the system (in an extreme case of warping the panels can tend to pop-out from their positions in the structure).
Department of Architecture Massachusetts Institute of Technology. K: 2006. L: 2005. Architecture. Unpublished. Shape Grammar in Education and Practice: History and Prospect. Thesis. Automation in Construction 13.S. L and Oxman. Oxman. California. 51-67(17). Sass. Sass. Seely. Thesis. Rulebuilding: Exploring Design Worlds through End-User Programming. Massachusetts Institute of Technology. M. Customized Digital Manufacturing: Concept to Construction Methods across varying product scales. Stiny. Sass. Digital Fabrication in the Architectural Design Process. Thesis.S. R: 1997. Department of Architecture. L with Shea. School of Electrical Engineering and Computer Science. Sutherland. J: 2004.S. Massachusetts Institute of Technology. C and Lee. September 21.D. University of California Press. Design by re-representation: a model of visual reasoning in design. 18. Sass. Design Production: Constructing free form designs with rapid prototyping. Ivan: 1963.S. Boca Raton FL. o Can grow. K and Powell. 4(1). Department of Architecture. International Journal of Architectural Computing. Design Computing of Complex-Curved Geometry using Digital Fabrication Methods. Environment and planning B 7(4): 409-462. A Physical Design Grammar: A Production System for Layered Manufacturing Machines. Ph. M. pp. Berkeley and Los Angeles. G and Gips. Department of Architecture. George: 1980. Knight. and can be flexibly changed. Eastman C: 1999. dissertation. Y: 2003. Thesis. Griffith. T: 2000. M. Materializing Design: the implications of rapid prototyping in digital design. . J: 1978: Algorithmic Aesthetics: Computer Models for Criticism and Design in Arts. A Wood Frame Grammar: A generative system for digital fabrication. M: 2005. L: 2007. M. G: 2004. Design Studies. Sacks. Loukissas. Massachusetts Institute of Technology. References Botha. Building Product Models: Computer Environments Supporting Design and Construction. Portugal.20 AUTHORS - o Can be scaled in a non-linear manner o Can be used with different materials. R: 2006. Sketchpad: A man-machine graphical communication system. Acknowledgements 6. 329-347. An alternative to post-rational views and strategies in digital design and fabrication. Stiny. M: 2006. January 2006. Paper presented at the ECAADE conference. The conceptual framework for a generative-fabrication software tool. Parametric 3D modeling in building construction with examples of precast concrete. Design Studies 27 325-355. The International Journal of Design Computing 2 (1999–2000). 291-312. Massachusetts Institute of Technology. CRC Press. R with Eastman. Kindergarten Grammars: designing with Froebel’s building gifts.