Figure 177 – uploaded by Axel Kilian

See full PDF downloadDownload figure

Image 1: Open Theater — Side view of the bearing structures — Figure 177 Image 1: Open Theater — Side view of the bearing structures

Related Figures (242)

Figure 1: (a) Hyperbolic Hexagon, (b) 7-story Scherk Tower, (c) Heptoroid (seven 4'"-order saddles). My interaction with Collins started when I encountered a photo of his Hyperbolic Hexagon (Fig.la) [COLLINS 1997]. Seeing his intriguing, highly structured sculptures, I wanted to understand their underlying generative paradigms. One way to interprete Figure la is to describe it as a ring of six consecutive hole-saddle combinations, like the ones in the center of Scherk’s 2™ minimal surface (Fig.1b) [SCHERK 1835].

Figure 2: (a) Totem 4 sculpture, (b) doubly-wound toroid. Although Sculpture Generator I is based on only one single geometrical module that gets bent, stretched, twisted, and reused in many different ways, it can produce an amazingly wide variety of different sculptural shapes. After I had the basic program running in 1995, I introduced several different paradigm extensions over the following years. The simple biped saddles was replaced with saddles of higher branching orders (Fig.1c). Affine stretching of the toroids produced totem-like sculptures (Fig.2a). Letting the hole-saddle chain loop around the toroidal ring more than once led to intricate interleaved structures (Fig.2b).

Figure 3: (a) Pax Mundi, (b) modulated 4-period Gabo curve. embedded in the surface of a sphere. The dominant undulations reminded me of the edges in sculptures by Naum Gabo, and I thus defined an “‘n-lobe Gabo curve” as a generalization of a baseball seam: a meandering curve completing 7 full cycles as it traverses around the globe along the equator. This curve was parameterized not only by the number of its cycles, but also by the amplitudes of the individual lobes, and by their width and pointiness (Fig.3b).

Figure 4: (a) Via Globi - Maloja, (b) Music of the Sphere. In this particular “sculpture generator” we also need to specify its cross section, and the way that it is rotated and scaled as it is swept along the base curve. Rather than writing another stand-alone program for generating sculptures of this kind, I used our modular modeling environment, Berkeley SLIDE [SMITH 2003], which already had a powerful sweep generator with all the necessary controls. Thus I just needed to add two modules for specifying the sweep curve on the sphere and for specifying a cross section. With these elements in place, it was then easy to generate a wide variety of such Viae Globi (“Roads-on-a-Sphere”) sculptures (Fig.4a). A few years later I could also easily accommodate a paradigm extension that moved the sweep curve away from the sphere surface and allowed it to make internal loops, in order to emulate Collins’ Music of the Spheres sculpture (Fig.4b).

Figure 5: (a) Volution_0, (b) Costa surface of genus 2. The Volution elements shown in Figure 5 are all based on twelve edge constraints in the shape of quarter circles, two each lying at opposite corners on the six faces of a cube. The suspended surfaces of different connectivity, ranging from genus 0 to genus 10, were inspired by the tabulation of triply periodic minimal surfaces found on Ken Brakke’s webpage [BRAKKE 2000].

Figure 6: (a) Volution-wall, (b) Knot-wall. The elements shown in Figure 5 not only make attractive abstract sculptures, but they also can be used as modular architectural building components. One obvious composition follows from the regular periodic surfaces shown by Brakke [BRAKKE 2000]. However, since many different surfaces of different genus can be suspended in the same set of curved edges on the cube surface, different elements can be mixed and matched with different orientations to construct a wide variety of architectural walls, reminiscent of the work by Erwin Hauer [HAUER 2004]. Figure 6a gives an example of such a modular assembly.

Figure 8: (a) Moebius bridge; (b) Moebius building. In case of the Moebius building, the cross section is kept vertical in the upper, S-shaped part, to accommodate several stories of apartments or offices. In the straight return path at ground level, the window facades of the upper portion turn into sky-lights for common function rooms such as, indoor atria, conference rooms, galleries, shopping malls, or sports facilities.

Figure 7: (a) MES, (b) MVS, and mixed optimization functional. Thus it is worthwhile to look for other functionals that might make a different tradeoff and lead to a different distribution of local curvatures. In the early 1990’s Henry Moreton explored Minimum Variation Surfaces (MVS), based on a functional that minimized the surface integral of the square of the change of curvature in the principal directions [MORETON and SEQUIN 1992]. It led to shapes with more distinct, nicely shaped toroidal arms (Fig.7b). Since then we have experimented with a few other functionals based on curvature changes. Pushkar Joshi has explored an MVS functional that also included mixed derivatives (Fig.7c), as well as weighted mixtures of the various functionals [JOSHI and SEQUIN 2007]. This work will eventually lead to an environment where a designer can choose from a variety of surface optimization styles that will best satisfy his or her sense of aesthetics.

Figure 9: (a) Pax Mundi, sagging; (b) final installation. One danger with using purely geometrical design tools that are not tied in with any physical simulation tools or any verification software for the intended fabrication process is that it is easy to forget the physical aspects of the emerging construction. In 2006 Collins and I received a commission to scale up the original, 2- foot diameter Pax Mundi wood sculpture to the 6-foot level and to turn it into a bronze sculpture for the H&R Block headquarters in Kansas City.

Figure 10: (a) The half-wheel master; (b) final EG-award trophy. Keeping the complete fabrication process in mind becomes even more important when one is asked to make many copies of the same object. This was the case in 2007 when I got the commission to design an award trophy in bronze to be handed out at the annual Eurographics conferences for a Distinguished Career Award, a Technical Contributions Award, and a Young Researcher Award. In total the conference management wanted about twenty copies to honor all past recipients, and they are planning to award about three more trophies in every coming year.

Figure 1: Various architectural models analyzed and modified by exploiting information on shape symmetries and regular repetitive patterns

Figure 2: To detect symmetries in geometric models, we uniformly) sample the boundary of the shape (left). Every pair of samples with compatible local surface geometry provides local evidence for a symmetry transformation (center). In this example we consider re- flections that are parameterized by an angle @ and the distance d to the origin. Accumulating such evidence using a clustering ap- proach yields the dominant symmetries of the model (right). The surface of the model is sampled uniformly with average sample spacing h. The user parameter h determines the scale of the small- est symmetric elements that we want to detect. For every sample

Figure 3: Schematic illustration of regular structures. The helix and spiral are generated by transformations that combine rotation with translation and scaling, respectively. The bottom row shows the three types of commutative 2-parameter groups that can be de- tected with our approach.

Figure 4: Large-scale symmetries detected in a digital model of the Sydney opera. The extracted symmetries include reflections, as well as general similarities that involve uniform scaling, rotation, and translation.

Figure 5: Structure discovery and model repair on a laser scan of a complex outdoor scene. The algorithm fully automatically discover: two translational grids within the acquired point cloud. Standard surface reconstruction yields an incomplete and inconsistent triangulatior shown in the zooms on the left. The models on the right have been created by augmenting the point set using replicated samples from the representative elements prior to reconstruction.

Figure 6: Conceptual differences between bottom-up structure discovery and top-down symmetry detection. The transparent bounding boxes (lower right) show the extracted symmetry hierarchy that can be utilized for symmetry aware part replacement as shown in the top right. The repetitive translational and rotational structures shown on the lower left support more fine-grain edits to individual structural elements.

Figure 8: The input model in the top left corner has been analyzed to reveal three dominant repetitive structures, illustrated in the top right. The zooms show the corresponding structural elements. A new design has been created by modifying the number of repetitions and replacing the repetitive elements with new geometry. Figure 7: Structure discovery and procedural modeling on a building facade. The regularity patterns of the model on the left have bee extracted automatically and can be modified by the user to alter the facade design as shown in the middle. For comparison, the image on th right shows the original model scaled along the horizontal axis.

Figure 9: Symmetrization illustrated on an architectural design study. The top row shows how one of the symmetric elements evolves during the optimization. After processing, the six-fold ap- proximate symmetry of the original model is perfect both in terms of geometry and meshing.

The key consideration for the architects was to retain overall geometric control of the stadium. This was achieved by using a combined model, the core component of which was a spreadsheet containing a: Il numeric parameters. This was accessed by a script file that described all geometric rules and relationships for constructing the stadium geometry. This package could then be issued to the engineers. The underlying geometric construction method used an array of similar curved sections arranged radially around the building footprint (figure 2). Variation in these sections was controlled by a set of control curves that mapped the horizontal o r vertical change of each of the points defining the section. Eac h sectional curve defined the centre line key structural roof members.

Lansdowne Road site (figure 1) was highly constrained by boundary conditions. These dictated rights-to-light planning restrictions and horizontal expansion limits defining a possible

Figure 3: Sectional Curves. Experience from structural engineers of steel facade constructior determined that facade geometry should be determined using tangential arcs. This reduced the range of possible types ot primitive geometric elements for defining the geometry and alsc indicated what parameters were needed. The precise descriptions of relationships between geometric elements emerged throug! development and use of the model. Initially the geometric relationships were judged aesthetically to not deliver enougl curvature to sections (figure 3). The relationships wer«

References Figure 5: Cladding panels.

Figure 6: Facade representations.

Figure 3: The Last 12 Steps of First Test In a testing phase we checked the functionality of this algorithm for simple 2 and 3-dimesional systems. Based on a single-span lattice truss, the regular diagonals are omitted. (Fig 1) So we started with a girder with a quantity of nine nodes at the upper and the lower chord. These nodes are possible connection points for diagonals. To reduce the number of possible solutions, the angle of each diagonal is by definition limited to above 18°. Despite a quantity of about 6“11 solutions is possible. The positions and the angles of the diagonals between the chords are generated by the algorithm.

In the next step we increased the number of nodes of the upper and lower chord and started a new process. The result was an irregular structure that features a lower deflection than the regular latticed girder. (Fig 5)

For generations with a low number of iterations, systems were evolved that were near to the optimal solution. (Fig 3) If the number of iterations is chosen high enough, the outcome was a traditional lattice girder. (Fig 4)

In a last test we changed from a 2-dimesional system into a 3- dimesional latticed girder grid. This structure is supported at four corners. The upper and lower chord follows a rectangular grid, diagonals are freely arrayed between these. Again the found solution possesses a lower deflection than a regular one. (Fig 6) Near the supported corners a concentration of members can be perceived, according to the concentration of lateral forces in this area.

Figure 7: Irregular Latticed Girder The described algorithm was then used for the structural design of a pedestrian bridge by FloSundK architecture. The loads are carried by two girders located at both sides of the footpath. These girders are twisted so their chords describe a hyperbolic paraboloid shell. Supports are located at four positions. As the decline of the diagonals density from one side to another was requested, the number of nodes at which diagonals are attached is also decreasing.

As diagonals are needed not only in the crossing points of the chords, horizontal moments in the chords were taken additional into account. A concentration of members occurs besides the rectangular hole in the middle. A high stiffness of the system was reached, while the weight of the structure was comparable to a regular system. (Fig 10)

To reach sufficient load bearing capacities, both girders have to be connected to form a space frame. These boundary conditions make it almost impossible to find a regular structure by a engineering approach. References The Roof of the Rhein-Main Arena in Frankfurt by Coop Himmelb()au covers an area of 90 by 120 m. The volume consists of a twisted box. For the roof an upper and a lower girder grid was defined by the architects. These grids are according to the shape of the box twisted to each other. Furthermore the lower grid is folded. (Fig 9)

Figure 1: Morphing Tilings, 2005 The experimental setup for the fagades presented here is based on an experiment launched in 2005, where self-deforming geometric tiles (Fig. 1, www.counton.org, National Centre for Exce lence in the Teaching of Mathematics) were automatically distributed using a simple PERL script in combination with the CAD- Software Vectorworks. Here the degree to which eac h tile is deformed is steered by the level of brightness of a bitmapped image, where both extremes of deformation are linearly linked with the brightness of the corresponding region of the image (0% to 100%). The position and size of the tile relative to the entire surface defines the region of the bitmap of which the brightness will be sampled. This setup requires that the proportions of the geometry of the surface have to correlate to those of the bitmap.

Figure 2: Bitmap-driven Morphing Tilings, 2005 2 Fountainhead

The geometry of the architectonic volume was determined, and the facade was depicted as a horizontal surface. Afterwards it was determined how the structure should react to each relevant parameter. These connected, often geometric changes in fagade elements were assigned a brightness level (0% to 100%) according to their deviation from the norm. These brightness values were then distributed over the fagade according to the parameter in question. (Fig. 3, Views to the courtyard, Fountainhead) Figure 3: Views to the courtyard, Fountainhead, KCAP, 2005 Based on the nature of the parameter and correlating geometric reaction, the relation of the brightness values could either run linearly or erratically.

The number of possible deformations is arbitrarily high, although the so-called boundary values have to be fulfilled in every deformation condition of a facade element. Since these boundary values cannot be verified automatically at this point in time, in practice this often leads to a restriction to a manageable number of parameters that steer the precise form of the elements. (Fig. 4, Sample element Fountainhead)

An evaluation of a number of parametric CAD programmes determined that Digital Project/CATIA by Gehry Technologies fulfilled the specific requirements the best. It is now possible to draw the geometric skeletons of fagade elements three- dimensionally and graphically evaluate all of the resulting geometries directly without using script at all. Since all of the essential decisions are now taken outside of the script level, the procedure becomes controllable by the architects:

The designer initially draws the fagade element parametrically, in form of a three-dimensional model. As a result, the architect is free to deform individual construction elements in any area, and the allowable movement can be linked in the end to a parameter. This parameter can be set externally at a later point in time through the script, which executes the tiling. As long as the number of bitmaps matches the number of parameters of the CAD model, the bitmaps as well as the CATIA model can be modified without resorting back to the script level. (Fig. 7, Example CATIA model from the Brave Tailor course) Figure 7: Digital Project Screenshot, Brave Tailor Course 2006

As soon as all parameters are encoded to a bitmap, a script written in PERL maps these parameters to the corresponding elements. The output is read in by a script written in CATIA which tiles the individual fagade elements. Since t his process requires very large resources depending on the complexity of the 3D model, individual scripts are generated for each surface of the volume. Finally, the complete facade can be exported and developed further with the software of one’s c Brave Tailor course) hoice. (Fig. 8, Results from the Figure 8: Resulting Fagades, Brave Tailor Course 2006

Figure 9: Facade Studies, MAS CAAD 2008

The final technical obstacle is the necessity to install a PERL environment. At the moment, we aim to translate the existing scripts to the now popular script language “Processing”, where a graphic user interface will make the procedure accessible to an even larger circle of people.

A number of concepts from the fagades presented here have been used for research projects, which investigate the translation of complex designs into fagades with parametric elements. Due to the tediousness of real construction processes, none of these projects has been completed at this time, however, it is becoming apparent that here as well, the generation of geometries using scripts should be replaced by their generation using parametric CAD software in the long term in order to be feasible by architects without specialized skills in computers. (Fig. 12, Stidpark, HUM & CAAD, 2006-2008) Figure 12: Siidpark, HdM Basel & CAAD, 2006-2008

Figure 1: circles & ellipses; offset Keywords: architecture, mathematical origami, curved creases, developable surfaces

Figure 2c: combined circles, ellipses & parabolas

Figure 4a: ellipses; offset on center Figure 3: circles; center shifts left and right

arose from our experiments. Perhaps most exciting is the wide variety of three-dimensional forms resulting from sometimes subtly different crease patterns, leaving a broad spectrum for design even within the context of pleating. Also intriguing is that shifting offset ellipses (as well as circles) alternately back and forth along a line, as shown in Figure 3 and 4c, results in a “twisted” folded form that lacks the mirror symmetry of the crease pattern. In contrast, shifting offset ellipses monotonically in one direction results in a mirror-symmetric form, as shown in Figure 4b.

A more negative example is the combination of three parabolas, shown in Figure 5, where it appears impossible to fold along all creases by a positive amount in the desired direction, resulting in a flat area. This outcome is not surprising, given the close proximity to a straight-crease design of concentric triangles, which behaves similarly. More interesting is that the closely related model shown in Figure 6, with two parabolas and a somewhat larger circular segment, folds nicely into a three-dimensional form with precisely the desired creases.

Figure 6: 2 parabolas & circle; scaled; midpt. in center

Figure 7: ellipses; offest w/ midpoint in center metal & polycarbonate The challenge regarding an architectural implementation is to find elastic materials that fold into these natural shapes, without showing additional creases, while being suitable for exterior applications. The proposed fabrication method is based on perforations, because a series of small holes can act as a guide for bending. S-shaped dashes for these perforations help metals bend easily [Ori]. Our successful experiments shown in Figure 7 were made of polycarbonate and steel cut with a water jet. This method also seems very promising for thicker sheets of aluminum.

Developable Surfaces with Curved Creases Figure 1: Top left: Reconstruction of a car model based on a felt design by Gregory Epps. Close-ups of the hood and the rear wheelhouse are shown on the left. The fold lines are highlighted on the car’s development. Top right and bottom: Architectural design. All shown surfaces can be isometrically unfolded into the plane without cutting along edges and can thus be texture mapped without any seams or distortions.

Figure 2: Two examples of paper models featuring curved folds that were created by David Huffman.

Figure 3: The car model of Figure I and its development (toy right). The patch decomposition into torsal ruled surfaces is showr using the following color scheme: planes are shown in yellow, cylinders in green, cones in red, and tangent surfaces in blue. Sam- ple rulings are shown on some patches of the windshield and the side window. Such a segmentation is essential for NURBS surface fitting and manufacturing.

Figure 4: A PQ strip (left) is a discrete model of a developable sur- face & (right). The intersections of edges piqi of adjacent planar quads generate the regression polyline r;. In the limit of a refine- ment process, this regression polyline becomes the regression curve r(t). Polylines C’, whose edges c;ei+1 intersect inner bisectors oj consecutive discrete rulings at right angles, are discrete versions of principal curvature lines, and serve for the definition of discrete curvatures. The unit normals to planar quads P; are denoted by nj. The basic optimization algorithm simultaneously optimizes a dis- crete developable surface M and its planar development P. To maintain isometry between corresponding faces of M/ and P, we originally let / be a quad-dominant soup of planar polygons M‘ in space. These polygons are isometric to the corresponding faces P* in the planar mesh P, see Figures 5 and 6. During the optimization, the polygon soup M/ will become a mesh via a registration proce- dure which bears some similarity to that used in the PRIMO mesh deformation tool [Botsch et al. 2006]. However, our optimization requires more sophistication since we have to simultaneously opti- mize the development P while satisfying various other constraints.

where the sum extends over all (6) combinations per vertex p € P, and over all vertices in P. Figure 5: Basic setup for the optimization when a reference surface D is used. Faces with the same color are congruent.

Figure 6: Top left: Initial polygon soup M. Top right: Develop- ment P. Bottom left: M after subdivision and optimization. Bottom right: M after three rounds of subdivision and optimization.

BotTscH, M., PAULY, M., Gross, M., AND KOBBELT, L. 2006. Primo: coupled prisms for intuitive surface modeling. In Symp. Geom. Processing, 11-20. in space and the side condition Fony is also not needed. Hence, the spatial sub-problem amounts to solving a sparse linear system, and subsequent application of the corresponding rigid body motion per face. Optimizing the development P is more involved since the weights w;; change in a non linear way as the geometry of P changes. Additionally we have a quadratic term Feony to maintain convexity as a side constraint. With the meshes scaled to fit inside a unit cube, we found \ = 1 and jp = 107“ to be good values to start the optimization.

3.2. Spitfire All cross-sections and outlines of this aircraft are composec of parts of elliptic curves. Deliberate restriction to a specific geometric method of construction. The tool is chosen tc achieve a highly homogenous shape. In contrast to the pre- vious case the choice of the tool is part of the design process.

Anonymous architecture, construction of a tower with the simplest means, without blueprint: shelves hammered _ to- gether. Very functional, significant. The tool acts as a trans- porter of the raw idea of a minehead, despite its limitations. It shapes the formal properties of the product.

The figure of Spider-Man swings in the silk threads of a spider in a stunning move along the truss of a bridge. This 3.3. Spider-Man

The Library building has a floor plan that is similar to an amoeba. It is the composition of a series of arcs, which seamlessly merge. It could be that the idea of geometry in this form is derived from the properties of the PolyArc—

3.4. Heaven A search on the internet for images with the word “render- ing” in the name reveals a great number of visualisations of buildings under bright blue skies with white clouds. Surpris- ingly, some of them resemble each other strongly. In some cases the tool used is the ArchiCAD background image li- brary, the image “heaven_clouds.jpg”. The tool is capable of lightening up any architecture with the brightest investor optimism. It adds content to the design.

3.5. Stairmaker State of the art architectural CAD software allows the user to work with data-enhanced parametric objects. With each new version of the software the parametric models included will become more comprehensive and more complex. The current version of ArchiCAD includes a tool for the design of stairs that knows fourteen different types of stairs. This tool adds content. The authors of the design of the stairs are software engineers.

In analogy to expressionist architecture in Germany in the twenties, in 1993 Philip Johnson designed a building for the Friedrichstrafe in Berlin. Unfortunately it was not built. The walls of the outer shell of the building are assembled from a series of double-curved surfaces, which are joined with creases. If we study the published drawings, we can clearly see that large parts of them are drawn by hand with techni- cal pens. Other parts look like a composition of printings of several screenshots. Actually, the drawings are a bricolage.

The design of one’s own task-specific design tools has itself to be considered as being part of the design process: to de- sign by tool design. Paul de Casteljau worked in the development department of Citroen. He did research on a mathematical model, which allows to accurately describe the comprehensive geometry of the bodywork of a car. In 1959 he succeeded to develop an algorithm as the foundation of his work. “De Casteljau’s algorithm” generates a type of parametric curve, which is today well known as Bézier curve. Pierre Bézier was working for Renault, he independently invented the same curve, and published it in 1962. It was bad luck for de Casteljau that his work at Citroen, in contrary to Bézier’s, had been sub- ject to total confidentiality until 1975 [MULLER, A. 1995]. In 1970, the Citroen GS hit the market. Its bodywork was, for the first time, completely describable with mathematics [PARIZOT, S. 1971]. De Casteljau reached his goal.

Logo is a general purpose programming language with a very simple syntax that offers the possibility of immediate graphic output and the properties of a high-level language

A tool to repeatedly chop a number of segments with a chopping knife, rendering nothing but powder. A tool to study mechanisms of destruction. A subsequent tool will chop three dimensional objects.

A tool that folds a square sheet three times. Predecessor of a tool that can fold origami figures. The intermediate stages of the folding movement are being materialized.

A tool that places rectangles of different size and proportion within a big rectangle without any gap. Predecessor of a tool that organizes spaces of various size in three dimensions.

A tool for the production of Voronoi diagrams [DE BERG, M. ET AL. 2000] for specific requirements. The initial data for the diagram was imported from Rhino to Logo. After the calculation the diagram was exported back to Rhino for fur- ther processing.

A tool for generating irregular polygonal surfaces with the main feature that all surfaces should have exactly four sides. An anti-triangulation-tool. References

A tool for reverse engineering the work of an artist. A string with the length of one kilometer is repeatedly knotted. The end product is one big knot. The tool uses a recursive proc- ess to insert knots.

A tool that produces the complex geometry of underground wormholes of an imaginary species of worms. Subsequent tools will be able to intervene specifically in the fabric of the city.

A tool for the production of Voronoi diagrams for specific requirements. After its calculation the diagram was exported to Form-Z for further processing.

Fig 1 — Neumunster Abbey and Jean-Jaures Metro Station Two projects, the glazed roof in the main courtyard of the Neumunster Abbey in Luxembourg [1] [2] [3] and the glazed roof of the Jean-Jaures metro station in Toulouse [2] [3] - designed in 1999 and 2001 respectively - use facetted surfaces of both quadrangular and triangular glass panels. In these two cases, the roof has not been created by assembling all different elements; rather, the structure and the glazing system are made of standardised elements.

4.1. The twisted glass of the Avignon TGV station Moreover, the establishment of the arcs implies that the glass panels are all different in plan, but are derived from two double- curved shapes: the spherical and the toroidal. In this way, it was possible to minimize the number of moulds and to produce all the glass panels using only a few oversized moulds. Annealed glass, which is laminated, was used in order to assure overhead security.

Fig. 4 - Avignon High Speed Train Station he panels are rectangular with a high aspect ratio as result of the cold-bending technique which allows the glass to form over the surface: the flat panes are elastically bent and forced onto the structure during mounting. This technique proved feasible since the stress due to twisting is negligible with respect to the stress due to the action of wind. The climate and the consequent thermal requirements demanded a double glazing, adding extra difficulties due to the sensitivity of the periphery joints to the effect of twisting. In order to contain the stresses in the sealant joints and to guarantee the long life of the glazing unit, a “pressure equalized DGU” (vitrage respirant) was developed and chosen in preference to the standard double-glazing. The sustainability of this project is based on the development of two techniques that were innovative at the time, 1997-2001: the cold-bending and the pressure equalization of the chamber of the double glazing.

Fig. 3 - Avignon High Speed Train Station

Fig. 6. - Strasbourg High Speed Train Station The resulting panel composition is laminated glass of two 6 mm-thick toughened panes. Due to the innovation and the large size of the skin envelope, the glazing technology and the mounting methodology were subjected to a severe testing and validation process run by the French building authorities.

Fig. 5 - Strasbourg High Speed Train Station The cold-bending technique also depends on the nature of the support. The arcs are not circular, so that the glass panels have different radii varying from approximately 11 to 30 meters. Therefore, the cold-bending process has been optimised for the various radii, in particular the tighter ones.

Fig. 7 — Louis Vuitton Fondation, Paris The choice to base the design on cylindrical developable geometry displaced part of the complexity from the skin to the structure. The feasibility of this project is the result of the integration and synergy between the glazing technology and the production constraints of both the fixing elements and the bearing beams, which twist in order to follow the surface. Of course, this complexity is managed using an advanced tool like Catia that takes advantage of the parametric nature of the software. Also, Catia generated a very precise model which included all the elements of the design and gave a very good representation of what it would be in reality. Catia conferred the advantage of being able to detect and solve during the design stage all the problems that might be encountered during the construction phase.

Fig 8 — Experimental geometries The first results of this research allow us to imagine new transparencies and non-conventional shapes, so that the complexities seen only in science fiction and in computer rendering are not far from being realisable, and may became a reality in the short- or medium-term.

Figure 1: Semi-discrete models as limits of discrete models. Par- tially subdividing quadrilateral meshes with vertices pi,; and pla- nar faces Pi,j Pi+1,j Pi4+1,j+1Pi,j+1 yields, in the limit, a D-strip model consisting of developable strips Y;. Each strip is bounded by edge curves p;(u) and pi41(u). We call the polygon with vertices Pi(u), p2(u),... a ruling polygon. Architectural freeform structures from single curved panels

Figure 2: Szervita Square, Budapest. A project designed by Zaha Hadid Architects. Example of approximating the outer shell by c D-strip model aligned with planar, parallel sections given by the three lowermost floor slabs. Sections, corresponding points usec for initialization and the resulting D-strip model are shown from top to bottom. The D-strip models may be used for a further ap- proximation using flat panels and cylinders. Conical strip models (Fig. 2): Each point p;(u) of an edge curve is the vertex of a rotational cone which is tangent to the two ad- jacent D-strips along their rulings. Hence, the tangent forms the

Figure 3: Left: A conical D-strip model is characterized by a sim- ple angle equality between the edge curve tangent and the rulings meeting there. Right: Conical models possess conical offset mod- els at constant distance d which appears between corresponding rulings and tangent planes.

Figure 4: Left: Assembly of wooden strips onto the framing for the interior of the Disney Concert Hall (courtesy Gehry Technolo- gies). Center and Right: Example of a simple geodesic strip model and its development. An edge curve of the geodesic model leads to oppositely congruent curves in the development (blue curve pair).

Figure 5: Geodesic D-strip models (in total five) which cover the interior of a freeform surface.

Figure 6: Box shell structure derived from an offset pair of conical strip models.

Figure 7: Multi-layer structure formed by two D-strip models in- terpolating parallel PQ meshes.

Figure 8: Positioning a curved I-beam with developable stem and developable horizontal bars along the edge curve of a conical model. be obtained via conversion to a circular model). Glass panels are aligned with further offset models. A technique for mounting the panels is sketched in Fig. 9.

Figure 9: Connecting the panels to the supporting beam. Acknowledgements

The intention was to create a space without orientation. A perimeter boarder is given by a rectangular box in which continuous surface is placed. This surface divides the space into cavities which could be either an exterior or interior part of the building. The smoothness of the surface is not only favoured for aesthetic reasons but is essential to obtain an efficient structural system. The shell type structure doesn’t

distinguish between wall and slab as their transitions are accruing continuously. By using a smoothing subdivision algorithm like Catmull Clark the surface is controlled by an array of polygonal quad facets which are providing the starting geometry for the process. Each facet is subdivided into a new set of vertices and then averaged between their adjacent neighbours. The crude starting grid densify and transforms — it is emerging. Keywords: Smoothing algorithms, Analysis automation

Crude mesh geometry

To gain full control over the process and the ability to extract specific data during the execution the algorithm was programmed from scratch by using Rhinoceros as visualisation engine. Specific requirements which were requested by the project’s geometry are added to the process. All vertices which are coinciding with the subscribing box are constrained on this perimeter and are only allowed to perform a 2D smoothing. After the smoothing is performed and a new set of geometry objects is created wherein topology and connectivity of the elements are detected and as user data attached to the drawing objects to create an extended geometric model. 4 Structural Model Generation Manipulation of the vertices

Performed Smoothing

The minimum data set which needs to be communicated with the architect is reduced to the crude mesh information. The developed tools enable to create the structural model in and optimized process and to gain almost an instant response to architectural design changes. In reverse structural optimized versions of the mesh geometry can be proposed and communicated back in the same way.

FE mesh from executed Input file The algorithm browses through all geometry elements of the extended geometric model extracting their attached topolog) and connectivity data and writes a simple ASCI input file which can be executed instantly in the FE package. Also loading patterns, support conditions and material data is already included in the processed model.

FE-Analysis showing increased structural performance

Structural reinforcement is placed over the geometry defining truss cage to facilitate off site fabrication as much as possible. Over the structural reinforcement, 3 layers of steel wire mesh are attached with appropriate cover distance to form the doubly curved formwork. Wire mesh is attached to the cage at 200mm pitch to control the geometry and to avoid the excessive deformation during concrete casting. This enables the 3D form to be approximated without expensive 3D tooling such as CNC milled Styrofoam. To develop a construction method to realize the free form geometry has been an equally challenging task for the design and engineering team. After a series of comparative studies, the team concluded that the Truss Wall System (TWS) developed by Asahi Glass Build-wall (AGB) is best suited for the project. TWS creates 3D surfaces from series of parallel 2D form defining non structural reinforcement trusses. Continuous 3D surface is divided in to parallel and radial zones based on the pre-smoothed crude mesh geometry.

Concrete is cast in situ and rendered manually after the structure is cured to the required strength. Rendering thickness of 25mm absorbs the construction tolerance and the out of plane deformation of wire mesh due to the hydrc static pressure of concrete prior to curing.

Full scale Mock up after pouring

Advanced technologies are used in building automation systems to control HVAC installations, lighting and other commonly dynamic building features. However, the ways in which building automation systems normally operate is not interactive, but pre- programmed and reactive.

Fig. 2 Samples of past works of Hyperbody Hyperbody has developed a unique approach towards investigating solutions for the development of interactive architecture. Our extensive theoretical research has always been instantly accompanied by creation of working prototypes that validate and illustrate in a tangible way investigated technologies and possibilities. Since the year 2003, conceptual working models of architectural objects have been built to exemplify in a playful way new possibilities of dynamic architectural creation.

An inflatable cushion with embedded fluidic muscles and a microcontroller can become an interactive pneumatic “brick” of a dynamic membrane. Capable of two-axial bending and extendable with a variety of sensors, connected in this way to its environment and to other elements, each brick can develop a wide variety of primitive behaviours. As a result, the whole structure consisting of a higher number of such elements can produce very intricate effects and reactions.

Fig. 3 Mem[e]brane concept render actively generate and modify various spatial conditions on its two sides.

Hyperbody includes the design and making of interactive installations in its education programme, where students on bachelor and master levels build interactive installations in close cooperation with Hyperbody researchers. This year the theme of those installations has been “interactive portals”, a dynamic, real- time data exchanging network of installations that in the applied manner explore the possibilities of creating spatial forms that connect distinct spaces. This project hasd been a collaboration with the faculty of industrial design at TU Delft and its outcomes are to be exhibited both in Delft and at the international architectural expo in Seville, starting in September 2008.

Fig. 6 Flap system, rendered concept Fig. 5 Flex system rendered concept and prototype

Fig. 8 Leaf portal prototype The other installation of the interactive portals series has been designed and prototyped as constituting of completely autonomous surfaces, capable of curling up from a flat to a fully folded state and dynamically changing their orientation. In this way spaces were formed spontaneously and could entirely disappear.

Fig. 9 Exhibition pavilion rendered concept On the path of its commercial development, the mem[e]brane concept can be quicker and more feasibly developed and validated on the partly limited experimental scale. For this, Hyperbody initiates a project which can lead to a prototype development with the function of a dynamic exhibition space. The project can result in a tangible outcome of an elaborate exhibition stand installation, planned to be finished in the beginning of 2009.

resources and lack of open access to preferred technologies still hinders the development of many potentially applicable solutions On the large scale agenda of the mem[e]brane development, Hyperbody, together with architectural design office ONL [Oosterhuis_Lénard], has initiated the project combining the cutting-edge research of Hyperbody with efficient project design and execution capabilities of the ONL office. Collaboratively designed A2 mem[e]brane is to become an unprecedented kind of object. Stretching along the edges of the A2 highway it will combine features of a sound barrier and a landscape body, while fulfilling many functional tasks such as sound protection, fine dust collection, visual separation, lighting and traffic control. However, in addition to it, the A2 membrane will also become an element which connects the social ecologies of areas surrounding the highway with the ecology of the highway itself. It will appear dynamically only if needed, as a consequence of the mediation between the swarm of cars passing through the highway and people inhabiting and using the surrounding spaces. In some aspects it will develop a barrier, in other it will become a connection. In all cases the A2 membrane will constantly, intelligently adapt to varying conditions of its environment and will dynamically interact with it, becoming a new kind of spatial inter-medium.

Overall there are four classes of panel types in TPI tessellation: The honeycomb in positive, the bow tie in negative curved areas, the degenerate rectangle in single curved areas, and a combination of half honeycomb half bow tie when generating points fall into regions of different Gaussian curvature sign. Figure 1: TPI hexagons generated on surfaces of positive and negative curvature.

Figure 2: The left configuration shows self-intersection which is repaired by diagonal edge swap on the right. Interpenetrations and selfintersections generally affect a group of four cells. It can be repaired by a common diagonal edge swap operation that reconnects the points in the generating triangulation (Figure 2, right).

While for the positively curved outer region of the torus the edges rotate in a well behaved and constrained manor, they tend to flip in the negatively curved inner part. A generating pair containing points where the edge is almost parallel to Np will not be suited to generate a well-behaved TPI mesh. The angle between the edge and the normal plane at py containing p; should therefore be maximized. Likewise for any point of a sphere it will be 90°. Figure 3: As the Normal plane NO rotates around the normal nO the angle a between the NO and Ni and the angle b between the perpendicular planes BO and Bi is varying.

Figure 4: Edge orientation at different position of a torus creates specific patterns.

7 Architectural Structures from TPIl meshes

Another focus of research next to the geometric generation of the meshes is their physical realization into prototypic structures. Using CNC controlled 3 axis milling and laser cutting technologies parameter controlled node, beam and _ panel components can be produced from flat stock material only and assembled to complex 3dimensional surface structures.

Figure 6: Planar hexagonal TPI mesh with offset mesh and symmetry planes for beam elements and node axis vectors.

Figure 9: Small and large scale prototypic structures. Figure 8: A prototype structure consisting solely of planar CNC cut elements (panels, struts, nodes).

Figure 1: Both quad meshes are circular meshes. The one on the left has a unique parallel spherical mesh, and the one one the right has no parallel spherical mesh and therefore no vertex offset.

Figure 2: A torus-like QCEP mesh without a vertex-offset mesh (left). The same surface but with one row removed for better illus- tration. Figure 3: Two QCEP meshes without vertex-offset meshes. The one on the left has non-homotopic quasi-spherical loops whose parallel spherical embedding loops are inconsistent with each other. The one on the right has odd-sided loops that are not quasi-spherical. The original models are from the repository of TOPMOD.

Figure 4: A torus-like circular quadrilateral mesh (left) with its unique parallel spherical mesh (right). Figure 5: Left: A hexagonal mesh before optimization; right: A quasi-circular hexagonal mesh after optimization with its discrete normal.

Figure 6: A QCEP polygonal mesh with three types of faces — quadrilateral, hexagonal and octagonal faces. This mesh also pos- sesses a face-offset mesh, since each of its interior vertices has va- lence 3.

Figure 1: 1:1 prototype showcasing the potential of wood joints for non standard element architecture Keywords: Digital production, CAD/CAM, Parametric design, complex form, mass customization, industry cooperation, design research, Case study

Fig.2: Initial Design Concept: a series of kinked beams create an overall surface with changing transparencies and patterns. The diagram shows possible sections through the construction shown in the rendering on the right. Starting from a real building project — the fagade of a large scale multi storey Parking lot, wherein a parametric concept using CNC wood manufacturing processes was proposed (Design: Martin Tamke and Blunck & Morgen Architects Hamburg) — the research project looked at ways to explore the link between design intent, formal and special expression and the realization process.

The parametric model not only allowed a direct link to production in the very first design process due to the embedding of fabrication specific parameters but as well the exploration of several variations within the design until certain predefined or evolving performance parameters were met. These parameters were later defined by tectonic, material, fabrication and aesthetic considerations. The exploration was conducted in an iterative process. Every design iteration led to a physical scale model, whose constituting elements and emerging nature could be discussed. These were distilled and emphasized in the next generation of models. By doing so a individual design language could be established. Fig.2: Setup of the parametric system from basic layout of the guiding rails and axis systems to solid shape representation.

Fig.3: Laser cut wood model directly processed from Fig 2 parametric system With a view to an intuitive handling (Burry 2005) the systems control via diagrammatic representations showed good results (called law curves in the tools underlying software package Generative Components). Whereas first models showed a direct and foreseeable reaction to changes of parameters, later ones with more and interdependent parameters showed more complex behavior. This led to the design of an abstract second order representation of the parameter driving law curves. This deflection showed good results, counterfeiting the fact that the addressing of specific areas and phenomena within the model became harder to predict the more parameters were in play.

Fig.4: Rendering and wood model of the designs last level internal calculations limited interactivity. This especially in the last step, when overall geometric freedom was introduced and the formerly straight carrying rails, were transformed into three- dimensional bending curves — creating massive topological change within the series.

Fig. 5. Places of diagonal beams: example of automatically generated 2" order geometry. Solid beams and 3 dimensional shaped geometries required more control. Therefore algorithms constructing the final geometries in real-time according local conditions have been introduced.

Fig. 6: Automatic calculation of cutouts by solver algorithms As the underlying parametric model was based on an abstract axis system the implications with material thickness had to be taken into account. For example a solver algorithm, courtesy provided by Axel Kilian, was used, which iteratively shifted the depth of the cutout between rail and beam towards a given optimal value.

The development of the process focused on a smooth functioning and seamless workflow. Thus we tried to use as much existing tool as possible and test their capacity before writing proprietary tools. Such as the Wood CAM tool HSB Cad, this could easily read our data. Its function was to define the different wood-joints on the predefined beam structure and write instructions for the wood joinery machine. A color-code modeler eased the identification of beams and the application of different craft based joints, as notches, cuts, dovetails and rafter joints. Fig.7: Fabrication process from the parametric geometric model, along the control and application of joints to the CNC machinery

Fig. 8: exploded view of the members with integrated joints and cuts

Figure 2- 5V Metal Roofing test The use of CATIA, or other parametric design software, could just as simply use a bow/banana truss or a space frame as construction methods, if only the form were linked back to the definition of these components. We propose that in fact this is a far more ethical and constrained method for deploying the software. We begin by defining and analyzing a system that we would like to explore. One of the first methods we attempted to use was conventional metal roofing, manufactured throughout the world and deployed in a myriad of types, environments and programs. Firstly, we have modeled a tessellated use of typical 5-V metal roofing, maintaining the typical interlocked connection along seams and through a bracket back to a structural system.

Figure 1- Gehry Technologies Digital Project (CATIA) model of Standing seam metal roofing bent (broken) to form an unconventional skin.

The alternative to using a predefined surface is to use the constraint systems built into the software to constrain distinct points to one another. Each constraint whether coincident, or planar, allow for distinct types of movement at each point in the system. Then the user can use this predefined “flat” model to articulate patterns and see immediately the effects that moving one point will have upon the rest of the system. As well, you can find moves that one may ask the system to do that the system simply cannot accomplish. The larger the system gets the more panels the more constrained the system becomes and the closer to completely flat the system wants to be. We imagine that this same system could function far more cleanly with a typical run of standing seam metal roofing. With standing seam metal roofing we will have to connect using brackets and very specifically locate the points of connection prior to construction. The system we proposed will cut the seam along each edge at particular intervals defined by the system. This geometrically defined and

Figure 4- Standing seam test on frame constrained system limits the form, by connecting the moves which are made along one edge of a surface by pulling or pushing on the opposing edge, to increase the area of the shape near the altered edge.

The concept of the GC platform provides opportunities for the development of structural design tools that are fast and flexible. However, GC has little structural analysis capabilities embedded. New features will have to be added, and a proper analysis method will have to be implemented that supports the parametric associative design concept. The StructuralComponents can work from a geometry deriving the stiffnesses or can work towards architectural and structural geometry by user-defined stiffnesses. The steps in the analysis are split up and translated to new feature in GC, which from now on are named StructuralComponents.

By making reference to the only solid piece of historical evidence for the geometry, a surviving photograph of the drawing of the elevation of the facade made in 1917 at the time Gaudi completed the last proposal for this elevation, it appeared that the position and distribution of the steps in this ascending stepping giant's causeway was not closely related to the changing intercolumniation below nor to the distribution of the hexagonal figures in the frieze. This greatly simplified the interface between his element and the rest of the assembly in the relational model. A simple constraint system could be set up to fit the element to the ower and upper limits of its ‘site’ in the model, maintain the inear pitch lines through the staircase, ensure the vertical coincidence in the height of certain repeating patterns of steps hrough the assembly and marry this with the curved profiles of the steps in plan. Figure 1: part of the 1917 photograph of the drawing of the Passion facade indicating the stepping ‘crestaria’ crowning the portal and upper colonnade of the west transept

Figure 2: The model of the stepping crest showing the changing steps that are subject to a variable growth function, adjusting the number and distribution of steps and the elevation showing linear pitch lines and plan showing curved profiles.

Figure 3: Soft stone prototype at 1:5 scale of one of the columns to show the geometry of the intersecting hyperboloid and paraboloids. There are two possible points of intersection on the surface for the lines of intersection between the hyperboloid and paraboloid, only one of which is ‘correct’ for each combination of variables controlling the surface shape. hyperboloid of one sheet for the central trunk of the column combined with eight paraboloid branches married along the straight lines common to these two different types of ruled surface. Similarly each column bonds with its immediate neighbouring columns through shared rule lines on the lapping paraboloid branches. This was developed in response to a detailed understanding of the geometric codex that Gaudi developed for the design for the sculptural surfaces of the church as well as direct reference to the photograph.

It is taken from an undergraduate student project. The context was an experimental research class bringing together architecture and civil engineering students to find ways to collaborate on design. The aim was to avoid the dual traps of structural pre rationalization (here’s a fine structural system - can you design using it?) and structural post rationalization (please make this design stand up now?). Could they negotiate a third way, ‘co- rationalization’ where the architectural design and structural consideration were concurrent in the design process? The semester long course was divided into three parts. In the first part architect- engineer partnerships explored technique, in the second, they applied the techniques they had developed in design projects, in the third part they were to start over and design for a challenging site. Figure 4: The Hybrid Cathedral with a surface from a gamma function mediating between congregational space and

Figure 5: Bridge based on a Jacobi Elliptic function (image by Steven Swain) For the third stage they chose as their site, the mouth of a Corio bay off Port Philip Harbour that had once been proposed for the site of a freeway bypass. They proposed an inhabited bridge in the tradition of the Rialto, Ponte Vecchio or Old London Bridge. In this case they speculated that the private public partnership would mean that the small scale private housing on the bridge would contribute to funding the huge causeway, bridging over a shipping canal at its midpoint.

Figure 7: Image of the inhabited bridge from the water (image by Steven Swain) Now they had an undulating shell structure, highly organic and variable in its form, several kilometres long. It could morph in response to a specific set of drivers and be transmitted between design participants in three short lines of function. It was this notational economy that ultimately delighted the protagonists and made them feel that they were somewhere on the track to revealing the secrets of a shared or co-rational design process.

Reasons for the use of sloping planes in architecture may be functional (as the basis for stairs, ramps and bleachers), structural (such as those proposed by Nervi, 1956), or related to visual perception (as suggested by Arnhein, 1975) and climate control. These reasons are taken into account in the present study. Figure 1. Paulistano Atletic Club — Paulo Mendes da Rocha — 1958

Figure 3: (a) Extruded volume; (b) revolved volume. Another possible classification of the buildings is related to their volume, which can be generated by two different geometric transformations that can be used in the generation of the volume: translation (Fig. 3a) and revolution (Fig. 3b).

townhouses, Mendes house, Prudente de Morais Neto house, and ITA’s hangar - if we ignore the rounded corners of the later (Fig. 4a, b, c and d). Figure 4: (a) Niemeyer’s CTA townhouses, (b) Mendes house, (c) Prudente de Moraes Neto house, and (d) ITA’s hangar.

The bridge is a double curved curved asymmetric concrete structure which contains all traffic requirements and generous pedestrian walkways. The parapet edges of the curl down to reveal a tiered space for pedestrians to pause and enjoy the views of the canal and new parks on either bank away from the immediate vicinity of the road. Alan Dempsey

The bridge is being constructed from a combination of in-situ and precast reinforced concrete and all 1100m2 of formwork is manufactured directly from digital model files. I will discuss how this process was developed from initial sketches, how initial models were constrained and optimized, and some of the issues which arose during construction.

The concept is to unite a number of different levels with a continuous perforated roof surface under which various pieces of built in and free standing furniture can be displayed with a high degree of flexibility. This roof surface also curves down to define an external courtyard located in the centre of the deep plan. However, the complex roof geometry is not just a formal and spatial device but also regulates the environmental performance of the building providing day light, ventilation, and rain water recycling.

This project is a house for a funiture collector on an undeveloped infill site in central Dublin and is at the detailed design stage. The site is exceptionally deep and narrow and has a number of contextual constraints, most notably the required set back of the main mass of the house from the front boundary.

night view of completed project This pavilion was commissioned by the Architectural A ssociation School for the tenth anniversary of the Design Research Lab and was completed in March. The project was the winning entry in a competition open to 354 graduates which required a small temporary structure manufactured from fibre reinforced concrete.

The project progressed from sketch design to construction documentation in ten weeks and required intense collaboration with the structural engineers to develop a range of parametric models and scripts to quickly optimise the form, evaluate structural solutions and manage the final digital fabrication of over 850 unique pieces of concrete and steel. The success of the project has resulted in it remaining open in London until October 2008.

Figure 1: Al Hamra Firdous Tower, Kuwait City (Byles and Oncina, 2008). Advances in architectural geometry have become instrumental in taking projects from a 3-D concept to reality. The intelligent 3-D model of the building or individual components can be sent to 3-D printers, used for cost estimating, shop drawings, and detailed coordination. This level of representation leads to enhanced understanding and commitment by all members of the client, design, and construction teams. However, the ability to apply the model directly to the construction process in either the manufacturing of components or formwork will be the most recognizable advancement of this new generation of buildings. This intersection of technology and environmental design will result in a range of architectural expression as diverse as the clients, sites, and design teams that create them. The following projects are examples of how far this multidisciplinary innovation is being pushed today and are intriguing representations of where the future may lead us (Byles and Oncina, 2008).

Free and unconventional digital experiments, such as imaginary design project proposals in competitions, sometimes also leads to "iconic"3D-artifacts, which have artistic value even as themselves. Good modern examples of digital-deconstructivistic and fluid. free-form architecture of virtual projects are for example shown by the recent works of Greg Lynn and Kivi Sotamaa (fig 5 and 6) ((Penttila, 2006).

Figure 4: Structural frames in Frank Gehry’s Experience Music Project in Seattle, produced by contouring (Kolarevic 2001).

The most common way for structural engineers to build up their geometry models is to define the centerlines of geometrical objects. In contrast to architects who represent the outer boundary of building elements such as columns, beams, walls or similar, structural engineers abstract geometric objects as simple centerlines and attach a thickness or predefined section types from a library of structural elements to them.

Figure 2: External CFD model Depending on the software in use, non rectilinear elements of the underlying building geometry have to be simplified into block objects to allow the thermodynamic analysis to be carried out. A sloping external wall would have to be reinterpreted as stepped block wall through manual input. For setting up geometry models for thermodynamic modeling on the exterior of a building, surrounding buildings need to be included in the model with their exterior walls drawn as surfaces to analyze overshadowing. Features that are smaller than 2 meter for walls or about 200mm for particular small scale elements do not get included in the model. Depending on this scale, some external shading devices do not get drawn, but they are abstracted as virtual surface representation. This is a process of equivalencing. Instead of drawing 20 horizontal slats one generates one equivalent vertical shading surface with a transmission factor.

Figure 3: CATT ® Acoustic v8 modelling interface 3D geometry models for Acoustic optimization and auralization is applied for testing the acoustic properties of interior spaces such as auditoria, meeting rooms, concert halls, and foyers. They are set up as coarse representation of the inner surface areas of a particular space for broad geometrical shaping or to simulate reverberation times from sound sources to their surrounding surfaces (Figure 3). The level of detail included in acoustic geometry models can not be defined uniformly as it depends on the frequency-range that is being investigated. Some acoustic- specific features such as deflection - screen need to be included in the model and represented with all their surrounding surfaces to ensure sound ‘bouncing; off them in the most accurate manner.

and the third type of analysis are simple tests for understanding heat- flux using radiation modelling. Whereas egress models do no require much appropriating of architectural models for their use (the analysis mainly needs to understand geometrical boundaries for escape-routes), smoke movement analysis works on a differen principle. Geometrical entities are either generated or (if imported from a third party model) subdivided on the basis of a user-defined grid that depending on the granularity of the information required for CFD analysis and the available computing-power. In order to analyze smoke movement within a building, fire engineers use the interior surfaces of a room to define obstructions in their 3D model that prevents smoke from spreading. Buildings are then se up as a series of connecting surfaces that get interpreted as volumes by the analysis software. Between the volumes there are interconnecting openings for doors and planar elements that si flush with the wall representing vents such as windows o1 mechanical piping. The wall thickness of the boundary surfaces of a volume is defined in the geometry model, according to the minimum requirements of the grid as seen in Figure 4. Figure 4: PyroSim®: interpretation of geometry

Image 2: Back view of the open theater

fig 2: Spiral traces connecting the inner bearings of the arcs

fig 3: Spiral traces connecting the outer bearings of the arcs

fig 4: Positions of centers — Radius of curvatures - Arcs

Departing from the principles, constrains and rules that apply to the geometry of regular single and double curvature configurations described in earlier publications (Liapi and Kim, 2004), new rules that comply to the helical surface geometry have been developed. Thus in a helical tensegrity structure, which is also a single curvature structure, the upper and the lower bases of adjacent units need to overlap along the radial direction to allow the structure to bend only one way (Figure |). Unlike to the geometry of a vaulted structure, the sizes of the units in a helical structure along the radial direction need to be reduced proportionally. Therefore, the center points of tensegrity units in addition to being shifted up and down, as is the case with vaulted tensegrity structures, need also to rotate. Figure 2 shows a flat configuration of a 4-unit cluster of a helical structure.

The amount of overlap between adjacent units as shown in Figure 3 varies from | to 0. These extreme configurations apply to both the upper base and the lower base overlap A zero lower overlap configuration is possible when the center line passes through the vertex of the lower base, as shown in Figure 3(a), On the other hand, if the center line passes through the vertex of the upper base, as shown in Figure 3(b), a zero upper based overlap can be created, Figure 3: Geometric limitations forthe upper and lower base overlaps

The process to be followed for the geometric construction of helical structures starts by determining the dimensions of the cable bases of consecutive units along the radial axis. To ensure proportional decrease in unit sizes, two guide lines with a given angle are drawn. The ratio of reduction in unit size (thus the actual size of individual units) depends on the angle of two guide lines. A high reduction ratio is associated with a wide angle between guide lines, and will cause a high stress to the overlap cables of the upper and lower bases between adjacent units along the circumference axis. The reverse is true for a low reduction ratio. Accordingly, the angle between guide lines needs to be taken into account when determining the sizes of consecutive units.

A geometric method for the construction of identical gaps between adlucent units in a clusters of 4 units is presented. Creating identical gaps between adjacent units along the radial direction is most critical to designing a helicoid structure; otherwise the constant rotation angle between units cannot be guaranteed.

Figure 1: deployability conditions

Figure 2: outer and inner SLEs "If we connect the basic SLEs patterns between them, in such a way that the compatibility of the movement of each piece is guaranteed, we obtain a complex system able to grow in one, two or three spatial directions, building a complex assembly with the same properties as the elements: expanding and folding abilities." [Escrig and Valcarcel, 1993, p.71] A three-dimensional flat structure made of square deployable units is a simple design task. Figure 2 shows a development of inner and outer SLEs on a common plane and the geometric relationships between the lengths of the bars. The outer SLEs defining the polygons are regular. Similar conditions like Figure 1 (a) should be satisfied for the bracing diagonals.

Figure 3 presents a model of deployable flat slab made of 10x5 square units. Figure 3: deployable flat slab

Figure 4: unit of an arbitrary double-curved SLEs structure In a double-curved structure different from a dome the lines cannot go through one single point. Another methodology has to be found, that can satisfy the deployability conditions. In a general case of an arbitrary unit like the one in Figure 4 these are at+b=c+d, d’+c’=e+f, e’+f’'=g’+h’ and h+g=a’+b’.

Figure 6: chain of elliptical gears, any 2D curve can be subdivided in this way and will be foldable The rolling of two ellipses can also be described as true elliptical gears. They can only be made to mesh properly if they are twins, and if they are rotated about their focal points [Chironis and Sclater, 2001, p.267]. If we link more such ellipses together we get a chain of elliptical gears, the rotation of each of them about the own focus can be controlled separately. The result is a two- dimensional SLEs structure. Its shape and curvature can be set arbitrary (Figure 6).

Figure 5: elliptical properties of a SLEs unit Using the properties of an ellipse we can construct a pair of SLEs in a pre-determined configuration with intermediate pivots on an ellipse. The resulting structure will ever fold properly (Figure 5).

Assuming that the ellipses underlying the structure are congruent with the same linear eccentricities and the major axes should be parallel. Hence the polygon with vertices the focal points of two such ellipses can be only a rhomb or a rectangle as a special case. Rectangular units can generate only a flat configuration.

Figure 8: SLEs structure approximating a translational surface [he prismatic SLE unit can be regarded as part of a pair of discretized planar curves, generating a translational surface. When the planes of these curves are orthogonal, a family of translational surfaces is generated which can be easily approximated by SLEs. This can be done by using the method described in the forgoing Chapter. Figure 3.26 shows such a structure. The result is a discrete surface with planar rhombic faces. The projections of these polygons on a horizontal plane are rectangles. The case study demonstrated in the following chapter is based on this assumption.

Figure 9: illustration of the actually investigated mehanism

Figure 1: PQ-design by simple geometric operations: The Sage by Norman Foster. Additionally, a simple strategy is to mix planar quads and triangles in areas where two triangles can be combined to flat quadrilateral faces. A prominent example for that is the freeform shell of the Milan Trade Fair by Massimiliano Fuksas, a rather complex looking shape generally segmented into triangles, where neighbouring coplanar triangles have been merged into planar quadrilaterals.

Figure 2: Floor slabs to be matched (red) as an example for architectural constraints on a PQ mesh. System lines. Polygons defined by corresponding edges of a quad mesh can be realized with different visual impact on the overall facade structure. We account for that by accordingly weighting the fairness of these polygons in the optimization.

The number of quads in the subdivided mesh depends on the used subdivision-algorithm. Although for our needs we use Catmull- Clark subdivision, the algorithm used doesn't matter much as long as it is a quad mesh subdivision method. Note that it is even possible to manually insert new system lines where needed, thus controlling the number of resulting panels. Figure 4: Architectural PQ Design created by the interactive approach of section 3.1, resulting from the initial mesh in Fig. 3.

Figure 3: Initial mesh (left) and initial mesh with (non-planar) subdivision preview. the most common CAD-design-tools. Note that this 'subdivision- preview' is used to only help estimate the shape of the planar result and does not guarantee planarity. The more planar the quads of the initial mesh are, the more likely the planar result will appropriately resemble the subdivision preview later on.

Manual surface remodeling has a major drawback: There are infinitely many possibilities to choose the coarse initial mesh. We have partially been addressing this problem by employing so- called ‘Subdivision Fitting’. Given an initial mesh with rough vertex positions, we use an optimization procedure to compute specific vertex positions, such that the subdivided mesh will be close to the reference surface and the quads become nearly planar. The use of subdivision helps in obtaining a smooth result, nevertheless we use fairness as an optional goal for this optimization. Figure 6: Subdivision fitting. A manually layed out rough initial mesh (second from left) is automatically matched to a freeform reference (left). Second from right: the resulting initial mesh for refinement, right: the subdivided planar result.

The manual remodeling approach tries to match the given surface best by means of the interactive design procedure mentioned before. Additionally, the given freeform-surface will be used as a reference in subdivision modeling (Fig 5). An appropriate coarse representation has to be created manually, by laying out an initial mesh resembling the given surface and editing its vertices such that the subdivided mesh will approximate it as good as possible. The similarity between the given surface and the subdivided initial mesh depends on the initial mesh's connectivity as well as on its resolution. Figure 5: Manual Surface Remodeling. An initial mesh (middle, red) is manually matched to a freeform reference (left). Right: the subdivided planar result.

Yet, one major limitation of this method is the possible large variation in size of the resulting planar quads. The optimization naturally tends to populate regions of higher curvature with more initial vertices. This biases equal distribution of quads, resulting in unequally sized panels which are in most cases unwanted for architectural applications. This circumstance can to a certain extent be controlled by considering the original surface's curvature when designing the initial mesh: Granting more detail to highly curved regions in the initial mesh will help distribute structure- lines more evenly. However, depending on the connectivity of the quad mesh, this could also introduce undesirable detail in other regions. Figure 7: Rendered result of planarizing 'the Opus' - facade.

Figure 8: Panelization by surface approximation. The close connection between conjugate curve networks on a smooth surface and PQ meshes leads to a promising approach for the approximation of freeform surfaces. In this case we (1) analyze the curvature flow of the given reference surface, (2) lay out a conjugate curve network on it and (3) extract a quad-dominant mesh resembling the conjugate curve network. Such meshes will be close to planarity and thus suitable for planarization [Sauer 1970]. Step 1 can be carried out using well known methods. Little work has been done to exploit the degrees of freedom for step 2 (Schiftner 2007]. Recent results on surface parameterization may be adapted to solve step 3 [Kaelberer et al. 2007]. Figure 8 shows an approximation of a complicated surface. Obviously, work remains to be done in order to make this approach suitable for architectural applications. The challenging task is to lay out a highly regular quad mesh which picks a suitable amount of detail of the reference surface.

Figure 9: Overview of PQ design strategies presented in this paper. Limitations of the presented approaches point out topics that need further investigation. An important part is to fully automate the creation of initial quad meshes for remodeling with PQ meshes. This needs a thorough understanding of conjugate curve networks on freeform surfaces on a global scale. Existing ideas include the use of a shape database which could be used to assemble an initial mesh using a decomposition of the freeform surface into known patches. Moreover, surface parametrization methods like [Kaelberer 2007] could be used. Subdivision fitting inherently lacks degrees of freedom given common architectural constraints like equally sized panels. Therefore, it is feasible to explore multiscale approaches for remodeling. As already mentioned, the layout of conjugate curve networks on freeform surfaces is an important part of approximation with PQ meshes, which has not yet been satisfactorily solved.

In a first step, those points are pushed onto the surface. Secondly, a process of dynamic relaxation is run, where every point on the surface tries to be the area centroid of the quadrangle described by its surrounding points. After these two steps, the process proceeds to the next iteration, so that in the end all points are lying on the trimmed surface. To create a point grid resembling a trimmed surface, an iterative process is applied, with the “untrimmed grid” as starting point: It is possible to detect, if a point lies on the trimmed surface or in one of the cut off regions. Thus it is also possible to identify the points neighboring the trimmed surface.

trimmed. Otherwise such gaps would simply be spanned by later applied geometric modules, being represented only by a slight anomaly in the grid. 2 Three-dimensional grids on stacks of trimmed NURBS surfaces

Another problem arises if some of the surfaces have points with multiple instances (due to trimmed gaps, see 1). These multiple This issue can be solved by connecting the corresponding surface points through interpolated curves and placing an arbitrary numbet of points along them using the t parameters.

The geometrical construction of the Cantor set can be explained as follows: Take a straight line segment, divide it into three parts of equal length and remove its middle third; divide again each of the resulting line segments and keep removing their middle thirds. If you repeat this for each of the new line segments, you will end up with the Cantor set. The Von Koch curve belongs among the first found and best known fractal objects. In 1904, the Swedish mathematician Helge Von Koch described it for the first time in [Koch 1904]. The

Figure 3: Sierpinski triangle and Von Koch curve according to Barnsley's IFS-formalism The strange properties of the aforementioned objects led the mathematicians to name them “‘monster curves’. In 1981, based on Hutchinson's fixed point theorem [Hutchinson 1981], Barnsley defined a formalism which was able to describe such objects in a deterministic way [Barnsley 1988]. His IFS-method (Iterated Function Systems) consists in a set of functions that are applied iteratively. In our case, a function is an affine geometric transformation. Iterative means that the construction is done step by step. The input of a construction step is the result of the step before.

Figure 2: Von Koch Curve Curve is constructed stepwise. Beginning from a straight line. there results a meandering curve with strange properties: The geometrical construction of the Von Koch curve is iterative, where each of the construction steps consists of four affine geometric transformations. The primitive is a section of a straight line, which is scaled, rotated and displaced by each of the transformations [T1...T4]. Per construction step, four duplicates are generated, of which each will produce four more duplicates in the next construction step.

Figure 5: IFS-Curve curve design: Control points (red) and subdivision points (blue) Whether a figure is smooth or rough depends only on the affine geometric transformations. The same curve might be smooth or rough. By changing the subdivision parameters, respectively the smoothness and the roughness can be adjusted. The input of the subdivision parameters is given by the position of what we call

Figure 7: Geometrically constrained rational IFS-surfaces The design possibilities of vector sums are limited compared to NURBS surfaces. In order to augment the design capabilities, we will employ methods of projective geometry. The IFS-formalism will be extended by the possibility of assigning different weights to its control and subdivision points (w # 1). The IFS becomes a rational object, where the single weights are not organized necessarily uniformly, but rationally.

Figure 6: Surface design by vector sum

Figure 8: Iterative Bezier Curve 4 Applications

In this example, we build a vault structure based on an iteratively constructed Bezier curve. The line sections which build up the curve will be replaced by raw sawn timber planks. The design of the global shape of the vault can be controlled via the control points. Thereafter, the curve will be subdivided into smaller parts until we will obtain adequate length for the constructional elements. On the one hand, the lengths of the elements should not be longer than the prevalent planks existing on the market. On the other hand, the subdivision should be fine enough to obtain a smooth rendering of the curve. In order to realise physical buildings out of discrete virtual geometries, the elements, which constitute the 3D-models, are replaced by constructional elements. For an iteratively designed curve, the line sections will be substituted by linear constructional elements, such as planks or beams. In the case of a discrete surface, we replace its faces by planar constructional elements (panels, plates etc.). The substitution of geometric elements by constructional elements poses a certain number of questions as the geometric figures do not have physical dimensions like thickness. We will first discuss the more demonstrative case of a two- dimensional figure: the Bezier curve.

Figure 10: Reduced Scale Model

We will show below the application of an iteratively constructed free-form surface as a panel construction. In this example, the faces that compose the surface are replaced by planar timber panels. The choice of the thickness of the timber panel is important as the virtual 3D-surface does not present any thickness. A volume model has to be derived from the surface model. First, we generate a parallel offset surface, which holds a constant distance to the initial surface. The distance corresponds to the thickness of the timber panel. Second, the bisector planes are calculated, we will use them later for the chamfer cut of the panels.

The question of how to partition a free-form object into a coherent set of constructional elements becomes obsolete because it is directly given by the iterative geometrical construction method. The digital chain, from design to production, is inherently optimized. This is an important cost and time factor for the production of free-form architecture.

Figure 14: Assembling of the partial prototype The assembled manufactured elements give an accurate rendering of the surface designed on the computer screen. This shows that practical realization of iteratively constructed surfaces becomes possible. It requires a relatively small planning effort. Several problems showed up during the manufacturing process because of the extremely low tolerances permitted by the perfectly fitting pieces. Big scale free-form buildings will probably loosen these tolerances, but the logistics and the montage will probably get more complicated. The efficiency of the method presented is proved insofar as the processing of the data, from design to production, last only a few moments.

Figure 13: Integrated manufacturing of the constructional elements

Figure 2: A four digit binary string controls the number of elements at every node of the upper mesh.

Figure 3: Cross and longitudinal sections of space frame from generation 215 in comparison to space frame with all possible elements.

Figure 4: Rendering of the final space frame with boxes representing spaces almost free from structure spaces and minimal deflections of the space frame could be satisfied.

Figures 1: photo during construction of Tower B (left), concept rendering of tower crown for Tower A (right) The project described in this paper is the uniquely shaped, glazed roof cap of the 364m tall Federation Tower in Moscow, Russia. The footprints of the twin-towers are triangular with arc-shaped sides. The smaller Tower B has been topped out by mid-2008 (images 1, left). The taller Tower A is under construction as this paper is written. The form of the tower's vertical envelope tapers as it increases in height. The roof cap, or “Tower Crown”, is the intersection of a cylindrical surface with the slanted vertical part of the building envelope. The cap for Tower A is envisioned to be made of glass and steel, as transparent and elegant as possible (images 1, right).

Figures 2: development of geometric concept Different to the more traditional design process where things are developed on a broader scale first and then progressed into details later, this project almost immediately looks for precision typically seen in the construction document phase to get the complex geometry correct from the start, images 2.

Figures 3: integration of roof structure with facade and drainage

Figures 5: virtual prototype development of glazing and framing Although all glass panels and all members of the support structure are geometrically unique, their forms follow similar development tules. Designing any “typical” detail in 3-D allowed the generation of parametric components that were then replicated throughout the whole roof grid, image 4.

Figures 4: MEPLA analysis to study forcing of glass panels into curved shape (left), structural FEM analysis of roof framing (right) A geometry that was developed based on a computational process naturally allowed rapid generation of various analysis models to design the structural steel, and the mechanics of the glass panes, images 3.

Figure 1. Workflow for processing architectural geometry. The general workflow for reconstruction starts by discretizing input NURBS surfaces as polygonal meshes. This initial stage serves as the basis for subsequent geometric reconstruction and is con can independent of trolled by the specified number of meshes. Alterna’ regard the meshed object as input to the reconstruction scheme where it originated from. The reconstruction scheme is defined by a number of parameters that reflec rela tionships reconstruction sketch out their intentions on structuring the underlying geometry. The reconstruc mes. the workflow. stage is interactive, where, potentially, tively, we the inter- from predecessor to successor shapes. The designers tion procedures are then applied to the d iscretized hed surface to produce alternative designs. Figure 1 illustrates

Figure 2: (Left) face:{ A, B, C ee D }; (Middle) face:{ T,, B, Tz, D }; (Right) faceSet(2):{{ A, B, T;, D }, {B, C, D, Tat} Figure 2 illustrates three different quad faces (shaded areas). The original input, face:{A, B, C, D}, is shown on the left hand side. Face:{T,, B, Tz, D} and faceSet(2):{{A, B, Ty, D}, {B, C, D, T.}} are alternative meshed faces that can be produced. The transition from one face (the input mesh quad) to single or multiple faces specifies a reconstruction scheme. Note that the face can be either convex or concave.

2.1.3. Fenestration Function

“Fenes” is the function identifier with “Num” number of Boolean values contained in the list. Each Boolean value maps to a corresponding face. The visibility of each face is tured on or off, according its value, 1 or 0, in the list. As Figure 4 illustrates, two different fenestration rules applied to the same set of faces generate different configurations.

A fter reconstructing the input surface, a post-processing procedure is suggested. This process involves a mesh offset operation on the polygonal faces and makes the surface with zero depth become < more realistic solid artifact, a panelized surface structure witl thickness. We can create offset meshes along each face normal by a given distance ’D”, as shown in the Figure 5. In this way, a solic volumetric surface can be created. Figure 5: Offsetting the surface by a given distance D.

An architecture example taken from one of Architect Santiago Calatrava’ s projects in 2000 was remodeled. The project, Windery Complex for the Bodega & Bebidas Group, features a curved roof composed of elementary rectangular tubes, as shown in Figure 6. We remodeled the curved surface of the roof as an example to explore potential variations by the procedural reconstructing schemes presented in this paper.

The initiated surface, created using NURBS, is illustrated in Figure 7. Although NURBS model complex freeform objects as mathematical representations, for the practical purposes, suc N as manufacture or fabrication, NURBS-based objects are usually are panelized. Motivated thus, we discretize the surface as quadrilateral meshes. Other meshes are possible, for instance, triangular meshes possess the same type of geometric informati including nodes, edges, planes etc. However, in this work, uon, we considered only quad faces. The number of quad meshes to replace the original NURBS surface controls the discretiza For this experiment, this number is set to 1000. tion. Initially, we applied the reconstruction scheme without any post face-offset operation and generated the final structure, shown in Figure 8. The reconstruction scheme applied here is the second iteration shown in Figure 3, in which the original face is replaced by the set with five faces. Also, the fenestration is represented as Fenes(5):{1, 1, 0, 1, 1}.

Figure 9: Surface derived from the reconstruction scheme and a post face-offset. Reconstruction scheme comprises the following face and fenestration niles: (1) face: {A BoC DL~>