122

integration through computationacadia 2011 _proceedings

This paper presents a methodology for the integration of fabrication constraints within the

architectural design process through custom written algorithms for fabrication. The method

enables the translation from three-dimensional geometry, or algorithmically produced data, into

appropriately formatted machine codes for direct CNC fabrication within a single CAD modeling

environment.

This process is traditionally one-way with part files translated via dedicated machine

programming software (CAM). By integrating the toolpath creation into the design package, with

an open framework, the translation from part to machine code can be automated, parametrically

driven by the generative algorithms or explicitly modeled by the user. This integrated approach

opens the possibility for direct and instantaneous feedback between fabrication constraints and

design intent.

The potentials of the method are shown by discussing the computational workflow and process

integration of a diverse set of fabrication techniques in conjunction with a KUKA 7-Axis Industrial

Robot. Two-dimensional knife-cutting, large-scale additive fabrication (foam deposition), robot-

mounted hot-wire cutting, and robot-assisted rod-bending are each briefly described.

The productive value of this research is that it opens the possibility of a much stronger network

of feedback relations between formational design processes and material and fabrication

concerns.

Keywords. Robotic fabrication; multi-axis; file-to-factory, open-source fabrication, parametric

modeling, computational design.

Formation Embedded Designa methodology for the integration of fabrication constraints into architectural design

Dave Pigram

University of Technology Sydney

Supermanoeuvre

Wes McGee

University of Michigan

ABSTRACT

123

1 Introduct ion / Background

The drawing is the t radi t ional mediator between design and construct ion. Since the

ancient Egypt ians, archi tects have employed codi f ied drawing systems to descr ibe thei r

bui ld ing designs (Bernstein 2004), especia l ly in order to have them constructed.Albert i ’s

15th Century descr ipt ion of the archi tectura l design preceding i ts own, di rect and

unchanged, real izat ion through bui ld ing (Carpo 2011) set the foundat ion for a h istor ical ,

and temporal , d iv ide between design intent and design real izat ion. This div ide was to

be t raversed only once, and in one di rect ion only.

In contemporary t imes, the major i ty of archi tectura l drawings have merely ‘evolved’ f rom

“ layered transparencies carefu l ly compl ied with penci l or ink [ into]… layered compi lat ions

of computer generated l ines, arcs and ci rc les” (Bernstein 2004). Drawings, at least

seemingly, must st i l l occur between the complet ion of the design and the commencement

of the construct ion. So-cal led ‘ fast-t racked’ projects are a minor pervers ion of th is, not

a contradict ion to i t . Thanks in part to technology t ransfer f rom other discipl ines, data-

r ich models are now seeing employment in the f ie lds of archi tecture and construct ion.

Bui ld ing Informat ion Model ing (BIM), computat ional s imulat ion and algor i thmic form

generat ion are exemplars of a t rend that ref lects the fact that as the formal and tectonic

constra ints of design have become more complex, the requirement that archi tects

understand the ul t imate means of product ion has become cr i t ical . This may include, for

example, the need to understand what input geometry wi l l be required by a speci f ic CAM

package; some require curves whi le others work f rom surface or mesh data. Algor i thmic

techniques have led to incremental steps towards ef f ic iency, such as unrol l ing formed

surfaces, s l ic ing sect ional contours, and nest ing parts into f lat or even 3D stock.

Meanwhi le, the t ranslat ion f rom conceptual design to construct ion documentat ion and

fabr icat ion remains a largely one-direct ional af fa i r.

Even in advanced pract ices heavi ly invested in dig i ta l fabr icat ion research, part- or shop-

drawings are st i l l predominant ly produced by specia l ists and translated v ia dedicated

machine programming software packages (CAM). This geometry-based dig i ta l workf low,

whi le h ighly advanced re lat ive to the drawing, reta ins the fundamental weakness that

the t ransfer of informat ion between plat forms incurs f r ict ion ( t ime and energy) and as

such is only l ike ly to occur so many t imes. I t a lso does not avoid the “reworking” that

f requent ly occurs when fabr icat ion constra ints are not adequately understood dur ing the

conceptual design phase.

Signi f icant work remains to be done to streaml ine the process f rom design to fabr icat ion.

2 Approach

The methodology proposed here is fundamental ly about removing the f issure between

the processes that br ing forth a design and the operat ions that wi l l lead to i ts eventual

fabr icat ion. The design and product ion process must be reconceived, not as a top-

down sequence, but as a bi-di rect ional cont inuum al lowing for mult ip le feedback loops

and negot iat ions between design intent ions and an array of other inf luences including

fabr icat ion constra ints and mater ia l propert ies as wel l as f inancia l and contextual

pressures.

Fundamental to enabl ing appropr iate feedback, is the development of a software

plat form that a l lows the designer to seamlessly migrate between design and fabr icat ion,

minimiz ing one-way t ranslat ion thresholds. This software plat form is bui l t on the

understanding that a construct ion drawing is s imply a graphical representat ion of a set

of instruct ions. Given th is, i t is c lear that i f the instruct ions can be precisely def ined,

communicated and understood through a di f ferent medium, then the drawing i tsel f is no

longer required.

“You are an organizer, not a drawing board art ist” (Le Corbusier in Sharp 1978).

To achieve th is we have establ ished an algor i thmical ly dr iven grammar of fabr icat ion

capable of incorporat ing an open-ended set of process-speci f ic var iables and

fabrication and prouction techniques

124

integration through computationacadia 2011 _proceedings

Figure 1. Multi-axis (robotic) waterjet system

Figure 2. 7-Axis industrial robot with axes labeled

(linear track is the 7th Axis)

Fig. 1

calculat ions. In the case studies descr ibed here th is has been enacted with in the 3D

Model ing envi ronment Rhino (McNeel Rhinoceros 3D), and i ts embedded scr ipt ing

language Rhino Visual Basic, but the conceptual f ramework is language and software

agnost ic. Vers ion 2.0 of the plat form is current ly being developed in the Python

language for open source re lease. By integrat ing the toolpath creat ion into the design

package, with an open f ramework, the t ranslat ion f rom part to machine code can be

automated, parametr ical ly dr iven by generat ive a lgor i thms, or expl ic i t ly modeled by the

user. This integrated approach del ivers the opportuni ty for di rect and instantaneous

feedback between fabr icat ion constra ints and design intent.

The software is developed in such a way as to provide a hierarchy to mot ion paths,

without being l imi t ing in appl icat ion. The capaci ty for adding addit ional object-or iented

data a l lows more complex processes to operate with s imi lar under ly ing geometry to

s impler operat ions. An example would be compar ing the machine instruct ions for

mult i -ax is mi l l ing with those of mult i -ax is water jet cutt ing (Figure 1) . Typical ly, mi l l ing

operat ions begin by loading the correct tool into the spindle, act ivat ing the spindle at

the correct rpm, and performing sequent ia l rapid, plunge, feed, and retract moves; these

di f fer only in speed. Toolpath geometry can be embedded with the correct tool data,

and the system wi l l automat ical ly change tools between paths. By compar ison, water jet

cutt ing re l ies on a very speci f ic sequence of tool act ivat ions at every path start and end,

in order to proper ly feed abrasive whi le prevent ing the cutt ing stream from marking the

mater ia l between cutt ing paths. Addit ional ly, more advanced water jet software var ies

the feedrate of the tool based on part geometry, in order to control taper ef fects. These

types of behaviors can be incorporated into the gener ic f ramework of the software,

a l lowing features to be expanded to sui t more complex processes.

Whi le these capabi l i t ies exist in the major i ty of CAM software packages, the abi l i ty to

embed these speci f ic at t r ibutes into the parametr ic model dist inguishes th is approach.

I t is worth not ing that s imi lar capabi l i t ies have long been sought by manufactur ing

industr ies. So cal led “ feature based machining” represents the main thrust of th is ef fort ,

whereby part models are embedded with design intent, exchangeable between var ious

Fig. 2

125

CAM packages, and reta in informat ion necessary to automat ical ly generate toolpaths.

As i t was pr imar i ly developed with in industr ies deal ing with sol id model ing programs and

parts sui ted to mechanical design, l i t t le, i f any, crossover has occurred with archi tectura l

fabr icat ion methodologies. Especia l ly in the case of a lgor i thmical ly generated geometry,

wi th a high degree of var iabi l i ty between parts, the capabi l i ty of automat ical ly t ranslat ing

design intent into machine codes holds the potent ia l for dramat ic ef f ic iency gains by

reducing programming t ime.

Feature or knowledge-based machining represents one type of integrated design and

manufactur ing technology, though arguably i t is best sui ted to parts with l imi ted var iat ion,

and i ts integrat ion into the “design” s ide is l imi ted at best. Programs such as Delmia and

Cat ia exempl i fy another approach. At i ts core, Cat ia is a parametr ic model ing system,

encompassing features s imi lar to BIM, sol id model ing, and feature-based model ing.

When fu l ly implemented i t a l lows the ent i re scope of vert ical integrated design and

manufactur ing. Parts are modeled using constra int-based techniques, and these models

are parametr ical ly l inked to bui l t- in CAM “workbenches” capable of outputt ing actual NC

code to v i r tual ly any standard CNC machine. Under the expanded Delmia PLM (Product

L i fecycle Management) package, ent i re manufactur ing operat ions including automat ion

and 4D ( t ime-based) process informat ion can be included, s imulated and control led. I t

is arguably one of the most extensive and compl icated “design” programs on the market.

Whi le advanced parametr ic model ing packages, such as CATIA, have been successful ly

adopted by contemporary archi tecture pract ice (Ceccato 2011) they do not readi ly of fer

the same level of a lgor i thmic integrat ion in the design process as the methodology

proposed here.

2.1 ROBOTIC FABRICATION

Robot ic fabr icat ion is an al l iance between gener ic equipment and custom processes;

robots were arguably the f i rst t ru ly open source fabr icat ion tools. The machines

themselves, in th is case a seven-axis industr ia l model, are re lat ively gener ic, and

whi le cont inual ly improving in performance, they are clear ly l inked to thei r 50-year-old

ancestor the Unimate. The promise of robot ic fabr icat ion has always been the abi l i ty

to perform a mult i tude of unique tasks f rom a common programming plat form; whi le

speci f ic manufacturers use unique syntax, the of f l ine programming techniques used

are consistent. The change in recent years is contextual , a resul t of the development of

computat ional design processes by innovat ive archi tecture pract ices and educat ional

inst i tut ions. The use of a lgor i thms to di rect ly control fabr icat ion tools is a natura l

progression; i t s imply requires an understanding of speci f ic fabr icat ion processes and

the abi l i ty to s imulate the k inemat ics of the machine tool (P igram and McGee 2011).

Figure 2 depicts the typical geometr ic conf igurat ion of a 6-axis robot on an external ra i l

( the seventh axis) .

2.2 SERIAL INSTRUCTIONS

Apart f rom the process- inf luenced geometry i tsel f , the ul t imate outcome of the method

descr ibed here is a ser ies of appropr iate ly formatted instruct ions. In the examples of

processes re lat ing to the Kuka robot, the instruct ions are wr i t ten in the Kuka Robot ic

Language (KRL), but a more common instruct ion standard is g-code, and these methods

have been recent ly extended to parse instruct ions into th is language for use with an

Onsrud 5-axis CNC router (not descr ibed in th is paper) .

The pr imary aspect of machine instruct ions common to a l l manufactur ing processes is

the requirement to def ine the locat ion of the machine tool and, for processes with more

than 3 axes of mot ion, i ts or ientat ion. As we wi l l see later in th is paper, there are many

more processes and machine speci f ic parameters that require control instruct ions, but

our method begins with the def in i t ion of the posi t ions themselves. This could happen

direct ly through an algor i thmic process, through expl ic i t model ing, or v ia a hybr id

approach where s imple a lgor i thms interpret predef ined geometry, usual ly toolpaths.

The f i rst step is to div ide the tool path into the requis i te number of points. This can

be done by using the start and end points of stra ight l ines, or by div id ing l ines and

fabrication and prouction techniques

126

integration through computationacadia 2011 _proceedings

Figure 3. Tool paths divided by number

Figure 4. Tool paths divided by length

Figure 5. Tool paths 2D aligned to a single curve;

Z-Axis is always straight down; X-axis is tangent to

the toolpath

Figure 6. Tool paths aligned using a reference

curve in fanning mode: the reference curve is

divided in the same proportions as the tool path,

creating a ruled edge - X-axis projection on the XY

Plane is constant

curves ei ther by length or by number as shown in Figure 3 and Figure 4 ( the densi ty

of points would typical ly be much denser than shown where curves are being l inear ly

interpolated). In the case of spl ines, a t ranslat ion step can be used to rat ional ize the

path into discrete l ines and arcs based on a def ined chord to lerance. Speci f ic machine

code standards wi l l support var ious methods of rat ional izat ion; of ten g-code simply

l inear izes the 3D curves into smal l st ra ight segments, whi le KRL al lows the programming

of cont inuous arcs on arbi t rary planes, giv ing the equivalent path greater accuracy with

fewer instruct ions, and subsequent ly smal ler f i le s izes.

The next step is to establ ish the tool or ientat ion. Figure 5 shows the or ientat ions for

a tool that is 2D-al igned, such as a kni fe cutter. For processes that ut i l ize t i l t ing tool

axes, a reference curve is of ten used to a l ign the tool, as in Figure 6 . In th is case the

Z-axis of the tool is the “working” di rect ion, whi le the or ientat ion of the X and Y of the

tool is arbi t rary re lat ive to the process. Relat ive to the robot ic system, the or ientat ions

are important however, so the user is prompted to provide a sui table or ientat ion ( the

project ion of the X axis is used to solve the extra degree of f reedom).

2.3 SIMULATION

For machines with greater than three axes of mot ion, the number of ways in which an

x,y,z-coordinate posi t ion can be achieved is inf in i te. Even when suff ic ient or ientat ion

informat ion is added, there are often numerous ways of posing a 6+ axis machine

that a l l sat isfy the given tool locat ion. (Brel l -Cokcan and Braumann 2010) This br ings

about opportuni ty, as wel l as responsibi l i ty dur ing fabr icat ion workf low planning. To

help determine the most appropr iate pose, a 3-dimensional s imulat ion of the fabr icat ion

system is calculated using inverse k inemat ics. In the case of s ingular i t ies, over-

rotat ions, or over-extensions, numer ic values for each jo int and l inkage provide suf f ic ient

informat ion to respond to. For issues of col l is ion detect ion posed geometry is required for

evaluat ion. This feedback provides cr i t ical informat ion to the designer, as understanding

the working envelope of the fabr icat ion process, especia l ly in compl icated mult i -ax is or

robot ic machining, wi l l have dramat ic ef fects on the feasibi l i ty of the process.

Fig. 5

Fig. 6

Fig. 3 Fig. 4

127

Figure 7. Inverse-kinematic wireframe model for

joint value calculation

Figure 8. Wireframe inverse-kinematic simulation

Figure 9. Sequenced operations within an

integrated workflow

Fig. 7

2.4 INVERSE KINEMATIC MODEL

The inverse k inemat ic mode l o f the 6-ax is indust r ia l robot has been crea ted us ing

t r igonomet r y to so lve fo r a l l jo in t va lues. Figure 7 shows the conf igura t ion.

Th is p rocess beg ins by ‘back ing-out -o f ’ the too l v ia a revers ing o f the d isp lacement

and ro ta t ion va lues in the too l ’s de f in i t ion , thus y ie ld ing the comple te pos i t ion

(coord ina te and or ien ta t ion ) o f Ax is 6 . The cente r o f Ax is 5 ’s ro ta t ion is ca l led Po in t

P and has a f i xed re la t ionsh ip to Ax is 6 . The va lue fo r Jo in t -1 is s imp ly ca lcu la ted

by pro jec t ing Po in t P ’s x ,y,z -coord ina te va lues to a wor ld-or ien ted XY p lane a t the

same he ight ( z -va lue ) as Jo in t -1 ’s ax is o f ro ta t ion and measur ing the ang le between

the de f ined zero pos i t ion on tha t p lane and the pro jec ted po in t . The cente r o f Jo in t

2 has a f i xed geomet r ic re la t ionsh ip to Ax is 1 . Va lues fo r Ax is 2 and 3 can be found

by comput ing the ang les o f a t r i ang le in p lane w i th Po in t P, the cente r o f Ax is 1 , and

the cente r o f Ax is 2 . The t r iang le conta ins one s ide l ink ing Jo in t 2 ’s cente r to Po in t

P, w i th the two s ide lengths g iven by the lengths o f L inkage 2-3 and L inkage 3-5.

Th is y ie lds jo in t va lues fo r A2, A3, and A5. F ina l l y the va lues o f A4 can be ca lcu la ted

by pro jec t ing both l inkage 2-3 and l inkage 5-6 onto a p lace tha t i s perpend icu la r to

l i nkage 3-5 and measur ing the ang le d i f fe rence between the pro jec t ions (ad jus t ing

fo r the A4 ’s de f ined zero pos i t ion ) .

A cumula t i ve w i re f rame sequence g ives a good ind ica t ion o f the robot ’s mot ion

(F igure 8 ) . A fo rement ioned jo in t l im i ts , reach l im i ts , and s ingu la r i t i es a re a l l ca l led

out w i th red geomet r y and an e r ro r descr ip to r i s a t tached to the o f fend ing jo in t a t

each pose.

2.5 PROCESS INTEGRATION

Centra l to the method proposed is the abi l i ty to integrate and respond to the propert ies

and constra ints inherent in speci f ic fabr icat ion processes and to remain open to the

embedding of future processes. Factors such as tool geometry, feed and speed rates,

and tool or ientat ion re lat ive to part geometry are a l l used to customize the creat ion

of formatted instruct ions. Figure 9 shows the or ientat ion geometry, p lunge/ lead- in,

toolpath, retreat/ lead-out, and traverse mot ion paths for one part of a complex ret iculated

surface structure. A ser ies of 6 operat ions across three tools including router, saw and

dr i l l paths shape the part wi thout requir ing a change in f ix ture, s igni f icant ly increasing

accuracy.

Tool changes are automat ical ly generated when a test recognizes that one posi t ion’s

tool def in i t ion di f fers f rom that of the previous. Figure 10 shows the successive moves

where a saw blade is loaded into a router spindle.

2.5.1 ROBOT-MOUNTED 2-DIMENSIONAL KNIFE-CUTTING (ROLLER CUTTER)

A custom-made mount for a 1- inch d iameter ro l le r-cut ter has been used to cut var ious

sheet s tock: fabr ic, rubber and the l ike (F igure 11) .

Th is was the f i rs t and s implest process thus far in tegrated in to the cont ro l sof tware.

Typ ica l ly, kn i fe-cut t ing machines ut i l i ze 4-ax is CNC cont ro l , which a l lows the b lade to

be steered to remain tangent ia l to the path. For fabr ic cut t ing the stock is f la t and

un i formly th in so no t i l t is requi red f rom the b lade (F igure 12) . The Z-ax is of the too l

is a lways or iented ver t ica l ly and the X-ax is fo l lows the tangent of the curve at each

point ( th is is arb i t rar y and based on the too l def in i t ion in the robot cont ro l ) ; mot ion

inst ruct ions can be created f rom a s ing le set of or iented input curves. Osci l la t ing kn i fe

t r imming is a lso used for process ing roto-molded components. In such cases the t i l t

o f the b lade is important , and a l lows complex 3D par ts to be t r immed; in that process

a second set of ( re ference) curves is used to def ine the b lade or ientat ion a long the

too l path.

2.5.2 LARGE-SCALE ADDITIVE FABRICATION (FOAM DEPOSITION)

An innova t i ve techn ique be ing deve loped a t the Un i ve rs i t y o f M ich igan tha t

resemb les a l a rge-sca le 3D p r i n te r, u t i l i zes rap id l y cu r i ng po l yu re thane foam

fabrication and prouction techniques

Fig. 8

Fig. 9

128

integration through computationacadia 2011 _proceedings

Figure 10. Automated tool change sequence

Figure 11. Custom roller-cutter mount and process

variables

Fig. 10

(F igure 13 ) . The s ign i f i can t advan tage th i s p rocess en joys ove r con tempora r y

3D p r i n t i ng techno log ies i s tha t the robo t i c moun t i ng enab les the depos i t i on o f

ma te r i a l f rom any d i rec t ion , and the foam i s f as t -se t t i ng and l i gh t enough to remove

o r s ign i f i can t l y reduce the need fo r supp lementa r y suppor t (depend ing on the

fo rm) . Th i s techno logy re l i es on the chemica l reac t ion be tween two-pa r t foam.

The two pa r t - foam m ix tu re i s m ixed i n the nozz le (F igure 14 ) and reac ts i n seconds ,

expand ing to a p red ic tab le s i ze .

Th i s i s an i nc red ib l y sens i t i ve p rocess , depend ing on speed, s tand-o f f , nozz le

se t t i ng , componen t tempera tu re and ang le o f i nc idence w i th p rev ious l y depos i ted

mate r i a l . Each o f these va r i ab les mus t be t i gh t l y con t ro l l ed to p roduce p red ic tab le

resu l t s . Even a t m in imum depos i t i on ra tes , the reso lu t i on o f the ou tpu t i s such tha t

seconda r y mach in ing (m i l l i ng ) i s requ i red to ach ieve accep tab le su r face f i n i shes .

Howeve r, w i th the goa l o f u t i l i z i ng foam e f f i c i en t l y, t he p rocess i s cons ide rab l y more

e f f i c i en t than m i l l i ng f rom so l id s tock .

The upper r i gh t foam tes t i n F igure 15 dep ic ts the resu l t o f a m ino r bu t compound ing

e r ro r i n the assumed expans ion ra te o f the foam. The expans ion ra te i n the s t ra igh t

sec t ions , whe re the robo t was mov ing fas te r, was l ess than tha t i n i t i a l l y ca lcu la ted .

A f te r a ce r ta in number o f l aye rs the e r ro r had compounded to the ex ten t tha t the

robo t was a im ing above the ac tua l l i ne o f depos i ted foam so i ns tead o f add ing the

nex t sequen t i a l l aye r, t he foam en t i r e l y m issed i t s t a rge t . As a resu l t t he l aye rs

ceased inc reas ing i n these sec t ions resu l t i ng i n the l a rge d ips shown in the image.

The ‘w igg le ’ wa l l i n the fo reg round (F igure 15 ) was a subsequen t tes t and c lea r l y

shows tha t the foam expans ion ra tes were be t te r app rox ima ted than i n the f i r s t t es t .

Howeve r, i t i s ev iden t f rom the non-p lana r uppe r su r face tha t expans ion was s t i l l no t

equa l i n a l l a reas . I n th i s case i t i s a resu l t o f reach ing a jo in t acce le ra t i on l im i t on

the t i gh t co rne rs and the re fo re no t ma in ta in ing a cons is ten t feed ra te . Cu r ren t l y the

p rocess i s unde r deve lopment , and w i l l i nco rpo ra te a c losed loop feedback ( r ang ing

feedback ) to adap t to changes o r e r ro rs i n the depos i t i on ra te .

F ina l l y, t he re l a t i ve l y smooth wa l l i n the upper l e f t shows the f i n i sh ach ieved when

the su r face o f the foam i s m i l l ed a f te r cu r i ng . As a p re - f i n i sh ing techn ique fo r

f ab r i ca t i on l a rge sca le too l i ng o r fo rmwork , the p rocess cou ld p rov ide cons ide rab le

ma te r i a l sav ings .

Fig. 11

129

Figure 12. Roller-cutter cutting sheet rubber on a

vacuum table

Figure 13. Foam deposition sequence

Figure 14. Foam deposition nozzle and process

variables

Figure 15. Early trials of foam deposition

Figure 16. Multiple exposure image of a hotwire

fabrication sequence

Figure 17. Custom fabricated hotwire bow and

process variables

Figure 18. An aggregation of custom cut ‘bricks’ is

robotically carved

Fig. 12

Fig. 13

2.5.3 ROBOT-MOUNTED HOT-WIRE CUTTING

A hotwi re cut ter cons is ts of a th in wi re ( in th is case 0.0126” th ick n ichrome) that is

heated v ia an e lect r ic cur rent to approx imate ly 200 degrees Cels ius and used to cut

po lystyrene or s imi la r types of thermoplast ic foam (F igure 16) . The foam vapor izes

just ahead of the wi re and as such very l i t t le force is requi red to push through the

stock. The robot-mounted bow-type hotwi re has a ser ies of speci f ic const ra in ts that

must be embedded wi th in the cont ro l sof tware. The pr imary const ra in t is speed, whi le

the pr imary s imulat ion feedback is the pos i t ion ing of the “bow” of the too l .

I f the wi re is moved too qu ick ly, the foam wi l l not vapor ize ahead of the wi re, and the

wi re wi l l drag through the foam, def lect ing and caus ing inaccurac ies. The ker f w idth

is d i rect ly propor t iona l to the speed, so cons is tency of mot ion is ext remely important .

Th is a lso means that you cannot stop the wi re a long a cut wi thout mel t ing excess ive

ker f in to the sur face.

Whi le the wi re i tse l f is rotat iona l ly symmetr ica l , the bow hold ing i t is not and much of

the work p lann ing is const ra ined by the pos i t ion of the bow re la t ive to the stock, any

j igs and the workspace ( tab le, f loor etc. ) I t is here where the s imulat ion of the robot

becomes inva luable in prov id ing rap id feedback, wi th a ser ies of response too ls to

appropr ia te ly modi fy problemat ic pos i t ions. As shown in Figure 17 , the bow used has

an ad justab le width in order to work in a var ie ty of vo lumetr ic enve lopes.

From the geometr ic s tandpoint , a l l resu l tant sur faces must be made up of ru led

sur faces. The input geometry is most of ten descr ibed by an upper and lower curve,

generat ing a ru led lo f t between curves. Th is type of operat ion is of ten re fer red to as

a swar f-cut t ing path, and a l lows par ts to be t ight ly nested in 3D. In the case of the

process stud ied, the ker f o f the wi re was bare ly 2mm. By ut i l i z ing a l l 7 axes of the

robot s imul taneous ly, la rge par ts can be machined in one step; the component in

Figure 18 is 12’ long.

2.5.4 ROBOT-ASSISTED BENDING

A custom robotical ly-assisted CNC bender has been constructed and is being tested as

an alternat ive to dedicated CNC forming equipment. This combination al lows for the serial

creat ion of accurately def ined bends (of consistent radi i ) , intervals and rotat ions. The

fabrication and prouction techniques

Fig. 14

Fig. 15

Fig. 16

Fig. 17

Fig. 18

130

integration through computationacadia 2011 _proceedings

robot provides the posit ioning whi le the purpose-bui l t bender imparts the large forces and

specif ic motions required for rotary-draw bending steel rod and tube stock (Figure 19).

Key to the successful operat ion of the robotic bending system is the proper choreography

of three clamps: a col let-gr ipper (Figure 20) mounted on the robot and two die clamps on

the bender. One die clamp is used to provide pressure during actual bending, whi le the

other two clamps alternate holding the stock in order to al low the robot to feed and rotate

i t into the appropriate posit ion. The clamps must be capable of providing enough pressure

and fr ict ion to counteract the torque result ing from the self-weight of the canti levering rod.

Figure 21 shows adjustments to the bend instruct ions that are required due to material

spring-back, a product of elast ic deformation. The left image depicts the desired f inal

form. The center image shows the actual result attained i f no al lowance is made for

springback and the image to the r ight shows the over-bends required to achieve the

original desired outcome (shown in pale grey).

2.6 FORM-FINDING

Form-f inding in architecture has tradit ional ly rel ied on exclusively structural inputs;

Antonio Gaudi’s hanging-nets and Frei Otto’s seminal work on minimal surfaces are the

famil iar, and just i f iably preeminent, examples. This has been at least part ly due to the

l imitat ion of analogmodes of computat ion: their f ixed relat ionship to immutable physical

laws and material propert ies. Denied the abi l i ty to expand, intervene-in, or tune the

parameters, architects deploying analog form-f inding techniques are either forced into

subservience to imperfect analogies between these factors and a larger set of extr insic

design intent ions, or, must remain content to l imit themselves to structural and material

investigat ions. When executed digital ly however, form-f inding processes can employ

computat ion towards broader architectural speculat ion, enabl ing a greater incorporat ion

of architectural consti tuency. In this way, programmatic, as wel l as material design goals

and fabricat ion concerns can negotiate in the elaborat ion of architectural experience,

ornament and performance (Pigram and Maxwel l 2010).

3 Findings

The fabricat ion grammar and algori thmic workf lows that support this methodology have

demonstrated that they are suff icient ly generic to be valuable to a wide range of processes,

whi le at the same t ime remaining suff icient ly adaptable to al low for the incorporat ion of

an open-ended set of highly specif ic fabricat ion processes and their respective famil ies

of constraints. The code thus represents a repository of acquired knowledge. This

knowledge is both empir ical, in the form of material propert ies, and methodological in the

form of the growing body of specif ic motion control and fabricat ion techniques.

The complete integrat ion of the inverse kinematic simulat ion within the environment of

Fig. 19

Figure 19. Custom fabr icated robot ic rod

bending system

Figure 20. Collet gripper and process variables

Fig. 20

131

geometry generat ion has been of central importance to the research. This not only al lows

for an expedited response to issues such as exceeding the work envelope or joint l imits, i t

more important ly opens the possibi l i ty of an algori thmical ly automated, user guided set of

responses. Addit ional ly, whi le not discussed in detai l here, the method has independently

shown that i t can be translated across instruct ion code languages and even machine

geometr ies.

4 Conclusions

The methodology described here enfolds design and fabricat ion within a feedback loop

whi le superceding the need for graphical construct ion drawings. I t does this by providing

an algori thmical ly dr iven workf low and fabricat ion grammar within the versat i le, accessible,

and commonly used model ing environment, McNeel Rhinoceros 3D.

A diverse col lect ion of fabricat ion processes and their accompanying family of constraints

have been successful ly integrated within the open-framework with a high degree of

continuity of workf low. Whi le most of these processes themselves have existed in the

manufactur ing industry for many years, the productive impact of their specif ic possibi l i t ies

and resistances on architecture remains an excit ing and contested terr i tory.

What is signif icant is not the value of a part icular machine, in this case an industr ial robot,

nor any specif ic fabricat ion technique. Nor is i t hoped that we wi l l move towards an era

where we can bui ld anything without col laborat ing with appropriate constraints. In fact

i t ’s quite the opposite. What is signif icant is the move away from stat ic object-centr ic

models, with neutral or determinist ic relat ionships to formation, towards operat ive models

where a given project’s physics – i ts way of material ly enter ing and occupying the world -

is intr insic to the design process (Pigram and McGee 2011)

It is hoped that this and other grammars of fabricat ion wi l l help architects return to a more

conversat ional, two-way relat ionship with making.

References

Bernstein, P. 2004. Digital representat ion and process change in the bui lding industry.

In Perspecta 35 - Bui lding Codes, The Yale Architectural Journal, eds. E. Huge, Inter leaf

128-129. London: The MIT Press.

Brel l-Cokcan, S., and J. Braumann 2010. ‘A new parametr ic design tool for robot mil l ing’.

In Life In:Formation - Proceedings of the 30th Annual Conference of the Associat ion for

Computer Aided Design in Architecture, 357-363. New York: ACADIA.

Carpo, M. 2011. The alphabet and the algori thm (writ ing architecture). Synopsis.

Cambridge, MA: The MIT Press.

Ceccato, C. 2011. ‘Galaxy Soho’. In Fabricate: Making Digital Architecture, eds. R. Glynn

and B. Shei l , 168-175. London: Riverside Architectural Press.

Pigram, D., and I. Maxwel l. 2010. ‘ Inorganic speciat ion: Matter, behavior and formation

in architecture’. In Contemporary Digital Architecture: Design and Techniques, D. Kottas.

Barcelona: Links Internat ional.

Pigram, D., and W. McGee. 2011. ‘Matter and making’. In Fabricate: Making Digital

Architecture, eds. R. Glynn and B. Shei l , 74-85. London: Riverside Architectural Press.

‘The 2003 Inductees: Unimate, The Robot Hal l of Fame Webpage’, The School of

Computer Science at Carnegie Mel lon University. Accessed December 5, 2010. “http://

www.robothal loffame.org/unimate.html”. Figure 21. The effects of springback

Fig. 21

fabrication and prouction techniques