Adaptive Pneumatic Shell Structures

Adaptive Pneumatic Shell Structures

Feedback-driven robotic stiffening of inflated extensible membranes andfurther rigidification for architectural applications

Paul Poinet1, Ehsan Baharlou2, Tobias Schwinn3, Achim Menges41CITA - Centre for Information Technology and Architecture | ICD - Institute forComputational Design 2,3,4ICD - Institute for Computational [email protected],3,4{ehsan.baharlou|tobias.schwinn|achim.menges}@icd.uni-stuttgart.de

The paper presents the development of a design framework that aims to reducethe complexity of designing and fabricating free-form inflatables structures,which often results in the generation of very complex geometries. In previousresearch the form-finding potential of actuated and constrained inflatablemembranes has already been investigated however without a focus on fabrication(Otto 1979). Consequently, in established design-to-fabrication approaches,complex geometry is typically post-rationalized into smaller parts and are finallyfabricated through methods, which need to take into account cutting patternstrategies and material constraints. The design framework developed andpresented in this paper aims to transform a complex design process (that alwaysrequires further post-rationalization) into a more integrated one thatsimultaneously unfolds in a physical and digital environment - hence the termcyber-physical (Menges 2015). At a full scale, a flexible material (extensiblemembrane, e.g. latex) is actuated through inflation and modulated throughadditive stiffening processes, before being completely rigidified with glass fibersand working as a thin-shell under compression.

Keywords: pneumatic systems, robotic fabrication, feedback strategy,cyber-physical, scanning processes

FROM PHYSICAL POST-PROCESSES TOCYBER-PHYSICAL FORMGENERATIONThe global design of a free-form inflatable structuretypically needs to be divided into developable strips(Figure 1 - left) or triangulated patterns in order toavoid any wrinkles that would diminish the aesthetic

qualities of the final product. Each piece needs to bepost-processed uniquely and separately (with differ-ent offsets and resizings) in order to consider differ-ent assembly situations (different valences) and var-ious pressure conditions. This post-process - mostof the time partly empirical - often leads to an accu-

SMART AND RESPONSIVE DESIGN - Volume 1 - eCAADe 34 | 549

Figure 1A traditionalpost-rationalizationof a free-forminflatablemembrane (left) isreplaced by acyber-physicalenvironment(right).

mulation of tolerances that are difficult to tackle andavoid during the assembly phase. The assembly pro-cess also presents some logistical difficulties as it isnecessary to sort all the pieces and find their respec-tive locations on the construction site.

The aim of this research is to create a self-supporting structure that is not given and further dis-cretized for fabrication purposes but generated fromthe material behavior which is modulated by exter-nal constraints and forces applied iteratively on theinflated membrane. Those local stiffening and mod-ulations would participate in the intricacy of the finaldesign that would allow creases and wrinkles.

Contrary to established methods where a givenform is meant to be perfectly approximated (AntFarm 1971), those local deformations are not seen asnegative outputs but as positive ones, as they fullyparticipate in the formation process (Menges 2010).The form would be modulated and generated untilit finds an equilibrium state between design targets(set by the user at the beginning of the process) andstructural requirements and performances. We canspeak here of a "steering of form" (Kilian 2014), a con-cept introduced recently that argues for digital simu-

lations to fulfill contradictory purposes. The "steeringof form" is mostly used in a digital context. The pre-sented research tries to translate it at full scale in acyber-physical environment (Figure 1 - right), wherephysical matter and digital analysis interact togetheruntil the inflated membrane satisfies both structuraland spatial conditions.

A cyber-physical setupTo achieve the stated purpose, the global setup (Fig-ure 2) consists of:

• An extensible membrane (e.g. latex) clampedto a steel frame (1x1 meter) that is fixed on avertical turntable KPF1-V500.

• An external 6-axis industrial robot arm KUKAKR 125/2 equipped with a carbon fiber ex-truder as an end-effector that helps tomodifythe stiffness of the inflated membrane.

• A scanning device (Microsoft Kinect V1)mounted near the rotating table that willrecord and analyze in the current shapeadopted by the membrane.

• An air compressor (connected to the steel

550 | eCAADe 34 - SMART AND RESPONSIVE DESIGN - Volume 1

Figure 2The global setup forfurther digitalfabrication andactuation of theinflated membrane.

frame) that can be controlled digitally from auser interface.

• A user interface developed within theGrasshopper plug-in for Rhinoceros 3D - com-puter graphics and computer-aided design(CAD) application software - that integratesboth the simulation and generation of roboticpaths, scanned data coming from the Kinectsensor and structural analysis performed bythe Grasshopper add-in Karamba3d.

Iteratively, the membrane is inflated. Each itera-tion is followed by a local stiffening and reinforce-ment achieved through the placement of carbonfiber threads helped by the robot arm and its end-effector. The latter is composed by a spool and twohooks mounted on a steel bracket that helps to keep

the fibers in tension.During the time of reinforcement, the setup ro-

tates according to the robotmovement inorder to laya carbon fiber thread along a path created betweentwohookingpoints placed along the steel frame (Fig-ure 5). This particularmethodhasbeen adapted fromthe fabrication process of the structural componentsof the ICD/ITKE Research Pavilion 2013-14 (Dörstel-mann et al. 2014).

Once the current shape is entirely reconstructedby the external scanning device (using the com-mercial software ReconstructMe and its SoftwareDevelopment Kit (SDK)) and further digitally post-processed, it is possible to run a structural analysis,which considers the physical scanned membrane inits final state as a self-supporting shell structure. Itis important to mention that during the fabrication


Figure 3Comparisonbetween the digitalsimulations (left)and the materialexperiments (right).

the membrane surface is under tension - the inversestructural state of the final shell. If it fulfills bothstructural performance and design targets asked bythe user, the process stops. If not, a new reinforce-mentpath is created andanewspecificpressure is re-leased. This form generation process continues untilall the conditions (spatial and structural) are fulfilled(Figure 7).

Once the form generation process stops, therigidification process starts. The latter consists ofspraying choppedglass and/or carbon fibers impreg-nated with resin on the final physical shape. In thecontext of the present research, the spraying processhas been achieved manually. However, the samerobot mounted with a different and suitable effec-tor could realize the same operation. When the resindries, the inflatedmembrane startsworking as a thin-shell under compression.

Existing precedents already proved the ability ofinflatable structures towork as compression shells af-ter rigidification processes. The Bubble Houses (Neff

1964) and the Binishells (Bini 1978) were the first pro-totypes that used shotcrete techniques for convert-ing tension based inflatable structures into compres-sion based thin-shells.

MATERIAL EXPERIMENTS AND DIGITALSIMULATIONSIn order to calibrate the system, both digital simula-tions and physical experiments have been made.

Initial physical experiments (Figure 3 - right)only involvedmaterial investigations which aimed toprove the ability of the latex to shape itself under pre-determined conditions (local stiffening) and to serveas a formwork during inflation. A small clamping de-vice has been used to conduct those experiments(Figure 4).

Parallel digital simulations (Figure 3 - left) provedthe potential of representing precisely the behaviorof inflated latex membranes under predeterminedconditions. In order to achieve this, the Grasshopperadd-in Kangaroo (Piker 2013) has been used to set-


Figure 4Physical setup forsmall scaleexperiments.

up differentiated stiffening strategies. First, the de-sired surface to inflate is discretized into a square gridwith each cell containing specific information relatedto its local stiffening condition (Otto1983). A stiffnessgradient is then applied through all the mesh and ismodulated with multiple attractor points. This samegradient is partitioned through a K-Means clusteringalgorithm which outputs data for further fabricationstrategy - in particular for the differentiated layeringtechnique (Figure 3).

In order to perform physical local stiffening, dif-ferent methods have been used, such as:

• Spraying glue on top of the membrane.• Differentiated layering technique (Figure 3).The different layers have been glued togetherusing latex gel.

• Laying tape locally on top of the membrane.• Laying carbon fiber thread locally on top ofthe membrane.

Laying carbon fiber threads directly on top of theextensible membrane allows a direct, quick and veryefficient differentiation of the material compared tothe gluing processes that need long curing time inorder to reiterate the process. Also, the third optioncaused a lot of problems related to the friction be-tween the tape and the membrane, which often ledthe latter to break. Thus, the last option - laying car-bon fiber thread locally on top of the membrane -showed the best potential to differentiate iterativelyand locally the extensible membrane. It has beenchosen and used in the following experiments (Fig-ures 5-6).


Figure 5Carbon fibers arerobotically placedon top of theinflated extensiblemembrane.


Figure 6Fabrication processof the finalprototype (left) andthe correspondingrigidified shell withits scanned dataalignment (right).

Figure 7Flowchartdescribing theoverall formgeneration process.

3D SCANNING AND FEEDBACK LOOPFrom the initial experiments that investigated onlymaterial performances, new prototypes have beenfabricated involving 3D scanning, incremental infla-tion and robotically controlled stiffening processes- without implementing any feedback strategy yet.The step-by-step process unfolds as follows:

• First, an initial inflation activates and tensionsthe membrane.

• Then, a predetermined path actuates andmodifies its current state.

• This path is reinforced until the membranefully stabilizes under the additional tensionexerted by the carbon fiber threads.

• The process presented here can be repeatedonce again along another predeterminedpath. During this process, the user couldchoosewhen to increase the pressure and theduration of inflation.

Further experiments have partially implemented thefeedback-loop strategy within the process. Beforestarting the latter, the user can assign a design tar-get in space (Figure 8) that the membrane needs toreach in order to achieve the desired spatial require-ment. After each iteration, the Kinect is automaticallytriggered and the turn-table - on which the mem-brane is standing - rotates 360 degrees. The shapeof the membrane is then digitally reconstructed andthe resultedmesh object can be used and controlledagainst two parameters: the initial design target


and structural requirements (self-supporting struc-ture under its own compression loads). If both are inaccordancewith the expectations of the user and themachine, the process stops. If not, it continues untilall conditions are fulfilled and themembrane finds itsequilibrium state.

Figure 8The membrane isiterativelyconstrained inorder to reach adesign target set bythe user at thebeginning of theprocess.

Man-machine interactionThe role of the user is to guide the robot actions dur-ing the fabrication process so the latter can achievethe design intentions dictated by the former. A de-cision made by the machine based on an externalsignal coming from the scanned data generated bythe Kinect might be contradictory with the user's ex-pectations. Therefore, the designer is able to inter-vene in the formgeneration process and to overwritethenext generatedfiber pathby an alternative action(e.g. path correction or new pressure input). Thus,the user is involved in the form generation processwhere he negotiates directly with the machine eachaction for further actuation (Negroponte 1969).

RESULTSFromthematerial experiments investigated in this re-search, the optionusing carbonfiber threads showedthe best potential to iteratively differentiate locally

the membrane (Figure 6). It has been used for thenext experiments involving 3D scanning and feed-back strategies that enabled both the user and themachine to reiterate the process after analysis.

From the different methods described above, ithas been possible to achieve successful prototypesand a 1x1 meter rigidified thin shell initially con-strained with carbon fiber threads (Figure 6 - right).These prototypes proved the feasibility of achievingself-supporting structures generated within a cyber-physical environment through iterative stiffening, in-flation and scanning processes.

DiscussionThe complexity of physical post-processes in tradi-tional design methods for inflatable structures canbe replaced by a cyber-physical setup that enablesthe user to iteratively interact and robotically modu-late the physical membrane until it fulfills design andstructural requirements.

The feedback loop strategy explained in this pa-per could be improved by letting the machine makemore autonomousdecisionswhen scanning analysis,fiber laying or inflation processes need to be oper-ated. The user would still be able to monitor or mod-ify the actions taken by themachine if the latter doesnot operate as one would expect.

Outlook and architectural applicationsThe experiments featured in this paper do not scaleup beyond the boundary reach of the robot-arm.One could solve this problemby usingmini-robots oreven drones that would operate in a similar manner(Figure 9). Successful architectural research projectshave shownpotential for fabricating rope bridges us-ingflyingmachines (Augugliaro et al. 2014). It ismostfeasible than such techniques could be investigatedfurther in order to interact directly with dynamicma-terial behaviors and processes.

Another type of architectural application mightbe component-based, where each module could befabricated within the design framework describedabove, before being aggregated together in order toproduce larger spatial structures.


Figure 9Expanding thecyber-physicalsetup described inthis paper by usingdrones in the formgeneration processof an inflatablestructure.

CONCLUSIONThis paper presented the potential for an au-tonomous cyber-physical fabrication setup that cangenerate differentiated pneumatic membranes andanalyze it during the process until the physical shapesatisfies both design and structural conditions. Themembrane is further rigidified and finally acts as athin-shell structure.

AcknowledgementThis paper presented the research development ofa master thesis project realized within the Integra-tive Technologies and Architectural Design Research(ITECH)master programat theUniversity of Stuttgart.

REFERENCESAugugliaro, FA, Lupashin, SL, Hamer, MH, Male, CM,

Henn, MH,Mueller, MWM,Willmann, JSW, Gramazio,FG, Kohler, MK and Andrea, RA 2014, 'The Flight As-sembled Architecture Installation: Cooperative con-struction with flying machines', IEEE Control SystemsMagazine, 34(4), pp. 46-64

Bini, DB and Morelli, GM 1978 'Cassaforme Pneumaticheper la Costruzione di Cupole in Cemento Armato',L'Industria Italiana del Cemento. Anno XLVIII, Giugno1978

Dörstelmann, MD, Parascho, SP, Prado, MP, AchimMenges, AM and Jan Knippers, JK 2014 'Integrativecomputational design methodologies for modulararchitectural fiber composite morphologies', Pro-ceedings of ACADIA 2014 Design Agency

Ant Farm, AF 1971, Inflatocookbook, Ant Farm, Sausalito


Kilian, AK 2014, 'Steering of form', in Adriaensses, SA,Block, PB, Veenendaal, DV and Williams, CW (eds)2014, Shell Structures for Architecture, Form Findingand Optimization, Routledge

Menges, AM 2010, 'FormGeneration andMaterializationat the Transition from Computer-aided to Compu-tational Design', DETAIL, Review of Architecture andConstruction Details, 4, pp. 330-335

Menges, AM 2015, 'The New Cyber-Physical Making inArchitecture, Computational Construction',Architec-tural Design, 85(5), pp. 28-33

Neff, WN 1964, Architecture of Southern California: A Se-lection of Photographs, Plans, and Scale Details fromthe Work of Wallace Neff, Rand McNally, Skokie, Illi-nois, United States

Negroponte, NN 1969, 'Toward a Theory of ArchitectureMachines', Journal of Architectural Education (1947-1974), 23, pp. 9-12

Otto, FO 1979, Growing and Dividing Pneus, Institut fürleichte Flächentragwerke (IL), Stuttgart

Otto, FO 1983, Air Hall Handbook, Institut für leichteFlächentragwerke, Universität Stuttgart

Piker, DP 2013, 'Kangaroo: Form finding with computa-tional physics', Architectural Design, 83, pp. 136-137


Adaptive Pneumatic Shell Structures

Documents

Transcript of Adaptive Pneumatic Shell Structures