Engineering Structures 21 (1999) 680690

Engineering analysis of tension structures: theory and practice

D.S. Wakeeld
*

Tensys Limited, 16 George Street, Bath BA1 2EN, UK

Abstract With the ever increasing popularity of tensile structures it is appropriate to review their engineering analysis requirements. This paper outlines the criteria for the selection of an appropriate analysis method. One such method, Dynamic Relaxation (DR), is presented in detail. Practical features of a DR implementation are reviewed, with particular reference to current project applications. 1999 Elsevier Science Ltd. All rights reserved.
Keywords: Tension structures; Dynamic relaxation; Membrane design; Membrane engineer

1. Introduction The last two decades have seen a rapid expansion in the growth of the number and range of applications of tensile architecture. There is now a well established base of projects, both temporary and permanent, ranging from simple free standing shade elements through to complex wall and roong elements that form an integrated part of a whole building (Fig. 1).

Fig. 1.

Inland Revenue Centre, Nottingham.

* Fax:

0044-0-1225-469-001.

This expansion in applications and possibilities through developments in materials and construction techniques has naturally encouraged a corresponding increase in the number of architects and engineers who wish to become active in the eld. It has not, however, reduced the specialist technical knowledge and experience necessary for the successful realisation of these projects in which form, structural performance and material application and fabrication skills are more closely interlinked than in any other building method. The design of lightweight stressed membrane surface structures has traditionally been set apart from that of more conventional structures. This is not because such design is particularly difcult, but rather that it is sufciently different to require specialist treatment. Unlike steel, concrete or timber building elements which may carry loads in bending, the principal forces in a tension structure are carried within the surface, either by membrane stress or cable tension, through to a supporting boundary structure. The detailed shape of the doubly curved surface structure is critical to its engineering performance. Coupled with the amount of installed prestress, it governs the magnitude and distribution of stresses and deections under applied loading. The architectural possibilities for the shape are also constrained by the need to establish a form that is physically possible, that is, in equilibrium in its installed and unloaded state. In practice this means that there should be no areas of wrinkled fabric or slack cables in the completed structure.

0141-0296/99/$ - see front matter 1999 Elsevier Science Ltd. All rights reserved. PII: S 0 1 4 1 - 0 2 9 6 ( 9 8 ) 0 0 0 2 3 - 6

D.S Wakeeld / Engineering Structures 21 (1999) 680690

681

There must, therefore, be close interaction between architect and engineer at the initial shape determination of the structure. This is known as the form-nding stage, and iterative renement of the form will continue throughout the design process. Tension structures undergo signicant surface movement in order to carry load, and any analysis must account for these relatively large displacements. This behaviour is non-linear in that the actual displacements are not directly proportional to the magnitude of the applied loading and may also include on/off effects such as membrane wrinkling and cable slackening. Accurate fabrication information for membrane, cables and support structure components must be provided to the contractor in order that the structure be correctly installed and prestressed on site. Tension structures are a high-precision pre-fabricated item, and accurate geometric processing and fabrication is essential. Detailed analysis of the installation procedure may also be necessary to aid the monitoring of the actual erection sequence on site. This is to ensure that the correct nal state of the building is achieved without the overstressing of either roof components or boundary structure during the process. Many of the early developments in tensile architecture were based upon physical models utilising soap lms (for minimal surfaces), stretch fabric and wire. In particular these were the foundation of the pioneering work undertaken by Frei Otto and colleagues at the University of Stuttgart [1]. Accurate physical models were used for deriving cutting patterns, and also to provide initial form data for subsequent load analysis by computer. Although physical models are still used, in particular by architects and engineers at the initial conceptual design stage, the principal design stages of form-nding, load analysis and patterning are now all undertaken by computer. An integrated suite of engineering analysis computer software forms the basic tool of what is termed membrane engineering. The role of the membrane engineer and the integration of his or her activities in the overall design process have been illustrated elsewhere [2,3]. In this paper it is intended to review the selection and application of appropriate analysis techniques from the viewpoint of a practising membrane engineer, with illustration by reference to recent and current projects. 2. Selection of appropriate analysis method and software As with most engineering structures, the computation of tension structures is based upon the nite element method. This involves the establishment of a numerical model to idealise the structure from a library of standard elements and to then compute the response of this model to varying load conditions.

There are a number of key points to be considered when selecting the appropriate tools. 2.1. Design of tension structures should not be constrained by the available software Functional, aesthetic and structural requirements should control the design rather than the imposition of the particular characteristics or limitations of the software. For example, a full range of surface stress controls should be available at the form-nding stage in addition to the option of a minimal surface form as determined by a uniform and constant stress eld (Fig. 2). 2.2. Control over surfaces stresses during form-nding The satisfactory long-term behaviour of a membrane structure requires either a uniform or smoothly varying distribution of stress within the warp and ll directions of the bres. This will minimise the chances of the development of wrinkles or local stress concentrations under load. As the stress distribution is a function of membrane surface relative curvatures, or shape, and vice versa, the availability of suitable controls during form-nding is essential. 2.3. Control over fabric weave orientation during form-nding and analysis Ideally the fabric weave directions should coincide with the directions of principal curvature of the surface. For example, the radial patterning of a simple conic tent, or the high point to high point seam trajectory of a parabolic four-point sail. This will optimise the stiffness of a given form, enabling easier initial stressing of the membrane and reducing surface deections under load. Although it is clearly difcult to optimise the seam

Fig. 2.

Mercedes-Benz A-Klasse Tour.

682

D.S Wakeeld / Engineering Structures 21 (1999) 680690

trajectories for more complex surface shapes, the analytical procedure should permit the clear specication of material orientation at the form-nding stage and pass this through to the assignment of element elastic properties for subsequent load analysis. The actual practical choice of whether the material warp or ll direction will follow a particular direction of principal curvature is governed by requirements of strength, installation strategy and fabrication economics. 2.4. Availability of a full library of element types Membrane should be numerically represented by triangular or quadrilateral nite elements, rather than by an equivalent grid of cable elements. At the form-nding stage, when element size and shape may change signicantly during iterations towards equilibrium, smoother stress control is possible by denition of a constant force per unit width. The alternative equivalent cable-net idealisation is insufcient, making accurate stress control difcult and inaccurate modelling in cases such as the interaction with boundary cables when one weave direction is essentially parallel to the boundary. In this latter case, unless an exceedingly ne cable mesh idealisation is employed, the edge cable will essentially only be loaded by cable elements perpendicular to the boundary. Under load analysis nite elements permit the introduction of cross stiffness terms representing the crimp interchange between the warp and ll bres and also enable the modelling of the shear stiffness of the coating. They also provide a convenient basis for the computation of updated wind and snow loading according to current deformed geometry. It should also be possible to model the stiffness of membrane seams. This can have a signicant effect when analysing, for example, a conical membrane where many seams come close together at the top pick up ring and effectively provide local reinforcement to the base membrane at this region of maximum stress under snow loading. A full range of cable, strut and geometrically nonlinear beam elements should be available to model the supporting structure. 2.5. Patterning and seaming The basic orientation of fabric has been discussed above. Individual panels are cut from rolls of material of known width. The cutting patterns themselves are obtained as a two-dimensional approximation to a strip of material taken from the curved three-dimensional structure surface. This may be achieved, for example, by the geometric unfolding of a set of sequential triangular nite elements into a plane. The use of geodesic seam trajectories helps to optim-

ise the use of material by providing a set of balanced patterns with equal material angles either side of a seam node. This eliminates the generation of banana shaped panels. Ideally the generation of geodesic seam trajectories, and consequent assignment of principal stress/weave directions, should be fully integrated with the form nding process. Cutting patterns are then generated directly from the form model. This ensures full compatibility between design intent and the realised structure, provided of course that the necessary material compensations have been measured and applied correctly. 2.6. Understanding of the method used The engineer should have a clear understanding of the software and its limitations. This includes the solution algorithm and how it will respond in particular situations, such as when a local region of the structure is unstable (either due to element collapse or physically inadmissible controls during form-nding, or to excessive membrane wrinkling or beam buckling under load analysis). Understanding and interpretation will be greatly helped if there is a clear physical analogy underlying the analysis method. It is also important to understand how constraint conditions and applied loading are applied and any inherent approximations. Tension structures, large or small, are non-linear systems which require careful engineering thought. The use of black box software wherein structures are designed and built without a full understanding of their behaviour should be resisted at all costs.

3. Available analysis methods Detailed reviews of numerical analysis methods for tension structures have been given elsewhere [4,5]. Two basic approaches to the overall problem of formnding and load analysis have been developed and applied in practice; matrix and vector methods. The matrix methods are typically an application of more standard non-linear structural analyses such as the NewtonRaphson method [12]. The structure overall tangent stiffness matrix is solved incrementally until convergence is obtained. Special controls limiting the maximum incremental deections and nodal residual forces may be required. The stress/strain relations for the individual components are coupled with the equilibrium and compatibility requirements for the complete structure. In vector methods the conditions of equilibrium and compatibility are decoupled until convergence to an equilibrium solution. The most common of these are Dynamic Relaxation (DR) [4,5] and the Scaled Conju-

D.S Wakeeld / Engineering Structures 21 (1999) 680690

683

gate Gradient Method [6]. The former has gained the most acceptance for the analysis of tension structures because of its clear physical analogy and ease of implementation of the necessary controls and constraints. Individual element stiffness relations are held separately, which greatly eases the specication of stress controls at form-nding and non-linear, stress dependent, elastic properties under analysis. It is similarly straightforward to introduce a wide range of boundary conditions that can themselves be dependent upon the current deformed state of the structure. A further method has been applied specically to the form-nding of tension structures. This is the Force Density method, described originally by Scheck [7]. Although used to date primarily with equivalent cablenet models [8], recent research in France has extended the concept to triangular surface elements, under the name of the Surface Stress Density Method [9]. The advantage of the method is that it provides a linearised solution to the equilibrium shape nding problem. The controlling element variable is that of force density or tension coefcient (for example T/l for a cable element of current tension T and length l). As noted in [9], the main drawback of density methods is that the nal distribution of stress is difcult to control. This can be overcome by iterating with updated force densities until the desired smooth stress distribution is achieved, but this would seem to negate the advantage of a linearised solution. Once a form has been found, a vector or matrix method must be used to analyse its response under load. As suggested earlier, the nal choice of method rests with the engineer and should be the one that he understands best and feels most comfortable with, assessed upon the criteria of Section 2. With ever increasing computer power at reducing cost, the relative efciency of each solution method is becoming irrelevant. What matters is the overall efciency of the tool when integrated into the design process. We continue to nd the Dynamic Relaxation method entirely appropriate to tension structure applications. In Tensys we develop and support our own software suite, which has been applied to over 500 built projects in the last seven years. The program capability is continually being extended to tackle new problems as they arise. The fundamental aspects of our DR application and how it meets our criteria are reviewed in the following sections with reference to actual projects.

increments t of Newtons Second Law of Motion applied to a loaded structure subjected to an imposed viscous damping [10]. On rst inspection this might seem a rather long-winded approach, but it does have signicant advantages when applied to the design of tension structures. Although originally applied to nite difference structural idealisations, the structural problem is modelled as a series of nodes at which coordinates are traced and masses assumed to be lumped. These nodes are interconnected by discrete nite elements representing membrane, cable, strut or beam components. At each stage of the iteration the current node residual forces are updated by considering individual elements in turn. Thus, the assembly of an overall stiffness matrix is not necessary, and the specication of element prestress during form-nding or non-linear stress/strain relations during load analysis is easily achieved. Belytschko et al. [11] utilised the concept of local coordinates for force calculations in non-linear transient dynamic analysis, with the origin of these element coordinates dened by one node and the axes xed relative to another. These local axes are thus free to translate and rotate with the structure as it undergoes deformation. The element displacements are dened as the minimum number of geometric variables necessary to completely dene the deformed conguration of that element, independent of rigid body motion which is treated automatically by the DR process. A set of associated natural element forces are determined from these basic displacements by means of the element natural stiffness relations. As well as simplifying understanding of the element, computer storage and time are saved by minimisation of the number of degrees of freedom per element. For example, the constant strain triangular membrane element, with basic displacements dened by its three side extensions, was rst proposed as an intermediate stage in a direct stiffness element derivation. It has subsequently been adopted in its natural form, with a [3 3] stiffness matrix, to the DR form-nding and analysis of prestressed membrane structures [13,14]. The iterative equations for the motion in direction x of any node i in space at time t are obtained directly from Newton (Force Mass Acceleration): Rtix MiAtix

which may be expressed in central difference form 4. Dynamic relaxation The static solution of both linear and non-linear structures subject to load may be regarded as the limiting equilibrium state of damped structural vibrations excited by that load. The physical basis of DR was initially perceived as the step-by-step solution, for small time Rtix
t Mi(Vix t/2

Vtix

t/2

)/ t

giving the recurrence relation for nodal velocities:

t Vix t/2

Vtix

t/2

tRtix/Mi

where Rtix is the x direction residual force for node i at

684

D.S Wakeeld / Engineering Structures 21 (1999) 680690

time t, Vtix is the associated node velocity, Mi is the node mass and t the time interval. The residual forces Rtix were computed for the then current node coordinates xti. An updated set may now be calculated from the incremented node coordinates: xti
t

xti

tVtix

t/2

Similar recurrence relations apply to all unconstrained degrees of freedom of the structure. The stability of a time stepping dynamic analysis will depend upon the selection of a suitably small time increment, which can be shown to be a function of the local node relative stiffness [4]. When only the nal static solution is sought, ctitious nodal masses may be used that optimise convergence for a chosen time interval, which is typically taken as unity for computational efciency. Separate masses may even be selected for each degree of freedom at a node, or square nodal mass matrices used which effectively optimise masses into the local principal stiffness directions [5]. For practical analyses, a single global mass factor is available to the user to maintain stability in cases of large initial deformation during form-nding or instability during load analysis. The detailed mass assignments are handled automatically. The original need to compute an optimised viscous damping coefcient has been eliminated by the adoption of a kinetic damping procedure [5,15]. When an oscillating body passes through a local approximation to its static equilibrium position, the plot of total kinetic energy against time passes through a local maximum. The total kinetic energy is traced as undamped iterations proceed, and all current node velocities are reset to zero whenever an energy peak is detected. The analysis continues, progressively eliminating the kinetic energy from various modes of vibration until the required degree of convergence is obtained. The process is automatic, requires no specication of damping factors, and can handle gross out-of-balance forces without the need for additional constraints. This means that gross geometrical inaccuracies and stiffness modications may be accommodated during the form-nding process. If one area of a particular structure should be numerically or physically unstable, the computation will typically equilibrate the remaining structure as far as possible. Intermediate visual examination of the model during analysis, coupled with a trace on the location of current maximum residual forces facilitates diagnosis of the underlying problem. The overall iterative process between kinetic energy peaks may be summarised as follows: (a) Initialise node velocities, residuals and previous kinetic energy to zero. (b) Update applied loading components. (c) Compute surface in-plane residuals for geodesic strings if membrane form-nding.

(d) Add applied loads into current residual force vector. (e) Add resolved element force components into residual force vector. (f) Reset residuals associated with constrained nodes to limit motion as required. (g) Update node vectors and coordinates. Calculate current structure kinetic energy. (h) If current kinetic energy > previous, then zero residual vector and return to (c). (i) If current kinetic energy previous, then adjust coordinates to value at KE peak. (j) Check residuals and return to (a) unless required degree of convergence obtained. The kinetic energy peak stages represent successively improving solutions to the problem, and provide a convenient point to reset element constraints and stiffnesses, for example the stability effects of current axial load on the momentcurvature relations of beam elements [16]. Coding and updating a Dynamic Relaxation program is a relatively simple and straightforward procedure. The particular advantage of the method is its clear physical analogy and ready adaptability without compromise to the wide range of modelling problems that arise on real projects.

5. Element and constraint library The following element types are available within the Tensys implementation. 5.1. Membrane elements These are triangular constant strain nite elements modelling orthotropic fabric behaviour. This simple element type requires a relatively dense modelling mesh but permits a straightforward introduction of non-linear material effects. Natural element stiffnesses used with the convected coordinate system account readily for geometric non-linearities. The numerical form generation of membrane structures seeks an equilibrium state in which the geometry and member forces of a structure are known. Material selfweight is generally neglected from this process, and the actual material properties of the elements need not be pre-assigned. The shapes dened are a function of relative stresses, and thus even the member forces may be scaled from the nal desired prestress level. Membrane shape is controlled directly by specied warp and ll stresses in constant stress triangular nite elements. Stresses may be varied smoothly in both warp and ll directions to obtain the desired form. The range of possibilities can vary from the uniform stress distribution of a minimal surface, such as the New Inland Revenue Centre (Fig. 1) and Edo Tokyo Museum (Fig.

D.S Wakeeld / Engineering Structures 21 (1999) 680690

685

3), to complex distributions in both warp and ll directions such as the temporary A-Klasse touring structure (Fig. 2) where the architect had clear visual and functional reasons for the particular shape he wished to achieve. The warp direction of the membrane is generally assigned to one side of the triangular element, and controlled on the surface in an overall sense by the introduction of geodesic string trajectories. These are determined simultaneously within the form-nding computation and also fundamental to the cutting pattern post-processing option. In some boundary areas, individual triangular elements may not have a side that is directly related to the warp direction dened by a geodesic string. In this case the element orientation is automatically assigned by reference to the nearest element in a sequence that has such an orientation. Prior to load analysis, the true elastic properties of the membrane are introduced. The theoretical unstressed element side lengths of individual membrane elements are back calculated from the equilibrium stressed geometry and specied stresses. It is assumed that material stiffness terms are based upon biaxial tests relative to the prestress condition. The stress/strain relations employed are:
w o w

where w is warp stress, o is warp prestress, f is ll w stress, o is ll prestress, w and f are warp and ll f strains measured from prestress state, EAw and EAf are warp and ll stiffness, respectively, and EAwf is cross stiffness (crimp interchange). The membrane elements themselves provide a convenient reference for the calculation of plan and surface applied loading. Pressure loading, whether due to snow, wind suction or pressure, or internal ination pressure may be conveniently updated according to current geometry at kinetic energy peak reset stages as the analysis proceeds. 5.2. Geodesic strings Geodesic trajectories are determined by introducing elastically or tensioned controlled strings into the membrane surface. By introducing these strings with high tensions in such a way that they only control the in-plane movement of surface nodes, the strings take up geodesic lines whose orientation is governed by their boundary end positions. Elastic control can be used to control the spacing of nodes along the string, whilst their out-ofplane residual force components are suppressed so as not to inuence the surface shape. Residual forces are also neglected at string ends so as not to inuence boundary forces or edge cable geometries. This procedure was originally developed so as to maintain stability of a model during form-nding with non-uniform warp and ll stresses. The numerical procedure for simultaneous geodesic calculation ts into stage (c) of the calculation loop as described above: 1. calculate geodesic string residuals {Rg} before all other element types; 2. calculate current node unit surface normal vectors {Vn} from the mean of normal vectors of adjacent triangular membrane elements weighted in inverse proportion to their area; 3. thus the normal, or out-of-plane, residual component at a node is Rn {Vn}T{Rg}; 4. for each geodesic string node, subtract the normal component from the global residuals: {R} {R} Rn{Vn}

EAw EAwf EAf EAwf

f f

w o f

if ( if ( if (

0.0)
w

and

0.0) 0.0)
f

0.0 0.0

5. set global residuals of string end nodes to zero; 6. continue residual summation for other element types. Geodesic strings are only active at the form-nding stage, although their topologies are used at the cutting pattern output stage. 5.3. Strut/cable elements These are two node line elements which may be controlled by specied tension/compression and/or elastic stiffness and slack length.

Fig. 3.

Edo Tokyo Museum Lounge Pavilions.

686

D.S Wakeeld / Engineering Structures 21 (1999) 680690

The form of boundary cables may be controlled by length or force. For length control, it is computationally most convenient to assign an (arbitrary) elastic property and unstressed length. For force control, one line element in the sequence making up a boundary is assigned a specic tension which is maintained throughout the iterative process. For elastically controlled cable elements the current member force is monitored at all stages and reset to zero should compression be detected. 5.4. Slip cables Slip cables are a sequence of line elements through dened nodes representing a friction-free cable in, for example, a pocket on the membrane surface. Under elastic control the common tension in all constituent elements is calculated from the total strain in the element sequence. The uncoupled nature of the DR process makes it particularly straightforward to implement elements of this type. 5.5. Warp seam stiffness

straight, two-node elements with additional rotational degrees of freedom assigned automatically. The orientations of the beam cross-section principal axes are dened by reference to a third node. The inuence of current axial load on the moment curvature relations is updated at kinetic energy peak stages as the analysis proceeds by application of the established s and c stability functions. The inuence of element bowing on axial load is also incorporated. Bending stiff beam elements may also be included in the form-generation process. In this case, however, real elastic properties should be used together with the desired rather than arbitrary levels of prestress in cables and membranes. An unstressed geometry for the beam elements must be dened as the starting point of the analyses, and this will be the reference for all subsequent force calculations in both form-nding and load analysis. Examples of the integrated use of bending elements include the Nottingham Inland Revenue Centre (Fig. 1), SAGA Amenity Building (Fig. 5) and the Atrium Wall for the Chicago Beach Tower Hotel development in Dubai (Figs. 6 and 7). 5.7. Node constraints

Warp seam stiffness is modelled by cable elements generated automatically along the geodesic seam lines. The inuence of seam stiffness in reducing membrane stresses is particularly signicant in the high point ring region of conical membranes with a radial seam layout (for example, the outer membrane of Rothenbaum Stadium, Fig. 4). 5.6. Beam elements Geometrically non-linear beam elements are included for enhanced support structure modelling. These are

Node constraints may be readily implemented in the DR iteration scheme by either resetting current velocities to zero for a fully xed degree of freedom, or resetting node coordinates according to a specic constraint requirement. This is undertaken at every iteration step as outlined in Section 4 above. Translational xity in global directions and rotational xity about global and arbitrary axes are available. In addition, nodes may be constrained to lie within arbitrary planes or on arbitrary vectors. As these planes and vectors are dened in terms of structure nodes, they can also

Fig. 4.

Rothenbaum Stadium, Hamburg.

D.S Wakeeld / Engineering Structures 21 (1999) 680690

687

Fig. 5.

SAGA Amenity Building.

Fig. 6.

Chicago Beach Tower Hotel.

move in space as the structure deforms if necessary. The movement of unconnected fabric over supporting arches is currently represented by the use of spherical constraint surfaces, and this principle may readily be extended to other geometric forms. These constraint planes and surfaces may also be set to allow lift-off of the nodes in one of the local surface perpendicular directions if the current residual force components are positive in that direction. Thus for the arch example, one can model possible fabric lift-off under wind suction. This was utilised in the load analysis of the Kawakita Ohasi Toll Gate project (Figs. 8 and 9), where the membrane was supported on a series of discrete discs parallel to the dome local surface. At the form-nding stage some manual intervention may be necessary for onoff contact problems. For the Edo Tokyo museum (Fig. 3) a PTFE glass membrane was stressed over a series of skeletal steel dome shapes, both bearing on the curved tubular members and then spanning freely between them. The modelling of the critical transition zones where the fabric lifted off an arch was handled iteratively by the examination of local graphics and node residual components.

688

D.S Wakeeld / Engineering Structures 21 (1999) 680690

Fig. 8.

Kawakita-Ohasi Toll Gate.

warp lines. In regions of high transverse curvature a better representation may be made by subdividing a physical panel into separate panels in the computer model. These are unfolded individually into a plane and subsequently sheared together at their common boundary. This is physically admissible as a solution because of the low shear stiffness of coated fabrics. On unfolding the patterns still represent the nal stressed geometry of the membrane. They must be compensated to allow for the actual stretching of the material from the unstressed as-cut state to the installed and stressed condition. The necessary warp and ll direction compensation factors are obtained from biaxial testing of the actual production material to be used. These tests will account for any initial construction stretch of the material, for the level of prestress to be utilised and for creep due to temperature and applied loading. Reduced compensation, or decompensation, may be applied at edge details to ease installation where the stiffness of local members is greater than that of the fabric. As well as patterning the membrane, boundary and support cable element lengths, pretensions and end ttings must be scheduled, together with the nal geometry and orientation of connection elements such as masthead details and steel corner plates.
Fig. 7. Chicago Beach Tower Hotel.

6. Cutting patterns and fabrication geometry The nal fabrication geometry for all aspects of a tension structure project should be derived from a single computer model. This is the nal form model that includes the detailed geometry of all system points and connection details as taken from the nal fabrication drawings. The geodesic seam lines on the membrane surface will have been adjusted such that their relative spacing does not exceed the useable width of the fabric. The three dimensional geometry of a particular fabric panel is then attened out to give the pattern shape by successively unfolding the triangular elements between adjacent

7. Erection analysis In addition to the usual form and design loading computations, it may also be necessary to numerically simulate the erection process in order to predict intermediate membrane stresses and support component forces. This is particularly necessary when the prestress condition is itself close to the working limits for certain elements, as was the case for the New Inland Revenue Centre, Nottingham (Fig. 1). The stability of a support structure may not be assured until all components of a membrane have been installed. For the SAGA Amenity Building currently under con-

D.S Wakeeld / Engineering Structures 21 (1999) 680690

689

Fig. 9.

Kawakita-Ohasi Toll Gate.

struction (Fig. 5), the stability of the principal braced steel arches has been examined by undertaking applied load analyses at all interim stages of installation as the main and side fabric regions are progressively added. Erection studies are undertaken in the reverse sequence of the intended process on site. Starting from the nal stressed state, the tension in stressing elements can be gradually reduced using under tension control in successive analyses. It is advisable to run sensitivity studies with reducing membrane stiffness properties as the prestress level reduces. Entire membrane regions can be effectively deleted from the model by simple commands in the system. For the Chicago Beach Hotel in Dubai (Figs. 6 and 7), erection analysis has been undertaken for the staged introduction of all support trusses and hanger rods prior to installation of the rst membrane. This study has included the post-processing of current deformed states to provide setting-out points at specic survey reference points on the steelwork. Allowance has also been made for the ongoing settlement of the concrete support structure due to creep and elastic shortening.

understood by the engineer. The use of black box software should be strongly resisted. Dynamic Relaxation is one approach that satises the selection criteria, and its implementation is described here in detail with emphasis on its ability to handle the requirements of an active membrane engineering ofce.

Acknowledgements Tensys acted as membrane engineers on behalf of KOCH Hightex GmbH who were roof contractors for all of the referenced projects: New Inland Revenue Centre Nottingham England Architect: Michael Hopkins and Partners Engineer: Ove Arup and Partners Edo Tokyo Museum Lounge Pavilions Tokyo Japan Architect: Kiyonori Kikutake Engineer: Gengo Matsui and ORS Contractor: Fudo Construction Ltd Mercedes Benz A-Klasse Tour 1997 Architect: Kauffmann Theillig and Partner Engineer: Sobek und Rieger Kawakita-Ohasi Toll Gate Japan Architect: Kanazawa Planning Research Co Ltd Engineer: Konishi Structural Engineers Contractor: Fudo Construction Ltd Rothenbaum Stadium Hamburg Germany Architect: Architekten Schweger and Partner

8. Summary The key points in the selection of engineering analysis software for tension structures are that it should not place any limitations on the design process, should have a clear physical analogy, and, most importantly, be fully

690

D.S Wakeeld / Engineering Structures 21 (1999) 680690

Engineer: Sobek und Rieger Chicago Beach Tower Hotel Dubai UAE Architect and Engineer: WS Atkins and Partners Overseas SAGA Amenity Building Folkestone England Architect: Michael Hopkins and Partners Engineer: Ove Arup and Partners

References
[1] Otto F. Tensile structures. MIT Press, Cambridge, MA, 1971. [2] Wakeeld DS. Membrane engineering: principles and applications. In: IASS Asia Pacic Conference on Shell and Spatial Structures, Beijing, 1996. [3] Wakeeld DS. Membrane engineering: current project examples. In: IASS International Symposium on Shell and Spatial Structures, Singapore, 1997. [4] Barnes MR. Form-nding and analysis of tension structures by dynamic relaxation. PhD thesis, University of London, 1977. [5] Wakeeld DS. Pretensioned networks supported by compression arches. PhD thesis, University of London, 1980. [6] Bucholdt HA, McMillan BR. Iterative methods for the solution of pretensioned cable structures. In: IASS Symposium on Tension Structures and Space Frames, Tokyo, 1971.

[7] Schek HJ. The force density method for form-nding and computation of general networks. Comp. Meth. Appl. Mech. Eng. 1974;3. [8] Grundig L, Baurle J. Automated cutting pattern dertermination and control for prestressed membranes. In: Symposium on Textile Composites in Building Construction, Lyon, 1990. [9] Maurin B, Motro R. Density methods and minimal forms computation. In: IASS Colloquium on Structural Morphology, Nottingham, 1997. [10] Day AS. An introduction to dynamic relaxation. The Engineer 1965. [11] Belytschko T. Efcient large scale non-linear transient analysis by nite elements. IJNME 1976;10. [12] Argyris JH et al. A general method for the shape nding of lightweight structures. In: International Conference on Tension Structures, London, 1974. [13] Barnes MR. Applications of dynamic relaxation to the design and analysis of cable, membrane and pneumatic structures. In: International Conference on Space Structures, Guildford, 1975. [14] Barnes MR. Form and stress engineering of tension structures. Structural Engineering Review 1994;6(34). [15] Barnes MR, Wakeeld DS. Form-nding, analysis and patterning of surface stressed structures. In: First Oleg Kerensky Memorial Conference, London, 1988. [16] Wakeeld DS. Dynamic relaxation analysis of pretensioned networks with exible boundaries. In: IASS World Congress, Madrid, 1979.