NM Tension structures Form and behaviour TA 660 . C3LU9x W J Lewis /2003 DavisTension Structures Form and Behaviour W. J. Lewis University of Warwick Graphic Design: A. J. Cartwright, University of Warwick = ‘L! ThomasTelfordPublished by Thomas Telford Puslishing, Thomas Telford Ltd, 1 Heron Quay, London E14 4JD. www.thomastelford.com Distributors for Thomas Telford 30oks are USA: ASCE Press, 1801 Alexander Bell Drive, Reston, VA 20191-4400, USA Japan: Maruzen Co. Ltd, Book Department, 3-10 Nihonbashi 2-chome, Chuo-ku, Tokyo 103 Australia; DA Books and Journals, 648 Whitehorse Road, Mitcham 3132, Victoria First published 2003 A catalogue record for this book is available from the British Library ISBN: 0 7277 3236 6 (© Wanda Lewis 2003 All rights, including translation, reserved. Except as permitted by the Copyright, Designs and Patents Act 1988, no part of this publication may be reproduced, stored in a retrieval system or transmitted in any form or by any means, electronic, mechanical, photocopying or otherwise, without the prior written permission of the Publishing Director, Thomas Telford Publishing, Thomas Telford Ltd, 1 Heron Quay, London E14 4D. This book is published on the understanding that the author is solely responsible for the state- ments made and opinions expressed in it and that its publication does not necessarily imply that such statements and/or opinions are or reflect the views or opinions of the publishers. While every effort has been made to ensure that the statements made and the opinions expressed in this publication provide a safe and accurate guide, no liability or responsibility can be accepted in this respect by the author or publishers. ‘Typeset by Keyword Publishing Services Printed and bound in Great Britain by MPG Books, BodminContents Preface Acknowledgements 1 Introduction 1.1 Definitions and classifications 1.1.1 Boundary tensioned membranes 1.1.2. Pneumatic structures 1.1.3. Pre-stressed cable nets and beams 1.2 Design process of tension membranes 1.3 Main features of tension membranes 1.4 Conventional roofing forms versus tension membranes 1.4.1 Common misconceptions relating to tension membrane design 15 Closing remarks References Form-finding 2.1 General concepts. Nature's ‘secrets’ 2.2 Concept of a ‘minimal surface’ — historical background 2.3 Form-finding methodologies 2.3.1. Physical models 10 ve 15 17 nig 20 20 23 26 28 itiContents 2.3.2 Computational models 29 2.3.3 ‘Soap-film debate’ 31 2.3.4 Form-finding, or form-dictating? 35 References 36 3. Geometrically non-linear behaviour. Solutions by commonly used numerical methods 39 3.1 Geometric non-linearity 39 3.2. Commonly used computational methods for the analysis of geometrically non-linear behaviour ray 3.3 Transient stiffness method 41 3.3.1. Static analysis of skeletal, ‘linear’ structures 42 3.3.2 Static analysis of skeletal, ‘non-linear’ structures — why iterative computations are necessary 45 3.3.3. Transient stiffness method applied to static analysis of tension structures. Role of geometric stiffness 46 3.3.4 Application of the transient stiffness method to form-finding of tension structures 48 3.3.5. Evaluation of the transient stiffness method 50 3.4 Force density method 51 3.4.1 Application of the force density method 52 3.4.2 Numerical example 53 3.4.3, Matrix formulation of the force density method 54 3.4.4 Evaluation of the force density method 57 3.5 Dynamic relaxation method 57 3.6 Computational static analysis versus form-finding 58 3.6.1 Out-of-balance forces in static analysis and in form-finding 58 3.6.2 Elastic and geometric effects in static analysis and in form-finding 58 References 59 4. Dynamic relaxation method 61 4.1 Dynamic relaxation method with viscous damping 62 4.1.1 Stability of tie iterative solution 63 4.1.2 Critical viscous damping coefficient 65 ivContents 4.2. Dynamic relaxation method with kinetic damping 67 4.2.1. Iterative process 68 4.2.1.1 Location of the point of maximum kinetic energy: correction for displacements 69 4.2.2. Application of the dynamic relaxation method with kinetic damping to static analysis of tension cable nets 70 4.3 Efficiency of the dynamic relaxation relative to the transient stiffness method 72 4.3.1. Simple net 72 (2x1) cable net cf: (2x2) cable net 72 Hypar net 73 Spatial net 74 Saddle net 75 Results 75 Discussion 7 44 Evaluation of the dynamic relaxation method 79 References 80 . Case studies of cable roof structures. Design issues: form-finding and patterning, 81 5.1 Introduction 81 52 Case studies 83 5.2.1 Spatial net 83 1.1 Geometry of the geodesic net and differentially stressed net 84 5.2.1.2 Behaviour of the geodesic and differentially stressed nets under imposed loading 85 5.2.1.3 Geodesic net versus differentially stressed form and the ‘soap-film debate’ 04 5.2.2. Saddle net 95 5.2.3 Poskitt truss 97 5.2.3.1 Load Case 1 99 2 Load Case 2 100 3. Accuracy in model manufacture|patterning 103 5.2.34 Conclusions 106 References 106Contents 6. Tension cables in suspension bridges. A case of form-finding vi 6.1 ‘Shape equation’ for an inextensible suspension cable 6.1.1 The parabolic approximation 6.1.2. The catenary 6.1.3 Constant stress cable 6.2 ‘Shape equation’ for an extensible suspension cable 6.2.1. Extensible (elastic) cable under self-zveight 6.2.2 Extensible cable under self-weight and deck weight 6.3 Numerical modelling of shape of suspension bridge cables 6.3.1 ‘Geometric’ case. Form-finding of an inextensible suspension cable under dead weight of the bridge 6.3.1.1 Inextensible cable. Form-finding for Load Case 1 - deck weight only 6.3.2 6.3.3, 6.3.1.2 6.3.1.1.1 6.3.1.1.2 6.31.14 Hanger arrangements 1a) and 1(b) — uneven number of hangers Finite difference analysis. Deck weight only. Hanger arrangement 1(a) — centre hanger present and end spacing halved Finite difference analysis. Deck weight only. Hanger arrangement 1(b) — centre hanger present and constant spacing throughout Analysis of results - Load Case 1 Inextensible cable, Numerical form-finding for Lond Case 2 ~ deck weight plus cable weight ‘Elastic’ case. Form-finding of an extensible cable under dead weight of the bridge 6.3.2.1. Extensible cable. Numerical form-finding for Load Cases 1 and 2 Comparisons between the ‘clastic’ and ‘geometric’ cases 6.3.3.1 Comparison of geometry 6.3.3.2. Comparison of load transfer 6.3.3.3. Practical aspects eae 6.4 General comments 6.4.1 Form-finding of suspension bridge cables References Estimate of temperature effect 110 112 pe) 114 115 noe) ae 119 119 122 122 125 126 127 128 129 aoe 136 136 136 138, 138 140 140 141Contents 7. Modelling of tension membranes 71 a 73 7A vi 7.6 77 78 ae 7.10 Introduction Surface discretization Surface discretization for use with the transient stiffness method: limitations of the approach 7.3.1 Line elements 73.1.1 Elastic stiffness formulation of a line element for use with the transient stiffness method 7.3.1.2 Geometric stiffness formulation of a line element for use with the transient stiffness method 7.3.2. Limitations of the approach Surface discretizations used with the dynamic relaxation method 7.4.1 Line elements 7.4.2 Triangular, ‘triple force’ elements 7.43 High-order elements 7.4.4 Noir-finite element formulations 744.1 Cubic spline method 744.2 Triangular mesh method Mesh control — implications for design 7.5.1 Case study: mesh control in form-finding and patterning 75.1.1 Analysis of results and consequences for design Patterning of membranes Line elements in modelling of stable minimal surface membranes Triangular elements in modelling of stable minimal surface membranes Numerical accuracy and criteria used for convergence Pre-processing References ‘Appendix I. Materials used for tension structures 143 143 143 144 144 149 bheya 153 aa wks 155 156 156 161 163 165 167 169 170 174 ss] 177 181 viiContents AI1 Fabric membranes AL2 Cable structures Appendix II, Minimal suzfaces AIL1 Calculus of variations AIL2 Variational problem: Euler-Lagrange equation for a minimal surface Appendix IIL Viscous damping in dynamic relaxation Index viti 181 182 185 185 187 aaPreface Tension structures offer wonderful design opportunities, due to their ability to span large distances with elegance and structural efficiency. This book covers two main types of tension structures: pre-stressed cable nets and beams, and membrane structures, used as roofing forms. The discussion of their behaviour is extended to include cables in suspension bridges, as they are just a special form of cable trusses. The most sensitive issues in the design of tension structures are the choice and definition of their structural form. Cable nets and membranes, in their tensioned state, adopt unique shapes that cannot be described by simple mathematical functions. They, quite literally, have to be found during a design stage, known as ‘form-finding’. This distinguishes them from conven- tional, rigid-type structures, whose shapes can be imposed. Because of their flexibility, tension structures exhibit load-shape interaction, which is visible. If their initial shape is computed on the basis of some preconceived idea of what the structure should look like, the actual shape may bear only some resemblance to the image on a computer screen. Minimum energy forms are observed in naturally occurring forms. This was a focus of study in an inter-disciplinary research programme SFB230 — Sonderforschungsbereich (Special Structures Group) under the direction of Frei Otto at the Institute of Lightweight Structures (IL), University of Stuttgart, Germany (1984-1995). The conclusion of that programme, in which the author had an opportunity to participate, was that principles of optimality and aesthetics of structures have their roots in nature. During the growth of a tree, for example, a constant surface stress devel- ops to ensure a uniform distribution of load along the surface of a stem or a branch. When a tree gets damaged, e.g. a branch gets broken off, the healing process minimizes the surface area of the wound. It is interesting to observe that the two features — a constant surface stress and a minimum surface area — can also be found in inanimate natural objects such as soap-films (the weight of the soap-film can be ignored). Soap-films are known, in math- ixPreface ematical language, as stable minimal surfaces. This book postulates that the principle of constant stress, which governs the formation of their shape, should be taken seriously as a principle of design for tension membrane structures, The aims of this book are to enhance understanding of tension structures from both a practical and theoretical point of view and to provide insights into the problems associated with computational modelling of their struc- tural form and behaviour. The most commonly used computational approaches are described, together with a brief evaluation of each method. The aesthetically pleasing examples of lightweight membranes illustrated in this book are the results of the interactive use of computer and physical models in a process that integrates both architectural and engineering skills. Form-finding is a theme running through all chapters of the book, focusing, on good and bad designs and discussing issues related to ill-chosen or ill- defined forms of tension structures in a practical context. Initially, the author had not envisaged the inclusion of form-finding calculations for suspension bridge cables, as these, according to the current literature, are structures of ‘known shape’. The case studies included in the book show that this is not necessarily true. ‘Patterning’ is another very important issue related to the manufacture of tensioned cable nets and membranes. Regrettably, a detailed exposition of the topic lies outside the scope of this book, and only a brief outline of the subject is given. Mathematically, patterning remains an unresolved issue, as it is not possible to ‘translate’ three-dimensional, undevelopable surfaces into two-dimensional cutting patterns, without distortion. This is an area of inten- sive research, made complex by the need to include non-linear and aniso- tropic material properties of structural fabrics, while addressing the issue of errors in surface geometry and minimization of wastage of material. A brief description of the scope of the book is given below. Chapter 1 gives an introduction to tension structures (with practical illus- trations), provides a summary of their main features, and, finally, sets out a discussion of certain misconceptions related to their design. Materials used for tension membrane and cable structures are described in Appendix I. Chapter 2 is devoted to the topic of ‘form-finding’. The significance of natural laws in finding optimal forms of structures is discussed, particularly in the context of soap-films (stable minimal surfaces). Some historical back- ground to the mathematical modelling of such surfaces is provided and the derivation of the basic result (the Euler-Lagrange equation (1760)) is given in Appendix II. Difficulties associated with obtaining an analytical solution to the Euler-Lagrange equation for an arbitrary boundary, or even making effective use of the Laplace-Young equation (1806) describing the same problem, emphasize the need for alternative approaches, involving num- erically oriented methods. A brief description of the ‘soap-film debate’ isPreface given in relation to the design of membrane structures. The chapter ends with the question: form-finding, or form-dictating? Chapter 3 introduces the concept of geometrical non-linearity, characteriz~ ing the behaviour of tension structures. It discusses the most commonly used computational methods in the modelling of structural forms and response to loading. The methods described include the transient stiffness, dynamic relaxation, and force density. Descriptions of each of the methods are fol- lowed by a brief evaluation. Chapter 4 gives the origin and a detailed description of the method of dynamic relaxation, which has been successfully applied to the problem of form-finding and static analysis of tensioned cable nets and membranes. The development of the method from viscous to kinetic damping is outlined, and the corresponding recurrence equations that enter the numerical algorithm are derived. The derivation of the critical viscous damping coefficient is given in Appendix IIL. The efficiency of the method, relative to the transient stiffness approach, is described by a study of numerical convergence using several cable net examples. Chapter 5 gives a simple illustration of load transfer in cables to demon- strate the need for a careful choice of the initial structural form of cable structures. It presents case studies involving several cable structures, includ- ing the Poskitt truss for which high-precision experimental measurements are available. The case studies are used to highlight the issue of form-finding, as oppose to form-dictating, raised earlier in Chapters 1 and 2. A study of the behaviour of a geodesic net (a cable equivalent of a stable minimal surface) is presented, and the issue of form-finding is discussed in the context of non- geodesic (differentially stressed) nets. This leads to the ‘soap-film debate’ once again, but this time supported by the results of a case study. Numerical modelling of the Poskitt truss is used to demonstrate the load- relieving capabilities of tension structures. The chapter ends with a discus- sion of the accuracy of the manufacture (patterning) of cable lengths, and potential problems of gaining control over the shape of the structure on site (with reference to experience with the Munich Olympic Stadium). Chapter 6 extends the concept of form-finding to suspension bridge cables. It is clear from perusal of acclaimed texts that suspension bridge cables are assumed to have a parabolic shape. Such an assumption is valid only for laterally flexible but inextensible and weightless cables. Occasionally, comments are made about catenary shapes. In reality, when the elasticity of cables is included, the resulting shape under the self-weight of the bridge is neither a parabola nor a catenary; it is a shape that has to be determined through non- linear analysis. The chapter starts with derivations of ‘shape’ equations for both extensible and inextensible cables, in an attempt to clarify the issue from the analytical point of view, before embarking on numerical analysis. Results of numerical form-finding for both extensible and inextensible cables are xiPreface given for a variety of span/dip ratios. In the case of inextensible cables, a variety of hanger arrangements are studied. The results of numerical form- finding of extensible cables are discussed within the context of potential problems that may be experienced on site. The issue of form-dictating is returned to at the end of the chapter, with a brief discussion of the London Millennium Bridge. Chapter 7 presents numerical modelling of tension membranes. It begins with surface discretization — a common stage in all numerical approaches. Mathematical formulations of line and surface elements are given for use with the transient stiffness and dynamic relaxation methods, respectively Limitations of the transient stiffness method are highlighted in the context of the line element formulation. A criticism of the high-order (24 degrees of freedom) elements for use in modelling of membrane structures is given. This is followed by a brief description of the recent non-finite element approaches to form-finding of stable minimal surface membranes. An approach to form-finding, which employs seam-line control in order to assist patterning, is described. The implication of such a methodology for design of membrane structures is discussed using a case study. Solved numerical examples of stable minimal surface membranes, validated by soap-film mod- els, are used to discuss the performance of line, triangular and high-order elements. Approaches to patterning are discussed in outline. The potential of non-finite element approaches (used in form-finding) to facilitate patterning is explained. Chapter 7 concludes with examples of stable minimal surface membranes solved using an integrated Formian/dynamic relaxation approach. xiiAcknowledgements The author wishes to acknowledge the use of some of the Figures provided by her former PhD students in the course of their doctoral studies: Dr P. D. Gosling (Figs. 7.17-7.18), and Dr T. S Lewis (Figs. 7.13-7.15, 7.19-7.21). Thanks are also due to my research fellow, J. $. Brew, for providing Figures 7.8-7.12. The picture on the cover of this book is by courtesy of IL - Institute of Lightweight Structures (now ILEK), University of Stuttgart. xiiiIntroduction The term tension structures covers a broad category of structures, which includes fabric membranes, pre-stressed cable nets, and cable beams in the form of trusses and girders. They are being used either as roofing forms, or bridge structures, and now a permanent feature of modern architecture, showing their potential for exploitation as efficient structures and aestheti- cally pleasing art forms. The essential elements of a membrane structure are: (i) a highly flexible fabric held under tension in order to generate stiffness in the surface; (ji) one- dimensional flexible elements, i.e. ties or cables, to create ridges, valleys and edge boundaries; and (iii) rigid support members sustaining compression/ bending. Cable structures are constructed using a combination of elements (i) and (ii). The pre-stress in fabric membranes can be achieved by applying a force at the boundaries, or by supplying an air pressure within the enclosure. Cable structures can be tensioned by applying direct axial forces to the cables, or by loading freely suspended cables with heavy cladding /decking. The latter describes structures falling into a category of suspension cable roofs and bridges. Fabric and cable have come to be used in conjunction with one another as major structural elements in designs that radically depart from traditional constructions. Their exploitation potential is immense, ranging from large span roofs over exhibition halls and sports stadia, to small canopies provid- ing shelter from rain or sun (Figs 1.1-1.8). Figure 1.1 shows Munich Olympic Stadium, which is one of the earlier examples of large tension roofing forms. The structure, designed by an inter- disciplinary team from Stuttgart, represents an outstanding architectural and engineering achievement of its time!" The cable roof, covering an area of 75000 m? required approximately 210 km of cable"! to make up the tension net, as shown. The project, because of its scale, marked a significant shift from physical model measurements towards computational methods for the prediction of the initial geometry and structural behaviour.Tension Structures. Form and behaviour ») Figure 1.1 Munich Olympic Stadium, 1972. (a) Aerial view, (b) view from inside.Introduction Further examples of tension structures are shown in Figs 1.2-1.8. The Millennium Dome, which is a cable net structure clad with tensioned fabric (Fig. 1.2), represents a departure from the usual anticlastic, or ‘saddle’ shape (characterized by a negative Gaussian curvature) that gives tensioned mem- branes their stability and stiffness. Instead, it has a synclastic shape (with a positive Gaussian curvature) and its structural action is tension, rather than compression that would be expected in a dome. The Millennium Dome, therefore, raises a question of ‘when is a dome not a dome?"*! A further discussion of this structure is presented in Section 1.4.1 and Chapter 2, Section 2.3.4. Figure 1.4 shows a membrane roof of a sports complex in East Germany. The conical membrane forms are pre-stressed by vertical antennae and edge cables. The roof is anchored to the ground by a series of slender masts, as shown. The project involved re-structuring of an old sanatorium. Tensile structures in and around Parque Expo '98 in Lisbon are shown in Figs 1.5-1.7. They cover an area of 4000 square metres. Their visual impact is most dramatic at night, as shown by Fig. 1.5 (b). The structure shown in Fig. 1.6 illustrates an unusual application of a tension membrane — it provides a bridge between two channel sections. The convertible umbrella structures''! shown in Fig. 1.8 are hydraulically operated and designed for environmental/climatic control; they provide shade during the day, and radiate heat during the night, when they are closed. The exact opening and closing times of the umbrellas depend on changes in temperature, which are monitored and fed into electronic con- trols, The structures were designed by an interactive physical and computa- tional modelling aimed at preserving a constant tension field in the membrane. Physical modelling involved making and photographing soap- film models made between boundaries geometrically similar to those in the real structures. Soap-film surfaces are characterized by a constant surface stress and their use as analogues in modelling of the umbrella structures resulted in a purity of form valued in Islamic culture. The issue of finding optimal forms of membrane structures and the role of soap-film models is discussed further in Chapter 2. This also provides a main focus of the book, as the theme of form-finding runs through almost every chapter. Tal 1.1 Definitions and classifications ‘There exist several categories of tension structures, as discussed below: (® boundary tensioned membranes (ii) pneumatic structures (iii) pre-stressed cable nets and beams.Tension Structures. Form and behaviour ‘proddony oing :uSts0q “dnysuouang sioBoy pampry -aampampey “090d “uopurT “yotame15 ‘smog untuayEYY ay, ZL 24Introduction Figure 1.3 Canopies around the area of the Millennium Dome, Greenwich, London, 2000. 1.1.1 Boundary tensioned membranes Boundary tensioned membranes are stressed by stretching the surface to meet the boundaries made of flexible, tension cables, or a rigid frame/ beam. Such structures are illustrated in Figs 1.1-1.8. The main difference between a boundary tensioned membrane and a pneumatic structure (discussed next) lies in the method of stressing the surface. Also, the former has zero excess pressure between the outer and the inner surface. The pre-stress levels in the surface must be of an appropriate level, to ensure against a possible loss of tension during the life span of the structure on one hand, and to allow the material to stay within the elastic range of deformations on the other. Under imposed loads, such as wind and snow, the stress in the surface can increase 6-8, even 10-fold. For these reasons, the design of tension membranes is aimed at keeping the initial pre-stress at approximately 1/20 of the breaking strength of the cloth. Further, it should be noted that the value of the breaking strength of fabrics is based on the strip tensile strength (measured uniaxially) of a clean, new, and dry cloth. In biaxial tests, fabrics may exhibit strengths approximately half that value. Their strength is also affected by temperature, moisture movements and creep. Therefore, manufacturers’ specifications of mechanical properties ofTension Structures. Form and behaviour b) Figure 1.4 Roof for a sports complex in Masserberg, East Germany, 1994. (a) Aerial view, (b) elevation view. Design: IPL Ingenieurplanung Leichthau. Membrane structure: Canobbio S.p.A., Milan. Photograph courtesy of Canobbio S.p.AIntroduction ») Figure 1.5 Exhibition area of Parque Expo '98 in Lisbon, Portugal. (a) Tensile roof. Architect: A. Barreiros Ferreira, (b) umbrella structures, Architect: M. Salgado. Design: IPL Ingenieurplanung Leichtbau. Membrane structure: Canobbio S.p.A., Milan. Photograph courtesy of Canobbio S.p.A.Tension Structures. Form and behaviour Figure 1.6 Suspended channel connection in Parque Expo '98 in Lisbon, Portugal. Photograph courtesy of Canobbio S.p.A. fabrics need to be confirmed through testing. The long-term properties of structural fabrics are a subject of continual research. More generally, research is needed to find inexpensive, fire proof, easily jointed and transportable fabrics and to understand their thermal behaviour. Fabrics for tensioned membranes can be made from natural fibres, such as cotton (canvas fabric), ceramics (glass fibre), and synthetic organic fibres, such as polyester. They are treated with layers of polymer coatings to improve water tightness of the surface and enhance self-cleansing properties. There is an enormous range of materials currently on the market, but the most commonly used are: PVC-coated polyester, Teflon-coated glass fibre one-coated glass.""> 16 171 A more detailed description ofIntroduction Figure 1.7 Entrance tent — part of the main entrance to Parque Expo ’98 in Lisbon Portugal. Architect: A. B. Ferreira. Photograph courtesy of Canobbio S.p.A. Figure 1.8 Convertible umbrellas 17x18 m, Profet’s Mosque in Medinah, Saudi Arabia, 1992. Design: Sonderkonstruktionen und Leichtbau (SL), Stuttgart. Photograph courtesy of SL.Tension Structures. Form and behaviour the fabrics used in the construction of lightweight tension membranes is given in Appendix I. The applications of some of the types of materials for tension membranes are illustrated in Figs 1.1-1.9. The roof for Munich Olympic Stadium (Fig. 1.1) consists of a pre-stressed cable net supporting acrylic glass cladding. The Millennium Dome (Fig. 1.2) was originally designed to be clad with a PVC- coated polyester, but was later changed to a Teflon-coated glass fibre (PTFE) to improve durability. The membranes shown in Figs 1.3-1.7 are made from polyester fabric coated with PVC, except for the channel connection, which was fabricated from PTFE. Some structures are fabricated from a special PVC-coated polyester with a PVDF top coat (Appendix D, to reduce dirt retention and protect the surface from harmful ultraviolet (UV) radiation. Tension membranes, which feature in Fig. 1.8, have been manufactured out of woven Teflon. 1.1.2 Pneumatic structures Pneumatic structures, or ‘air houses’ (Fig. 1.9) are thin membranes stressed by internal air pressure generated by fans." '*! Their shape is very strongly affected by the difference between the external and internal pressures, which change continually, as a result of temperature variations, wind and snow load conditions. It is difficult to ensure adequate internal pressure through- out the life span of the structure, due to unpredictability of environmental factors, or failure of internal control systems responsible for maintaining the internal pressure. Air houses are designed!® 1°! to maintain an internal pressure between 0.2 KN/m? and 0.55 kN/m”. The design values of snow loads are typically between 1.2 KN/m? to 2.4 kN/m? (on plan) and hence much higher than an inflation pressure. It is therefore recommended that heating of the internal air space is undertaken to melt the snow and prevent dimpling (Le. reversing of the surface curvature) of the fabric, which would otherwise occur. Apart from the local damage to the skin of the membrane caused by the accumulation of snow, or wind forces, or both, an overall instability of the structure may also occur, due to wind excitation activating the large mass of air inside the membrane. The performance of pneumatic structures depends also on the grade of fabric used in construction. Good quality fabric has to be selected to give an adequate factor of safety against the propagation of cuts, or flaws likely to exist in the material. A factor of safety of 6, or more is recommended, which implies that the stresses in the fabric, at any time, should be at least 6 times lower than the minimum breaking strength (the minimum breaking strength corresponds to a value below which no more than 1% of breaking strength test results may fall). In addition, a factor of safety of 1.5 is recommended for loads." 10Introduction ») Figure 1.9 Air supported structure. Swimming pool in Bologna, 1994. (a) External view, (b) view from inside. Photograph courtesy of Canobbio S.p.A. 11Tension Structures. Form and behaviour The popularity of pneumatic structures continues, because they provide cheap alternatives to permanent structures. However, they are sometimes designed without an input from structural engineers, which calls for tight quality control measures. 1.1.3 Pre-stressed cable nets and beams The next category of tension structures, pre-stressed cable nets and beams, form, jointly with tension membranes, the main topic of study in this book. There are very strong similarities between the behaviour of tension mem- branes and cable nets. A pre-stressed cable net is, to an analyst, a discrete- type membrane, or a membrane represented by a system of cable, or line elements, as discussed in Chapter 7. The term ‘tension structure’ will be used to describe both cable nets and membranes Cable nets can be stressed directly using rigid supports such as compres- sion ring beams (Fig. 1.10 (a)), or flexible edge cables with supporting masts and tie backs (Fig. 1.10 (b)). They can take the form of a suspended structure stabilized by means of a heavy roof cladding (Fig. 1.10 (c)), or can be encased in concrete. The opportunity for creating new structural forms is immense.!"? 113 141 If q pre-stressed cable net constitutes a part of a con- crete shell, it is no longer a lightweight tension structure and the advantage of flexibility is lost. However, because of their apparent visual lightness, the shapes of cable net structures are copied into rigid forms, as was the case with concrete ‘tents’ constructed in the Middle East in 1991.'"'! Pre-stressed cable beams are two-dimensional versions of cable nets, and their main application lies in industrial buildings. Cable beams can be classified further into cable trusses and cable girders. Cable trusses, illustrated in Fig. 1.11 (a) and (b) have their top and bottom cables, as well as the internal members (known as hangers) all working in tension. The main load carrying elements, i. the cables, have a much larger cross-sectional area than the hangers. The shape of the truss and the initial level of pre-stress must be so designed as to prevent hangers going into compression under imposed load. Cable trusses can also be utilized as sus- pended structures stabilized by a heavy cladding, or as pre-stressed forms incorporating solid, rather than textile type cladding. A well-known example of a cable truss is a suspension bridge (Fig. 1.11(c)). Cable girders, illustrated in Fig. 1.12, incorporate struts as their compres- sion members. The resulting forms are less flexible than cable trusses and, in principle, no different to tensegrity systems/''®! which consist of cable elements and bars; the bars may cross each other, but are connected only to the cables. There are strong similarities between the behaviour of tense- grity systems and cable nets, and, therefore, the two types of structures may be considered as belonging to a group called ‘cabled structures’. The pos- LRIntroduction Figure 1.10 Pre-stressed cable nets: (a) spanning a ring beam, (b) spanning onto edge cables and (c) supporting heavy cladding.Tension Structures. Form and behaviour aS ®) ~ -_- Figure 1.11 Pre-stressed cable trusses: (a) inclined hangers, (b) vertical hangers, and (c) suspension bridge structure. sibility for creating hybrid forms by combining a cable truss and girder into one cable beam is illustrated in Fig. 1.13. ‘The amount of pre-stress in a cable net structure depends on the allowable deformations and fatigue strength of individual members. Excessive pre- tensioning, of cables to reduce deformations is expensive, and, therefore, it is important to derive the stiffness of the structure through changes in struc- ture's geometry, arising from the alteration of boundary configurations. In order to reduce costs, a certain amount of movement of the structure has to Figure 1.12 Pre-stressed cable girder. 14Introduction Figure 1.13 Cable girder/truss. be allowed, and therefore, the roof cladding design must incorporate this requirement for flexibility.!"7! Materials used for the construction of cable structures are discussed in Appendix I. Cables!" 117, “8 are usually made from spiral steel wire strands, or steel ropes containing steel or synthetic fibre cores, as illustrated in Fig. AL2.1. 1.2 Design process of tension membranes In this section and in the remainder of the book the term ‘tension membrane’ is used to describe boundary stressed structures (cable net and fabric) rather than pneumatic membranes. In general, the design of tension membrane structures follows a three-stage procedure: @) the ‘form-finding’ i) the ‘patterning’ (iii) the ‘static analysis’. One common problem that tension structures share is their shape definition. With regard to boundary tensioned membranes, their configuration is not given by an obvious mathematical function. Consequently, the designer is faced with the problem of ‘form-finding’, ie. finding the basic static shape of the structure (under pre-tension forces only), before detailed analysis, invol- ving, imposed loads such as snow and wind, can commence. Thus, the ‘form- finding’ stage addresses the question of the surface geometry of the tension membrane spanning a given boundary configuration. In this case, a physical model can be used for visualization purposes, but a detailed description would entail problems such as error magnification, if measurements of the model were to be scaled up toa full-size structure. Further, a measurement of the actual tension field in the surface can be tedious and inaccurate. Hence, for more accurate form-finding of tension membranes, iterative computa- tions are carried out, which amount to a gradual adjustment of the surface geometry until it becomes compatible with the state of static equilibrium and 15Tension Structures. Form and behaviour the level of pre-stress in the structure. A more detailed exposition of the form-finding issues is given in Chapter 2. The ‘patterning stage’ allows a three-dimensional shape of a tensioned sur- face to be translated and relaxed into a two-dimensional cutting pattern, to enable the manufacture of the membrane. The pattern has to be made out of strips of fabric of 2-3 m wide, with minimum waste of material and distortions of the fabric, This problem can again be solved using small-scale models, which would provide useful information, but, as indicated above, the direct use of the results for a full size structure could be questionable. Computational methods employed for cutting patterns involve a certain amount of flattening of the surface and this leads to inaccuracies. Further errors may be introduced when simplifying assumptions are made in order to produce smooth cutting lines automatically. In conclusion, an analysis through computation and phy- sical modelling used interactively is believed to be the most effective approach to both form-finding and patterning Static analysis is a continuation of the form-finding process, in which stresses and deformations are calculated under imposed external loads. The process begins with the form-found shape of the structure, initially in equilibrium, and with a given surface stress distribution. Static loads are subsequently applied, and a new state of equilibrium is found. The proce- dure has to follow an iterative computation, in which displacements of the structure and stresses resulting from the applied loads have to be deter- mined, as explained in Chapter 2. The problem is geometrically non-linear, but the non-linearity is not as pronounced as in the case form-finding. Hence, the convergence of the numerical solution is easier to obtain. The dynamic analysis of lightweight tension structures, not listed under the three-stage design procedure, is typically based on data obtained from wind tunnel experiments involving small-scale models. The wind loading is established with the help of pressure coefficients determined from such experiments. Although codes of practice provide pressure coefficients for conventional buildings, these cannot be used for doubly curved shapes of tension structures, and hence the need for wind tunnel experiments However, the small-scale models created for the purpose of wind tunnel testing are usually made of rigid material, incapable of giving a true representation of the pressures that may develop on the surface of a flexible membrane. Conservative values of ‘equivalent static loads’ are used to simulate wind action/"* '*! and this approximation has proved to err on the safe side. The major difficulty in the analysis of tension structures lies in establishing not only the design values of wind loads, but also snow loads. Although snow and wind modelling is cost-effective for large structures, it cannot be afforded for smaller structures where experience, judgement, and code pro- visions must be used. Tension membranes represent a considerably more 16Introduction complex aeroelastic problem than that encountered in aircraft wing design, for example. With an aircraft wing, structural deflections may be assumed to be small, and because the flow is one over a surface of shallow curvature, the flow may be assumed to be ‘attached’. Consequently, the appropriate aero- dynamic model is a linearized perturbation of the unsteady flow equations. With a tension membrane, because of the curvatures involved, flows will not necessarily be attached, and surface deformations resulting from wind load- ing, may well be large. It follows that the classical approach to aeroelastic modelling is likely to be of restricted significance in membrane structures, and there is a considerable need for further research into the interaction of surface geometry, wind loads and structural response. 1.3 Main features of tension membranes The characteristic features of tension membranes are summarized below. ( A tensioned surface spanning an arbitrary boundary configuration adopts its own, unique shape, consistent with the chosen boundary, material properties and state of pre-stress (which can be constant or varied). The shape of a tension membrane cannot be dictated, it has to be found. The ‘form-finding’ process is the starting point in the design process. (ii) The structure accommodates loads through changes in surface ten- sions and very large displacements, producing geometrically non- linear behaviour. Even when the working loads are within the elastic limit, the deflections are of such magnitude that the resulting changes of the overall geometry must be accounted for in the analysis. (iii) The level of pre-stress in a fabric membrane is usually kept low to ensure against tearing under imposed loads. Compression and/or bending stresses develop in a supporting structure. (iv) The structural behaviour of tension membranes is characterized by low bending and shear stiffnesses. (v) The cost of the computational analysis, unlike that of a conventional structure, forms a significant part of the total cost of the whole design project. These characteristics highlight the main differences between the design of flexible, fabric roofing forms and conventional, rigid structures. 1.4 Conventional roofing forms versus tension membranes The growing interest in tension structures can be attributed to two factors. Firstly, they offer the designer scope for dramatic artistic expression, and, aeTension Structures. Form and behaviour secondly, they provide efficient, lightweight load-carrying systems. They conform to a doctrine: ‘for an efficient structure use tension rather than compres- sion, and either in preference to bending."-"! The above statement has to be seen in a broader context of structural optimization. Functionality can sometimes dictate the type of structural action necessary to accommodate it. This is the case with CUBIC struc- tures,!'°l which harness bending in floors and roofs in order to provide space and access to services, while maintaining slim construction. Structural design requires both architectural and engineering skills to pro- duce a functional, safe (and sometimes even aesthetically pleasing) product. In the design of a conventional, rigid-type building made of traditional materials such as steel, concrete, timber or masonry, the engineering and architectural aspects of the structure can be quite easily separated, as explained in the succeeding sections. Rigid-type structures do not deform very much, or at least not sufficiently, as to affect their overall geometry. Their behaviour is descriked as ‘linear’, because they deflect proportionally to the loads applied, ie. have a linear load—displacement relationship.* Often, the term ‘linear’ is used to describe just the stress-strain character- istics of the material. Structures which obey a linear load-displacement rela- tionship will automatically be characterized by a linear stress-strain relationship (Hooke’s law) for the material. However, the opposite may not be true. In the case of cable structures, for example, the material may be assumed to behave in a linearly elastic manner, but the behaviour of the structure will, in general, be geometrically ‘non-linear’, unless very high levels of pre-stress are imposed to limit the deflections. A cable structure behaves in a ‘geometrically non-linear’ manner, which means that 1 fold changes in the applied loads do not result in 1 fold changes in deflections. In general, the shape of a conventional, ‘linear’ structure remains (to a very good approximation) essentially unaltered, when the structure is subjected to loading within its design specification. At the analysis stage, which follows the conceptual design, an engineer perceives the structural form as a system of simple, interconnected elements that need to be checked for stability, deformation and stress under imposed loads (self-weight, wind, snow, etc). The main tool used for this purpose is a ‘mathematical model’, ie. a set of formulae relating external loads to internal stresses and deformations of the structure. Numerous mathematical models exist for different types of structures, load applications, and boundary conditions. They are based on theories developed from observation and experiments concerning, the beha- viour of materials, structural elements, and complete structures. *The term ‘linear’ behaviour is used in this book to describe structures working within the elastic limit and subjected to loading below critical buckling loads. 18Introduction During the analysis stage of a ‘linear’ structure, the engineer carries out standard calculations for the sizing of the main load carrying elements. This may take several iterations, before correct sizes are obtained. As already mentioned, during the design process, the shape of the structure is rarely questioned — unless it threatens the structure's suitability for a given pur- pose. The architect remains responsible for the overall structural form, even though he/she may have little comprehension of the complexities of the mathematical modelling involved. It is the responsibility of an engineer to make the structure safe. Therefore, the two roles — that of an architect, who sees the structure purely in terms of geometry and purpose, and that of an engineer, who understands it in terms of load paths and deformations — can be quite separate. In contrast, the design of lightweight tension structures requires full inte- gration of both engineering and architectural skills. This is dictated by the need to understand the interaction between the shape of the structure and load patterns that may develop. Very large deflections (of up to 1 m), can take place during the life span of the structure. The resulting changes of the overall geometry, known as ‘geometrically non-linear’ behaviour, cannot be ignored in the analysis. The governing non-linear equations, describing the shape of the membrane under pre-stress and, subsequently, the behaviour under the applied loads, cannot be solved using conventional methods, applicable to ‘linear’ structures. Thus, the stress/displacement analysis has to be carried out using iterative numerical procedures. 1.4.1 Common misconceptions relating to tension membrane design One of the misconceptions concerning tension membrane structures is that they are negatively characterised as ‘cheap’. This belief may stem from the perception that fabric membranes are not as durable, or robust as conven- tional structural forms made from concrete, steel or masonry. In view of this, it is necessary to continue to improve design processes of tension structures and enhance the quality of the end product. With the growing expertise in the field of design and construction, and new, more durable materials appearing on the market, the competitiveness of tension membranes against conventional roofing forms will continue to grow. It is generally not appreciated that the initial shape and the final config- uration of a tension membrane under loading are not known. This leads to another common misconception that ‘the form of a fabric structure can fit any architectural vision’. A good example illustrating this point is the Millennium Dome in Greenwich, designed as a part of a sphere made up of tensioned fabric cladding and cables. As stated earlier, tension membranes naturally adopt a ‘saddle’, or anticlastic shape (unless their boundaries lie in one plane). Departing from this form and choosing a spherical shape for the 19Tension Structures. Form and behaviour Millennium Dome has led to problems associated with local saddling of the fabric between the membrane supports, presenting a danger of ‘ponding’, i.e. accumulation of snow and water on the surface of the membrane.''?"! Steps taken to remedy the situation have led to a complex design, with the pro- liferation of stays and their supports to enable the fabric to approximate the spherical curvature. The issue of choice of the structural form of the Millennium Dome is discussed further in Chapter 2, Section 2.3.4. 1.5 Closing remarks Research carried out by the European Consortium''~! showed that at the time of the Millennium, approximately 4-5 million square metres of land in Europe were covered by membrane structures. This figure compares with 1 million square metres in 1988.'"?"! The growth of tension membrane struc- tures indicates that they compete with the conventional roofing forms, despite complexities involved in their design. Efforts are being made to simplify the design process and this may encourage the growth of temporary and semi-permanent membrane structures, with a variety of applications: from light canopies providing shelter from harmful UV radiation, to protect- ing animals in exposed areas against the elements. Past experience demonstrates that tension structures can provide excellent reliability and service, if they are properly designed, operated and main- tained. For this to be possible, it is crucial to develop a good understanding of the principles of optimal structural forms of tension structures. A more detailed discussion of the subject is presented in Chapter 2. References [1.1] Forster, B. (1994). Cable and membrane roofs —a_ historical survey. Structural Engineering Review, 6, No. 3-4, 145-174. (1.2). Leonniaror, F. and Scutaicn, J. (1972). Structural design of roofs over the sports arenas for the 1972 Olympic Games: Some problems of prestressed cable net structures. The Structural Engineer, 50, No. 3, 113-119. (1.3). Waker, D. and Aopis, B. (1997). Happold. The Confidence to Build. Happold Trust Publications Limited. (1.41. Catalogue (1995). The Work of SL. Sonderkonstruktionen und Leichbau, Rohrerstrasse 140, D-70771 Leinfelden-Oberaichen. Germany. (1.5]. Lippeut, W. I. (1998). Structural fabrics for foils. 1st Oleg Kerensky Memorial Conference. I. Struct. E., June, 13/5-17/5. [1.6]. Harpo. et al. (1987). The design and construction of the Dyplomatic Club, Riyadh. The Structural Engineer, 654, No. 1, 15-26. 1.7]. Discussion. (1987). The design and construction of the Dyplomatic Club, Riyadh. The Structural Engineer, 65A, No. 10, pp 377-382 [1.8]. Patterns 5. (1995). Buro Happold Consulting Engineers. 20Introduction [1.9]. Lippett, W. L (1994). Minnesota Metrodrome. A study on the behaviour of air supported roofs under environmental loads. Structural Engineering Review, 6, No 3-4, 215-235. 11.10]. ScuLaicu, J., BERGEMANN, R. and SoseK, W. (1994). The air-inflated roof over the Roman amphitheatre. Structural Engineering Review, 6, No 3-4, 203-214. [1.11]. Lippet, W. I. (1993) Air-supported structures. The Structural Engineer, 71, No. 14, 247-249. {1.12}. BucriioLor, H. A. (1985). Introduction to Cable Roof Structures. Cambridge University Press. [1.13]. Capasso, A., Majowrrcki, M. and Pinto, V. (1993). Le Tensostrutture a Membrana per L'architettura. Maggioli Editore. [1.14]. Majowigcki, M, (1994), Tensostrutture. Progetto E Verifica. CREA. 11.15]. Hupson, P. and Torrine, B. H. V. (1991). The design and construction of reinforced concrete tents in the Middle East. The Structural Engineer, 69, No. 22, 379-385. [1.16]. Viwnay, O. (1990). Cable Nets and Tensegric Shells. Ellis Horwood. [1.17]. Pfeifer catalogue. Seilbauwerke. Lightweight Cable Structures. Pfeifer Seil- und Hebetechnik GmbH & Co., Mammostrasse 1, P.O. Box 1754, D-8940 Memmingen, Germany. [1.18]. Gumsine, N. J. (1983). Cable Supported Bridges. John Wiley & Sons. [1.19]. Apois, B. (1994). The Art of the Structural Engineer. Artemis. {1.20}. Kusix, L. A. (2002). Recent developments in CUBIC structures. Space Structures 5. Thomas Telford, 1, 347-354. [1.21]. Barnes, M. and Dickson, M. (2000). Widespan Roof Structures. Thomas Telford. [1.22]. Final Report to the European Commission (2000). Framework IIL CRAFT Project. Contract BRST.CT98-5166. Proposal BES2-2498: Design and Manufacture of Minimum Energy Forms of Lightweight Tension Membranes (DELITE). [1.23]. Berns, S. (1988). Architectural fabric membrane roofs. Construction Specifier, 41, No. 11, 112-120.] Form-finding 2.1 General concepts. Nature’s ‘secrets’ Cable structures and fabric membranes adopt unique shapes under tension. Such shapes, or forms, are not known a priori; they cannot usually be described by simple mathematical functions, and, therefore, require the designer to get involved in a process known as ‘form-finding’. This require ment distinguishes tension structures from other structural types, such as those made of concrete, steel (other than cable), or masonry, where it is possible to dictate the shape at the outset of the design. In general, the process of form-finding should yield optimal structural shapes: shapes that would satisfy the functional requirements and atten- dant durability and strength at a minimum cost. As noted at the outset of Chapter 1, one of the most attractive features of tension structures is their ability to offer aesthetically pleasing, even dramatic effects. However, the aesthetic aspect of any design is often the most disputed one. Research carried out at the Institute of Lightweight Structures (IL) at the University of Stuttgart during 1964-1991, under the direction of Frei Otto, focused on structural forms, which obey a ‘lightweight principle’; that is, use a minimum amount of material while maximizing, through the shape optimization process, their stability and strength. This process consisted of predominantly experimental work on small-scale models, carried out by interdisciplinary teams of architects, biologists and engineers. They engaged in an iterative process of improvement and selection, in which aspects such as symmetry, perspective, and proportion were all considered, in addition to the optimization criteria and the structure’s function. The results were quite surprising — even though the research teams did not expect to create perfect objects, aesthetic aspects were inherently present in them.” They emerged naturally, in, what was considered a ‘self optimi- zation’ process. It was concluded that the path to achieving an aesthetic 23Tension Structures. Form and behaviour product is never straightforward and that the ‘lightweight principle’ has its roots in nature. The topic of natural structures has interested biologists, mathematicians, and architects for many years. D’Arcy Thomson, in his book On Growth and Form?! examined a variety of natural objects and discussed principles behind their formation. Shells, for example, grow in size, but do not change their shape, maintaining constant similarity during growth. Their shape can be described by equiangular, or logarithmic spiral, described by: @=klogr (21) where, with reference to Fig. 2.1, it can be seen that sectors cut off by suc- cessive radii, at equal vectorial angles grow continuously without changing, the shape. The vector angles about the pole are proportional to the logarithm of the successive radii. Interestingly, the logarithmic spiral is also a shape observed in natural objects, such as tiger claws. This allows the external loads to be transmitted axially. Research carried out by Mattheck,?* into the mechanics of tree growth, also stresses the importance of natural laws for the design of man-made structures. Trees are highly optimized objects, characterized by a minimum weight, with sufficient strength to carry all relevant load cases. The mechan- ism of tree growth ensures a constant stress distribution on the surface of the stem and branches of the tree. The principle of homogeneous surface stress distribution, which exists naturally in trees, has been applied to the design and optimization of engineering components subjected to fatigue stresses. In this case, the process of optimization consists of allowing a mechanical com- ponent to ‘grow’ into a shape of constant surface stress." 25! Because of the presence of a three-dimensional stress field, a constant value of von Mises stress is used in the process of optimization of these ‘biological components’. The von Mises stress is calculated from the maximum-distortion-energy Figure 2.1 Logarithmic spiral 24Form-finding theory/**! which relates a three-dimensional stress field to a single value of stress, as in: a (1/v2) Ve = 0)? + (02 — 03) + (oy — 037 (2.2) where, 03, 03, and 3 are the local principal stresses. & is the equivalent uniaxial stress, which can be compared with the yield stress in the von Mises yield criterion. A value above the yield stress indicates ductile failure of the material. The von Mises stress equation implies that the initial yield surface expands uniformly and in proportion to the root mean square differ- ences of the principal stresses. Mattheck demonstrated that engineering components optimized according to the biological rule of constant surface stress are characterized by minimum weight. His work served as inspiration to a group of researchers in Australia who developed a method known as ESO — Evolutionary Structural Optimisation.’””! The method relies on removing inefficient (lightly stressed) material from a structure, in an iterative process involving finite element modelling. The basis of the material removal is the value of the von Mises stress relative to some assigned value It is clear that in the case of shape optimization of engineering com- ponents, the requirement of a constant surface stress cannot be fulfilled for all possible load cases. While living structures (trees) can compensate for changeable load conditions by adaptive growth, engineering. struc- tures are not able to do so. Consequently, decisions have to be made as to the most critical load combinations. The need to establish a design principle that would incorporate different load cases and load combina- tions in the process of shape optimization/shape finding, is an area of active research. The origin of Mattheck’s work lies in observation of trees and tree growth. When a tree gets damaged, ie. a branch gets broken off, the healing process minimizes the surface area over the wound. The two features described: a constant surface stress and a minimum surface area can be found in non-living objects, such as soap-films. Soap-films serve as examples of ideal lightweight tension membranes. They are laminae that adopt a configuration consistent with a (local) minimum potential energy arising from surface tension. Because the potential energy is a minimum, the configuration is stable and can be realized by a soap-film. Furthermore, such a configuration is characterized by a minimum sur- face area. In mathematical language, a soap-film is referred to as a stable minimal surface, The subject of the mathematics of minimal surfaces has been continually explored, since its first introduction in the eighteenth century, 7Tension Structures. Form and behaviour 2.2 Concept of a ‘minimal surface’ — historical background The earliest mathematical equation describing minimum area surfaces using the calculus of variations was derived by Euler (1744).'°*! It was reformu- lated later by Lagrange (1761)?! and since then became known as the Euler-Lagrange equation. It has proved very difficult to find closed forms of solutions to this non-linear second-order partial differential equation, par- ticularly for general boundary configurations containing sharp corners. However, mathematicians have discovered many properties of minimal surfaces, together with two well-known classes of solutions, which are the catenoids and the right helicoids (shown in Figs 2.2 and 2.3, respectively). The full derivation of the Euler-Lagrange equation, based on calculus of variations, is given in Appendix II. Work carried out independently by Young, (1805) and Laplace (1806) showed that for a curved fluid surface, the relationship between the excess pressure across the surface at any point and the radii of principal curvatures is given by!210 211, 212] iL gach =o(—+— 3 reo(t) es where p is the pressure difference between the inner and the outer surface, « is the surface tension, and 7; and rz are the maximum and minimum radii of principal curvature, that occur in planes perpendicular to each other and are both perpendicular to the tangent plane of the surface. In the case of soap-films, o is known to be constant, and if the excess pressure p is zero (i.e. the soap-film is not a soap bubble), then eqn (2.3) is reduced to 14 1,19 (24) non Figure 2.2 A catenoid. 26Form-finding Figure 2.3 A helicoid. This result connecting the principal radii of curvature of a minimal surface was discovered by Meusnier (1785) who used a purely mathematical argu- ment, which showed that the Euler-Lagrange equation could be re-expressed. in terms of the radii of curvature p; and p2 at any point on the minimum surface defined by any orthogonal pair of axes"! Thus, 1 PL Pe (2.5) Introducing the mean curvature H as the arithmetic average of the principal curvatures: pT aaa) eo it can be said that at each point, a surface of minimum area must have a mean curvature of zero. It should be noted that the feature of zero mean curvature is not unique to minimum area surfaces, such as soap-films. They are a feature of ‘minimal surfaces’, of which soap-films are a special case, because they are surfaces which are also stable and possess a minimum area. The term ‘minimal sur- face’, used since the earliest days of the subject, can be a source of confusion. The central point is that the Euler-Lagrange equation for the area functional (Appendix I), is a necessary condition to extremize the functional, ie, the resulting area may be a maximum, or a minimum, but it is always subject to the condition of zero mean curvature. Therefore, mathematicians use the aTension Structures. Form and behaviour term ‘minimal surface’ to describe surfaces of zero mean curvature.?" It would have been helpful, if the subject had been termed ‘extremal surfaces’ rather than ‘minimal surfaces’. However, to draw the distinction, when it is important to do so, mathematicians refer to surfaces of minimum area as stable minimal surfaces; and surfaces of maximum area as unstable minimal surfaces; a terminology that is drawn from the physics of soap-films. In sum- mary: every soap-film is a minimal surface, but not every minimal surface is a soap-film. In order to avoid confusion, the term ‘stable minimal surface’ rather than ‘minimal surface’ is used throughout this book. Many questions related to soap-films, including that of how many stable minimal surfaces can span a given contour, were addressed by a Belgian physicist Plateau (1873).?"5! By numerous experiments he demonstrated that every contour of a single closed curve bounds at least one soap-film However, the challenge to prove this phenomenon mathematically became so great that mathematicians recognized it as the ‘Plateau problem’. It was not until 1930 that this problem was significantly advanced." It was shown that for each simple closed curve there exists a surface of minimum area The role of soap-films (stable minimal surfaces) in form-finding of tension membrane structures is discussed in Sections 2.3.2 and 2.3.3. An overview of the subject can be found in Refs [2.17] and [2.18]. 2.3 Form-finding methodologies In general, the process of form-finding of surface stressed membranes involves: (@ the construction of small-scale physical models using soap-film, fabric, or paper, and/or (ii) the development of computational models using a variety of numer- ical techniques. 2.3.1 Physical models Physical models continue to be used at various stages of design, mainly as means of communication between the members of the design team and as verification of computational results. However, the limitations of physical modelling, noted in Section 1.2 of Chapter 1 and discussed below, are the main reasons for the increasing application of computational methods in preference to expensive hand-made models, particularly when numerous changes, or refinements of the surface geometry have to take place. The design of the German Federal Pavilion for the Montreal Expo’ 1967 was based on measurements of small-scale models made of soap-film and 28Form-finding fabric. The process required a sequence of models, refined with painstaking care and precision, to ensure that the measured geometry of the surfaces could be used for the static analysis phase and construction drawings. The prototype of the structure was built on campus at the University of Stuttgart and subsequently used to house IL— The Institute of Lightweight Structures. The structural design of roofs over the sports arenas for the 1972 Olympic Games in Munich (Chapter 1, Fig. 1.1) also involved the construction of hand-made models. A model of the Stadium was constructed to a scale of 1:125, but conventional measurements proved too inaccurate for the genera- tion of cutting patterns for the full-size structure.”"! The problem of accu- racy was solved at the experimental stage by taking photogrammetric measurements of the model, and smoothing the scatter in data numerically, to give what was believed to be a good estimate of the equilibrium config- uration. However, certain unforeseen problems occurred during the con- struction of the roofs. It was found that inaccurate pre-cutting of cables on site led to a variation of forces in the cable net of up to 60%, compared with the predicted values. A simple calculation showed that a systematic error of 0.75 mm in every mesh width of 750 mm would give a total error of up to 50 mm ina 50 m length cable. The longitudinal errors thus obtained had a direct effect on strains. As a result, a substantial redistribution of forces in the cables took place after the structure was installed. The design of the Munich Olympic complex stimulated further research into computational modelling of tension structures, which, as already indi- cated, has largely replaced extensive use of expensive physical modelling. However, computational modelling alone has not produced uniformly good results. Experience gained so far indicates that a combination of computa- tional and physical modelling is a far more promising strategy than exclusive use of either methodology. Although theoretical modelling has grown stea- dily in sophistication, its limitations need to be clearly understood. It is still vitally important that the design team possesses analytical and visualization skills necessary to produce structurally efficient and aesthetically pleasing solutions. Physical models convey a visual impression instantaneously, from any perspective, and more truthfully than a computer image. The examples of tension membranes given in Chapter 1, Fig. 1.8, are illustrations of the quality of design that can be achieved by a mixed approach, involving com- putational and physical models. 2.3.2 Computational models Computational models of tension structures represent a set of numerical and graphical data describing structure's shape, stresses and deformations 29Tension Structures. Form and behaviour under certain load conditions. The data are obtained from numerical algorithms describing an iterative process of geometric adjustment of a tensioned surface, until a static equilibrium (i.e. the equilibrium of all forces acting on the surface) is reached. As with all numerical methods, an inspired guess of ‘what the surface should look like’ is the starting point of this process. The ‘guessed forny is unlikely to satisfy the condi- tions of static equilibrium and hence iterations are needed to reach the final result. The term ‘form-finding’, used in relation to computational modelling, can mean one of the following: (i) finding an optimal shape of a tension membrane (e.g. a stable minimal surface) (ii) finding a shape of a tension membrane that is in static equilibrium, but does not necessarily have constant surface stress (iii) finding a shape, which approximates the state of full static equili- brium (common in some design methodologies combining form- finding and patterning). In the first case, the boundaries of the structure are imposed, and the numerical algorithm is corstructed in such a way as to allow the surface to adopt its own geometric configuration, according to the natural principle of constant surface stress. In the second case, the condition of constant stress is relaxed, as in the case of the force density method discussed in Chapter 3. The resulting structural form is in static equilibrium, but, as indicated above, the surface stress is not necessarily constant. Finally, case (iii) represents an attempt to shorten the design path by utilizing the same mesh, or at least some of it, for both form-finding and the cutting pattern development. In order to achieve that, mesh distortions during the form-finding process are controlled. This is car- ried out in various ways. One method, described in Chapter 7, Section 7.4.3.2, assumes that the positions of the seam lines in the x-y plane are known at the outset and vemain unchanged as a result of form-finding. Consequently, the movement of the nodes in the x-y plane during the actual process of form-finding is prevented. This methodology ensures equilibrium in the z-direction only. The shape of the resulting surface may approximate the final (fully equilibrated) surface quite closely, but, the smooth lines on the surfaces are not the lines of full static equili- brium, and therefore should not be used as seam lines to generate cutting patterns. A more detailed analysis of the problem of mesh control and its implication on the final design is given in a case study presented in Chapter 7, Section 7.5.1. 30Form-finding 2.3.3 ‘Soap-film debate’ Equation (2.3) can be exploited in form-finding of tension membrane struc- tures. A repetitive calculation of curvatures until the pressure difference across the surface is (almost) zero would render, ultimately, a stable minimal surface structure. Such a form-finding methodology is outlined in Chapter 7, Section 7.4.1. If a membrane structure has a shape of a stable minimal surface, i-e., it is characterized by a constant pre-tension in every direction, and its thickness and density are uniform, then the amount of material required to construct it is minimal. Such a structure can be referred to, in an engineering context, as obeying the ‘lightweight principle’ (concept introduced in the opening para- graph of section 2.1). Unfortunately, ‘soap-film analogy’, or the use of soap-films as analogues for computational form-finding of tension membranes (Fig. 2.4) is not uni- formly accepted. It is argued that stable minimal surfaces are ideal struc- tures, but for one load case only, ie. their own surface tension. This argument is extended further by stating that there is a need to control the structural form by means of differential pre-stressing, in order to limit the deflections of the structure arising from additional loads, such as snow and wind. At first sight such an argument may appear plausible, but it ignores Figure 2.4 A soap-film model — an analogue for a membrane canopy. Photograph cour- tesy of Institute of Lightweight Structures, (IL), Stuttgart. 31Tension Structures. Form and behaviour various options available io the designer. A stable minimal surface has its shape determined by the boundary configuration and the uniform loading applied around the boundary. Varying the boundary configuration and, possibly, the level of constant pre-stress in the surface, gives the designer a considerable degree of flexibility in meeting the specified load cases. In a number of actual designs, however, rather than repeating the process of form-finding for different 20undary configurations, and levels of pre-stress, an easier option of controlling the structural form, by uneven pre-tensioning, is chosen. Such a strategy, in general, is unwise, as explained below. To optimize the fatigue/creep characteristics of fabric, it is clearly desir- able to have both warp and weft fibres (Fig. 2.5(a)) in the same, or nearly the same state of stress, so that the consequences of randomly applied loading are essentially born equally by both sets of fibres and, as a result, their ‘ageing’ is comparable. If the initial surface, prior to the application of static and dynamic loads, is in a state of constant surface stress, the resulting form is a saddle shape, in which the curvatures are equal and opposite (Fig, 2.5(b)), and the ratio of tension in the hogging to sagging directions is equal to one. On application of external loading, fibres aligned in the sagging direction will undergo an increase in tension, while those in the hogging direction, a corresponding decrease. The result is an uneven distribution of stress, but this in itself does not, as is sometimes claimed, invalidate the argument for the design of constant surface stress structures (stable minimal surfaces), for the following reasons. Assuming, initially, that the fabric is made of an isotropic material,* but the shape of the membrane arising from the process of form-finding is such that the tension field in the surface is not constant, it follows that fibres which are tensioned more, i.e. have a greater stiffness, will attract more load than those with lower tension, as shown in Fig. 2.6(a). As noted earlier, if the warp fibres follow the sagging curvatures, and the weft fibres, the opposing hog- ging lines, the structure will resist externally applied loads (such as snow) by an increase in tension forces in the sagging, direction and a corresponding reduction in the hogging direction. This is true for both types of structures ie. those with a non-uniform and uniform initial stress fields, respectively. However, what is apparent from Fig. 2.6(a) is that, in the case of differential pre-stressing (i.e. when the ratio of tensions in the sagging to hogging direc- tions is not equal to one), the difference in tension forces produced in the two sets of fibres by the external loads can be much greater than in the case of the “Research into developing isotropic materials for membrane structures continues. Certain fabrics, such as the ‘précontraint technique’ developed by Serge Ferrari (B.P. 54-38352 La Tour-Du-Pin Cedex, France) are deemed to be isotropic, i.e. char- acterized by the same elongation in warp and weft directions. aeForm-finding a) , 8 a5 aE mi Ean @ oe ee ae OE oe ae aE aE @ 7a a Ea boa 2 ee | a a Ea Ea a i i = ‘ie a” a” war direction aa ‘Warp Figure 2.5 (a) Warp and weft directions in fabrics, (b) alignment of fibres in a surface. initial constant tension field. This argument is reinforced by a case study presented in Chapter 5, Sections 5.2.1.1-5.2.13. It should also be noted that imposed loads are temporary and, in their absence, the structure designed as a stable minimal surface remains in its optimum state of uniform tension. It is simply impossible to optimize a structure for every conceivable load case. Tension membranes that give 33Tension Structures. Form and behaviour Stress d Displacement Strain °) Figure 2.6 (a) Idealized load-displacement relationship for unequally stressed fibres (iso- tropic material), (b) stress-strain relationship for anisotropic material (warp and weft directions), (c) Ferari ‘Precontraint’ treatment of fabrics: crimping followed by coating rise to very uneven stresses in the fabric are predisposed to differential ageing” of fibres and earlier rupture than would have occurred in a lower, more equitable distribution of stress. The argument presented above becomes mote itivulved when anisotropic fabrics are considered. Generally, the weft direction is weaker than the ‘warp. Therefore, pre-tensioning both sets of fibres to the same level would mean that the stresses in the weft direction are of a higher proportion of the break- ing strength of these fibres compared to the warp. It is therefore necessary to set the pre-stress at an appropriate level. Biaxial tests carried out on samples of anisotropic material show that for the same value of pre-stress applied in warp and weft directions, different levels of straining in the respective direc- “It is recognized that fatigue stresses in organic fibres can generally be ignored, but the same is not necessarily true for glass, or polymeric fibres. In cases where high pre- stress is present (‘compacting’ the fibres together) and there are stress fluctuations coupled with the changes of the angle of the weave, fatigue damage could occur. 34Form-finding tions occur (Fig. 2.6(b)). In fact, at a certain low value of pre-stress, the warp yarns, being initially straight, undergo crimping, which results in their rela- tive contraction, rather than expansion, as measured on plan; a phenomenon known as ‘crimp interchange’. The problem of crimp interchange is more pronounced in fabrics subjected to differential pre-stressing. The difference in the level of straining between the warp and weft fibres under the initial pre- stress has to be taken into account at a patterning stage, where appropriate compensation factors, expressed as percentages of unstrained lengths of material in the two respective directions, are applied. In order to exercise a maximum degree of control over patterning problems, the effects of crimp interchange should be minimized as far as possible. This, to some extent, has been addressed by Ferrari who developed special treatments for their fabrics. The treatments amount to pre-stressing the fibres so that their crimp in both directions is comparable, and subsequently applying plastic coating to both surfaces of the fabric, as shown in Fig. 2.6 (c). As a result, the same, or nearly the same stiffness in both warp and weft fibres is achieved, which helps to maintain a fairly uniform distribution of stress during the life-span of the structure. However, another practical aspect that needs to be considered is that in the case of an inappropriate cutting pattern, coated fabrics being stiffer, may not be as ‘forgiving’ as, for example, canvas. It should be noted that, in general, patterning cannot generate surfaces of exact constant tension, because of material properties, and, to some extent, the weight of the fabric. The latter is known to be small (estimated to give a change in the initial tension field of around 1%). The geometry of the structure is not noticeably affected by the weight of the fabric, unless the membrane surface is flat and/or has unsuitable pretension levels (the typically used pre-stress values are: 1.5 kN/m for PVC/ polyester, and 2.5-4.5 KN/m for PTFE fabrics). Theoretically, the weight of the fabric could be modelled numerically by assigning elastic properties to the form-found shape (with its initial tension) and loading the structure with just the weight of the fabric, in order to predict the initial shape and tension field ‘precisely’. However, such an exercise may prove futile, as, ultimately the effect may be swamped by much greater uncertainties in the material properties of the fabric. Fabric weight is usually included at the static analysis stage, together with other imposed loads. Despite all these limitations, patterning can produce good results, demonstrated by low variations in the initial tension field of the structure. This is possible only if form-finding is carried out to achieve a stable minimal surface form. 2.3.4 Form-finding, or form-dictating? Many designers are reluctant to follow the soap-film analogy in form-finding of tension membrane structures. Their judgement is affected by some of the 35Tension Structures. Form and behaviour practical difficulties outlined above, concerns over the control of deflections, and sometimes by the failure to appreciate the form-finding aspect that would lead to optimal shapes of tension structures. As stated earlier, a membrane tensioned between non-planar boundaries adopts a ‘saddle’, or anticlastic shape. The Millennium Dome represents a deliberate departure from such a form, in order to fulfil an architectural concept. Further to the comments made in Chapter 1, Sections 1 and 1.4.1, it should be noted here that in addition to the numerous masts and ties used to maintain the required shape of the Dome, radial cables had to be pre- tensioned to about 2/3 of the peak tension to ensure an adequate level of stiffness.?° Saddling could not be eliminated altogether and it remains to be seen how the Dome will behave under snow loading, It is estimated that 700 km of cable was used in the structure (compared to 100 km in the Munich Olympic Stadium) Dictating the shape of a tension structure beyond what is known to be its optimal form, i.e. an anticlastic shape, may have unforeseen consequences and, long term, may prove to be futile; over a period of time, the structure will creep towards its minimum energy form, in an attempt to equalize stresses. In the case of the Millennium Dome, the minimum energy form towards which the structure will have a tendency to creep, is a flat surface (a special case of a stable minimal form) The feature of constant surface stress present in many natural objects, such as soap-films or trees, arises from the process of minimization of the potential energy of surface tension. Designs which aim to implement (as far as is practically possible) this biologically inspired ‘lightweight principle’ produce stable, durable, and aesthetically pleasing structural forms (Fig. 1.8), which use a minimum amount of material. Such characteristics are the key metrics of sustainable development. It is recognized that the architecture of the minimal is one of the future directions, not the only direction. Form-finding is the subject that needs to be studied in a context broader than just the efficiency of the structural form, or its visual impact; it is possible to create structures, which are simply inter- esting, dramatic looking sculptures. Good architecture is more important than beautiful architecture. Beautiful architecture is not necessarily good.2?"! References [2.1]. Orro, F. (1979). The aesthetic. IL 21,141-190. [2.2]. D'Arcy THomson, W. (1917). On Growth and Form. Cambridge University Press. 36Form-finding [23]. Marturck, C. (1990). Why they grow, how they grow: the mechanics of trees’. Arboricultural Journal, 14, 1-17. (2.4). Martusck, C. and Mouvennauer, H. (1990). An intelligent CAD method based on biological growth. Fatigue Fractures of Engineering Materials and Structures, 13, No. 1, 41-51. [2.5]. Matrnec, C. and Burkarot, S. (1990). A new method of structural shape optimisation based on biological growth. International Journal of Fatigue, 12, No. 1, pp. 185-190. [2.6]. Bupynas, R. G. (1977). Advanced Strength and Stress Analysis. McGraw-Hill Book Company, 1977. [2.7]. Xie, Y. M. and Stevens, G. P. (1997). Evolutionary Structural Optimisation. Springer, London. [2.8]. Ever, L. (1952). Opera omnia. Orell Fiissli, 24, 25. 12.9]. Lacrance, J. L. (1889). Mécanique Analytique. 4th edn. Gauthier-Villars. [2.10]. Isenserc, C. (1978). The Science of Soap-films and Soap-bubbles. Tieto Ltd. [2-11]. Lartace, P. S. (1839). Mécanique Céleste. Impr. Imperiale, supplement to 10th Book, 1806. See also Laplace P. S., Oewores. In English, annotated translation by N. Bowditch, Little and Brown, IV. [2.12]. Youne, T. (1807). A course of lectures on natural philosophy and the mechanical arts. J. Johnson, I, 468-469. [2.13]. Meusnier, J. B. M. C. (1776). Memoire sur la courbure des surfaces. Mén Mathém. Phys. Acad. Sc. Paris, prés. par du Savans. Verlesen, 10, 477-510 (2.14). Hivpesranot, $. and Trowaa, A. (1984). Mathematics and Optimal Form. Scientific American Library. [2.15]. Patra, J. A. F. (1873). Statique expérimentale et théorique des liquides sommis aux seules forces moléculaires. Gauthier-Villars, Trubner et Cie, F. Clerune, Vols 1 and 2. [2.16]. Ravo, T. (1933). On the problem of Plateau. Ergebnisse der Mathematik und ihrer Grenzgebiete. Springer-Verlag, Berlin. [2.17]. Lewis, W. J. (1999). Lightweight tension structures — an aesthetic integra- tion of geometry and mechanics. Part | — The role of minimal surfaces and soap- films. Mathematics Today (the IMA Bulletin), 35, No. 1, 10-16. (2.18). Lewis, W. J. (1999). Lightweight tension structures — an aesthetic integra- tion of geometry and mechanics. Part Il — Finding the form of minimal surface. Mathematics Today (the IMA Bulletin), 35, No. 3, 80-84 [2.19]. Leontiarot, F. and Scuiaicn, J. (1972). Structural design of roofs over the sports arenas for the 1972 Olympic Games: Some problems of prestressed cable net structures. The Structural Engineer, 50, No. 3, 113-119. [2.20]. Barnes, M. and Dickson, M. (2000). Widespan Roof Structures. Thomas Telford. [2.21]. Orro, F. and Rasch, B. (1995). Finding Form. Towards an Architecture of the Minimal. Axel Menges. Bid3 Geometrically non-linear behaviour. Solutions by commonly used numerical methods 3.1 Geometric non-linearity As stated earlier (Chapter 1, Section 1.4), tension structures exhibit geome- trically non-linear behaviour: ie., even when working within the elastic limit, their deflections can be so large that the resulting changes of the overall geometry must be taken into account. Geometrically non-linear behaviour is not unique to tension structures, and, in order to understand the problem, it may be helpful to consider a simple structure, which is not pre-tensioned, but which can exhibit large displacements under applied loads Example. Two-bar structure The structure, shown in Fig. 3.1, consists of two linear elastic bars pin-jointed at the centre and at the supports, and separated by a distance of 2b. The initial, unstrained length of the bars is Lj, and the current length is given by Liv €r Where ey, is the extension of each bar. Vertical equilibrium gives P= 2Ke,, sin 6’ (3.1) Vy Ent are where K is the elastic stiffness of the bar, equal to EA/L;, i.e. the product of Young’s modulus E and cross-sectional area A, divided by the element length L;,, and 6’ is the angle the bars make with the horizontal, after defor- mation. 39Tension Structures. Form and behaviour Figure 3.1 Two-element structure. Vertical load. Further manipulation of eqn (3.1) gives P =2Ke,, ¥ 2K (8+) Sich Ge (3.2) Li8 +h) y@+ hy +02 ig un is = 2K(1 - horn =2K/5+h which shows that the relationship between the load P and the displacement 5 is non-linear. It is now possible to show under what conditions the load-displacement relationship becomes linear. Further manipulation of the bracketed term gives d4h- A tD. VE + 25h +I + (3.3) , a-12 =54n— ln + 0), 25h + 8 Vitae | + +E Using the binomial expansion of the term in square brackets, and noting, that (Li)? =I? + 0°, the right-hand side of eqn (3.3) becomes 1 Any wo+h-@+h[1- (26h+35) +...) a =—s CP 40Geometrically non-linear behaviour. Solutions by commonly used numerical methods to the first order in 5. The substitution of expression (3.4) in eqn (3.2) gives P P=2K——s8 (3.5) (Lin) which describes a linear load-displacement relationship, since h? /(Lj,)° is a constant). The example illustrates the point that, for large displacements, the load- displacement relationship is non-linear, while for small displacements (suffi- ciently small to allow the expansion (eqn 3.4) to be truncated after the first- order term) it is linear. The geometrically non-linear behaviour of structures can be solved by a variety of numerical methods and techniques. The most common computa- tional approaches are discussed in the next section. 3.2 Commonly used computational methods for the analysis of geometrically non-linear behaviour The most common, but distinctly different, numerical methods for the solution of geometrically non-linear response of structures are (i) the transient stiffness method (ii) the force density method (iii) the dynamic relaxation method. In addition, there are non-linear gradient vector methodologies developed by Buchholdt*" 7! for application to pre-stressed cable networks, which are based on the minimization of the total potential energy of the system. These include the method of steepest descent and the conjugate gradients method. It has been found that these methods were less powerful than the Newton- Raphson technique. However, the conjugate gradients method is currently being used in conjunction with the force density method discussed in Section 34. A description of each of the three methods listed above, together with a brief evaluation of their advantages and drawbacks, is given in the following sections. 3.3 Transient stiffness method The method evolved from the conventional, small displacement theory, which assumes a linear dependence of deflections upon forces in the structure. The methodology is traced with the help of examples presented below. 41Tension Structures. Form and behaviour 3.3.1. Static analysis of skeletal, ‘linear’ structures In the simplest case of a structure consisting of one element, as shown in Big. 3.2(a), the load-displacement relationship can be described by a scalar equation: Ki=P (3.6) a) ca reo — y 2 Element of cross-sectional area A Figure 3.2 (a) One-element structure, (b) two-element structure. Generalized load. 42Geometrically non-linear behaviour. Solutions by commonly used numerical methods where K is the axial stiffness, given as a force per unit length of the element, and 4 happens to be equal to the elongation of the element. For an elastic elongation, the stiffness K is equal to the elastic stiffness K- as in eqn (3.1). The displacement of the structure is observed in local and global systems of axes, which, in this case, can coincide.* With the stiffness K and applied load P known, the displacement of the structure is uniquely defined. For a slightly more complex structure, consisting of, say, two elements, as shown in Fig. 3.2(b), the same load-displacement relationship may be true, provided the displacements are small, i.e. when the difference between the rotations @ and 6” is insignificant. However, even for this relatively simple structure, it is no longer convenient, or indeed practical, to measure the displacements along the lines of each element. Therefore, in addition to the local axes, a global co-ordinate system is set up to measure the deflection 5 in terms of its x°, y° and z° components in the Cartesian co-ordinate system. These components are typically referred to as u, v, and w displacements, respectively, which constitute the three ‘degrees of freedom.’ (In the example under consideration the » component happens to be zero). Consequently, everything is measured in terms of the x°, y° and 2° components, becomes a vector of nodal displacements {6}, as in {a)=| 0 (3.7) w the axial stiffness of each member is entered into a 3 x 3 global stiffness matrix [K], as in 44,412, 13 UK] = | 21,420, 423 (3.8) 51,432, 433 and the external load is given as the load vector {P}, expressed in terms of its x°, y°, and z° components at each node of the structure: Pye (P} = | Py (3.9) Pro “The global system of axes, denoted by x°, y°, 2°, and its orientation, are usually independent of a structure’s geometry. The orientation of the local x, y, z system varies, since one of the local axes is usually aligned with the direction of each struc- tural member. 43Tension Structures. Form and behaviour Thus, for a more complex structure, the global stiffness matrix [K] is used to relate the column vector of nodal displacements {6} to the external load vector {P}, as in [K]{6} = (P} (3.10) Equation (3.10) represents a set of static equilibrium equations which has to be satisfied at every node of the structure. The ‘nodes’ are certain character- istic points of the structure, such as points of intersection of structural ele- ments. For simplicity, the suffices indicating the node number (suffix j), and the direction according to the Cartesian co-ordinates (suffix i), are omitted from the formulation. The left-hand side of eqn (3.10) represents the internal force vector, which has to balance the externally applied load vector {P}. The coefficients ay;, 42... @in of the global stiffness matrix [K] carry the dimension of force per unit length, and can be determined from experimental measurement, but it is more common to calculate them using standard struc- tural theories and algebraic manipulations. Such manipulations involve tri- gonometric transformations of the local co-ordinate system into a global system, using a transformation matrix constructed from direction cosines. Once the transformations are completed, the displacements and internal forces in the structure can be expressed in terms of their global x°, y®, and 2° components, rather than in terms of the local extensions and forces. The transformations are simple but tedious, and not recommended for hand calculation, particularly for structures with degrees of freedom greater than 3. The individual stiffness matrices for each member contribute to the forma- tion of the global stiffness matrix. Where the members share a common node, their stiffness is entered as a sum of the relevant coefficients in the global stiffness matrix. The size of the stiffness matrix is determined by the number of unknown nodal displacements. In the example illustrated in Fig. 3.2(b), the displacement v and the y® component of the load vector are both equal to zero, and hence [K] can be reduced to a 2 x 2 stiffness matrix. The analytical procedure described above is known as the ‘stiffness matrix method’. It is a method that features very strongly in modern structural analysis. In essence, the method provides global stiffness matrix equations, which describe the load-displacement relationship for each node of the structure. The main limitation of the stiffness matrix method is its inherent assump- tion that the relationship between forces and displacements is linear, i.e. that the structure displaces in proportion to the loads applied. With the initial geometry of the structure known and the applied loads estimated, the static analysis process is reduced to finding the unknown small displacements and deformations. The internal forces in the structure are found from the stress~ strain relationship, i.e. Hooke’s law for the material. The analysis of such 44Geometrically non-linear behaviour. Solutions by commonly used numerical methods ‘inear’ structures does not involve form-finding —the shape of the structure is imposed at the outset and is assumed to change very little under the applied loading 3.3.2 Static analysis of skeletal, ‘non-linear’ structures — why iterative computations are necessary The condition of static equilibrium described by eqn (3.10) cannot be directly applied to structures which exhibit large displacements (a phenomenon referred to as geometrically non-linear behaviour). With reference to the example in Fig. 3.2(b), the problem of geometric non-linearity arises when the rotations # and @’ are significantly different. In order to be able to use eqn (3.10) for the case of geometric non-linearity, it is necessary to proceed with the solution in a stepwise manner, as explained below. If the deflections of a tension structure were to be calculated from eqn (3.10) using just the initial stiffness matrix [K] as (6) = [K""(P) (3.11) then the condition of static equilibrium would clearly be violated, because the displacement (6), being large, would alter the initial configuration of the structure — the configuration for which the stiffness matrix [K] was origin- ally calculated. Therefore, a new updated stiffness matrix would have to be found, based on the new geometry. This situation triggers off the process of correction of displacements and re-calculation of the stiffness matrix, as explained below. If k denotes an iterative step corresponding to the current geometry of the structure, {X};, on which the calculation of the stiffness matrix (K}; is based, then the displacement vector [6] is: (8}41= IK IP} (3.12) and the new geometry is (Xe = (Xb thea (3.13) If the displacement is large, the new stiffness matrix [K],,, multiplied by the displacement vector found from eqn (3.12) would generate an internal load vector, {P}, which does not balance (P}. Thus Uhr Bhs (Phir (3.14) The difference between the external load vector (P} and the internal vector {P}, represents the out-of-balance force vector, or ‘residual’ {R}, as in (Rhr= [0P] = (Pha (3.15) 45Tension Structures. Form and vehaviour The residual force vector yields a correction, or increment for the displace- ment vector {6}, say {Ad}, as in (Ab )cu1= DKA Reg (3.16) This increment in the displacement vector is subsequently used to update the geometry of the structure. Subsequently, a new stiffness matrix is calculated, which, together with new residuals will generate a new increment of dis- placements. This then sets up an iterative process of calculations, which continues until the residuals are reduced to (almost) zero, ie. the static equilibrium is reached. Equation (3.14) represents a ‘transient form’ of the static equilibrium equa- tions. The numerical procedure presented above is known as the ‘transient stiffness method’ Accordingly, the stiffness matrix [K], is referred to as a transient stiffness matrix or instantaneous stiffness matrix. Although the numer- ical procedure is formulated in terms of the stiffness matrix changing with each iteration, it has been found that the convergence of the numerical solu- tion is improved by keeping [K] constant for a small number of consecutive (iterative) steps. The transient stiffness method relies on a linearization process of the non- linear system of equilibrium equations and is critically dependent on the assumption of small displacements. 3.3.3 Transient stiffness method applied to static analysis of tension structures. Role of geometric stiffness In the case of tension structures, the assumption of linear behaviour may be valid, if the pre-stress forces are high and the applied loads are relatively small. In general, such a relationship does not hold, and the non-linearity of the force-displacement relationship, as evidenced by the dependence of the stiffness matrix upon the displacement, requires iterative calculations In order to model the behaviour of tensioned structures, the presence of pre-stressing forces must be taken into account. This is achieved by means of the geometric stiffness. The role of the geometric stiffness is best explained with the help of a simple example. Example A structure shown in Fig. 3.3 consists of two cables pre-tensioned by the force Ty. Due to symmetry, only half of the structure is shown. Each cable has a pre-tensioned length L,,, and is subjected to the external load P, which produces a displacement ¢, as shown. If the original, unstrained length of each cable is L;, and the initial extension corresponding to the pretension force T,, is ¢;, then 46Geometrically non-linear behaviour. Solutions by commonly used numerical methods 8 Figure 3.3 Two-bar structure. Role of geometic stiffness. AE ee = (3. Tem ™ (3.17) ie. the product of the direct elastic stifiness EA/L;, and the extension of the cable e%,. When an external load P is applied, there is a further increment of member tension AT,, and a corresponding extension ¢;,, Le. a en tein) = Tn + AT n (3.18) and hence ‘ Affn SAT, (3.19) Due to symmetry, horizontal equilibrium is automatically satisfied and hence only vertical equilibrium of the structure needs to be considered; thus (Tm + Tm) , (3.20) where Ls, is the current member length, given as Lh, = (Lyy+€°m) + &, = Ly, + Gu Where L,, is the initial strained length. Substituting eqn (3.19) into (3.20) and re-arranging terms gives é as pa (es tete lo (e+ Kee )e (3.21) where K = T,,/L%, is defined as the geometric stiffness (explained below), and Ki. is the direct elastic stiffness. 47Tension Structures. Form and behaviour The geometric stiffness allows one to express a change in the nodal force components, arising from prestress and the altered geometry of the structure given by the displacement 5. This change can be significant, even if very little elastic straining took place. This is purely a geometric effect. The structure shown in Fig. 3.3 is very simple, and together with eqn (3.21) serves the purpose of illustrating the nature of the force-displacement rela- tionship in pre-stressed structures. Due to the simplicity of the case consid- ered, in which there is only one unknown displacement, both the geometric and elastic stiffness can be represented in scalar form and calculated directly from member properties. The application of the transient stiffness method to multi-dimensional systems requires the force-displacement relationship to be represented in the form of a global stiffness matrix comprising both the elastic and geometric effects. A detailed exposition of the elastic and geometric stiffness of cable (line) elements used in discretization of three- dimensional surfaces, as originally proposed by Argyris,°*! is presented in Section 7.3.1 of Chapter 7. The transient stiffness method discussed so far relates to the static analysis problems of tensioned cable nets/membranes and rigid structures exhibiting geometrically non-linear behaviour. The method can also be applied to the form-finding of tension structures, as shown in the next section. 3.3.4 Application of the transient stiffness method to form-finding of tension structures At the outset of the computational process of form-finding of tension struc- tures, a crude initial structural shape is assumed, together with the values of pre-stress forces in the surface. Initially, the structure is not in static equili- brium, because the assumed shape is unlikely to be compatible with the assumed forces in the structure. A simple cable structure, described below, can illustrate the problem. Example A structure consisting of four cables with a pre-tensioned force T,, in each cable is shown in Fig. 3.4. The initial ‘guessed’ shape of the structure does not produce static equilibrium. Consequently, the correct shape has to be deter- mined through finding the unknown co-ordinates of the internal node 5. As a first step, the initial global stiffness matrix is calculated from the assumed geometry. Subsequently, the tension force in each member is resolved into their x°, y°, and z° components at the common node. The sum of the components of the member forces produces the internal load vector {P}. With reference to Fig. 3.4, it can be seen that the contribution of any one member sharing a common node to the internal load vector is 48Geometrically non-linear behaviour. Solutions by commonly used numerical methods Yar a) b) Figure 3.4 A pre-tensioned cable structure: (a) General arrangement and (b) orientation of members in global co-ordinates. Ty, COSGy;, Where the suffix i= 1, 2, and 3, corresponds to the x°, y®, and 2° global directions, respectively. Hence, 4 DH Tn C08 On . [ee P, re in=[t]-[B} & Tn 608 Ona (3.22) Be Pp mi ET 608 Ong For static equilibrium to be satisfied, the right-hand side of eqn (3.22) must be zero. Since no external load acts directly on node 5, the internal load vector (P) constitutes the out-of-balance, or residual force vector (R} = (P). Once this residual is defined, it is possible to proceed with the iterative compu- tation, as in the case of static analysis. The residual force vector enables the generation of incremental displacements {Ad} according to eqn (3.16). Subsequently, the surface geometry can be modified, leading to a new stiff- ness matrix for the structure. In summary, the iterative procedure for the form-finding, process consists of the following steps: (i) assumption of the initial geometry (ii) calculation of the stiffness matrix (based on the assumed initial geometry, but subsequently calculated from updated geometry) (iii) calculation of the residual force vector as (Rha = {Pheu (iv) calculation of the incremental displacement vector {A6) from eqn (3.16) (v) correction of surface geometry due to {Ad} (vi) calculation of the new stiffness matrix and returning to step (iii). 49Tension Structures. Form and vehaviour In form-finding of stable minimal surfaces represented by line elements, the pre-tension forces T,, remain constant. The process of form-finding consists of an iterative adjustment of the initially assumed surface geome- try, until the value of the out-of-balance force vector (residual) is reduced to (almost) zero. The residual is equal to the internal load vector {P} calcu- lated from the assumed tension field and current configuration of the structure. 3.3.5 Evaluation of the transient stiffness method The transient stiffness method (the instantaneous stiffness method) is described in the work of Argyris, Siev, Haug, and others. 34: 35.36.37, 381 It was developed to facilitate form-finding and patterning of Munich Olympic roofs'**! simultaneously with the force density method proposed by Linkwitz.°*) It is possible to draw an analogy between the method of transient stiffness matrix and the Newton-Raphson method 3) 37 30! for the solution of systems of non-linear equations. Indeed, the transient stiffness method is really a vector version (relating force to deflection) of the well- known scalar Newton-Raphson method. The finite element method and stiffness matrix approaches dominate mod- ern structural analysis. Their success can be attributed wholly to the advent of computers, as they are totally unsuited to hand calculations; even for small systems of equations, numerous matrix manipulations have to be performed to convert local element solutions to the global solution. For large systems, if matrix inversion is used, the solution may be prone to divergence, or yield useless results, due to the computational round-off errors. Round-off errors are magnified through lengthy arithmetic operations and possible ill-condi tioning of the stiffness matrix. Note, a matrix is said to be ill-conditioned if it contains some coefficients which are orders of magnitude greater (or smaller) than other coefficients. The problems likely to arise in the inversion of ill- conditioned matrices are well known. These include ‘swamping’ of the effects of small terms, or ‘loss of significance’ in the case of small differences between large numbers containing too few figures to maintain accuracy.°"1! Therefore, during the computation of the matrix inverse, round-off errors caused by arithmetic operations on very small and very large numbers can be disastrous. To mitigate round-off, techniques such as scaling of the stiff- ness matrix"! 3"! are recommended, but they all add to the computational effort. It is well-known? >"! that if the size of the matrix is nx 1, then the total number of arithmetic operations required for matrix inversion (and hence the computing time) is proportional to n°. It follows that for the large-scale calculation required for tension structures, matrix inversion should be avoided. Solution procedures which minimize the number of arithmetic operations should be employed instead. 50Geometrically non-linear behaviour. Solutions by commonty used numerical methods Further problems with the transient stiffness method can be encountered, because of the formulation of the element stiffness, which, as stated earlier, assumes small displacements and small rigid body rotations of the structure. In the case of form-finding, where gross changes of surface geometry may take place in the initial stages of computation, the condition of small displa- cements and rotations can only be fulfilled by scaling down the size of the iterative step. The accuracy of the solution is affected by a possible fictitious straining of elements, introduced by the fact that the direction cosines enter- ing the stiffness matrix formulation ‘lag one iteration’ behind. This problem is specific to the transient stiffness method and is examined in more detail in Chapter 7, Section 7.3. Thus, it should be realized that the numerical approaches relying on the formulation of the stiffness matrix as proposed in Ref [3.3], and subsequently reproduced in many textbooks/$" $10 3141 have this limitation. 3.4 Force density method The method of force density was developed in response to the need for computational modelling of structures for the Munich Olympic complex. The method uses a linear system of equations to model static equilibrium of a pre-tensioned cable net under prescribed force/length ratios. Earlier work due to Siev ef al." 5! also focused on the applicability of linear systems of equations to modelling equilibrium states of non-linear systems, such as pre-stressed cable nets. However, the solution to form-finding pro- blems proposed there was dependent on a number of simplifying assump- tions, such as a constant value of horizontal component of pre-tension force in each cable, and the cables lying in vertical parallel planes, with the two sets of planes perpendicular to each other. It was observed that with the pre- tension in each cable assumed constant, the system of equilibrium equations was linear with respect to the unknown displacements in the vertical direc- tion. The resulting shapes of pre-stressed networks were, of course, restricted by the assumptions made in order to acquire the solution. The force density method, originally proposed by Linkwitz and reported in references [3.9, 3.15-3.18] is less restrictive than that proposed by Siev. The method relies on an ingenious mathematical assumption that the ratio of tension force to length of each cable can be constant. This transforms a system of non-linear equations to a set of linear equations that, in principle, could be solved directly. However, due to the size of the matrix and its sparsity, the formulation does require the use of indirect, iterative methods of solution, such as the method of conjugate gradients.) At this point it is worth noting that the form-finding process modelled by the force density method does not equate with finding shapes of stable oeTension Structures. Form and behaviour minimal surface membranes; it simply allows one to generate shapes of tension structures that are in static equilibrium, The method, however, has the potential to be developed into a numerical tool for finding stable minimal surfaces, as discussed in Section 3.4.4 The principles of the force density method are illustrated with the help of the following example. 3.4.1 Application of the force density method The structure shown in Fig. 3.4 can be viewed as a branch of cables extracted from a larger network. The equation of equilibrium of forces at node 5 can be arrived at by resolving the tension forces T,, of the connecting members into the x°, y®, and z° global components. This is achieved simply by multiplying T,, by the direction cosines, defined as the ratios of the member lengths PFOjectioNS Lys, Lmy, and Ly,z to the actual lengths L,. Since each member m is defined by two end nodes i and k, the respective direction cosines are (3.23) In the current example, m = 1, 2,3, and 4, Hence, the equilibrium of forces at node 5 is given by the following system of equations: Ti@t 38) , T8 ~ 38) | TsQ5 = 28), Tal = 38) Ly Lb 1 Ly Tay = 5), Taly8 — ys) , Ta(yd — ys) + + <4 (3.24) Lp ts im 28 — 28) | To(24 — a where Py, Py, Py» are the components of the external load vector acting at node 5. In the above system of equations, the member lengths L,, (m = 1,...4) are non-linear functions of the co-ordinates of points (nodes), which define the members. The equations form the basis of the direct approach reported in reference [3.17], which subsequently led to the development of the force density method. By introducing constant values of tension coefficients, or force densities qn, defined as a ratio of member force to member length — namely qn, = T»:/Ly, — the system of equations becomes linear: 52Geometrically non-linear behaviour. Solutions by commonly used numerical methods quad = 38) + qolx8 — x8) + qa(ax§ — x8) + gard — 28) = Pro guy — ¥8) + aaly3 — y8) + 43(8 — Ys) + Ga(Ys — 5) = Pye (3.25) ule} — 28) + qa(2 — 28) + 93(23 — 25) + qal2a — 28 Equation (3.25) can be expanded to include the remaining branches of a larger structure, and take into account nodal connectivity, as well as known co-ordinates of the fixed (boundary) points."1°*"8! The resulting system of linear equations can be solved for the unknown co-ordinates of the network directly,” as illustrated by the next example. 3.4.2 Numerical example A simple structure is to be created from four pre-tensioned cables (Fig. 3.5). ‘The co-ordinates of the anchorage points of each cable are known, but the final configuration of the structure that satisfies static equilibrium is not. Therefore, the form-finding problem is reduced to the question of, where do the four cables meet? For simplicity, the co-ordinates of the anchorage points are given in Fig. 3.5 as non-dimensional integers. It is possible to find the unknown co-ordinates of joint 5 from the system of equations given by eqn (3.25), if values for force density are assumed. Equating the external loads Py, Py, and to P» zero, and re-arranging terms, gives qi +42 + 43+ a)¥8 + 41X1 + 4223 + 43x3 + uxt = 0 (3.26) and similarly for the remaining co-ordinates and y$ and 28. Assuming, for simplicity, a constant value of force density for all branches of the network to be equal to, say 1, leads to 4x8 -12+34+4+1)=0 4y3 -10+34+4+1)=0 (3.27) 428-15 +14+14+1)=0 which results in the co-ordinates of joint 5 to be 2.5, 2.25, and 2.0 units. It can be seen that the unknown co-ordinates of the structure, calculated by the force density method are simply a function of assumed force densities. In general, if the value of q,, increases or decreases by the same amount, no change in the co-ordinates of the structure would result. Hence, in order to achieve an alternative configuration of the network, different values of force densities have to be prescribed for each branch, or cable. This allows the generation of a range of three-dimensional shapes, giving the designer “The reader should verify that when this full set of equations is written down, there are always enough equations to solve for the unknown co-ordinates,Tension Structures. Form and behaviour Node 4 = (1, 1, 1) Figure 3.5 A simple form of cable structure. scope for improvement of the initial concept of the structural form. A typical value of force density used for all inner cables is one, and a value inversely proportional to the cable length is assigned for boundary cables, when such cables have irregular lengths.!°'® +"! It is clear that unequal lengths of cables (elements) give unequal forces in them, if a constant value of force density is used. Therefore, if a tension structure is to have a constant tension field, the co-ordinates calculated by the force density method would have to be such that the lengths of the branches remain equal. Hence, as noted earlier, the method cannot be used directly to find minimal surface membranes for which the condition of constant value of force, rather than force density, would have to be pre- served in the computation. 3.4.3 Matrix formulation of the force density method For the purpose of computer calculation, the system of equations given by eqn (3.25) needs to be represented in a matrix form. As a first step, one can observe that the projected lengths of members [px Lynye NG Lyy2 Which enter the eqn (3.23) can be expressed using the connectivity matrix [C] and the nodal co-ordinates, as in 54Geometrically non-linear behaviour. Solutions by commonty used numerical methods Figure 3.6 Network branch. -u{*} = 1c) F oye =[ -{% | =(CIY} (3.28) fz =U = : | =(CHZ} For the network of cables shown in Fig. 3.6, the connectivity matrix is con- structed in such a way as to clearly indicate which node numbers belong to which members. Thus, entering the node numbers i, k along the horizontal, and the member numbers along the vertical, gives connectivity matrix [C], shown in Fig. 3.7. m Figure 3.7 Connectivity matrix, (C]. @ aTension Structures. Form and behaviour Hence, with reference to Fig. 3.7, the projected lengths of members in the x direction can be expressed as ae eee cel xf x8 Ix {_]o 100 -1}fxe[_|e-2 ,,,rt~zr—S:sCdCsCssc 6.29) is 0001 -1f{x¥ e—8 and similarly in the y and z directions, giving the general form shown in eqn (3.28): (Linx) = [CHX} {Ling} = [CHY} (3.30) {Lz} = [CHZ} The x, y and z components of internal forces Quix, Qmys Qnz in cable members can be expressed as a product of force densities Q and projected member lengths, which in matrix notation give [Qn] = [Q]Lmed [Qhy] = [QLny) (3.31) [Que] = [QH Linz} where [Q] is a diagonal matrix of force densities a : (=| ° 2 A (3.32) % The equilibrium of forces at each node is given by the sum of the x, y and z components of internal forces Q,, in members connecting to the node, which must balance the external load vector {P} acting at the nodes of the structure. Hence Yn = Pre YX Qny = Pye (3.33) ae Substituting for Qwns, Qmyr Qnz from eqns (3.31) and (3.30), and noting that the summation in eqn (3.33) can be carried out by pre-multiplying the internal force components Qn, Qny, and Qn by ICI", leads to the matrix form of eqn (3.33) as follows: 56Geometrically non-linear behaviour. Solutions by commonly used numerical methods (CTQICHX) = Pv [CH TQICIY) = Py ey (CH LQUCKZ} = P-» One can verify this result by considering equilibrium in the x direction only, of the five nodes present in the cable network shown in Fig. 3.5: 1 0 0 0 Of 0 ot. ae Pre {cy OIC aq 0 7 6 Geet 0 00 1), 8 3% a i a a dG 35 Pao (3.35) The equilibrium equations in the remaining directions can be expanded in a similar manner. 3.44 Evaluation of the force density method The force density method can be used to generate, quite effectively, the initial shapes of cable nets and membranes. The method requires only the co-ordi- nates of the boundary points and values of the force densities to be specified, and research has shown that trivial values of force densities, such as 1, produce reasonable results. $171 The method can be developed to satisfy the condition of constant pre- stress and any other additional constraints, such as the preservation of rec- tangular or equidistant meshes in the structure. Such requirements have already led to the development of a non-linear force density method, which involves additional theoretical formulations" >"! and iterative pro- cedures. Nevertheless, even in its original formulation, as presented here, the method plays an important role in facilitating the design process of tension structures. 3.5 Dynamic relaxation method Unlike the previous two methods, the method of dynamic relaxation does not have to rely on matrix manipulations for the solution of the system of non-linear equilibrium equations. The very origins of the method can be traced to Sir Richard Southwell," who developed a relaxation method for the solution of partial differential equations via finite difference approx- imations. The method, intended for hand calculation, was, and still is, con- ceptually appealing and straightforward. Its development into the dynamic relaxation method is due to Otter,'>?"! who, in 1964, carried out an analytical 57Tension Structures. Form and tehaviour investigation of complex pre-stressed concrete nuclear vessels. A year later, the method was applied by Day’?! to the analysis of portal frames. Its first application to static analysis of cable networks is due to Day and Bunce!*?*! in 1970. Since then the method has gone through various stages of develop- ment and it is now widely recognized that its application to modelling of pre- stressed cable nets and membranes is very successful. For this reason, the method has been adopted in this book as the main numerical technique for the form-finding and static analysis of tension structures. The full description and evaluation of the dynamic relaxation method is given in Chapter 4. 3.6 Computational static analysis versus form-finding 3.6.1 Out-of-balance forces in static analysis and in form-finding In the case of static analysis, the structure is assumed to be initially in equili- brium, which means that the internal load vector (P} is zero. The out-of- balance forces arise as a result of the externally applied loads {P}, which disturb the equilibrium. The difference between the two types of load con- stitutes the out-of-balance force vector in the structure. Both the geometric and elastic stiffness have to be used to determine the internal forces, to reflect the initial pre-stressed state and subsequent elastic straining of the structure due to the externally applied loads. The final surface geometry is that for which {R} = {P} — (P} is zero. In the case of form-finding, the out-of-balance forces arise as a result of an assumed geometry, which is unlikely to be compatible with the force dis- tribution assumed for the structure. The force distribution represents a ten- sion field set at some pre-defined level, which should not be exceeded during the life-span of the structure. For all the nodes other than the support nodes, the out-of balance force vector is equal to the internal load vector {P}, calcu- lated from the tension field using current configuration of the surface. At the boundary, static equilibrium is satisfied by equating the internal force vec- tors to the boundary forces. The final configuration of the surface is that for which {R} = {P} is (almos’) zero at all internal nodes. 3.6.2 Elastic and geometric effects in static analysis and in form-finding Unlike the static analysis stage, the computational form-finding process does not require the elastic effects to be included. The basic reasons for this state- ment are as follows: (i) Since the only applied load is pre-tension, it is sufficient to resolve the tension forces at each node and proceed to relax these during the 58Geometrically non-linear behaviour. Solutions by commonly used numerical methods iterative adjustment of the geometry of the structure, until static equi- librium is reached. (ii) The inclusion of elastic effects (straining) in form-finding is possible, but this (explained with a help of a cable structure), will have the following consequences: e lack of control over the values of the cable forces, which will vary, and may ultimately exceed the specified safe, or permissible load levels necessity for adjustments of the unstrained cable lengths, in order to maintain the required tension levels in the structure (and prevent problems stated above) necessity for further iterations, in order to restore static equilibrium, which gets violated by the adjustment of unstrained lengths of cables. The procedure thus described is tedious, involving iterations within itera- tions. Consequently, the inclusion of elastic straining is not recommended for form-finding of tension structures. Elastic effects are normally considered at the stages of patterning and static analysis. For reasons of simplicity, much of the discussion concerning computa- tional modelling of tension structures has concentrated on pre-stressed cable nets. However, cable or line elements may be used to approximate membrane surfaces, and this is discussed further in Chapter 7. Consequently, computational procedures, adopted in the form-finding and analysis of pre-stressed cable nets, also apply to tensioned fabric structures (membranes). References [3.1]. BuchHotpr, H. A. (1968). The analysis of cable nets. Journ. Inst. Math Applics., 4, 339-358. 3.2]. Buchinouor, H. A. and McMitay, B. (1973). A non-linear vector method for the analysis of vertically and laterally loaded cable assemblies. Proceedings of the Institution of Civil Engineers, 55, 211-228. (3.3). Arcyris, J. H. and Scriarpr, D. W. (1972). Large deflection analysis of pre- stressed networks. Proceedings ASCE, Journal of the Structural Division, ST3, 633- 654. 13.4]. Arcyris, J. H., ANcELoroutos, T. and Bicuat, B. (1974). A general method for the shape finding of lightweight tension structures. Computer Methods in Applied Mechanics and Engineering, 3, 135-149. 13.5]. Suv, A. (1963). A general analysis of prestressed nets. International Association of Bridge and Structural Engineering, 3, 283-292. [3.6]. Stev, A. and E1eiman, J. (1964). Stress analysis of prestressed suspended roofs. Proceedings ASCE, Journal of the Structural Division, ST4, 103-121. 59Tension Structures. Form and behaviour [3.7]. Hau, E. and Powstt, G. (1971). Analytical shape finding for cable nets. Proc. TASS Symposium, Part II on Tension Structures and Space Frames, 83-92. [3.8]. BroucHton, P. and Noumsaro, P. (1994). The Analysis of Cable and Catenary Structures. Thomas Telford. (3.9]. Linewitz, K. and Scuek, H. J. (1972). A New Method of Analysis of Prestressed Cable Networks and Its Use on the Roofs of the Olympic Game Facilities at Munich. IABSE, Amsterdam. [3.10]. Leonarn, J. W. (1985). Tension Structures. Behaviour and Analysis. McGraw- Hill, New York. [3.11]. Crrapra, S. C. and Canate, R. P. (1985). Numerical Methods for Engineers, 2nd edn, McGraw-Hill. [3.12]. Wiuxinson, J. H. and Reiscu, C. (1971). Linear Algebra. Handbook for Automatic Computation. Springer, Berlin. [3.13]. Fox, L. (1964). An Introduction to Numerical Linear Algebra. Clarendon Press, Oxford. [3.14]. Meek, J. L. (1991). Coinputer Methods in Structural Analysis. E.&F.N. Spon. [3.15]. Linxwitz, K. and Scrtk, H. J. (1971). Einige Bemerkungen zur Berechnung von vorgespannten Seilnetzkonstruktionen. Ingenieur - Archiv. 40, 145-158. (3.16). Scuirx, H. J. (1974). The force density method for form-finding and com- putation of general networks. Computer Methods in Applied Mechanics and Engineering, 3, 15-134. [3.17]. Linxwrrz, K. (1999). Formfinding by the direct approach and pertinent strategies for the conceptual design of prepressed and hanging structures. International Journal of Space Structures, 14, No. 2, 73-87. [3.18]. Linkwrrz, K. (1999). About formfinding of double-curved structures Engineering Structures, 21, 709-718. [3.19]. Sincer, P. (1994). Analogies between minimal surfaces and membrane constructions (numerical part). Natiirliche Konstruktionen, SFB 230, Part 9, 107-111. [3.20]. Soutuwett, R. V. (1946). Relaxation Methods in Theoretical Physics. Oxford University Press. [3.21]. Orrer, J. H. R. (1964). Computations for prestressed concrete reactor pres- sure vessels using dynamic relaxation. Nuclear Structural Engineering, 1, 61-75. (3.22]. Day, A. S. (1965). An introduction to dynamic relaxation. The Engineer. Technical Contributors Section, 220-221. [3.23]. Day, A. 8. and Bunce, J. H. (1970). Analysis of cable networks by dynamic relaxation. Civil Engineering and Public Works Review, 383-386. 604 Dynamic relaxation method The method of dynamic relaxation in its early stages of development was perceived as a numerical, finite difference technique. It was first used to analyze shell structures," then skeletal and cable structures," *"! and plates.!"® 4191 The method relies on a discretized continuum in which the mass of the structure is assumed to be concentrated (lumped) at given points (nodes) on the surface (Fig. 4-1). The system of lumped masses oscillates about the equilibrium position under the influence of out-of-balance forces. With time, it comes to rest under the influence of ‘damping’. The iterative scheme reflects a process, in which static equilibrium of the system is achieved by simulating a pseudo-dynamic process in time. In its original form, the method makes use of viscous damping, as described in the following section, yy WY iy Wr I i y A Ao, iy WY SN SS y _ N CN “> WS SS KS \ S \ ZZ \ KX) \ Figure 4.1 Discretized continuum. 61Tension Structures. Form and behaviour 4.1 Dynamic relaxation method with viscous damping The equation from which the method has originated is the equation of motion, which, for a discretized system is Py = {= Ké] +My, + CB, (4.1) where subscript ji refers to the jth node in the ith direction in a discretized system, Pj; is the vector of external loads, [> Ki]; is the vector of internal loads (with K representing stiffness and 5 displacements), C is the coefficient of viscous damping, and 4,,4 are the vectors of nodal accelerations and velocities, respectively. Introducing the nodal residual forces, Rj;, as the difference between the external P,j and internal P;; force vectors, gives R= Pi - [DO Ks}, - (4.2) = Pi — Pi Hence, from eqn (4.1): Ry = Mjid,; + C8 (4.3) iO Equation (4.3) states that the motion of a system is produced by the out-of- balance forces. Equation (4.2) shows that static equilibrium of the system requires these forces to come to zero. Equation (4.3) can be approximated by centred finite differences, in which the acceleration term is represented by the variation of velocities over the time interval At, and the velocity term as an average over the same inter- val.2""| Thus, with 1 denoting the time interval at which variables are calculated, the residual forces at time increment nAt are given by std | ant or + a Ri = Ma CD (4) Equation (4.4) gives the recurrence equation for velocities: My _C ia Ri it AE i MC +m, ,t (4.5) py oe = The velocities are then used to predict displacements at time +1: itt = 6h + dy" Ar (4.6) The iterative process consists of a repetitive use of eqns (4.5) and (4.6); it is a process that continues until the residual forces are close to zero. The 62Dynamic relaxation method residuals are calculated from eqn (4.2), which, in the iterative scheme, takes the form Ri=P- [DY xs | = Pj — (Pil"" 4.1.1 Stability of the iterative solution In order to use the iterative scheme presented above, it is necessary to make assumptions regarding the values of masses, damping, coefficients, and the time increment. Since the objective is not to follow the actual motion of a real structure but to determine the position of its static equili- brium, the values of the masses and damping coefficients can be chosen on the grounds of convenience. The following is a summary of results given in Ref. [4.11] Both the differential equation (4.1) and difference equation (4.4) should have the values of masses M and damping C, so chosen as to ensure that the recurrence scheme does indeed converge to the static equilibrium. For linear problems, the stability of the iterative scheme depends on the values of At, K and M defined earlier, and in particular on /411, the largest eigenvalue of the matrix M"'K. For stability: es Ate (4.8) To minimize computation time, a value close to this limit (but not too close) should be used. In practice this limit is not known, but if M is chosen as a diagonal matrix whose elements are « times those of the leading diagonal of the stiffness matrix K then 2/a is a near upper bound on j;. With 4; * 2/a, the maximum time increment for stable iteration is Mmae % V20 = py (4.9) The time increment can be chosen arbitrarily, because the problem is not considered in real time. If At is taken as unity, the ratio M/K is equal to 1/2. Choosing a =} gives 1; ~ 4 and a stability limit of At ~ 1 from eqn (48), Clearly, At =1 is a convenient choice, and since the upper bound is approximate and the actual problem is non-linear, a value of « larger than } would ensure At = 1 to be safely within the limit. An alternative derivation of this same stability limit can be found in Ref. [4.8]. Although the analysis of the stability of the dynamic relaxation algorithm is couched in terms of the stiffness matrix of the structure, this has been done 63