Newcastle University e-prints

Date deposited: 5th April 2013

Version of file: Author final

Peer Review Status: Peer reviewed

Citation for item:

Gosling PD, Bridgens BN, Zhang L. Adoption of a reliability approach for membrane structure
analysis. Structural Safety 2013, 40, 39-50.

Further information on publisher website:

http://www.elsevier.com/

Publisher’s copyright statement:

The definitive version of this article, published by Elsevier, 2013, is available at:

http://dx.doi.org/10.1016/j.strusafe.2012.09.002

Always use the definitive version when citing.

Use Policy:
The full-text may be used and/or reproduced and given to third parties in any format or medium,
without prior permission or charge, for personal research or study, educational, or not for profit
purposes provided that:

 A full bibliographic reference is made to the original source

 A link is made to the metadata record in Newcastle E-prints
 The full text is not changed in any way.

The full-text must not be sold in any format or medium without the formal permission of the
copyright holders.

Robinson Library, University of Newcastle upon Tyne, Newcastle upon Tyne.

NE1 7RU. Tel. 0191 222 6000
Adoption of a reliability approach for
membrane structure analysis
P.D. Gosling1, B.N. Bridgens1, *, L. Zhang1
1
School of Civil Engineering & Geosciences, Newcastle University, Newcastle-upon-Tyne, NE1 7RU, UK
*
Author for correspondence: ben.bridgens@ncl.ac.uk, +44(0)191 222 6409

Abstract
Membrane structures, including tensioned fabric and pressurized foil, feature in high profile and large
span projects throughout the world for well-appreciated architectural and functional reasons. From an
analysis and design perspective these types of structure exhibit high levels of material and geometric
non-linearity. Current design practice is use of a permissible stress approach using stress reduction
factors with values typically between 4 and 8. Design guidance is limited and the magnitude and
nature of these stress factors varies widely.

Recently CEN TC250 Working Group 5 has been established to write a standard for Membrane
Structures for inclusion in Eurocode 10. In this paper we apply the principles of BS EN 1990:2002
“Eurocode - Basis for Structural Design” to membrane structures and explain the implications for
analysis and design. The key feature of this approach is the mathematical and numerical rigor and
consistency in predicting the safety of membrane structures as an explicit function of inherent
uncertainty in applied loading, structural form and material properties. This new approach to
membrane structure analysis is demonstrated through two examples, which show the application of
reliability analysis in achieving the objectives of “Eurocode - Basis for Structural Design” and highlight
the further work that is required before this approach can be fully utilized by industry.

Keywords: Membrane structure; tensile fabric; stress factor; reliability analysis; safety index; Eurocode

1 1
1 MEMBRANE STRUCTURE DESIGN – CURRENT PRACTICE
1.1 FUNDAMENTALS OF FABRIC STRUCTURE DESIGN

Fabric canopies are one of the earliest forms of roofing, and have been used for traditional forms of
construction for thousands of years. However, modern fabric structures using synthetic materials have
only been in use for about 50 years. A fabric membrane acts as both structure and cladding, thereby
reducing the weight, cost and environmental impact of the construction. These structures combine
striking architectural forms with high levels of structural stability and durability, with expected lifespans
in excess of thirty years depending on the type of material used (Figure 1). Key to achieving these
high levels of performance is accurate modeling of the form and behaviour of the structure, including
incorporation of the inherent uncertainties in all aspects of the design process.

Architectural fabrics have negligible bending and compression stiffnesses, and therefore fabric
structures must be designed with sufficient curvature to enable wind and snow loads to be resisted as
tensile forces in the plane of the fabric [1, 2]. The low tensile and shear stiffnesses of the membrane
material result in large strains and significant changes to the structural form under load. The modified
membrane curvature results in a change in the geometric stiffness of the structure, and hence an
iterative large-displacement geometrically non-linear analysis is required to determine the equilibrium
configuration of the structure for each load case.

To resist both uplift and down-forces (typically due to wind and snow respectively) the surface of the
canopy must be anticlastically double-curved, i.e. saddle shaped rather than dome shaped. Fabric
structures are pre-tensioned (usually referred to as ‘prestressed’) to ensure that the fabric remains in
tension under all load conditions and to reduce deflections. The low weight of the fabric means that
gravity or ‘self weight’ loading is often negligible. Consequently, tensile fabric is frequently more
structurally efficient and cost-effective for large span roofs than conventional construction methods.

2 2
Figure 1. Tensile fabric has been used for diverse, architecturally striking structures for over 50 years,
examples include Schlumberger Research Centre, Cambridge, UK, 1983 (top), a complex cable supported
fabric form [© Andrew Dunn (www.andrewdunnphoto.com)]; Hampshire Cricket Club, UK, 2001, (centre) –
classic tensile architecture consisting of multiple conic canopies [© Buro Happold / Mandy Reynolds];
and Dalton Park, UK, 2002 (bottom) – multiple conic canopies provide a light but sheltered environment
for shoppers [© Ben Bridgens].

Unlike conventional structures, the shape of a tensile structure cannot be prescribed, but must take a
‘form-found’ shape determined by equilibrium and geometric constraints. Prior to the development of

3 3
numerical methods, experimental techniques that exploited the properties of surface tension of a soap
film were used to visualize and determine feasible membrane structure forms (Figure 2) [4-6]. The
‘form found’ shape ideally takes the form of a minimal surface which joins the boundary points with the
smallest possible membrane area and has uniform in-plane tensile stresses throughout [7, 8]. True
minimal surfaces cannot be formed between all boundary conditions, but a pseudo-minimal surface
can be developed for a fabric membrane by accepting increased stresses in the region where the
soap bubble would have failed. This reduces the limitations on the forms that can be created but
results in reduced structural efficiency.

The design of membrane structures is not codified in Europe, and only recently has limited design
guidance been compiled [9]. International standards do exist [10] but provide broad design principles
rather than a detailed analysis and design methodology.

Figure 2. Soap-film model of a uniformly stressed membrane structure with cable boundaries (image
reproduced with permission of Institut für Leichtbau, University of Stuttgart).

Architectural fabrics consist of woven yarns to provide strength, with an impermeable coating to
provide waterproofing and stabilise the weave. Warp yarns run along the length of the roll and are
typically fairly straight, with highly crimped weft (or fill) yarns woven across the roll (Figure 3). The
most common material combinations are PVC coated polyester yarns and PTFE coated glass-fibre
yarns, with silicone coated glass-fibre fabric becoming increasingly popular and PTFE coated ePTFE
(Tenara) providing very high light transmission. A combination of non-linear stress-strain response of
the component materials (yarn and coating) with the interaction of orthogonal yarns, results in complex
(non-linear, hysteretic, anisotropic) material behaviour [11]. Full quantification of the response of
coated woven fabrics to in-plane loading (biaxial and shear) is time consuming and costly, and
arguably has not yet been achieved. It is common practice to use assumed stiffness values for a given
fabric material [2], but the actual stiffness may differ by a factor of between two and five from these

4 4
assumed values [12]. Elastic moduli, Poisson’s ratios and shear stiffness are independent and are not
constrained by conventional limits and relationships for anisotropic materials.

Figure 3. Architectural fabric cross-sections showing highly crimped, woven yarn bundles encased in
PTFE or PVC coating (left); interaction of warp and fill yarns in woven fabric results in highly non-linear
stress-strain behaviour dependent on the ratio and magnitude of the applied in-plane biaxial stress.

1.2 DESIGN REQUIREMENTS

There are four fundamental design requirements for membrane structures:

 stress limits based on material strength and stress factors,

 deflection,

 avoidance of ponding,

 avoidance of slackness and/or wrinkling.

Fabric strength is routinely determined by manufacturers from uniaxial tensile tests [13], however it is
well known that the presence of a flaw such as a tear or a severe crease can dramatically reduce the
in-service strength of the fabric. A 40mm tear in a 400mm wide uniaxially loaded panel may reduce
the strength of the fabric in a direction normal to the tear by 75% [14]. However, the specific reduction
in strength is dependent on a number of factors including coating & yarn types, weave, temperature,
moisture content, and loading rate. The significance of these factors has not yet been established.
Transportation and deployment on site can also have significant impacts on the strength of the
canopy, particularly for fabrics with brittle glass-fibre yarns.

Large displacements (e.g. +/- 500mm) are acceptable in membrane structures and it is difficult to
define meaningful displacement limits. Typical design requirements are that the membrane does not
clash with other parts of the structure and that the curvature of the membrane does not invert. The
analysis results may show acceptable (positive) tensile stresses, but if the curvature has inverted the
membrane will have passed through a state of slackness en route to the equilibrium position. This may
result in flapping and crease-fold damage to the fabric, and even structural instability.

5 5
Ponding is the accumulation of rain, snow or melt-water in hollows in the membrane surface, which
can lead to fabric failure. It is vital to ensure that fabric structures maintain positive drainage under all
load conditions. This check is carried out following analysis with unfactored loads (due to geometric
non-linearity), but this means that the subsequent design may only just avoid ponding with a factor of
safety close to one, with no consideration of uncertainty and variability in the material properties,
prestress levels, construction tolerances and the analysis itself. Whilst deflection limits are typically
considered to be a serviceability condition, clashes with structural elements, ponding and slackness
could all result in failure of the membrane structure, hence deflection should be considered to be an
ultimate limit state requirement for this type of structure.

Due to the negligible compressive and out-of-plane bending stiffness of architectural fabrics, when the
tensile stress reduces to zero in any direction the membrane surface will start to buckle and wrinkling
will subsequently appear with further compressive strain. Wrinkling is not typically considered in
analysis, but is a common problem during installation when the mobilization of shear deformation is
not sufficient to achieve the required level of curvature. Wrinkling typically results in alternating regions
of high and low stress, with potentially much higher stresses than predicted by the membrane
analysis. Wrinkling is therefore unacceptable for both aesthetic and functional reasons, and can be
considered to be both a serviceability and ultimate limit state. Wrinkling criteria have been proposed in
terms of either principal stresses or principal strains, but it has been found that for geometrically non-
linear analysis with anisotropic materials a combined stress and strain criteria is more appropriate [15]
(Table 1).

Stress state Wrinkling Membrane state

Minimum principal stress > 0 No Taut

Maximum principal strain ≥ 0 and

Uniaxial Wrinkled
minimum principal stress ≤ 0

Maximum principal strain ≤ 0 Biaxial Slack

Table 1. Wrinkling criteria based on principal strain and stresses [15]

1.3 CURRENT DESIGN PRACTICE: STRESS FACTOR APPROACH

The geometrically non-linear behaviour of membrane structures renders the limit state method
employed in the design of conventional structures (e.g. steel and concrete) to be inappropriate. A
‘stress factor’ or ‘permissible stress’ approach is employed in which the fabric strength is reduced by a
factor (frequently incorrectly described as a ‘factor of safety’) to give a permissible stress value. The
maximum fabric stress values from an analysis with un-factored (characteristic) loading are compared
to this permissible stress value. The stress factor must account for all uncertainties in the structure –
including variability of material properties, loading, long term material damage and degradation and
construction tolerances. Any lack of confidence in the results of the analysis must also be reflected in
this single factor.

6 6
“With development in science and technology, the element of ignorance can be largely eliminated,
while the uncertainties, being changed in form and magnitude, can never be removed” Freudenthal,
The Safety of Structures [16].

Current practice assumes that these uncertainties are accounted for via a suitable combination of
individual stress factors to give a single ‘composite factor’. Engineering groups across a range of
countries have adopted alternative stress factors that have been derived using a number of different
approaches (Table 2). Whilst it is not within the scope of this paper to discuss the merits of each
approach, it is clear that the range of stress factors is large and reflects a lack of consensus at both
European and international levels. Recent work on development of a European standard for
Membrane structures for inclusion in Eurocode 10 via CEN250 Working Group 5 highlights the need
for harmonization of these disparate design practices.

Whether or not it is explicitly stated, the magnitude of all of these stress factors is driven by the
knowledge that fabric strength is severely reduced by the presence of a tear [14]. The large magnitude
of the factors, combined with the common misnomer of referring to them as ‘safety factors’, gives false
comfort to the Design Engineer that a large margin of safety has been incorporated in the design, and
that this will accommodate any other uncertainties that may not have been explicitly considered. For a
given design case the behaviour and performance of a membrane structure may be affected by a
series of uncertainty sources, each with different individual and combined influences, making the
effective selection of a single value (or narrow range) very difficult. If the stress factor does not
account for all of these uncertainties, the ‘safety factor’ may be in fact be one or less.

Catastrophic failures of a small number of tensile fabric structures have shown that in some cases this
approach is not reliable. In the 4th February 1999 issue of New Civil Engineer a special feature
appeared under the title “Creative tension – wiring into tented structures”. The main article opened
with the following - “Are giant tents domed? Tensile fabric structures are among the most spectacular
of the 20th century – and the most controversial”. This statement was made with reference to the
collapse of the new fabric roof over Montreal’s Olympic Stadium in early 1999 (New Civil Engineer,
28th January 1999). The actual cause of the failure was identified as design based and not arising
from accidental damage.

7 7
 Stated uncertainty incorporated
Source Stress factors
within the stress factor

Fabric and testing consistency;

International Association of Shell
Calculation accuracy; Loading,
and Spatial Structures Working
4.2 – 6.0 (warp)5.0 -7.0 (fill) fabrication and installation
Group 7 recommendationfor
uncertainty; Environmental
pneumatic structures [9]
degradation; Unforeseen factors

American Society of Civil Engineers Life cycle factor; Strength reduction

4.0 – 7.8
55-10 [10] factor; Load combination factor

Fabric & seam quality; Structure scale

(probability of defects); Pollution &
French Design Guide [9, 17] 4.5 – 7.0
environmental degradation (including
quality of finite element analysis)

Fabric
Uniaxial strength, modified depending
4.9 – 6.4 Permanent load
on whether structure is loaded
German standard (DIN 4134) [18] 2.9 – 3.2 Wind load
biaxially or uniaxially; Load factor;
with reduction factors according to 4.4 – 5.1 Snow load
Material factor, accounting for seams
Minte [9, Section 6.2.3, 19] Seams
and connections; Load duration;
6.7 – 9.5 Permanent load
Pollution and degradation;
3.5 Wind load
Temperature
4.9 Snow load

4.5 wind
Draft Italian Code [9] [20] No details provided
3.75 snow

Factor of 2.5 for simultaneous wind

Chinese Technical Standard [21] 5
and snow loading

Membrane Structures Association

8 for sustained loads
of Japan guide for ‘Specific No details provided
4 for temporary loads
Membrane Structures’ [9]

Table 2. Stress reduction factors for tensile fabric structure design according to different organisations
and countries.

8 8
2 APPLICATION OF RELIABILITY ANALYSIS TO MEMBRANE STRUCTURES
BS EN 1990:2002 “Eurocode - Basis for Structural Design” [22] provides the underlying principles for
the Structural Eurocodes and the assumptions on which they are based. The theoretical base for the
Eurocodes is reliability theory, with reliability defined as the ‘ability of a structure or a structural
member to fulfil the specified requirements, including the design working life, for which it has been
designed. Reliability is usually expressed in probabilistic terms’ [22]. The development of a standard
for membrane structure design within CEN250 Working Group 5 falls within the context of Eurocode
“Basis for Structural Design”. Use of reliability analysis for membrane structures would bring their
design in line with ‘conventional’ building materials, would fit within the widely accepted Eurocode
framework, and would allow closer integration of the design of fabric structures and the supporting
structure. Reliability requirements are defined according to the importance and design life of the
structure. Depending on the consequence of structural failure and design life, membrane structures
can be categorized according to Table 3 and hence the required safety index, , can be determined.
For each reliability class (RC) there is a specified annual probability of failure, which results in differing
safety indexes for 1 year and 50 year reference periods. These are typical timescales used for
temporary and permanent structures throughout the Eurocodes.

Annual failure Recommended minimum

Consequences Classes or Reliability Reference values for reliability index β
probability
Classes period (ultimate limit states)
[23]

RC3: High consequence for loss of

human life, or economic, social or -7 1 year 5.2
10
environmental consequences very 50 years 4.3
great.

RC2: Medium consequence for loss of

human life, economic, social or -6 1 year 4.7
10
environmental consequences 50 years 3.8
considerable.

RC1: Low consequence for loss of

human life, economic, social or -5 1 year 4.2
10
environmental consequences small or 50 years 3.3
negligible.

Table 3. Target reliabilities for different building categories, from “Eurocode - Basis for Structural Design”
combining Table B1 Definition of consequences classes and Table B2 Recommended minimum values
for reliability index β (ultimate limit states) [22].

If used as a basis for structural design, the principle of structural reliability must be followed through
the entire fabric structure design process. The safety or reliability requirements of a structure are
related to different types of limit state – e.g. ultimate, fatigue, and serviceability (Table 4) – as used in
other structure types.

Limit State Target reliability index, , for reference period of

9 9
 1 year 50 years

Ultimate 4.7 3.8

Fatigue - 1.5 – 3.8*

Serviceability 2.9 1.5

* Depends on degree of inspectability, reparability and damage tolerance.

Table 4. Target reliabilities of RC2 building type by limit states, reproduced from “Eurocode - Basis for
Structural Design” Table C2 Target reliability index β for Class RC2 structural members [22].

Calculation of membrane structure reliability will require a substantial amount of material uncertainty
information, and a robust and efficient reliability estimation approach which can reasonably combine
the material and geometric information with the characteristics of the membrane structure
performance. In addition to high quality material information, of equal importance is the essential
requirement for a high fidelity analysis capability. A comparative ‘round robin’ analysis has recently
been carried out by the Tensinet Analysis & Materials Working Group (www.tensinet.com;
tensinet.amwg@ncl.ac.uk), in which twenty organisations independently analysed four simple,
precisely defined tensile fabric structures to identify epistemic uncertainty associated with the analysis
[24]. The paper concludes that: “The results show very high levels of variability in terms of stresses,
displacements, reactions and material design strengths, and highlight the need for future work to
harmonise analysis methods and provide validation and benchmarking for membrane analysis
software” [24].

The reliability of a structure is based on performance requirements or limit state functions. Within the
context of membrane structures, these concepts are not new with “limit state functions” such as
maximum permissible stress or maximum displacements routinely used in design. A typical design
question currently used may be of the type: “Has the stress at any point within the structure as predicted by
a numerical model exceeded the (factored) permissible stress?”. The equivalent reliability-analysis-based
question would be of the type: “What is the probability that the stress at any point within the structure as
predicted by a numerical model has exceeded the fabric strength?”. The data arising from fabric strength
tests may be in the form of a stated minimum value or a statistical representation comprising a mean,
standard deviation, and higher statistical moments or coefficients depending on the best-fit
distribution. The principal difference between the stress associated with the stress factor approach
(FOS) and the reliability-based approach (Pf) is not the means by which the values are determined
from the material tests, but in the representation and application of the data. For example, in the case
of the stress factor approach, FOS would be either the minimum stated value or the mean value
perhaps reduced by a pre-selected number of standard deviations. This value is then further reduced
by a stress factor as described above (Table 2). In contrast, when undertaking a reliability analysis,
the full statistical information which describes the performance of the fabric can be used. A substantial
programme of testing will be required to provide statistically meaningful distributions of results, and in
addition existing test data (for example manufacturers’ databases of test results) will be able to be
utilized more fully.

10 1
0
Similar questions to those regarding the fabric strength may also be posed relating to deflections,
ponding and wrinkling. Formulation of these limit state functions are not as straightforward as strength
criteria. Deflection of the fabric due to wind loading may be considered to be a serviceability limit state,
with consequences such as noise or occupant distress rather than structural failure. However, if
deflections are limited to avoid clashes with structural elements, this would become an ultimate limit
state with varying limits at different points on the structure. Similarly, deflections due to snow loading
represent an ultimate limit state when ponding may occur which would result in structural failure.
Ponding checks require an assessment of slope rather than absolute deflection, which presents a
further complexity for the analysis. Clearly in addition to development of reliability analysis tools,
substantial understanding and experience is required to formulate the correct limit state functions.
Straightforward examples of strength and deflection criteria are given in Table 5.

Design Philosophy
Design criteria
Stress factor Reliability

  FOS Pf – 
MAX MAX
Strength (warp & fill) =0

Deflection (e.g. 160mm at a

  160mm 160mm – 
MAX MAX
given location to prevent clash =0
with steelwork)

 = maximum stress; 
MAX MAX
= maximum deflection;
FOS = permissible stress; Pf = limiting stress

Table 5. Example comparability between stress factor & reliability principles.

In current fabric structure design practice stress factors are assumed to account for a range of
uncertainties. Unlike the use of partial factors in the design of steel and concrete structures, the loads
remain un-factored and uncertainties arising from variability of the loading are included in the
“permissible” stress value. The complete probabilistic content of the analysis is replaced by a factored
deterministic analysis. Structural performance is assessed after the analysis has been completed, with
performance criteria assessed by the Design Engineer. Assessment of the contribution made by each
of the input parameters (e.g. strength, biaxial and shear stiffness etc.) to the performance of the
structure is only possible with a time consuming parametric study involving multiple analyses. The
performance of the structure is considered on a point-by-point basis – for example the maximum
stress at any location is checked against the permissible stress. A structure with large areas of high
stress would be treated in the same manner as a structure with only a localized area of high stress.

In contrast to the stress factor approach, a reliability-based analysis requires the statement of
performance criteria before the analysis is undertaken. A reliability-based analysis makes use of
statistical information to quantify the uncertainty content of each parameter. This statistical information
is obtained directly from test data. As more tests are carried out on a particular material, an
increasingly accurate statistical description of the material will be developed. New uncertainties may
be introduced as information becomes available to refine the quality of the structural simulation, for
example by inclusion of time-dependent behavior including creep and relaxation. The statistical
information will typically be in the form of mean and standard deviation values. The coefficient of
variation is defined as the ratio of the standard deviation to the mean, giving a non-dimensionalised

11 1
1
transferable measure of uncertainty. In a reliability-based analysis, the uncertainty measures are
combined as part of the analysis methodology to assess the most onerous design condition for the
structure. The combination of the uncertainty measures is a function of the uncertainty measures
themselves and the structure, and will change as the uncertainty and structural parameters (e.g.
geometry, loading, etc.) change.

The output from the reliability analysis is a probability of failure, pf , or safety index, , and a design
point [25]. The probability of failure may be compared with the acceptance criteria specified within
“Eurocode - Basis for Structural Design” (Table 3 & Table 4). The design point is the unique set of
values of the uncertain variables that combine to make the most onerous condition for the structure
when assessed against the stated engineering performance criteria (or limit state function), for
example strength, deformation, etc. The physical location of this point on the structure is also
identifiable from the evaluation of the limit state function.

The probability of failure may also be read as a safety index, , which is the number of standard
deviations that the design point is located away from the mean design. For example, in “Eurocode -
Basis for Structural Design”, the minimum value of  for an ultimate limit state is 3.8 (Table 4). In other
words, following the reliability analysis if the design point were more than 3.8 standard deviations
away from the mean then the structure would not meet the requirement of the code. Placing this within
a general engineering context, the same principles are used in manufacturing where the concept of
the six-sigma approach is applied, in which a product must perform within 3 standard deviations of
the mean to be acceptable.

For complex structures, explicit mathematical solutions of the probability of failure are not normally
available. Approximate reliability methods are required to make the problem tractable [25]. These
include first and second-order reliability methods (FORM and SORM [26], Figure 4), and other
approximation methods, such as response surface approaches [27]. Coupled with the Hasofer-Lind
transformation [28], FORM has emerged as one of the most effective reliability analysis methods in
estimating the probability of failure.

12 1
2
Figure 4. Schematic sketch of FORM and SORM approximation in the standard normal space, reproduced
from Valdebenito et al [25].

In a FORM analysis, the values of the limit state functions may be obtained in the usual way through a
deterministic structural analysis, with derivatives of G(Xs) calculated either analytically or numerically,
effectively forming a sensitivity analysis of the structural performance with respect to uncertainty.
Therefore, as with a Monte Carlo simulation, the FORM reliability analysis makes use of an existing
deterministic analysis tool, but with the benefit of drastically improved efficiency. In essence, the
FORM reliability analysis may be viewed as a computational tool that “wraps around” a standard
analysis – there is no requirement to rewrite existing deterministic codes to be able to undertake a
FORM analysis.

The limit state function is derived from the state equations, typically in terms of displacements or
stresses. The stresses are functions of strains which are non-linear functions of the displacements
and, for the cases presented in this paper, material constants which are independent of the
deformation state. The limit state function, therefore, will typically be non-linear in terms of
displacements. Derivatives of the limit state function are obtained with these non-linearities explicitly
included as variables and are obtained using the chain rule, for example. Within the FORM algorithm
the limit state function and associated derivatives are evaluated at equilibrium of the structure. The
non-linearities are, therefore, captured explicitly at this state and contribute to the FORM calculations.

Apart from an immediate application to FORM, the sensitivity analysis is also important in enabling the
effects of the random variables on the membrane structure response to be elucidated. For variables
that have significant effects on structural failure, any epistemic uncertainty may be reduced by
collecting additional information through testing, thereby increasing the proportion of aleatoric
(intrinsic) uncertainty and producing a more accurate statistical description. High sensitivity values
may also be used to direct design changes, influence material selection, or to inform manufacturing
tolerances and specifications.

13 1
3
Variables having little influence on structural reliability may be considered to be deterministic to save
computational effort, without compromising the accuracy of the reliability estimate. The influence of
this strategy on the safety index  at the MPP can be readily assessed with a single subsequent
analysis. Sensitivities may be calculated either analytically if possible (subject to availability of access
to the analysis coding), semi-analytically if explicit analytical sensitivities cannot be derived
mathematically, or using approximate numerical methods such as finite differences.

Combining sensitivity analyses and FORM to estimate the probability of failure of a system with an
implicit limit state function uses computed information about the value and gradient of the limit state
function at the design point at convergence of the analysis, and an iterative optimisation scheme to
determine the safety index  [29]. The advantage of this approach is that solving the limit state
equation G(Xs) = 0 is avoided, so that the reliability analysis can be undertaken for a structure with a
complicated or implicit limit state function, given that the partial derivatives of G(Xs) with respect to the
uncertain variables are available.

3 MEMBRANE STRUCTURE RELIABILITY ANALYSIS: EXAMPLES & DISUCSSION

In this section two examples of the application of reliability analysis to membrane structures are
presented. These analyses have been carried out using Fortran code developed for this work, based
on the principles described above. The first example is a hyperbolic paraboloid, commonly referred to
as a ‘hypar’, the simplest form of membrane structure and commonly used in practice (Figure 5). The
hypar has been used as a straightforward but realistic structure that can be used to efficiently explore
the effect of differing levels of variability in the analysis parameters. The second example is a case
study based on an existing fabric structure, with the probabilistic analysis compared to the
deterministic analysis results which were used for the actual design. The purpose of this case study is
to examine the safety factors currently applied to fabric structure design consistent with the reliability
requirement of the Eurocode 0. It also demonstrates how to apply the reliability analysis tool to the
realistic design of a fabric structure.

14 1
4
3.1 HYPERBOLIC PARABOLOID (‘HYPAR’)

Figure 5. Hypar structures are characterised by alternating high and low points with cable supported
edges. Hemmingsway Hotel, Kenya, 2002 (top), a classic 4 point hypar; Cayman International School,
Cayman Islands, 2006 (bottom), an elegant 6 point hypar. All images © Architen-Landrell Associates.

In this simple but realistic example the probability of failure of a pre-tensioned hypar constructed from
lightweight PVC coated polyester fabric is considered. A combination of deterministic (single valued)
parameters (geometry, cable properties and prestress levels, Figure 6) and statistically defined
parameters (loading and fabric properties, Table 6) has been adopted. The choice of which values to
describe statistically can be decided for each structure and will depend on available statistical
information.

15 1
5
Figure 6. Hypar example

Xj Distribution Mean value Standard deviation (COV = 0.1)

Ef (kN/m) Normal 1000 100

Ew (kN/m) Normal 1000 100

Gwf (kN/m) Normal 30 3.0

vwf = vfw Normal 0.1 0.01

2
F (kN/m ) Normal 1.0 0.1

σper (kN/m) Normal 10 1.0

Table 6: Distributions and parameters of the random variables for the hypar example

An assumed material strength of 10 kN/m has been used, equivalent to a lightweight PVC coated
polyester fabric with an ultimate tensile strength of 50 kN/m with a typical stress reduction factor of 5
applied. In this example the hypar has been subjected to uniform wind uplift only, hence for the
chosen patterning direction the fill stress will always be greater than the warp stress. It follows that the
limit states for this structure can be defined as:

Limiting warp stress G1(Xs) = not required for this example

per –  per = ult / 5 = 10 kN/m

f
Limiting fill stress G2(Xs) = max

 – 
p p p
Wrinkling criteria G3(Xs) = min per per =0

Deflection criteria G4(Xs) = Dall – Dmax Dall = 35 mm

The allowable displacement (Dall) is an arbitrary choice that would usually be defined by specific
issues (clashing with steelwork or ponding). In this example the value of 35mm is based on the span
of the structure (4m) divided by 120.

16 1
6
For this example all random variables are assumed to be normally distributed with the standard
deviation of each variable initially set to be 10% of the mean value (i.e coefficient of variation = COV =
0.1). This gives safety indices of 2.30 for fabric failure (first column of Table 7), 1.98 for wrinkling
(Table 8) and 3.09 for displacement (Table 9). Safety index requirements from the Eurocodes (Table 3
& Table 4) are typically 3.8 for ultimate limit states and 1.5 for serviceability. The safety index of 2.3 for
fabric failure clearly does not meet this requirement. Wrinkling is acceptable if it is considered to be a
serviceability limit state, and the safety index for displacement is acceptable for either limit state.

A conventional, deterministic analysis of the same structure (i.e. COV=0 for all variables) gives a
maximum fabric stress (fill direction) of 7.6 kN/m and a maximum displacement of 11.5 mm.
Comparing the maximum stress to the material strength with a stress reduction factor of 5 gives a
factor of safety of 1.32. The displacement is well within the limit of 35mm. It is significant that using a
deterministic analysis the structure would have been deemed to be acceptable, whereas with the
introduction of variability (COV=0.1) in the analysis parameters the structure does not meet the
fundamental requirement of the Eurocodes.

17 1
7
 Safety indices

Xj
COV = 0.1 COV = 0.15 COV = 0.20 COV = 0.25 COV = 0.30 COV = 0.40

Ew 2.301 2.286 2.251 2.231 2.192 2.089

Ef 2.301 2.271 2.235 2.219 2.146 2.055

Gwf 2.301 2.301 2.301 2.300 2.300 2.299

vwf 2.301 2.301 2.301 2.301 2.301 2.301

F 2.301 2.149 1.984 1.814 1.777 1.403

σper 2.301 1.596 1.213 0.977 0.817 0.616

Table 7: Safety indices for the fabric failure limit state with changing parameter variability

Safety indices

Xj
COV = 0.1 COV = 0.15 COV = 0.20 COV = 0.25 COV = 0.30 COV = 0.40

Ew 1.976 1.904 1.831 1.662 1.542 1.316

Ef 1.976 1.762 1.563 1.401 1.262 1.041

Gwf 1.976 1.976 1.975 1.975 1.973 1.973

vwf 1.976 1.976 1.976 1.976 1.976 1.976

F 1.976 1.479 1.182 0.971 0.828 0.633

Table 8: Safety indices for the wrinkling failure limit state with changing parameter variability

Safety indices

Xj
COV = 0.1 COV = 0.15 COV = 0.20 COV = 0.25 COV = 0.30 COV = 0.40

Ew 3.086 3.067 3.061 3.054 3.045 2.996

Ef 3.086 2.486 2.183 1.725 1.515 1.167

Gwf 3.086 3.086 3.084 3.082 3.076 3.071

vwf 3.086 3.084 3.083 3.081 3.079 3.069

F 3.086 2.382 1.956 1.681 1.426 1.057

Table 9: Safety indices for the deflection limit state with changing parameter variability

18 1
8
Not all of the random variables will have the same level of effect on the reliability for a specific limit
state. In certain cases, only a subset of the random variables (e.g. loading and material strength) play
significant roles in the reliability analysis, and the reliability is not sensitive to the other variables even
though they may have high levels of uncertainty. To investigate the sensitivity of the probability of
failure to the variability of each analysis parameter, the coefficient of variability (COV) of each random
variable has been increased in turn whilst leaving the COV of all variables at the base-line value of
0.1. The resulting safety indices are given in Table 7, Table 8 and Table 9 for the three limit states.

As might be expected, the safety index for the failure limit state is more sensitive to the variability of
permissible stress and applied load rather than the fabric elastic properties. For the wrinkling and
serviceability failure criterion the permissible stress is not considered. For both of these limit states the
dominant variables are the applied load and the fill stiffness, with variability of other parameters having
minimal effect on the performance of the structure. Further analysis of a wide range of ‘typical’ fabric
forms will enable key parameters to be identified for reliability analysis of different structural types,
enabling the costly and time consuming process of determining statistical information for analysis
parameters to be focused on the most significant variables.

3.2 IMPLEMENTATION IN REAL STRUCTURES: DONCASTER EDUCATION CITY CRECHE CANOPY

The Doncaster Education City Creche Canopy is a double-conic canopy designed by Arup and
fabricated by Base Structures (Figure 7) in Doncaster, UK. The structure is a modest size twin conic
2
with an area of 425m , but has a complex form with high levels of curvature that required radial cables
to achieve the required form without excessive membrane stress or deflection that could result in
ponding. The one millimetre thick PVC coated polyester fabric is supported by two steel head rings
suspended from inclined masts, with radial booms supporting articulated corner clamp plates and
tensioned boundary cables to the perimeter. The membrane prestress is 1.5kN/m in warp and fill
directions.

The original design was carried out using Oasys GSA (General Structural Analysis) Fabric
(www.oasys.com). This industry standard software enables simulation of the initial membrane form-
finding followed by a deterministic geometrically non-linear analysis. In line with current industry
practice stress reduction factors of between 5 and 10 were used to calculate the required fabric
strength based on the stress results from the un-factored load combination cases.

19 1
9
Figure 7. Doncaster Education City Creche Canopy – general arrangement drawings (top) [courtesy of
Arup] and images of the completed structure (bottom) [courtesy of Base Structures].

20 2
0
Figure 8. Finite element simulation using 166 linear strain triangle elements, with example displacement
results.

The reliability based analysis of the membrane (Figure 8) requires definition of limit state functions and
statistical descriptions of all parameters. Four limit state functions can be defined for the canopy:

Limiting warp stress G1(Xs) = ult –

Limiting fill stress G2(Xs) = ult –

Wrinkling criteria G3(Xs) = – =0

Deflection criteria G4(Xs) = Dall – Dmax Dall = – 160 mm

G1(Xs) and G2(Xs) require that the maximal warp and fill stresses ( and ) at any location in
the membrane are less than the measured, statistically defined, strength of the fabric, ult. Limit state

function G3(Xs) describes a basic wrinkling criterion in which the minimum principal stresses, , at

all locations are positive definite as given by . A simple deflection criteria is represented by G4(Xs)
in which an allowable displacement, Dall is compared with the (local) maximum Dmax. These four limit
states are sufficient for this example; a more extensive and detailed set of criteria could be developed
if required to consider local issues such as ponding.

Carrying out sufficient testing to determine the mean and standard deviation values for all material
properties would be uneconomic for any given project. Some manufacturers carry out extensive
21 2
1
testing in-house for quality control and development purposes, but this data is commercially sensitive
and is not published. A move towards reliability based analysis will generate demand for
manufacturers to provide more detailed material properties on their data sheets, including mean and
standard deviation information. This move would be motivated by the competitive advantage in
showing that one manufacturer’s product was more consistent than another. Currently there is little
benefit in demonstrating consistency of manufacture, but once this information is included in the
analysis then specifications will be able to demand tight manufacturing tolerances and consistent
material performance.

For this work it was not practicable to undertake a statistically significant set of tests for all relevant
parameters. To demonstrate the reliability analysis methodology, mean values based on limited test
data have been used, with standard deviation values based on the test values and an assumed
normal distribution (Table 10). This approximate method for interpreting small numbers of data values
may be very useful as a pragmatic approach for filling in gaps in the available statistical information
about a material.

22 2
2
 Sample
Statistical Mean Coefficient Values used in
Variable standard
distribution value of variability original design
deviation

Ultimate tensile strength,

Warp = 84,
ULT (undamaged material) Weibull 76.9 2.7 0.035
Fill = 80 a
(kN/m)

Torn fabric strength Assumed

21.7 1.8 0.083 ULT / 4 b
(kN/m) normal

Biaxial warp stiffness, Ew Assumed

850 246 0.29 600 c
(kN/m) normal

Biaxial fill stiffness, Ef Assumed

924 109 0.12 600 c
(kN/m) normal

In-plane shear stiffness Assumed

15.0 5.0 0.33 30 c
(kN/m) normal

Assumed
Poisson’s ratio (warp-fill) 0.45 0.06 0.13 0.8 c
normal

Assumed
Poisson’s ratio (fill-warp) 0.53 0.21 0.40 0.8 c
normal

Prestress 1.5 kN/m

2 2
Applied snow load (kN/m ) -0.55 kN/m

Applied wind uplift 2

2 0.6 kN/m
(kN/m )

Structural steelwork
tolerances; fabric
Not considered
patterning & fabrication
tolerances

Notes
a
Tensile strength values provided by the manufacturer,
b
Generally accepted 75% strength reduction with a small tear,
c
Typical elastic constants used for PVC coated polyester.

Table 10. Statistical description of membrane analysis parameters for Doncaster Creche Canopy analysis

The strength of the fabric with a small tear has been used [14], based on six wide panel tear tests.
This is a conservative approach that is consistent with the current use of stress factors – it is assumed
that the fabric is torn at any given location. Further development of the reliability analysis presented
here would enable the spatial distribution and probability of a tear occurring to be included in the
analysis: what is the likelihood that there will be pre-existing damage in the area of maximum stress?
This will enable benefits from specifying a high level of inspection and maintenance to be incorporated
in the design with a reduced likelihood of damage at a given location at a given time. Conversely, for
inaccessible structures the implications of limited inspection and maintenance can be understood.

23 2
3
Biaxial stiffness values for architectural fabrics are typically closely repeatable between tests, but the
values vary significantly dependent on the applied stress ratio, magnitude and history [11, 30]. Current
best practice is to consider all stress ratios together to give a single set of elastic constants which
approximate the non-linear fabric behavior [31] and are compatible with standard analysis codes. By
considering best-fit elastic constants to different stress ratio ‘zones’ the variability of the stiffness
parameters has been explored (Table 11) and incorporated in the reliability analysis. This method
could be further generalized by considering elastic constants derived from small ‘patches’ based on a
surface fit to the test data.

Biaxial stress Warp Fill Poisson’s Poisson’s Quality of

ratios used to modulus modulus ratio ratio (fill- fit of
calculate elastic (warp-fill) warp) elastic
(kN/m) (kN/m)
constants constants
to test
2
data (R )

All stress ratios 841 931 0.38 0.65 0.973

0:1, 1:3, 1:2 370 895 0.23 1.16 0.996

3:4, 1:1, 4:3 986 1030 0.32 0.37 0.995

2:1, 3:1, 1:0 1132 967 0.51 0.17 0.997

1:3, 1:2, 3:4 693 957 0.32 0.68 0.996

4:3, 2:1, 3:1 1051 950 0.45 0.31 0.996

Mean 845 955 0.37 0.56

Sample standard
280 44 0.10 0.36
deviation

Note: best-fit elastic constants determined using differential strain

minimization [31] for ‘increasing-load’ data only, with residual strain during
test removed [32] and with the reciprocal relationship not applied (i.e. 4
independent elastic constants).

Table 11. Best-fit elastic constants for varying load ratios

24 2
4
For this example the geometry of the supporting structure is deemed to be deterministic, i.e. the
boundaries conditions are taken to be certain. In addition the level of prestress and the initial shape of
the canopy (determined by the accuracy of the cutting patterns and fabrication) are considered as
deterministic single values. However, if sufficient information is available each of these parameters
could be described statistically. The advantages of including construction and fabrication tolerances in
the analysis are two-fold. The impact of these uncertainties on the behaviour of the structure is
incorporated in a single analysis, ensuring that these are considered in design without the need for
multiple analyses and parametric studies to determine the worst case. The benefit of reducing
tolerances and improving fabrication quality is reflected in the analysis results, providing an incentive
to improve quality, resulting in lower variability, more certainty in the design and hence a more
efficient, cost-effective design. The importance of quality control, management and supervision for
structural reliability are discussed in detail by Carpenter [33] and the Standing Committee on
Structural Safety [34].

Ongoing work at Newcastle University on shear testing of architectural fabrics has shown that shear
stress-strain behavior is non-linear, hysteretic, and severely dependent on previous shear deformation
[35]. Repeated shear cycles result in degradation of the shear stiffness. This is reflected in the large
standard deviation shown in Table 10, but further work is required to fully elucidate the complex shear
response of architectural fabrics, and then to provide a statistical description.

For design of structures under the Eurocode framework, characteristic snow and wind loads with a 50
year mean return period are calculated [36, 37] and these values are then multiplied by partial factors
[22]. Due to large displacements and consequent geometric non-linearity, it is standard practice to
apply un-factored characteristic loads in the analysis of membrane structures. It is important to
consider the purpose of each partial factor when deciding whether it should be applied to the
membrane structure loading or analysis results. In “Eurocode - Basis for Structural Design” there is a
distinction between the following partial factors for loads (referred to as ‘actions’):

γf Partial factor for actions, which takes account of the possibility of unfavourable
deviations of the action values from the representative values

γSd Partial factor associated with the uncertainty of the action and/or action effect model

γF Partial factor for actions, also accounting for model uncertainties and dimensional
variations (γF = γSd × γf).

“Eurocode - Basis for Structural Design” distinguishes between actions (the applied load) and action
effects (the resultant force, moment etc. within the structure). Brief guidance on the application of
partial factors in the case of geometric non-linearity is provided for the combined factor γF : (p.42-43)
[22]:

“(4) For non-linear analysis (i.e. when the relationship between actions and their effects is not
linear), the following simplified rules may be considered in the case of a single predominant
action:

a) When the action effect increases more than the action, the partial factor γ F should be
applied to the representative value of the action.

25 2
5
 b) When the action effect increases less than the action, the partial factor γ F should be applied
to the action effect of the representative value of the action.

NOTE Except for rope, cable and membrane structures, most structures or structural
elements are in category (a).”

Case (a) refers to P-∆ effects where structural deformation has a destabilizing effect. In this case the
partial factor should be applied to the load, to give the worst case P-∆ effect. For membrane structures
structural deformation typically increases the stiffness of the structure and makes it work more
efficiently. Taking the simplest example of a cable with a uniform linear load forming a semi-circular
arc between two supports, the tension is equal to the load multiplied by the radius of curvature [38]. If
the cable is extensible and the load remains constant, the radius of curvature will decrease as the
cable extends and hence the cable tension will reduce, as described in case (b). In this case the
conservative approach is to apply the partial factor to the action effect, and this represents current
practice where loading uncertainty is usually incorporated in the stress factor (Table 2).

The safety indices provided by the reliability analysis for the fabric failure limit state (1.3 and 1.0 for
warp and fill directions) are significantly lower than the typical Eurocode requirement of 3.8. A
deterministic analysis would have shown that the design was (just) acceptable with a minimum factor
of safey of 1.02, but with the inclusion of the high levels of variability used for this example the design
does not meet Eurocode requirements. In addition to using the best estimate of parameter variability
(Table 10), the Doncaster Creche Canopy has also been analysed with uniform, lower levels of
variability applied to all parameters (right hand side of Table 12). With a coefficient of variability of 10%
the minimum safety index increases to 4.0, which is acceptable under the Eurocode framework.
Further decreases in variability to 5% and 1% give corresponding increases in safety index.

Safety index from reliability

Deterministic analysis using stress factor
analysis, using torn material
approach
strength

Best COV COV COV

Ultimate
estimat = 0.1 = 0.05 = 0.01
Materi tensile Allowab
Maximum Stress Factor e of
al strength - le
stress reductio of variabili
directi provided by stress
(kN/m) n factor safety ty
on manufactur (kN/m)
(Table
er (kN/m)
10)

Warp 16.4 84 5 16.8 1.02 1.3 5.2 11.2 44.3

Fill 10.7 80 5 16.0 1.50 1.0 4.0 8.6 33.8

Table 12. Comparison of deterministic stress-factor approach and reliability based analysis for fabric
failure limit states. Doncaster Creche Canopy with prestress plus wind uplift applied.

26 2
6
There are two reasons why the safety indices using ‘best estimate’ variability are well below the
Eurocode requirements – high levels of variability in the fabric stiffness values and the use of the torn
fabric strength. Stiffness values taken from different zones of the non-linear response surfaces result
in highly variable elastic moduli and Poisson’s ratios (COV up to 40%). This high level of variability is
appropriate if elastic constants are used to approximate the material behavior. However, this level of
variability has been shown to be unacceptable under the Eurocode framework as the resulting safety
indices are too low. The alternative is to use a more complex, non-linear material model (for example
using look-up tables of strain-stress data, or using neural network response surfaces). This would lead
to a step change in the variability measure as it would only have to account for the variability between
tests, not between different load conditions within each test.

The use of the torn fabric strength as the permissible stress in the reliability analysis is overly
pessimistic and arguably does not allow the true safety index to be determined. This approach
assumes that there is a 100% likelihood of a tear being present at any given location on the structure.
Combined with other uncertainties this results in an unacceptably low safety index. The true benefit of
the reliability analysis can only be realized when we have an understanding of the probability of a tear
being present at any given location. A meaningful assessment of this probability should be based on
surveys of existing structures, and should include consideration of structure and material type,
maintenance, age and fabric handling during installation. This approach will enable benefits from
specifying a high level of inspection and maintenance to be incorporated in the design with a reduced
likelihood of damage at a given location at a given time. Conversely, for inaccessible structures the
implications of limited inspection and maintenance can be understood.

4 CONCLUSIONS
A key parameter for the design of fabric structures is the ‘stress reduction factor’ which reflects the
severe reduction in fabric strength in the presence of a tear, combined with any other uncertainties
such as degradation and material variability (Table 2). A wide range of values are used by different
countries and organizations, making consistent design and checking problematic. A major benefit of
reliability based analysis of fabric structures is that it removes the need for this single, all-
encompassing stress factor. The value and analytical elegance of a reliability analysis means that
uncertainties in all analysis parameters can readily be included in a mathematically consist manner as
the test or simulation data becomes available. The individual contribution of uncertainty in each
analysis parameter to the probability of failure of the structure can be assessed, and probabilities of
failure can be compared against widely accepted values given in “Eurocode – Basis for Structural
Design” to assess the adequacy of the design (Figure 9).

27 2
7
Figure 9. Reliability analysis principles and practice, after Melchers [39]

Within the context of recently begun work of CEN250-WG5 in contributing to Eurocode 10, the
preceding examples demonstrate the application of reliability concepts to the analysis of membrane
structures and the feasibility of adopting these principles. A clear step that must be taken is to provide
guidance for situations where reliability analysis capabilities are not available. This may be achieved,
in principle, through the normal route of code calibration (Figure 10). The most significant differences
in applying this process to membrane structure design is in replacing partial factors with stress factors
and in the myriad of realised forms that membrane structures can take. In adopting the philosophy of
“Eurocode - Basis for Structural Design”, it is anticipated that as a minimum it may be necessary to
classify structures by geometrical form (e.g. conic, hypar, barrel vault [3]) and support system (e.g.
masts, cables, arch) to achieve consistency in the definition of appropriate stress factors. This
approach may prove sufficient for certain classes of membrane structures, whereas for more unusual
or hybrid forms it is expected that a full reliability analysis will be required and that the analysis will be
subject to a safety index constraint.

28 2
8
Figure 10. Process for calibration of code safety-checking format, reproduced from Melchers [39]

For consistent application of Eurocode principles to fabric structures there must be consensus on
whether each limit state is classed as ultimate or serviceability, with very different safety index
requirements for the two limit states (Table 4). Deflection criteria are typically considered to be a
serviceability limit state, but for membrane structures excessive deflections can result in failure. In
addition a rigorous ponding criteria must be established, with checks for ‘hollows’ in the membrane
surface required rather than overall deflection measures, and wrinkling checks must be incorporated in
the analysis.

For the Doncaster Creche Canopy analysis presented above the safety indices fall well below the
acceptable Eurocode values for two reasons – high levels of variability in the fabric stiffness values,
and the use of the torn fabric strength. This clearly points to the developments that are required before
tensile fabric structures can be designed under a reliability framework. Improvements in fabric material
models, assessment of tear damage to existing structures, and increased material testing are required
to provide consistent and meaningful statistical descriptions of the key analysis parameters. These
requirements should drive a range of improvements across the industry: improved analysis
capabilities, increased consistency of design and manufacture, routine material testing, and better
specified installation and maintenance.

29 2
9
5 ACKNOWLEDGEMENTS
Financial and material support from Architen Landrell Associates (www.architen.com), Arup
(www.arup.com), UK Engineering and Physical Sciences Research Council (www.epsrc.ac.uk),
SergeFerrari (www.ferrari-architecture.com), and Tensys (www.tensys.com) are gratefully
acknowledged. Permission from Arup to make use of the double-conic canopy example is also
gratefully acknowledged.

BIBLIOGRAPHY
[1] Berger H. Form and function of tensile structures for permanent buildings. Engineering Structures.
1999;21:669-79.
[2] Bridgens BN, Gosling PD, Birchall MJS. Tensile fabric structures: concepts, practice & developments. The
structural engineer: journal of the Institution of Structural Engineers. 2004;82:21-7.
[3] Bridgens B, Birchall M. Form and function: The significance of material properties in the design of tensile fabric
structures. Engineering Structures. 2012;44:1-12.
[4] Bach K, Structures IfL. Forming bubbles: Institute for Lightweight Structures; 1987.
[5] Dao TT, Fomenko AT. Minimal surfaces, stratified multivarifolds, and the Plateau problem: American
Mathematical Society; 1991.
[6] Gass S. IL 25: Form Force Mass 5 - Experiments. Stuttgart: Karl Kramer 1990.
[7] Bletzinger K-U, Ramm E. Structural optimization and form finding of light weight structures. Computers &
Structures. 2001;79:2053-62.
[8] Linhard J, Bletzinger K-U. "Tracing" the Equilibrium - Recent Advances in Numerical Form Finding.
International Journal of Space Structures. 2010;25:107-16.
[9] Forster B, Mollaert M. Design guide for tensile surface structures. Brussels: Tensinet; 2004.
[10] ASCE/SEI. 55-10: Tensile Membrane Structures. 2010.
[11] Bridgens BN, Gosling PD, Birchall MJS. Membrane material behaviour: concepts, practice & developments.
The structural engineer: journal of the Institution of Structural Engineers. 2004;82:28-33.
[12] Gosling PD, Bridgens BN. Material Testing Computational Mechanics A New Philosophy For Architectural
Fabrics. International Journal of Space Structures. 2008;23:215-32.
[13] BSi. BS EN ISO 1421. Rubber- or plastics-coated fabrics. Determination of tensile strength and elongation at
break. British Standards Institute; 1998.
[14] Happold E, Ealey TA, Liddell WI, Pugh JW, Webster RH. Discussion: The design and construction of the
Diplomatic Club, Riyadh. The structural engineer : journal of the Institution of Structural Engineers. 1987;65A:377-
82.
[15] Roddeman DG. Force transmission in wrinkled membranes. Eindhoven: Eindhoven University of Technology;
1987.
[16] Freudenthal AM. The safety of structures. ASCE Transactions. 1945:1157-98.
[17] ITBTP. Recommendation pour la conception des ouvrages permanents de couverture textile. Annales des
Batiments et Travaux Publics. Modified 1998 ed1997.
[18] DIN. Tragluftbauten - DIN 4134. 1983.
[19] Minte J. Das mechanische Verhalten von Verbindungen beschichteter Chemiefasergewebe: Aachen; 1981.
[20] UNI. EN 13782:Temporary Structures - Tents - Safety. 2006.
[21] CECS. 158: Technical Specification for Membrane Structures. China2004.
[22] BSi. Eurocode - Basis of structural design. BS EN 1990:2002 + A1:2005. Brussels: British Standards
Institute; 2002.
[23] Duckett W. Risk analysis and the acceptable probability of failure. The structural engineer : journal of the
Institution of Structural Engineers. 2005;83:25-6.
[24] Gosling PD, Bridgens BN, Albrecht, Alpermann, Angeleri, Barnes M, et al. Analysis and design of membrane
structures: results of a round robin exercise. SUBMITTED TO Engineering Structures.

30 3
0
[25] Valdebenito MA, Pradlwarter HJ, Schuëller GI. The role of the design point for calculating failure probabilities
in view of dimensionality and structural nonlinearities. Structural Safety. 2010;32:101-11.
[26] Ditlevsen O, Madsen HO. Structural Reliability Methods. Chichester: John Wiley & Sons; 1996.
[27] Rubinstein RY, Melamed B. Modern simulation and modeling: Wiley; 1998.
[28] Hasofer AM, Lind NC. Exact and Invariant Second-Moment Code Format. Journal of the Engineering
Mechanics Division-ASCE. 1974;100:111-21.
[29] Rackwitz Rd, Flessler B. Structural reliability under combined random load sequences. Computers &
structures. 1978;9:489-94.
[30] Galliot C, Luchsinger RH. A simple model describing the non-linear biaxial tensile behaviour of PVC-coated
polyester fabrics for use in finite element analysis. Composite Structures. 2009;90:438-47.
[31] MSAJ. MSAJ/M-02-1995. Testing Method for Elastic Constants of Membrane Materials. Membrane
Structures Association of Japan; 1995.
[32] Bridgens BN, Gosling PD. A new biaxial test protocol for architectural fabrics. In: Motro R, editor. IASS 2004
symposium: shell and spatial structures from models to realisation. Montpellier: IASS; 2004. p. 246-7.
[33] Carpenter J. Reliability: the practical application of Eurocode BS EN 1990. The structural engineer : journal of
the Institution of Structural Engineers. 2011;89:18-23.
[34] SCOSS. The assumptions behind the Eurocodes – revised September 2010. Standing Committee on
Structural Safety; 2009. p. 1-6.
[35] Jackson AL, Bridgens BN, Gosling PD. A new biaxial and shear protocol for architectural fabrics. In: Domingo
A, Lazaro C, editors. IASS Symposium 2009: Evolution and Trends in Design, Analysis and Construction of Shell
and Spatial Structures. Valencia2009. p. 2167-79.
[36] BSi. Eurocode 1: Actions on structures. Part 1-4: General actions — Wind actions BS EN 1991-1-
4:2005+A1:2010. Brussels: British Standards Institute; 2005.
[37] BSi. Eurocode 1: Actions on structures. Part 1-3: General actions — Snow loads BS EN 1991-1-3:2003.
Brussels: British Standards Institute; 2003.
[38] Borgart A. An Approximate Calculation Method for Air Inflated Cushion Structures for Design Purposes.
International Journal of Space Structures. 2010;25:83-92.
[39] Melchers RE. Structural reliability analysis and prediction: John Wiley; 1999.

31 3
1