You are on page 1of 14

ARTICLE IN PRESS

Journal of Constructional Steel Research (

Contents lists available at ScienceDirect

Journal of Constructional Steel Research


journal homepage: www.elsevier.com/locate/jcsr

Statistical evaluation of the lateraltorsional buckling resistance of steel I-beams,


Part 1: Variability of the Eurocode 3 resistance model
C. Rebelo a , N. Lopes b , L. Simes da Silva a, , D. Nethercot c , P.M.M. Vila Real b
a

Universidade de Coimbra, Department of Civil Engineering, Rua Lus Reis Santos, Plo 2 da Universidade, 3030-788 Coimbra, Portugal

Universidade de Aveiro, Department of Civil Engineering, 3810 Aveiro, Portugal

Imperial College London, Department of Civil Engineering, London SW7, UK

article

info

Article history:
Received 13 February 2008
Accepted 2 July 2008
Keywords:
Eurocode 3
Lateraltorsional buckling
Model uncertainties
Safety factors

a b s t r a c t
The EN version of part 1-1 of EC3 introduces significant changes in the evaluation of the lateraltorsional
buckling resistance of unrestrained beams, as well as alternative design procedures. Simultaneously,
as a part of the preparation of the National Annexes of EC3 part 1-1 and the establishment of the
corresponding NDPs (Nationally Determined Parameters) it becomes necessary to define the partial
safety coefficients for the bending resistance of beams when lateraltorsional buckling is a potential
failure mode. In this paper the methodology for the resistance evaluation of beams subjected to instability
is briefly described and the results are compared with FEM numerical results for the same elements.
Subsequently, to assess the accuracy of the alternative design formulae, a statistical analysis of the results
is performed on the basis of EN 1990-Annex D. A methodology is proposed for the evaluation of the design
procedures allowing for the uncertainties in the resistance model. Results are presented for a wide set
of beam geometries and loading cases and a proposal for the definition of the partial safety factor rd is
presented for the various methods, in line with the target failure probability of EN 1990.
2008 Elsevier Ltd. All rights reserved.

1. Introduction
The Eurocode programme has been developed within a probabilistic framework, whereby all design procedures should lead
to a suitable target failure probability [2]. This target probability
naturally varies according to the importance of the structure in
terms of its consequences of failure and the intended design working life [19]. However, given the complexity and unique character
of civil engineering structures, the implementation of new or updated design procedures must be viewed with extreme care. Thus,
the probabilistic verification of the target probability cannot exclusively rely on the statistical evaluation of a representative sample
of experimental results but must be carefully weighted (mostly in
a subjective way) with previous practice.
Recently, during the conversion of Eurocode 3 from ENV to EN
status, the project team introduced in EN 1993-1-1 [3] significant
changes in the evaluation of the lateraltorsional buckling resistance of unrestrained beams. These changes include the possibility of choosing, at national level, two alternative procedures, as

Corresponding author. Tel.: +351 239797216; fax: +351 239797217.


E-mail addresses: crebelo@dec.uc.pt (C. Rebelo), nuno.lopes@ua.pt (N. Lopes),
luisss@dec.uc.pt (L. Simes da Silva), d.nethercot@imperial.ac.uk (D. Nethercot),
pvreal@ua.pt (P.M.M. Vila Real).

well as specifying some parameters that affect the level of safety


of the formulas for some slenderness ranges. Although these new
methods were developed over a number of years at several leading
research institutions, they rely on numerical calibration of some
parameters, and the underlying target failure probability was
never explicitly calculated. Furthermore, it is also the responsibility of the National Standards Bodies to define the partial safety
coefficients for design (M0 and M1 ), within the establishment of
the NDPs (Nationally Determined Parameters), that besides variability of material properties also cover the variability of the design
procedures.
Classically, any codified procedure should be assessed against
experimental test results because they supply physical evidence
that in principle is unquestionable. Nowadays, this is no longer the
unique validation criterion, mainly because of the two following
reasons.
Statistical evaluation and the subsequent probabilistic inference require, in strict terms, a significant number of test results
to ensure that the sample adequately represents the probabilistic
distribution of the population. Because of economical reasons this
is not usually possible in structural engineering.
Because of the complexity of some problems, where the resistance of a structure depends on many variables, the available
historical record of experimental results does not include the

0143-974X/$ see front matter 2008 Elsevier Ltd. All rights reserved.
doi:10.1016/j.jcsr.2008.07.016
Please cite this article in press as: Rebelo C, et al. Statistical evaluation of the lateraltorsional buckling resistance of steel I-beams, Part 1: Variability of the Eurocode 3
resistance model. Journal of Constructional Steel Research (2008), doi:10.1016/j.jcsr.2008.07.016

ARTICLE IN PRESS
2

C. Rebelo et al. / Journal of Constructional Steel Research (

Notations

Mb,Rd
Mcr

E
R
LT
LT ,0

Design buckling resistance moment


Elastic critical moment for lateraltorsional buckling
Nlin
Number of numerical tests considering all values
below Rm R
Pf
Probability of failure
Q
Factor related to Vr
Ri
Relation between re,i and rt ,i
R
Random variable related to Ri
Rm
Mean of R
The same as Rm , using the linearization of the lower
R m
tail
V
Coefficient of variation of the error term
Vfy
Yield stress coefficient of variation (correspondent
to CoV coefficient of variation)
Vr
Coefficient of variation taking into account the
variableness of the model and the material
Wpl,y
Plastic section modulus about the y-y axis
a, b, c , d Class indexes for buckling curves
b
Width of the cross section
f
Modification factor for LT
fy
Yield stress of steel
kc
Correction factor for moment distribution
h
Height of the cross section
l
Length of the beam
r
Resistance
rk
Characteristic value of the resistance
rd
Design value of the resistance
re
Experimental resistance
re,i
Experimental resistance of a specimen
rm
Resistance value calculated using fy,m and Rm
rt , i
Theoretical resistance of a specimen
x
Axis along the beam
y
Lateral geometrical imperfection
LT
Imperfection factor

In Section 2: Correction factor for the lateral torsional buckling curves for rolled sections; Otherwise: Reliability index
M
Partial safety factor for a material property, also taking into account model uncertainties and dimensional variations
M0
Partial factor for resistance of cross-sections whatever the class is
M1
Partial safety factor for a material property, also taking into account model uncertainties and dimensional variations
m
Partial factor for a material property, from its
characteristic value
Rd
Partial factor associated with the uncertainty of the
resistance model
R
Standard deviation of R
R
The same as R , using the linearization of the lower
tail
i
Observed error term for test specimen i, obtained
from the relation between Ri and Rm
i
Value correspondent to the logarithmical transformation of i

Standard deviation of the variable



The same as , using the linearization of the lower
tail

Cumulative distribution function of the standardized normal distribution

LT
LT

LT ,mod

FORM (First Order Reliability Method) sensitivity


factor for resistance
Standard deviation of the action effect
Standard deviation of resistance
Value to determine the reduction factor LT
Plateau length of the lateral torsional buckling
curves for rolled sections
Non dimensional slenderness for lateraltorsional
buckling
Reduction factor for lateraltorsional buckling
Ratio of moments in segment
Modified reduction factor for lateraltorsional buckling

measurement of some important variables.1 It is further noted


that, for some phenomena, the test procedure may itself condition
the results, as is often the case for stability tests of very slender
members.
As a consequence, the concept of generalized experimental
results is nowadays widely accepted. These experimental results
may be the outcome of physical laboratory tests or may result
from advanced numerical analyses using FEM. In this latter case,
the numerical procedure should have a degree of sophistication
one order of magnitude greater than the code design procedure
that it is supposed to validate (currently, a typical example
is GMNIA Geometrically and Materially Nonlinear Analysis
with Imperfections included for stability problems). Physical
laboratory tests still play a major role for the calibration of the
advanced numerical models but a large number is no longer
necessary to ensure the validity of the statistical evaluation.
The assessment of a target failure probability must include uncertainties in loading, material properties, geometrical properties
and the design procedure itself [11]. Carrying out a probabilistic
analysis that includes all these variabilities is still nowadays impossible because the required size of the sample would be in the
range of 106 to 107 . This is recognized in Annex D of EN 1990 [2]
that allows the separate evaluation of the uncertainty related to
loading and resistance.
Focusing on the latter aspect of the lateraltorsional resistance
of beams, Annex D of EN 1990 also separates the variability of
the design procedure and that resulting from the variability of the
material properties. This separation is of great practical interest
because, coupled with a reliable methodology for the probabilistic
evaluation, it ensures consistency across different design checks
and greatly reduces the number of required calculations.
Part 1 of this paper presents a probabilistic assessment of the
alternative design procedures prescribed in part 1-1 of Eurocode
3, following the methodology of Annex D of EN 1990 [2]. This
probabilistic assessment is based on a large number of numerical
simulations carried out by the authors [18], using the finite
element program SAFIR [7] and does not include in the evaluation
the variability of loading, material and geometrical properties to
allow for an unbiased comparison of the various methodologies.
Part 2 of this paper [12] presents a probabilistic evaluation of the
influence of the variability of the material properties of steel on the
lateraltorsional buckling behavior of steel beams and assesses the
required partial safety factor M1 for the resistance that meets the
eurocode reliability targets.

1 It is noted that very often, the relevance of some of these variables was only
established as a result of those investigations.

Please cite this article in press as: Rebelo C, et al. Statistical evaluation of the lateraltorsional buckling resistance of steel I-beams, Part 1: Variability of the Eurocode 3
resistance model. Journal of Constructional Steel Research (2008), doi:10.1016/j.jcsr.2008.07.016

ARTICLE IN PRESS
C. Rebelo et al. / Journal of Constructional Steel Research (

Table 1
Imperfection factors
Buckling curve

Imperfection factor LT

0.21

0.34

0.49

0.76

Table 2
Choice of the buckling curve
Section

Limits

Buckling curve
General case

Special case

I hot rolled

h/b 2
h/b > 2

a
b

b
c

I welded

h/b 2
h/b > 2

c
d

c
d

Other section

Fig. 1. Deformed shape of a steel I-beam.

2. Eurocode 3 design procedures for lateraltorsional buckling


resistance of beams

and

2.1. Introduction

LT =

The lateraltorsional buckling is an instability phenomenon


induced by the compressed flange of unrestrained beams subjected
to bending around the major axis. Fig. 1 illustrates the deformed
shape of a steel I-beam subjected to lateraltorsional buckling.
The lateraltorsional buckling design curves proposed in the
ENV version of the Eurocode 3 only took into consideration the
loading type in the determination of the elastic critical moment,
not accounting for the additional beneficial effect resulting from
the reduction of the plastic zones, directly related to the fact
that the bending moment diagrams are variable along the beam,
leading to over-conservative results in beams not subjected to
uniform bending distributions [1]. EN 1993-1-1 [3] presents two
approaches for the design check of lateraltorsional buckling
effects of I-beams [9]. One of them is identical to the previous
procedure proposed in earlier versions of the norm and is denoted
General Case. The other procedure, Special Case is newly
introduced and is intended for use for hot rolled or equivalent
welded sections. As for other international regulations [8,14,
13] that already considered this effect, this procedure introduces
a correction factor f , which takes into account the effect of the
bending moment diagram and considers new buckling curves that
take into account the torsional stiffness of the beam.

where Wpl,y is the plastic bending modulus, fy is the yield strength


of steel and Mcr is the elastic critical moment for lateraltorsional
buckling. The imperfection factor LT is defined in Table 1
according to the buckling curves given in Table 2.
For hot rolled sections and equivalent welded sections, the
determination of the reduction factor can be alternatively obtained
according to the Special Case using the following expression:

2.2. Evaluation of the lateraltorsional buckling reduction factor


The design buckling resistance moment of a laterally unrestrained steel beam with Class 1 and 2 cross-sections is determined, according to EN 1993-1-1 [3], by the following expression
Mb,Rd = LT Wpl,y fy /M1

(1)

where the lateraltorsional buckling reduction factor LT can be


determined using two different procedures.
According to the General Case procedure, this factor is
determined as follows:

LT =

LT +

2
LT

2
LT

with LT 1.

(2)

The non-dimensional slenderness for lateraltorsional buck LT , and the parameter LT are given by:
ling,

s
LT =

Wpl,y fy
Mcr

(3)

LT =

i

1h
2
1 + LT LT 0.2 + LT
2

LT +

with
2
LT


2
LT

(4)

LT 1
LT

(5)

LT

where

LT =

i

1h
2
1 + LT LT LT ,0 + LT .
2

(6)

The values of LT ,0 and are established in the national annexes,


provided that the limits of a maximum of 0.4 for LT ,0 , and a
minimum value for of 0.75 are respected. These limits are the
values recommended in the EN 1993-1-1 [3] and were used in this
study.
The buckling curves used in this method differ, in some
situations, from the curves of the General case, as shown in
Table 2, where h denotes the height and b the width of the crosssection.
2LT ,0 no lateraltorsional
Whenever LT LT ,0 or MEd /Mcr =
buckling check is required, both for the General Case and the
Special Case procedures.
According to EN 1993-1-1 [3], when the Special Case is used, a
modified reduction factor may be adopted to take into account the
different bending diagrams, using the expression

LT ,mod =

LT
f

with LT ,mod 1,

(7)

where the f -factor is given by


f =1

1
2

h
2 i
(1 kc ) 1 2 LT 0.8

with f 1.

(8)

The correction factor kc is given in Table 3.


In order to compare the influence of the load parameter f
together with the General Case procedure, the modified reduction
factor was also computed for the General Case procedure using
Eqs. (2) and (7). The use of the f -factor together with the General
Case procedure is here addressed as General Case/f .

Please cite this article in press as: Rebelo C, et al. Statistical evaluation of the lateraltorsional buckling resistance of steel I-beams, Part 1: Variability of the Eurocode 3
resistance model. Journal of Constructional Steel Research (2008), doi:10.1016/j.jcsr.2008.07.016

ARTICLE IN PRESS
4

C. Rebelo et al. / Journal of Constructional Steel Research (

Table 3
Correction factor kc for the different bending moment diagrams
Bending moment diagram

Table 4
Parametric study
kc

Parameters

1
1.330.33

Fabrication

rolled
welded

688
643

0.86

Section

HEA500
IPE220
IPE500

463
434
434

0.94
Moment diagram

177

195

203
Fig. 2. Simply supported beam subjected to non-uniform bending.

3. Parametric study
195

3.1. General description


Using the same strategy that was adopted in previous work by
the authors to develop the LTB buckling curves for beams under fire
conditions [16,17], a simply supported beam with fork supports,
illustrated in Fig. 2, was studied. Regarding the bending moment
variation along the member length, five values (1, 0.5, 0, 0.5
and 1) of the ratio (see Fig. 2) were investigated as well as a uniformly distributed load and a mid span concentrated load, corresponding to the bending moment diagrams shown in Table 4.
Hot rolled sections and equivalent welded sections were
analysed. The chosen sections were HEA500, IPE220 and IPE500
representative, respectively, of the following relations between the
height, h and the width b: h/b < 2, h/b = 2 and h/b > 2.
Each of those profiles was then studied combining the following
parameters:

203

179

179

Slenderness

< 0.2
0.2 < 0.4
0.4 < 1.2
1.2

157
161
665
348

Steel grade

S235
S355
S460

472
437
422

the steel grades: S235, S355 and S460;


the type of production: hot rolled or equivalent welded
sections

the type of bending moment diagram or load distribution along


the beam.
The studied cases correspond to 1331 numerical tests that
resulted from the analysis of several beam lengths for each
combination of the parameters, covering a wide range of nondimensional slendernesses, representative of practical situations.
Table 4 summarizes the parameters and number of simulations.
In the numerical simulations, a lateral geometric imperfection
of sinusoidal type with a maximum value of l/1000 [5] given by the
following expression was used:
y (x) =

l
1000

sin

x
l

Total

1331

(9)

where l is the length of the beam. An initial rotation around the x


axis with a maximum value of l/1000 radians at mid span was also
introduced [15,17].
The adopted residual stresses, that present different typical
patterns for rolled and welded sections, are considered constant
across the thickness of the web and flanges. For hot-rolled sections,
a triangular distribution as shown in Fig. 3a, with a maximum value

Fig. 3. Residual stresses: C compression; T tension. (a) Rolled sections; (b)


Welded sections.

Please cite this article in press as: Rebelo C, et al. Statistical evaluation of the lateraltorsional buckling resistance of steel I-beams, Part 1: Variability of the Eurocode 3
resistance model. Journal of Constructional Steel Research (2008), doi:10.1016/j.jcsr.2008.07.016

ARTICLE IN PRESS
C. Rebelo et al. / Journal of Constructional Steel Research (

of 0.3 235 MPa was chosen [4,15]. For the welded sections, the
distribution shown in Fig. 3b, that has the maximum value of fy
(yield strength) [6,17] was used.
The evaluation of the lateraltorsional buckling reduction factor
using the three methods described above, the General Case
GC, the Special Case SC and the General Case/f GC /f ,
was made and the results were compared with the numerical
results obtained with the finite element program SAFIR [7]. This
is a program for geometrical and material non-linear analysis,
developed at the University of Lige, Belgium. This program, widely
used by several investigators, has been validated against analytical
solutions, experimental tests and numerical results from other
programs, and it has been used in several studies that led to
proposals for safety evaluation of structural elements, already
adopted in Eurocode 3.
3.2. Numerical results
To illustrate the main conclusions of this study only selected
combinations, among all the possible parameter combinations, are
presented in Figs. 46, further details being available in [18]. These
figures compare the numerical results obtained with the program
LT ,0 = 0.2), the Special Case with
SAFIR, the General Case (
LT ,0 = 0.4 and = 0.75, and the General Case/f . They are
representative of the values h/b < 2, h/b = 2 and h/b > 2, for hot
rolled sections, equivalent welded sections and for the steel grade
S235 and S460. The figures show that the Special Case presents
some unsafe results. On the other hand the General Case/f shows
a good agreement with the numerical results and is always on the
safe side. The results also show that the General Case is generally
over-conservative for non-uniform bending moment diagrams.

procedure and the accompanying partial safety factors should


be calibrated such that the reliability levels of representative
structures are as close as possible to the target reliability index.
According to EN 1990 [2], structures are classified according
to consequence classes (CC) that are directly linked to reliability
classes (RC) with associated minimum values for the reliability
index , which is related to the probability of failure Pf by
Pf = ()

(10)

where is the cumulative distribution function of the standardized Normal distribution.


For consequence class CC2, corresponding to medium consequence for loss of human life, and considerable economic, social
or environmental consequences (for example, residential or office
buildings), the corresponding reliability class RC2 is characterized
by a minimum reliability index = 3.8 for a fifty years reference
period (typical design working life of building structures), corresponding to a minimum probability of failure of about 104 .
For the purpose of calibrating a new design procedure and
establishing the partial safety factors that yield the global target
reliability, it is thus necessary to separately calibrate the partial
safety factors related to uncertainties on the resistance side only.
EN 1990 establishes that, for a given reliability index ,
P (r rd )= (r )

(11)

where r denotes the resistance, rd is the design resistance and r


may be taken as 0.8 provided that the standard deviation of the
action effect (E ) and resistance (r ) are not too different (0.16 <
E /r < 7.6). This means that, for reliability class RC2

4. Methodology for the statistical evaluation of resistance


models

Pfr = (0.8 3.8) ' 103 .

4.1. Theoretical background

According to EN 1990 [2], uncertainty associated with the


design resistance is, in its most common format, expressed as

In the framework of a partial safety factor format, the


establishment of the partial safety factors and the inherent safety
of the associated code design procedures may be carried out
using two alternative methods [2]. The first consists of the choice
of the partial safety factors on the basis of calibration to a
long experience of building tradition, while the second relies on
statistical evaluation of experimental data and field observations
carried out within the framework of a probabilistic reliability
theory.
The first method is still widely used and it was the leading
principle for most of the partial safety factors and the factors
proposed in the currently available Eurocodes. This means that,
whenever a new design procedure is proposed, the resulting
structure or structural members and connections should be similar
to those that would be obtained using the former design procedure.
The usual exception occurs whenever the new design procedure
takes into consideration a previously unaccounted failure mode or
additional variables.
Whenever a new design procedure clearly attempts to more
accurately estimate the failure of a structure or sub-structure,
presenting significant gains over the previous design procedure,
the second method must be used. Failure to do so would mean that,
in order to match previous practice, the new partial safety factors
would end up significantly higher.2 For this second case, the design

2 This is, however, the policy that some countries have adopted for the ongoing
shift from national codes to the Eurocodes.

rd =

rk

rd m

rk

(12)

(13)

where

rd is a partial factor covering uncertainties in the resistance


model, plus geometric deviations if these are not modeled
explicitly;
m is the partial factor for the material or product property from
its characteristic value;
rk is the characteristic value or alternatively the nominal value
of the resistance;
rd is the design value of the resistance.
For the purpose of calibrating a new design procedure, it is
possible to separate the scatter related to material properties
[11] and to completely isolate the uncertainty associated with
the resistance model. Given that the experimental resistance
corresponds to the result obtained from a numerical (FEM)
simulation, it can be compared with the theoretical resistance
rt obtained by applying the design model to the same situation.
If nominal values for the material characteristics are used in the
computations, the scatter associated with uncertainties in material
properties does not exist, thus making it possible to address the
calibration of rd directly, assuming that the geometric deviations
are included in the design model.
For the purpose of calibrating new safety factors both rd and m
must be obtained or alternatively, M must be directly assessed.

Please cite this article in press as: Rebelo C, et al. Statistical evaluation of the lateraltorsional buckling resistance of steel I-beams, Part 1: Variability of the Eurocode 3
resistance model. Journal of Constructional Steel Research (2008), doi:10.1016/j.jcsr.2008.07.016

ARTICLE IN PRESS
6

C. Rebelo et al. / Journal of Constructional Steel Research (

Fig. 4. Numerical results for the HEA 500, representative of h/b < 2.

Please cite this article in press as: Rebelo C, et al. Statistical evaluation of the lateraltorsional buckling resistance of steel I-beams, Part 1: Variability of the Eurocode 3
resistance model. Journal of Constructional Steel Research (2008), doi:10.1016/j.jcsr.2008.07.016

ARTICLE IN PRESS
C. Rebelo et al. / Journal of Constructional Steel Research (

Fig. 5. Numerical results for the IPE 220, representative of h/b = 2.

Please cite this article in press as: Rebelo C, et al. Statistical evaluation of the lateraltorsional buckling resistance of steel I-beams, Part 1: Variability of the Eurocode 3
resistance model. Journal of Constructional Steel Research (2008), doi:10.1016/j.jcsr.2008.07.016

ARTICLE IN PRESS
8

C. Rebelo et al. / Journal of Constructional Steel Research (

Fig. 6. Numerical results for the IPE 500, representative of h/b > 2.

Please cite this article in press as: Rebelo C, et al. Statistical evaluation of the lateraltorsional buckling resistance of steel I-beams, Part 1: Variability of the Eurocode 3
resistance model. Journal of Constructional Steel Research (2008), doi:10.1016/j.jcsr.2008.07.016

ARTICLE IN PRESS
C. Rebelo et al. / Journal of Constructional Steel Research (

4.2. Methodology
Annex D of EN 1990 [2] provides a methodology for the
evaluation of the reliability of design models, by comparing the
results obtained using the design model with experimental results.
In the case of lateraltorsional buckling of beams the resistance
(Eq. (1)) depends on three main variables, which are the reduction
factor LT related to lateraltorsional buckling, the plastic flexural
modulus, Wpl,y and the steel yield stress fy . The reduction factor is
computed according to the EC3 procedures presented in Section 2
and is the main source of the model uncertainties which affect
the design buckling resistance moment. The flexural modulus
Wpl,y only depends on the geometrical imperfections at the crosssectional level i.e. departures from nominal dimensions. On the
other hand, the yield stress contributes alone to the uncertainties
concerning the material properties.
Although the computation formulae for the reduction factor,
as they are defined in EC3, depend on the yield stress, this is a
weak dependence and both variables can be treated statistically
as independent within a steel grade and separate safety factors
for different steel grades can be obtained. As a consequence, it
is acceptable that the model uncertainty source and the material
uncertainty source are treated separately in the following analysis.
Concerning the reduction factor LT , its variability will be assessed
by comparing the values obtained from the design code formulae
with the results from FEM analysis for the large number of design
situations referred to in Section 3 of this paper.
The probabilistic evaluation will, therefore, consider the three
variables in expression (1) as 3 independent random variables.
Concerning the variability of the material properties, it will be
assessed in Part 2 of this paper. Finally, any variability of the
flexural modulus will be neglected in this paper.

where R is the random variable whose ith realization was


mentioned above. Its mean value (Rm ) and standard deviation (R )
can be estimated by
n
1X

Rm =

n i=1

and R =

Ri

n
X
(Ri Rm )2

n 1 i=1

(16)

where n is the sample size.


If the points representing pairs of corresponding values
(rt ,i , re,i ) are plotted on a diagram they will show some scatter
around the estimator line defined by re = Rm rt . For each
R
experiment, an error term i = R i can be so defined such that
m
re,i = Rm rt ,i i .
According to Annex D of EN 1990 [2], using the values of i an
estimated value for the coefficient of variation of the error term,
V , can be obtained. Taking into consideration the logarithmic
transformation i = ln i and the corresponding variance, 2 , the
coefficient of variation of the error term is given by:

q
V =

2)
(

1.

(17)

Finally, the design value of the resistance and consequently


the safety factor for the design formulae are obtained from the
following expressions, given that a sufficiently large number of
experiments is available (n > 100):
2
rd = Rm rt e(kd,n Q 0.5Q )

(18)

and

rd =

1
Rm

2
e(kd,n Q 0.5Q )

1.0

(19)

where
4.3. Statistical procedure for safety factor evaluation
Q =
The methodology for the evaluation of the reliability of design
models establishes that, after developing the design model to
obtain the theoretical resistance of a specimen i, rt ,i , this can be
compared with the experimental resistance, re,i .
To assess the goodness of fit of the procedures for the evaluation
of the LateralTorsional buckling factor it is possible to compare
the theoretical values obtained with the application of the
three methodologies described before with the corresponding
experimental results obtained from numerical simulations using
SAFIR. A typical realization of this comparison is carried out using
a random set i of parameters both in the SAFIR model and in the
design procedure under consideration in order to obtain the ratio
Ri =

re,i
rt , i

(14)

Since the design procedure will not give exactly the same result
as the numerical simulations, Ri is, in general, different from unity.
When seeking economical and reliable design procedures, the
scatter of Ri should be minimized and Ri should be not less than
unity in a probabilistic sense.
Considering the population of beam elements subjected to
lateraltorsional buckling, a single realization of this set gives the
experimental resistance re,i . From a probabilistic design point of
view the procedure used for the calculation of the resistance should
give a sufficiently low value rd , so that the probability for the
random variable re to take values lower or equal to that design
value is very low. That is
P (re rd ) = P


re

rt

rd

= (r )


=P

re
rt

rd



1
=P R
rd
(15)

ln V2 + 1

and kd,n = 3.04.

(20)

During the numerical statistical evaluation the compatibility


of the test population with the assumptions made in the design
procedure must also be analyzed. Dividing the total test population
into sub-sets, for instance defined according to the steel grade,
section or load type, or to the slenderness ratio, permits an
assessment of which parameters have most influence on the
scatter and allows greater accuracy in the evaluation of the
error terms, although attention must be paid to the number of
experiments in each sub-set, which should not be too small.
It is often found that the frequency distribution of R cannot
be described by a uni-modal function [10]. This can be checked
by plotting the observed quantiles of the variable against
the theoretical quantiles for a chosen probability distribution.
Fig. 7 shows two of these plots and the corresponding straight
line representing the theoretical Normal distribution with the
parameters, mean and variance, calculated with all of the available
data points. Although this plot should show a good agreement
between the real probability distribution and the uni-modal
theoretical one, this is not usually the case, because the presence
of non identified sub-sets results in bi-modal or multi-modal
probability functions.
Whenever safety is the issue instead of the accuracy of the
design procedures, it is the lowest part of the distribution that
is at most concern In this case, the values for the uni-modal
distribution parameters can be estimated from a linear regression
performed at the lower end of the probability function, given
that at least 20 data points are used. If this linearization is
performed, new values in Eqs. (16) and (17) are obtained mean
value R m standard deviations R and that should then be
used in Eqs. (19)(21). In the present research this linearization

Please cite this article in press as: Rebelo C, et al. Statistical evaluation of the lateraltorsional buckling resistance of steel I-beams, Part 1: Variability of the Eurocode 3
resistance model. Journal of Constructional Steel Research (2008), doi:10.1016/j.jcsr.2008.07.016

ARTICLE IN PRESS
10

C. Rebelo et al. / Journal of Constructional Steel Research (

Fig. 7. Observed vs. theoretical quantiles of the results concerning the application of the Special Case (SC) and the General Case/f (GC/f ) methods to evaluate the LT .

was performed considering only the Nlin values below (Rm R ),


whereas Rm and R are obtained using all the available values.
Naturally, the statistical procedure described so far is equally
valid for a direct assessment of the partial safety factor M (thus
simultaneously assessing the influence of the design model and
the material variability). However, this would require a much
larger sample size, as was already discussed. A two step procedure
whereby the influence of the material variability is incorporated
a posteriori is much more efficient in terms of number of
simulations. This can be achieved in two alternative ways, both
starting from the same basic point: it is required to characterize
statistically the relevant material properties of steel. In this case, it
is necessary to evaluate appropriate statistical distributions for the
yield stress of steel.
The first alternative procedure uses Eq. (13) assuming that the
safety factor related to the material variability (m ) is independent
from the model safety factor (Rd ). In this case, m is calculated
according to the yield stress mean value fy,m and to the yield stress
coefficient of variation Vfy according to:

m =

fy,nom
fyd

fy,nom
fy,m (1 1.64Vfy )

(21)

The global resistance safety factor M follows directly from


M = Rd m .
The alternative approach is to use Eqs. (1), (19) and (20) to
evaluate the global safety factor M1 including the variability of the
model and the material in the coefficient of variation. The design
resistance is given by:
My,b,Rd = LT ,nom Wpl,y fy,nom /M1 .

(22)

According to Annex D of EN 1990 [2],


2
My,b,Rd = My,b p =0.001 = Rm LT ,nom Wpl,y fy,m e(3.04Q 0.5Q )
fr

(23)

so that, comparing both equations, the safety factor is given by

M1 =

fy,nom
rm

2
e(3.04Q 0.5Q )

where
Q =

ln(Vr2 + 1), Vr2 = (V2 + 1)(Vf2y + 1) 1


= V2 + Vf2y

and rm = Rm fy,m .

(24)

Fig. 8. Histograms and best-fit normal distributions of the ratio R computed


with the FEM results (SAFIR) and the results from the General Case (GC), the
General Case/f (GC/f ) and the Special Case (SC) design methods.

5. Statistical results concerning design models


5.1. Accuracy of the design procedures
Using the set of results obtained for the 3 design procedures
presented in Sections 2 and 3 leads to the 3 histograms of Fig. 8
where the best fit for the normal distribution probability of R is
also represented. The tendency of the Special Case method (curve
on the left side) to produce unsafe results is evident since a great
number of cases correspond to R < 1. For this method both the
mean value and the coefficient of variation are substantially lower
than for the other two methods. The curve on the right side of Fig. 8,
corresponding to the best fit normal distribution of the General
Case method, has the greatest mean and coefficient of variation.
Using the scatter plots obtained with the results from the FEM
simulations and the application of the design methods, it becomes
possible to analyze visually the dispersion and the regression lines
to look for any systematic deviation in the resistance model (Fig. 9).
The General Case (GC) method gives results clearly on
the safe side, but with a great dispersion, especially when the
lateraltorsional reduction factor approaches unity. When the
influence of the loading type is taken into account through the
introduction of the correction factor f (GC/f ) the scatter is
reduced, mainly for the higher values, keeping the tendency to
remain on the safe side.
Analyzing the plot obtained for the results of the Special Case
(SC) method, it is obvious that the scatter reduces relative to the
other two methods; however the values do not remain exclusively
on the safe side.

Please cite this article in press as: Rebelo C, et al. Statistical evaluation of the lateraltorsional buckling resistance of steel I-beams, Part 1: Variability of the Eurocode 3
resistance model. Journal of Constructional Steel Research (2008), doi:10.1016/j.jcsr.2008.07.016

ARTICLE IN PRESS
C. Rebelo et al. / Journal of Constructional Steel Research (

11

Fig. 9. Results from the application of the General Case (GC), the General Case/f (GC/f ) and the Special Case (SC) design methods plotted against the FEM results (SAFIR)
all data, n = 1331.

5.2. Accuracy among sub-sets

5.3. Safety factor evaluation for design models

The mean and standard deviation of R are plotted (see Fig. 10)
and compared among the sub-sets defined in Table 4, except for
the slenderness, for which the category number in the plot axes
LT < 0.1, (2) LT [0.1 0.2[, . . . ,
represents: (1)
(13) LT 1.2.
Concerning the type of fabrication the results show lower
standard deviations, and consequently better accuracy, for rolled
sections. If the section type is the set splitting criterion, the
differences between sub-sets do not seem to be relevant. On the
other hand, if the moment diagram is the criterion, for constant
bending and for concentrated or distributed load the scatter of the
ratio R is lower for all methods.
One point of great interest in this analysis arises when the
slenderness is used to split the data set. As can be seen in Fig. 10d,
the scatter as well as the mean value of R grow with the slenderness
for all design methods. Only the Special Case method in the
range 0.20.5 can be considered an exception, since the R-mean
diminishes assuming unsafe values in that range.
For all sub-sets the above conclusions remain true. The Special
Case method produces R-values with lower mean and lower
standard deviations than the other two design procedures. From
the same point of view, the General Case/f procedure behaves
better than the General Case procedure, showing lower standard
deviations, although not very clearly in some sub-sets.
It should also be emphasized here that the quality of a design
method relies essentially on the low scatter (low variance) of its
results for every situation of use. In contrast with the correction
of the mean, for which the parameters in the design model can
be corrected individually or globally through the safety factor, the
scatter can not be corrected with the safety factor rd .

The evaluation of the safety factors is made according to the


statistical procedure introduced above, more precisely using Eq.
(19). This evaluation is made for the global data set and for the subsets previously defined, and is condensed in the Appendix where
the mean of the ratio R(R m ) and the standard deviation of the logerror term ( ) together with the safety factor computed for each
sub-set are given. The results are summarized in Table 5 where the
safety factor concerning the design model are given for each of the
defined sub-sets and for the whole set of R values.
Because a safety evaluation is only concerned with the lower
tail of the statistical distribution and the procedures that were
described are only valid for normal or lognormal distributions,
the statistical characterization was performed based on the Nlin
(see Appendix).
From Table 5 it can be concluded that the safety factor is
lower for the results obtained from the General Case method
and higher when the Special Case method is used. Particularly
LT < 1.2
for the most common medium slenderness 0.4 <
the safety factor for the Special case is substantially higher
than for the other two methods. The same conclusion is valid for
welded HEA500 section, constant moment diagram, any type of the
considered loading and steel grades S355 and S460.
6. Conclusions
Firstly, it should be stated that this study only considers
standard hot rolled profiles or equivalent welded sections.
Therefore it is not possible to make any statements about the use of

Please cite this article in press as: Rebelo C, et al. Statistical evaluation of the lateraltorsional buckling resistance of steel I-beams, Part 1: Variability of the Eurocode 3
resistance model. Journal of Constructional Steel Research (2008), doi:10.1016/j.jcsr.2008.07.016

ARTICLE IN PRESS
12

C. Rebelo et al. / Journal of Constructional Steel Research (

Table A.1
Sub-set

Rm

R m R

Nlin

R m

(V )2

rd

Rolled

GC
SC
GCf

688
688
688

1.15323865
1.03352796
1.0927916

0.13806738
0.07021712
0.11721294

1.0151713
0.9633108
0.9755787

152
38
39

0.996350567
0.954749356
0.966629096

0.01040157
0.01268934
0.00848921

0.0001082
0.00016103
7.2069E05

0.0104016
0.0126893
0.0084892

1.04
1.09
1.06

Welded

GC
SC
GCf

643
643
643

1.29068743
1.06022386
1.21539482

0.22147187
0.11146482
0.20084432

1.0692156
0.948759
1.0145505

154
52
144

1.011067296
0.921782
0.99311928

0.0262404
0.02346081
0.01174987

0.0006888
0.00055056
0.00013807

0.0262404
0.0234608
0.0117499

1.07
1.17
1.04

HEA500

GC
SC
GCf

463
463
463

1.14608122
1.01093733
1.08094226

0.1494636
0.07036146
0.12297605

0.9966176
0.9405759
0.9755

54
37
2a

0.987661148
0.913761049
0.967897457

0.00959256
0.02333067
0.01107653

9.2021E05
0.00054447
0.0001227

0.0095926
0.0233307
0.0110765

1.04
1.18
1.07

IPE220

GC
SC
GCf

434
434
434

1.24694296
1.05824239
1.18400133

0.19871971
0.09656007
0.18542829

1.0482232
0.9616823
0.998573

84
20
72

1.007288007
0.950230377
0.986495134

0.0187493
0.01041606
0.01009237

0.0003516
0.0001085
0.00010186

0.0187493
0.0104161
0.0100924

1.05
1.09
1.05

IPE500

GC
SC
GCf

434
434
434

1.27080958
1.07246544
1.19586786

0.21210634
0.10009339
0.18512526

1.0587032
0.972372
1.0107426

101
49
106

1.00301672
0.958348923
0.988275241

0.02075897
0.01268823
0.01372386

0.00043103
0.000161
0.00018836

0.020759
0.0126882
0.0137239

1.06
1.08
1.06

Load_1

GC
SC
GCf

177
177
177

1.11321734
0.99901427
1.11321734

0.13327633
0.06452302
0.13327633

0.9843
0.936
0.9843

6a
19a
6a

0.980505254
0.90836245
0.980505254

0.00331169
0.02368914
0.00331169

1.0967E05
0.00056133
1.0967E05

0.0033117
0.0236891
0.0033117

1.03
1.18
1.03

Load_0.5

GC
SC
GCf

195
195
195

1.25374846
1.08007851
1.20959543

0.19831888
0.12301298
0.20504021

1.0554296
0.978
1.0045552

34
4a
31

1.000895779
0.964489465
0.982802005

0.02463557
0.01320336
0.00986785

0.0006071
0.00017434
9.7379E05

0.0246356
0.0132034
0.0098678

1.08
1.08
1.05

Load_0

GC
SC
GCf

203
203
203

1.25824795
1.0692224
1.17320369

0.19962576
0.09253945
0.1832793

1.0586222
0.982
0.9899244

50
14a
25

1.002578614
0.970795415
0.975833699

0.02129611
0.01174105
0.0126769

0.00045363
0.00013786
0.00016072

0.0212961
0.011741
0.0126769

1.06
1.07
1.07

Load_0.5

GC
SC
GCf

195
195
195

1.29383646
1.07487307
1.16929629

0.23165482
0.10158946
0.19398138

1.0621816
0.9865
0.9865

52
5a
6a

1.013519105
0.977105127
0.977105127

0.02185546
0.00737268
0.00737268

0.00047778
5.4358E05
5.4358E05

0.0218555
0.0073727
0.0073727

1.05
1.05
1.05

Load_1

GC
SC
GCf

203
203
203

1.25450649
1.03476193
1.11477065

0.21060681
0.08034618
0.16596528

1.0438997
0.969
0.975

50
4a
1a

1.013372423
0.958009474
0.964223818

0.01222202
0.01297099
0.00814303

0.00014939
0.00016826
6.6311E05

0.012222
0.012971
0.008143

1.02
1.09
1.06

Load_conc

GC
SC
GCf

179
179
179

1.18841723
1.03560942
1.14223941

0.15171587
0.07141979
0.15091137

1.0367014
0.976
0.9935

35
13a
12a

0.999409523
0.954331502
0.987796795

0.01316907
0.02073073
0.00614095

0.00017344
0.00042986
3.7712E05

0.0131691
0.0207307
0.0061409

1.04
1.12
1.03

Load_dist

GC
SC
GCf

179
179
179

1.15478124
1.02383907
1.1368519

0.14629237
0.07368109
0.14885796

1.0084889
0.954
0.9947

29
18a
4a

0.996037813
0.928067499
0.990557312

0.00524172
0.02434184
0.00267002

2.7476E05
0.0005927
7.129E06

0.0052417
0.0243418
0.00267

1.02
1.16
1.02

Slend_Vlow

GC
SC
GCf

157
157
157

0.99386053
0.99386053
0.99386053

0.01136661
0.01136661
0.01136661

0.9824939
0.9824939
0.9824939

25
25
25

0.974284606
0.974284606
0.974284606

0.00924593
0.00924593
0.00924593

8.5491E05
8.5491E05
8.5491E05

0.0092459
0.0092459
0.0092459

1.06
1.06
1.06

Slend_low

GC
SC
GCf

161
161
161

1.02812526
0.98447786
0.99735833

0.03882568
0.01571417
0.02064445

0.99072
0.9687637
0.979

16a
22
12a

0.985859826
0.955549192
0.973193021

0.00454299
0.01136278
0.00532458

2.0639E05
0.00012912
2.8352E05

0.004543
0.0113628
0.0053246

1.03
1.08
1.04

Slend_med

GC
SC
GCf

665
665
665

1.23585221
1.02413349
1.12142602

0.16403321
0.0666092
0.11531677

1.071819
0.9575243
1.0061092

103
71
104

1.034332293
0.930541045
0.984429

0.0231975
0.02524549
0.01459488

0.00053827
0.00063754
0.00021303

0.0231975
0.0252455
0.0145949

1.04
1.16
1.06

Slend_High

GC
SC
GCf

348
348
348

1.37912131
1.14139475
1.35339194

0.16808192
0.10835544
0.16016238

1.2110394
1.0330393
1.1932296

71
46
62

1.166007574
1.002643238
1.148557207

0.03158324
0.02649604
0.03123922

0.000998
0.00070229
0.00097637

0.0315832
0.026496
0.0312392

0.94
1.08
0.96

S235

GC
SC
GCf

472
472
472

1.1858685
1.02664583
1.11526466

0.18031622
0.07036293
0.14725243

1.00555228
0.9562829
0.97519975

84
25
6a

0.991260561
0.941273873
0.966512282

0.01097289
0.01252572
0.01056144

0.00012041
0.00015691
0.00011155

0.0109729
0.0125257
0.0105614

1.04
1.10
1.07

S355

GC
SC
GCf

437
437
437

1.22862287
1.0507611
1.16123876

0.19481519
0.09339547
0.17451607

1.0338077
0.9573656
0.9867227

100
29
38

1.002178328
0.928987835
0.976522458

0.01394286
0.02464223
0.00848823

0.00019442
0.00060742
7.2053E05

0.0139429
0.0246422
0.0084882

1.04
1.16
1.05

S460

GC
SC
GCf

422
422
422

1.2481092
1.06405627
1.18358586

0.2072784
0.11051061
0.19338873

1.0408308
0.9535457
0.9901971

92
24
48

1.002985917
0.919894476
0.978825934

0.01705769
0.02895726
0.00974691

0.00029101
0.00083887
9.5007E05

0.0170577
0.0289573
0.0097469

1.05
1.19
1.05

Total

GC
SC
GCf

1331
1331
1331

1.21963952
1.04642463
1.15202065

0.19556195
0.09339283
0.17414002

1.0240776
0.9530318
0.9778806

280
71
57

0.997914857
0.928192464
0.968379396

0.01277661
0.02417035
0.00904198

0.00016326
0.00058438
8.1761E05

0.0127766
0.0241704
0.009042

1.04
1.16
1.06

When Nlin 20, a sub-set of 20 results is used.

Please cite this article in press as: Rebelo C, et al. Statistical evaluation of the lateraltorsional buckling resistance of steel I-beams, Part 1: Variability of the Eurocode 3
resistance model. Journal of Constructional Steel Research (2008), doi:10.1016/j.jcsr.2008.07.016

ARTICLE IN PRESS
C. Rebelo et al. / Journal of Constructional Steel Research (

13

Fig. 10. Mean values with vertical bars denoting 1 standard deviation of the ratio R plotted against the sub-sets defined by the Production Type (a), the Section Type (b),
Loading Type (c) and the Slenderness (d).
Table 5
Safety factors using the linearization of the lower tail of the distribution function; statistic details given in the Appendix
Partial safety factor rd

Sub-set

General case

Special case

General case/f

Fabrication

Rolled
Welded

1.04
1.07

1.09
1.17

1.06
1.04

Section

HEA500
IPE220
IPE500

1.04
1.05
1.06

1.18
1.09
1.08

1.07
1.05
1.06

Conc.
Distrib.

1.03
1.08
1.06
1.05
1.02
1.04
1.02

1.18
1.08
1.07
1.05
1.09
1.12
1.16

1.03
1.05
1.07
1.05
1.06
1.03
1.02

Slenderness

< 0.2
0.2 < 0.4
0.4 < 1.2
1.2

1.06
1.03
1.04
0.94

1.06
1.08
1.16
1.08

1.06
1.04
1.06
0.96

Steel grade

S235
S355
S460

1.04
1.04
1.05

1.10
1.16
1.19

1.07
1.05
1.05

1.04

1.16

1.06

Loading

Total

=1
= 0.5
=0
= 0.5
= 1

the General Case/f for I profiles with dimensions that fall outside
the range of the hot rolled sections.
The qualitative analysis in the first part of this paper consistently shows a higher resistance to lateraltorsional buckling for
the Special Case. Additionally, it is clear that this method exhibits
some unsafe results for intermediate values of the slenderness LT .
This fact is true for the choice of the parameters LT ,0 and made

here, within the limits of the permitted values and corresponding


to the recommended values in EC3.
The methodology presented to estimate the safety factor
rd , concerning the uncertainties in the resistance model, and
to evaluate the quality of the design formulae permits further
conclusions. When all the 1331 cases are considered, that is, no
linearization is performed (see Fig. 10), the Special Case method

Please cite this article in press as: Rebelo C, et al. Statistical evaluation of the lateraltorsional buckling resistance of steel I-beams, Part 1: Variability of the Eurocode 3
resistance model. Journal of Constructional Steel Research (2008), doi:10.1016/j.jcsr.2008.07.016

ARTICLE IN PRESS
14

C. Rebelo et al. / Journal of Constructional Steel Research (

produces R-values with lower mean and lower standard deviations


than the other two design procedures (Rm = 1.046; R =
0.093) and could be considered, therefore, the most accurate. The
General Case/f (Rm = 1.152; R = 0.174) procedure is more
accurate than the General Case procedure (Rm = 1.220; R =
0.196), exhibiting a lower mean and standard deviation.
When only a small number of the most non-conservative
cases in each sub-set are considered, the situation is considerably
different concerning the standard deviation. This implies that the
safety factor necessary to guarantee the safety level related to the
design models is, in general, much lower for the General Case
method in comparison with the Special Case. If a safety factor
rd = 1.1 is included in this design methodology, corresponding to
a shift of the probability curve to the right, the consequence is that
on average it becomes equivalent to the General Case/f method.
If the safety factor rd = 1.0 is to be maintained, the most suitable
design method would be then the General Case/f .
However, is not clear whether EN 1993-1-1 strictly allows the
use of the factor f for hot rolled or equivalent welded sections
with the General case. To overcome this difficulty the authors
LT ,0 = 0.2, = 1 and
recommend using the Special Case with
the same buckling curves as for the General Case. This methodology
was adopted by the Portuguese National Annex of EN 1993-1-1 and
corresponds to use of the General Case with the factor f for hot
rolled or equivalent welded sections.
Finally, it should be noted that the partial safety factor M (see
Eq. (13)) may still be safely taken as 1.0 for all methods (or even
lower for the General Case or the General Case/f ) if the statistical
distribution (mean and standard deviation) for the actual yield
stress of steel divided by the corresponding nominal values shows
a high mean and a low standard deviation. This will be explored in
Part 2 of this paper [12].
Appendix. Calculation of the safety factor RD concerning the
model uncertainties
Table A.1 summarizes the calculation of the safety factor rd
concerning the model uncertainties. In the third column are given
the total number of values N included in the sub-set defined in the
first column. The other columns include the mean value Rm and
standard deviation R of the sub-set. The number of values Nlin and
the corresponding mean value R m and standard deviation are
used to perform the computation of the safety factor rd according
to Eqs. (17)(20).

References
[1] Boissonnade N, Greiner R, Jaspart JP, Lindner J. Rules for member stability in EN
1993-1-1. Background documentation and design guidelines, ECCS Technical
Committee 8 Stability. Publication 119. ECCS; 2006.
[2] CEN, European Committee for Standardization. EN 1990:2002, Basis of
structural design, April 2002. Brussels. 2002.
[3] CEN, European Committee for Standardisation, EN 1993-1-1:2005, Eurocode
3: Design of steel Structures Part 1-1: General rules and rules for buildings.
Brussels (Belgium). 2005.
[4] ECCS European Convention for Constructional Steelwork, Technical Committee 8 Structural Stability, Technical Working Group 8.2, System, ultimate
limit state calculation of sway frames with rigid joints. 1st ed. 1984.
[5] CEN, European Committee for Standardisation. prEN 1090-2 Execution of Steel
Structures and aluminium Structures Part 2: Technical Requirements for the
execution of steel structures. 2007.
[6] Chen WF, Lui EM. Stability design of steel frames. CRC Press; 1991.
[7] Franssen J-M. SAFIR. A thermal/structural program for modeling structures
under fire. Engineering Journal, AISC 2005;42(3):14358.
[8] Lui EM, Ge M. Analysis and design for stability in the US an overview. Steel
and Composite Structures 2005;5(23):10326.
[9] Nethercot DA, Gardner L. The EC3 approach to the design of columns,
beams and beamcolumns. Steel and Composite Structures 2005;5(23):
127140.
[10] Rebelo C, Simes da Silva L, Vila Real PMM, Lopes N. Statistical evaluation of
the Eurocode 3 design rules for lateraltorsional buckling of I-beams. In: Mota
Soares CA et al., editors. III European conference on computational mechanics
solids structures and coupled problems in engineering. 2006.
[11] Sedlacek G, Muller C. The European standards family and its basis. Journal of
Constructional Steel Research 2006;62:104759.
[12] Simes da Silva L, Rebelo C, Nethercot D, Marques L, Simes R, Vila Real P.
Statistical evaluation of the lateraltorsional buckling resistance of steel Ibeams Part 2: variability of steel properties. Journal of Constructional Steel
Research (2008), doi:10.1016/j.jcsr.2008.07.017.
[13] Trahair NS, Bradford MA, Nethercot DA. The behaviour and design of steel
structures to BS5950. Spon Press; 2001.
[14] Trahair NS, Bradford MA. The behaviour and design of steel structures to AS
4100; 3rd ed. Australian. E&FN SPON, 1998.
[15] Vila Real P, Lopes N, Simes da Silva L, Piloto P, Franssen J-M. Towards a consistent safety format of steel beamcolumns: Application of
the new interaction formulae at ambient temperature to elevated temperatures. International Journal of Steel and Composite Structures 2003;3(6):
383401.
[16] Vila Real P, Lopes N, Simes da Silva L, Franssen J-M. Lateraltorsional buckling
of unrestrained steel beams under fire conditions: Improvement of EC3
proposal. Computers & Structures 2004;82:173744.
[17] Vila Real P, Lopes N, Simes da Silva L, Franssen J-M. Parametric analysis of the
lateraltorsional buckling resistance of steel beams in case of fire. Fire Safety
Journal 2007;42(67):41624.
[18] Vila Real PMM, Lopes N, Simes da Silva L, Rebelo C. Numerical validation of
the Eurocode 3 design rules for lateraltorsional buckling of I-Beams. In: Mota
Soares CA et al., editors. III European conference on computational mechanics,
solids, structures and coupled problems in engineering. 2006.
[19] Vrouwenvelder ACWM. Developments towards full probabilistic design codes.
Structural Safety 2002;24:41732.

Please cite this article in press as: Rebelo C, et al. Statistical evaluation of the lateraltorsional buckling resistance of steel I-beams, Part 1: Variability of the Eurocode 3
resistance model. Journal of Constructional Steel Research (2008), doi:10.1016/j.jcsr.2008.07.016

You might also like