Engineering Structures 199 (2019) 109624

Contents lists available at ScienceDirect

Engineering Structures
journal homepage: www.elsevier.com/locate/engstruct

Design of structural steel members by advanced inelastic analysis with strain T

limits
Andreas Fieber , Leroy Gardner, Lorenzo Macorini
⁎

Imperial College London, SW7 2AZ London, UK

ABSTRACT

Structural steel design is traditionally a two step process: first, the internal forces and moments in the structure are determined from a structural analysis. Then, a
series of design checks are carried out to assess the strength and stability of individual members. The structural analysis is typically performed using beam finite
elements, which are usually not able to capture local buckling explicitly. Instead, the assessment of local buckling and rotation capacity is made through the concept
of cross-section classification, which places class-specific restrictions on the analysis type (i.e. plastic or elastic) and defines the cross-section resistance based on
idealised stress distributions (e.g. the plastic, elastic or effective moment capacity in bending). This approach is however considered to be overly simplistic and
creates artificial steps in the capacity predictions of structural members. A more consistent approach is proposed herein, whereby a second-order inelastic analysis of
the structure or structural component is performed using beam finite elements, and strain limits are employed to mimic the effects of local buckling, control the
spread of plasticity and ultimately define the structural resistance. The strain limits are obtained from the continuous strength method. It is shown that not only can
local buckling be accurately represented in members experiencing uniform cross-sectional deformations along the length, but, by applying the strain limits to strains
that are averaged over a defined characteristic length, the beneficial effects of local moment gradients can also be exploited. The proposed method is assessed against
benchmark shell finite element results on isolated members subjected to bending, compression and combined loading. Compared to conventional steel design
provisions and even to existing advanced design approaches utilising second-order elastic analysis, the proposed design approach provides consistently more accurate
capacity predictions.

1. Introduction from the continuous strength method [3] and are used to mimic the
effects of local buckling. Hence, advanced inelastic analysis using beam
Beam finite elements are typically used for the structural analysis of finite elements with strain limits can be used to analyse and design
steel frames due to their computational efficiency and ability to accu- structures comprising cross-sections of any class, with failure defined as
rately capture the overall elastic-plastic load-deformation response of the lower of (1) the peak load factor reached during the analysis or (2)
the structure. However, local cross-section deformations (i.e. local in- the load factor at which the strain limit is first reached.
stabilities in either the elastic or inelastic range) are not accounted for The benefits of advanced analysis are widely recognised [4–10]. For
in conventional beam elements. Hence, to overcome this limitation, example, member stability is captured directly in the analysis, thereby
steel design specifications, such as EN 1993-1-1 [1], specify class-de- eliminating the need for member checks and effective lengths. Cur-
pendent restrictions on the cross-section resistance and analysis type rently, the most common approach is to carry out a second-order elastic
through the concept of cross-section classification. As such, plastic analysis with frame and member imperfections, followed by cross-sec-
analysis and design methods, which allow for the beneficial effect of tion checks. This avoids the need for the calculation of effective lengths
force and moment redistribution, are limited to structures composed of [11,12] and provides a more accurate representation of the distribution
compact (Class 1) cross-sections. The effect of strain hardening is gen- of forces and moments around the structure than is achieved through
erally ignored though. Structures composed of more slender cross-sec- first-order analysis due to the fact that the changes in stiffness arising
tions (i.e. non-compact (Class 2 and 3) or slender (Class 4)) must be from geometric nonlinearities are captured. However, by accounting for
analysed elastically and step-wise resistance functions are employed to plasticity (i.e. by performing a second-order inelastic analysis with
limit the capacity of the cross-sections and members. imperfections), yet greater accuracy is achieved. Furthermore, by in-
Recently, a more consistent design approach based on geometrically cluding strain limits, both the spread of plasticity throughout the cross-
and materially nonlinear structural analysis, also referred to as ad- sections and the level of plastic redistribution through the structure can
vanced inelastic analysis, using beam finite elements together with be controlled to a level that is appropriate for the slenderness of the
strain limits has been proposed [2]. The strain limits are determined constituent cross-sections. The accuracy of this design approach is

⁎
Corresponding author.
E-mail address: [email protected] (A. Fieber).

https://doi.org/10.1016/j.engstruct.2019.109624
Received 8 March 2019; Received in revised form 19 August 2019; Accepted 2 September 2019
Available online 03 October 2019
0141-0296/ © 2019 Elsevier Ltd. All rights reserved.
A. Fieber, et al. Engineering Structures 199 (2019) 109624

assessed herein against benchmark shell finite element models and [1].
conventional design according to EN 1993-1-1 [1] for structural steel Current design by advanced analysis is usually based on a second-
members subjected to bending, compression and combined loading. order elastic analysis (GNIA) with equivalent geometric imperfections
While individual members are the focus of the present paper, applica- used to account for the effects of initial member out-of-straightness and
tion of the proposed method to indeterminate systems and frames is residual stresses, followed by a cross-section check applied to the most
described in [2]. heavily loaded section [11,12]. No member buckling checks or effective
lengths are required. However, current design by GNIA still relies on
the concept of cross-section classification to determine the imperfection
2. Traditional steel design approaches
magnitude and the corresponding cross-section interaction surface.

Traditionally, the strength and stability of structural steel members

is verified through a series of design checks. Typically, only the limit 3. Benchmark shell finite element modelling
states that are not directly accounted for in the analysis need to be
checked. A key component to the design of steel members in many Benchmark shell finite element (FE) models, against which the
modern specifications is the classification of cross-sections. To facilitate
proposed beam FE design approach is later compared, are presented in
the design process, EN 1993-1-1 [1] specifies four cross-section classes, this section. Validation of the shell FE models against a wide range of
whereby the classification of a cross-section is based upon the most
experiments from the literature [17,20–28] on hot-rolled I-sections and
slender plate element within the cross-section. square/rectangular hollow sections (SHS/RHS) is first described. A
In the case of bending, Class 1 cross-sections can reach their plastic
thorough parametric study is then presented to assess the proposed
moment capacity Mpl and have sufficient rotation capacity to be used in design approach, as well as current EN 1993-1-1 capacity predictions,
plastic design. Class 2 cross-sections can also attain their full plastic
for isolated steel members subjected to bending, compression and
moment resistance, but are deemed to possess insufficient rotation ca- combined compression plus bending (see Section 5).
pacity for plastic design; structures composed of Class 2 cross-sections
must therefore be analysed elastically. Local buckling limits Class 3 and
4 cross-sections to their elastic Mel and effective Meff moment re-
3.1. Modelling approach
sistances respectively. For Class 4 cross-sections, effective section
properties are calculated on an element-by-element basis using the ef-
Shell FE models of individual structural members were developed
fective width method, without considering the effects of element in- using the finite element package Abaqus [29]. Four-noded shell ele-
teraction [13]. Typical moment-curvature relationships for the four
ments with reduced integration, denoted as S4R in Abaqus [29], were
classes of cross-section defined in EN1993-1-1 [1] are shown in Fig. 1. used. The modelled cross-sections were subdivided into 12 elements
The influence of the bending moment distribution on member sta-
along both the web depth and flange width. The number of elements
bility is accounted for in EN 1993-1-1 [1] through the use of equivalent along the member axis was defined such that the element aspect ratio
uniform moment factors. However, the beneficial effect of local mo-
was close to unity. The full load-deformation behaviour of the struc-
ment gradients on local buckling (i.e. on the cross-section strength) is
tural members was captured through geometrically and materially
not considered. This is despite the fact that experiments [14–17] have
nonlinear analyses with imperfections (GMNIA), which were solved
shown that members under moment gradients exhibit greater cross-
using the modified Riks arc length algorithm [30].
section resistances than those under uniform bending, and the fact that The characteristic yield plateau and strain hardening behaviour of
moment gradients are present in most practical applications. This
hot-rolled steel was modelled by the quad-linear material model de-
shortcoming is addressed in the proposed approach. veloped by Yun and Gardner [31] – see Section 4.2.2. The engineering
The flexural buckling resistance of compression members is tradi-
material stress–strain response was converted into true stresses and
tionally assessed using column curves. A reduction factor is applied to strains to account for the change in cross-sectional area captured by the
the cross-section strength (i.e. Npl for Class 1–3 and Neff for Class 4
shell finite elements during the geometrically nonlinear analysis. A
cross-sections) to account for member buckling in the presence of initial Poisson’s ratio = 0.3 was assumed in the elastic range, while the
bow imperfections and residual stresses. Column curves are simple to
Abaqus default value of = 0.5 was used in the plastic range. For the I-
use for isolated members with idealised boundary conditions. At the and H-shaped cross-sections, residual stresses were modelled explicitly
frame level though, the determination of the effective lengths of in-
as an initial stress condition, assuming the ECCS [32] distribution for
dividual members has been a source of uncertainty among practising hot-rolled sections as shown in Fig. 2. The residual stresses were al-
engineers [12,18,19]. Structural steel design specifications typically
lowed to equilibrate during a separate analysis step prior to loading.
account for the combined effects of compression and bending through Based on the experimental observations of Wang et al. [28], residual
interaction factors, such as those presented in Annex B of EN 1993-1-1
stresses were not included in the hot-rolled SHS/RHS models.
Pin and roller support conditions were achieved by coupling all
nodes of the member end cross-section to a master node located at the
centroid of the cross-section. Concentrated loads were applied uni-
formly over the depth of the web for the I-sections and over the depth of
both webs for the SHS/RHS. To prevent web crushing, web stiffeners
with a thickness equal to the web were modelled at the locations of
point loads and supports. For the I-sections, the web was coupled to the
flanges using *MPC, BEAM constraints, thereby accounting for the gap
of half the flange thickness between the two plates [33,34]. Sinusoidal
member out-of-straightness and local cross-section imperfections were
included in the shell FE models, as shown in Figs. 2 and 3 for I-sections
and SHS/RHS respectively. The half-wavelength Lb,cs of the local im-
perfections was predicted using the expressions provided in [35], with
the nearest integer number of half-wavelengths fitted into the in-
dividual member lengths.
Fig. 1. Typical moment-curvature responses for Class 1–4 cross-sections.

2
A. Fieber, et al. Engineering Structures 199 (2019) 109624

Fig. 2. Geometry, local geometric imperfections and residual stress pattern adopted in shell FE models of I-sections. Note that f y = 235 N/mm2, regardless of the
strength of the modelled material, in line with [32].

Fig. 3. Geometry and local geometric imperfections adopted in shell FE models of SHS/RHS.

3.2. Validation of shell FE models Fig. 4(a). The thicknesses of the fillet elements were defined such that
the total cross-sectional areas were equal to the reported values. For
The shell FE models were validated against the results of 52 ex- square and rectangular hollow sections, the rounded corner geometry
periments on hot-rolled I-sections and SHS/RHS from the literature. was modelled explicitly by means of five elements in each corner, as
Different loading configurations were considered, including beams in shown in Fig. 4(b).
three- and four-point bending [17,20–23], columns of varying slen- A summary of the shell FE validation, including the mean and
derness [28,24] and beam-columns [25–27]. coefficient of variation (CoV) of the shell FE to test failure load ratios, is
To compare the shell FE results with the experimental results, the given in Table 1. Generally, the shell FE models were able to accurately
measured cross-section geometry, material properties and imperfection predict the experimental failure loads for both I-sections and SHS/RHS
magnitudes were used during the validation of the models. If no mea- with an average FE-to-test ultimate load ratio of 0.98 and corre-
sured imperfection magnitudes were provided, the values from Section sponding CoV of 0.074.
3.3 were assumed. Similarly, if full stress-strain curves were not pro- Typical load-deformation curves obtained from tests and the shell
vided, the measured yield and ultimate stress values were used as input FE models for an I-section subjected to four-point bending [21] and a
parameters for the quad-linear material model described in Section SHS subjected to three-point bending [17] are shown in Fig. 5, where
4.2.2. Different material properties were applied to the flanges and web pl and pl are the elastic portions of the total rotation (three-point
when the corresponding experimental data was available. For the I- bending) or curvature (four-point bending) at midspan corresponding
sections, the fillets at the web-flange intersection were modelled by two to the plastic moment Mpl . The shell FE models accurately capture the
additional elements with increased thickness [25,34], as shown in complete load-deformation response for both the 10WF25 and the SHS

3
A. Fieber, et al. Engineering Structures 199 (2019) 109624

Fig. 4. Assumed cross-section geometry for shell FE models for the validation (top) and parametric (bottom) study respectively: (a) I-sections with and without fillets
and (b) SHS/RHS with and without rounded corners.

Table 1
Comparison of the experimental and numerical ultimate capacities for hot-rolled I-sections and SHS/RHS subjected to different loading configurations.
FE/ test

Reference Steel grade Load configuration No. of tests Mean CoV

I-sections
Byfield & Nethercot (1998) [20] FE430A Beam (4pb) 6 0.91 0.038
Lee & Galambos (1961) [21] ASTM A-7n Beam (4pb) 2 0.99 0.033
Sawyer Jr. (1961) [22] S235 Beam (3pb) 6 0.99 0.063
Lay (1965) [23] ASTM A441 Beam (3pb) 1 0.98 –
Sfintesco (1970) [24] S235 Column 6 0.91 0.073
Yun et al. (2018) [34] S235 & S355 Stub beam-columns 6 0.92 0.054
van Kuren & Galambos (1961) [26] ASTM A7 Beam-column 2 0.95 0.067
Aglietti et al. (1964) [27] A35 Beam-column 1 0.99 –
SHS/RHS
Wang et al. (2016) [17] S460 Beam (4pb + 3pb) 6 1.00 0.064
Wang & Gardner (2017) [28] S460 Column 16 1.05 0.039

Total 52 0.98 0.074

Fig. 5. Shell FE model validation against four-point bending test on a hot-rolled Fig. 6. Shell FE model validation against beam-column tests A-5 and A-7 on a
I-section reported by Lee [21] and three-point bending test on a hot-rolled SHS 4WF13 cross-section, as reported by van Kuren and Galambos [26].
reported by Wang et al. [17].

increased bending capacity due to the beneficial effect of the local

100 × 100 × 5 beam tests. Failure was governed by in-plane bending, moment gradient is captured by the shell FE models, as shown in Fig. 5.
and eventually inelastic local buckling, in both the experiments and the The normalised load-deformation curves of the beam-column ex-
corresponding shell FE models. In the case of three-point bending, the periments of van Kuren and Galambos [26] on a 4WF13 cross-section,

4
A. Fieber, et al. Engineering Structures 199 (2019) 109624

as well as the responses of the corresponding shell FE models, are imperfections were reported in [26]. A similar observation was made by
shown in Fig. 6. Following the loading sequence employed in the tests, Kucukler et al. [36].
the shell FE models were first loaded to a specified level of axial force;
the end moment at the base of the member was then increased in- 3.3. Benchmark modelling
crementally. The shell FE models accurately predict the test results,
with a mean FE-to-test ratio of 0.95 and CoV of 0.067. The small dis- The validated shell FE models were used to generate benchmark
crepancy between the FE and test results for specimen A-7 may result data, against which the accuracy of the proposed method of design by
from the assumed imperfection magnitude being greater than that of advanced inelastic analysis using beam finite elements could be as-
the test specimen, noting that no member out-of-straightness sessed. For the parametric study, the effects of the fillets were ignored

Table 2
Overview of key load cases and varied parameters considered during assessment of the proposed method of design by advanced inelastic analysis.

5
A. Fieber, et al. Engineering Structures 199 (2019) 109624

in the I-sections and the rounded corner geometry was neglected in the design approach that replaces the concept of cross-section classification
SHS/RHS, as shown in Fig. 4. This ensured that the geometric proper- by a continuous relationship between the cross-section slenderness and
ties of the shell FE models were directly comparable to the beam FE deformation capacity [3]. The CSM consists of two fundamental com-
models. The cross-section geometry of the beam FE models was defined ponents: (1) a base curve that defines the peak compressive strain that a
using the Abaqus command *BEAM SECTION, with the SECTION cross-section can endure csm and (2) a material stress-strain model. The
parameter set as I or BOX. The member out-of-straightness was assumed continuous strength method was initially developed to account for the
to be a half sine-wave in shape with a magnitude of l/1000 , where l is highly nonlinear material behaviour of stainless steel [3,37,38]. The
the member length. The recommended local plate imperfection from CSM has also been developed for the design of aluminium alloy struc-
Annex C of EN 1993-1-5 [13], having a periodic sinusoidal shape with a tural elements [39,40]. Recently, Yun et. al. [41,42] showed that the
half wavelength of Lb,cs and a magnitude of b/200 , where b is the width CSM also provides better capacity predictions than traditional design
(based on centreline dimensions) of the respective plate, was assumed. methods for hot-rolled carbon steel I-sections and SHS/RHS. Since the
For I-sections subjected to major axis bending, out-of-plane displace- extension of the base curve to slender cross-sections [43], the CSM now
ments were restrained at regular intervals, equal to the local buckling covers cross-sections of all four classes. The following subsections
half-wavelength, along the flange centrelines. The same modelling ap- provide relevant background information on the continuous strength
proach has been implemented successfully in previous studies [25,33]. method and explain how the CSM is incorporated into an advanced
An overview of the key load cases and the varied parameters con- analysis framework.
sidered for the assessment of the proposed method of design by advanced
inelastic analysis is provided in Table 2. For each case, three methods of
analysis and design were considered: (1) geometrically and materially 4.2.1. Base curve and cross-section slenderness ¯ p
nonlinear shell FE modelling with imperfections, serving as the benchmark The CSM base curve predicts the normalised peak compressive strain
against which the proposed and Eurocode approaches could be evaluated, csm / y that a cross-section of given local slenderness p , defined below,
¯
(2) the proposed approach of design by advanced inelastic analysis using can endure prior to failure, where y = fy /E is the yield strain. The base
beam finite elements with strain limits and (3) design using simplified curve has been calibrated against tests on stub columns and beams in
analysis with beam finite elements plus Eurocode 3 [1] cross-section and four-point bending and thus implicitly includes the influence of geo-
member checks. The beam lengths were varied to provide a range of metric imperfections and residual stresses at the cross-sectional level.
moment gradients defined by the ratio of the shear span Ls to the local The base curve, shown in Fig. 7, is split into two parts. For non-
buckling half-wavelength of the full cross-section Lb,cs . The lengths of the slender cross-sections with ¯ p 0.68, the CSM strain limit is defined by
columns and beam-columns were defined with reference to the non-di- Eq. (1), while for slender cross-sections with ¯ p > 0.68, local buckling
mensional slenderness ¯ y = Afy / Ncr,y and ¯z = Afy / Ncr,z for major and occurs prior to yielding and the CSM strain limit is given by Eq. (2).
minor axis buckling of Class 1 to 3 cross-sections, where Ncr,y and Ncr,z are Two upper limits are defined for the cross-section deformation capacity
the elastic critical buckling loads of the member about the major and csm / y in Eq. (1). The first limit on is a project specific design
minor axis, with corresponding effective buckling lengths Lcr,y and Lcr,z parameter that defines the permissible level of plastic deformation. For
respectively. For Class 4 cross-sections, the effective area Aeff was used to example, where extensive plasticity is tolerable at the ultimate limit
calculated the non-dimensional slenderness, in line with EN 1993-1-1 [1]. state and a suitably ductile steel is being used, a high value of say
Some common additional load cases were also considered, such as I-sec- = 30 could be specified. Conversely, for serviceability checks, in
tions subjected to uniformly distributed loads and minor axis bending. An which material yielding is not desirable, a value of = 1 could be used.
overview of all results is given in Table 5 in Section 5.5. A value of = 15 is generally recommended to prevent excessive de-
formations and to ensure that the material ductility requirements of EN
1993-1-1 [1] are satisfied. The second limit of C1 u / y prevents over-
4. Design by advanced inelastic analysis using beam finite
predictions of material strength when the simplified CSM resistance
elements with strain limits
functions are used [41,44].

4.1. Introduction
csm 0.25 C1 u ¯p
= but , for 0.68
¯ p3.6 (1)
A novel approach for structural steel design based on geometrically and y y

materially nonlinear analysis including imperfections, also referred to as

GMNIA or advanced inelastic analysis, with strain limits has recently been
proposed [2] and is developed further herein. In the proposed approach, the
continuous strength method (CSM) [3] strain limits are used to mimic local
buckling in beam finite elements and hence, structures with cross-sections of
all classes can be designed using advanced analysis; previously this has only
been possible with more sophisticated shell FE models. In conjunction with
a standardised material model for hot-rolled steel (see Section 4.2.2), the
strain limits control the spread of plasticity through the depth of a cross-
section, define the level of moment redistribution within a structure, and
provide a rational means to exploit the beneficial effects of strain hardening.
Additionally, the strain limits eliminate the need for effective section
property calculations in the design of Class 4 cross-sections. Failure is de-
fined as either (1) the peak load factor reached during the analysis or (2) as
the point at which the CSM strain limit is reached, whichever occurs first.
Details of the proposed method of design by advanced inelastic analysis are
outlined in the following subsections.

4.2. The continuous strength method (CSM) and strain limits Fig. 7. The CSM base curve (shown for = 15) provides a continuous re-
lationship between the cross-section slenderness ¯ p and the deformation ca-
The continuous strength method (CSM) is a deformation based pacity expressed in terms of the normalised peak compressive strain csm / y .

6
A. Fieber, et al. Engineering Structures 199 (2019) 109624

0.222 1 the stress corresponding to the intersection of the third and fourth
csm
= 1 for ¯ p > 0.68 segments of the material model, as shown in Fig. 8. Throughout the
¯ p1.05 ¯ p1.05
y (2) present study, grade S355 steel is used and represented by the quad-
The cross-section deformation capacity csm / y depends on the material linear material model, with a Young’s modulus E = 210000 MPa, a yield
properties and the cross-section slenderness. The cross-section slen- stress fy = 355 MPa and an ultimate stress fu = 510 MPa.
derness ¯ p is a dimensionless parameter that quantifies susceptibility to E for y
local instability and is defined by Eq. (3), where fy is the yield stress and fy for <
y sh
cr,cs is the elastic local buckling stress of the full cross-section, in-
= fy + Esh ( sh ) for < C1
cluding the beneficial effects of element interaction. sh u
fu fC1 u
fC1 + ( C1 u) for C1 <
fy u u C1 u u u
(4)
¯p =
cr,cs (3) fy
sh = 0.1 0.055 but 0.015 sh 0.03
Various methods exist to calculate the elastic critical buckling stress fu (5)
of a cross-section. Numerical methods (e.g. the finite strip method
performed in software such as CUFSM [45]) can fully capture the web- fy
u = 0.6 1 but u 0.06
flange interaction and provide the most accurate prediction. Alter- fu (6)
natively, plate interaction can be accounted for through modified
buckling coefficients k developed for standard steel profiles subjected to fu fy
Esh =
pure compression or pure bending [46]. However, most structural C2 u sh (7)
members are subjected to a combination of loads. For this reason, the
C1 and C2 are experimentally calibrated material constants required to
authors have recently developed closed-form expressions to predict the
define the material model, and are given by Eq. (8) and (9) respectively.
full cross-section local buckling stress of standard steel profiles sub-
jected to compression, bending and combined loading [47]. The un- sh + 0.25( u sh )
C1 =
derlying concept is that the buckling stress of the full cross-section lies u (8)
between the buckling stresses of the isolated critical plates with simply-
sh + 0.4( u sh )
supported and fixed boundary conditions. Conservatively, standard C2 =
(9)
plate theory can be used to estimate the cross-section slenderness based u

on the most slender isolated plate of the cross-section with simply- Note that the second strain limit specified in Eq. (1) confines the
supported boundary conditions, though this approach is not re- cross-section response to the first three segments of the quad-linear
commended. Herein, the functions developed by the authors [47] have material model [41,48]. This limit exists primarily to simplify the CSM
been used to predict the cross-section local bucking stress cr,cs and resistance functions for hand calculations. However, this limit is not
hence the cross-section slenderness ¯ p . required if the full stress-strain curve (i.e. all 4 stages) is used in the
advanced analysis, though may be retained to avoid excessively high
4.2.2. Material model strains from developing.
A quad-linear material model has been developed by Yun and
Gardner [31] to represent the stress-strain response of hot-rolled carbon 4.3. Application of CSM strain limits to advanced inelastic analysis
steel in a standardised manner. The model has been calibrated against
coupon test data collected from around the world and uses derived 4.3.1. Location for application of strain limits
functions to define the yield plateau length and the strain hardening Application of the CSM design philosophy lends itself to an ad-
behaviour, as shown in Fig. 8. Only three commonly available para- vanced inelastic analysis framework, since numerical integration is
meters are required to define the full stress-strain curve in the devel- performed at each increment of the analysis. Through the aforemen-
oped model: the modulus of elasticity E, the yield stress fy and the ul- tioned material model and application of the strain limits, it is possible
timate stress fu . The stress-strain relationship is defined over the full to define the CSM cross-section and deformation capacities directly in
strain range by Eq. (4), where sh is the strain at which strain hardening the analysis. The method was implemented herein using linear
commences defined by Eq. (5), u is the ultimate strain defined by Eq. Timoshenko beam finite elements, designated as B31 and B31OS [29],
(6), Esh is the strain hardening modulus defined by Eq. (7), and fC1 u is in Abaqus (see Section 5 for further details). However, in general, the
method can be implemented in any structural analysis software capable
of allowing for both geometrical and material nonlinearity, provided
the beam element formulation is compatible with the studied problem,
the mesh is suitably refined (see Section 4.4) and outer-fibre strains can
be extracted.
The determination of the CSM cross-section capacity for I-sections
and SHS/RHS is shown schematically in Fig. 9. The strain distribution
at failure is defined by the peak compressive strain that a given cross-
section can endure csm , which is obtained from the CSM base curve.
The assumed material model defines the corresponding stress dis-
tribution, from which the cross-section capacity can then determined by
integrating the stresses over the area of the cross-section. CSM re-
sistance functions have been developed for a range of cross-section
shapes [41,42,49,50], but for many cross-sections, the exact analytical
expressions derived by explicit integration become overly complicated
and hence simplified expressions are used for design purposes [41].
Fig. 8. Quad-linear material model developed by Yun and Gardner [31] to Such simplifications are of course not required if numerical integration
represent the yield plateau and strain hardening behaviour of hot-rolled carbon is performed, as is the case in an advanced plastic zone (fibre) analysis
steel in a standardised manner. using beam elements.

7
A. Fieber, et al. Engineering Structures 199 (2019) 109624

Fig. 9. Application of the CSM strain limit ( csm at the extreme fibre or csm,cl at the centreline of the section wall thickness) to the compatible design strain ( Ed at the
extreme fibre or Ed,cl at the centreline of the section wall thickness), as well as the corresponding stress distribution at failure, for (a) I-sections and (b) SHS/RHS; the
case of pure major axis bending of compact cross-sections benefiting from some strain hardening is shown here.

In the calculation of CSM cross-section resistances, the strain limit csm is

applied to the peak compressive strain within a cross-section csm (i.e. the
strain at the extreme fibre). However, beam finite element section points, at
which individual strain outputs may be obtained, are usually located along
the centreline of the wall thickness of the cross-section. In most cases, the
centreline of the wall thickness is located at a distance of half the plate
thickness away from the extreme fibre. This issue is found to be negligible,
provided compatibility between the strain limit and the considered output
strain is satisfied. For example, if the cross-section slenderness is defined
with reference to the stress at the extreme fibre, the corresponding strain
limit should be applied to the extreme fibre strain. Alternatively, the cen-
treline strain limit csm,cl can be applied to the compatible centreline design
strain Ed,cl , as shown in Fig. 9. The difference between the extreme fibre
and centreline approach is negligible, since the marginally lower strain at
the centreline membrane is offset by a corresponding reduction in strain
limit. Note that in the case of pure compression, as well as I-sections sub-
jected to minor axis bending, the extreme fibre and centreline strains are
equal. It follows that in an advanced inelastic analysis framework, the cross- Fig. 10. Illustrative example of the proposed method of design by advanced
section can withstand the required strain demands if the design strain is less inelastic analysis on a simply-supported IPE 180 beam subjected to four-point
than or equal to the CSM strain limit, as given by Eq. (10) for the extreme bending: the shell FE model directly captures local buckling and reaches a peak
fibre strain or by Eq. (11) for the centreline strain. load, while the beam FE model, without applied strain limits, follows the path
defined by the stress-strain relationship of the material model. Application of
Ed
1.0 for extreme fibre strains the CSM strain limit mimics local buckling in the beam FE model and accurately
csm (10) predicts the bending and deformation capacities of the beam.

Ed,cl
1.0 for strains at centreline of wall thickness
csm,cl (11) strain limit for an IPE 180 of grade S355 steel subjected to pure bending
is 14.3 y . Application of this strain limit accurately predicts the point of
failure at a moment capacity of 1.08Mpl in the beam FE model and also
4.3.2. Illustrative example provides a good estimation of the shell FE deformation capacity, as
The benefits of including the CSM strain limits into a beam FE ad- shown in Fig. 10. For comparison, EN 1993-1-1 conservatively predicts
vanced inelastic analysis are best demonstrated through an illustrative failure at the plastic moment capacity for this Class 1 cross-section, and
example. The normalised load-displacement response of a simply-sup- defines its rotation capacity as infinite.
ported IPE 180 beam subjected to four-point bending is shown in Fig. 10
for both the shell and beam FE models. The shell FE model undergoes
extensive plasticity and strain hardening, and reaches a peak moment of 4.4. Strain averaging approach
1.11Mpl . Failure is caused by inelastic local buckling, recalling that out-
of-plane failure has been prevented by the provision of lateral restraints. The CSM base curve has been calibrated based on experimental results
The beam FE model on the other hand cannot capture local buckling and on cross-sections under uniform compression (i.e. stub columns) and
the cross-section response in bending follows that defined by the con- uniform bending (i.e. the central region of laterally restrained beams
stitutive behaviour of the material. As a result, no peak load is reached under four-point bending). These loading configurations represent the
during the analysis and, without supplementary cross-section checks, the most severe case of uniform strain along the length of the member. In
ultimate moment capacity would be grossly overestimated. The CSM other cases, the local stability of the cross-section is enhanced by the

8
A. Fieber, et al. Engineering Structures 199 (2019) 109624

Fig. 11. Schematic representation of the strain averaging approach for the case of n = 2 — the CSM strain limit csm is applied to the strain averaged over the two
elements that lie entirely within the local buckling half-wavelength Lb,cs (i.e. Ed,Lb = ( 1 + 2 )/2 ).

presence of a strain gradient (i.e. the local moment gradient along the in the finite element model. For this reason, the maximum element length
member length). To account for the beneficial effects of local moment must be less than or equal to the local buckling half-wavelength (i.e. n 1).
gradients, the CSM strain limit can be applied to strains that are averaged The effects of moment gradients can be conservatively ignored by applying
over a characteristic length along the member, rather than simply (con- the strain limit to the peak strain – note that the capacity predictions are still
servatively) to the peak strain. The length over which the strains are affected by the element size though, as discussed in Section 4.4.2.
averaged in the proposed approach is the local buckling half-wavelength
Ed,Lb
of the full cross-section Lb,cs (aligning with the treatment of local buckling 1.0
csm
on a full cross-section basis in the stress domain in the definition of local n
1
slenderness), since local buckling is the limit state controlling the re- where Ed,Lb = n i and n 1
sistance of the cross-section in both the elastic and inelastic regimes. The 1 (12)
local buckling half-wavelength is also a distinct calculateable character- The application of the strain averaging approach is shown sche-
istic of a cross-section under a given loading condition. A similar concept matically in Fig. 11 for the case where n = 2 . Assuming that first-order
has been explored by Lay and Galambos [51,52], who argued that inelastic beam finite elements with a single integration point located at the
local buckling in Class 1 cross-sections (i.e. the formation of plastic hinges) centre of the element are used, the continuous strain distribution is
occurs when the compression flange has fully yielded along a length equal approximated by the beam FE model in step-wise increments as shown
to the inelastic local buckling half-wavelength for the case of uniform in Fig. 11. The considered member is able to withstand the applied
moment. Likewise, for the case of inelastic local buckling occurring ad- design loads if the averaged strain over the local buckling half-wave-
jacent to a fixed boundary (e.g. in a cantilever), a region equal to the full length Ed,Lb = ( 1 + 2 )/2 is less than the CSM strain limit csm . Note
inelastic local buckling wavelength is required to yield. that the strain contribution 3 from element 3 is not considered in the
The strain averaging approach is an extension to the cross-section strain averaging procedure, since element 3 lies partially outside Lb,cs . Cur-
check given in Eq. (10) and (11). Instead of considering the peak com- rently, the strain averaging approach is only applicable to uniform
pressive strain along the member, the strain contributions i of all n ele- members, but further research on tapered members is underway.
ments located completely within the critical local buckling half-wavelength The location of the critical strain averaging length typically corre-
Lb,cs are considered. Assuming equal length elements, the strain averaging sponds to the region with the highest bending moment Mmax . Note
approach is expressed in Eq. (12), where Ed,Lb is the averaged design strain. though, at the system level, the beneficial effect of high local moment
If the length of elements located within Lb,cs are not equal, the weighted gradients in some regions may result in lower stressed regions under
average strain must be determined. The strain averaging approach also more uniform bending being critical. Note also that the presence of
reduces the sensitivity of strength predictions to the mesh density employed stiffeners locally constrains the shape of the cross-section and hence the

Fig. 12. Bending moment diagrams, strain distributions and typical locations of critical strain averaging regions for (a) cases with web stiffeners and (b) cases
without web stiffeners.

9
A. Fieber, et al. Engineering Structures 199 (2019) 109624

4.4.1. Local buckling half-wavelength

The local buckling half-wavelength of a full cross-section Lb,cs can be
obtained using numerical methods such as finite strip analysis (e.g. CUFSM
[45]) or approximate analytical expressions [35]. The expressions pre-
sented in [35] are based on the same concept as used to predict the full
cross-section local buckling stress [47] and are hence used herein as they
require minimal additional calculation effort. Note that a conservative
prediction of the full cross-section local buckling stress tends to lead to an
over-prediction of the associated half-wavelength [35]. However, this over-
prediction of Lb,cs that extends the region over which strains are averaged is
off-set by the corresponding reduction in strain limit due to the higher local
slenderness arising from the lower local buckling stress. It has been found
that bending capacity predictions are not overly sensitive to fluctuations of
± 25% in the averaging length, with the effects being smallest for cross-
sections with 0.4 ¯ p 0.55. This tolerance is well within the accuracy
level of the predictive half-wavelength expressions [35], and thus does not
affect the bending capacity predictions significantly.
Fig. 13. Normalised bending capacity for IPE 200 beams subjected to different
levels of local moment gradient. 4.4.2. Element size requirements
The finite element (FE) method is based on the discretisation of a
local buckling half-wavelength is located to either side of the stiffener. continuous system into small units, whose individual response can ea-
The bending moment diagrams, strain distributions and location of the sily be defined. Clearly, the FE mesh density directly influences the
critical averaging length for common design examples are shown in accuracy of the predicted response. While Eurocode 3 [1,13] provides
Fig. 12. In uniform members with web stiffeners, the critical averaging detailed guidance on the imperfections to be included in a structural
length Lb,cs is usually adjacent to Mmax , in the region with the shallower analysis, specific guidance is not currently provided on the meshing
moment gradient. If no web stiffeners are used, the averaging length requirements of FE analysis. Annex C.3(1) of EN 1993-1-5 [13] merely
Lb,cs is centred around the maximum moment. states that a convergence study should be undertaken to establish the
The influence of the local moment gradient on the ultimate bending mesh size. The proposed strain averaging approach explicitly defines
capacity of a series of IPE 200 beams is shown in Fig. 13. The moment the maximum permissible element length as the local buckling half-
gradient was controlled by varying the member length. The shell FE models wavelength Lb,cs of the cross-section, since the strain averaging ap-
capture an increase in bending capacity of around 15% as the length of the proach cannot be implemented with larger element sizes.
shear span Ls is decreased from 35Lb,cs to 5Lb,cs . Similar benefits are pre- The effect that the element size has on the bending capacity pre-
dicted by the beam finite element models in conjunction with the proposed diction of beams subjected to moment gradients is demonstrated with
strain averaging approach. EN 1993-1-1 on other hand ignores the effects of reference to an IPE 160 cross-section subjected to three-point bending
local moment gradients and predicts a constant bending capacity for all in Fig. 14, where Mu is the ultimate moment capacity predicted by the
members. Note that, in the comparisons presented in Fig. 13, the positive different design approaches. Some account for the effects of local mo-
benefit of high local moment gradients outweighs the negative influence of ment gradients is often unintentionally taken in traditional design when
high shear, though this is not always the case, as discussed in Section 4.5. the design moments are obtained from FE analysis, since the element

Fig. 14. Strain averaging approach and effect of finite element size on the capacity prediction of an IPE 160 cross-section in three-point bending. Provided the beam
element length is less than Lb,cs , accurate and consistent resistance predictions are achieved.

10
A. Fieber, et al. Engineering Structures 199 (2019) 109624

integration point is typically located at the mid-point of elements (i.e. at

a distance of up to half the element length away from the true peak
moment). With decreasing element length, the EN 1993-1-1 capacity
prediction approaches a lower bound, which, in this case, can be ver-
ified through hand-calculations. The strain averaging approach pro-
vides a more rational and consistent method to account for local mo-
ment gradient effects. For the considered IPE 160 beams, the local
buckling half-wavelength is Lb,cs = 138 mm. As the element length de-
creases, the strain contributions of more elements are included in the
strain averaging procedure given by Eq. (12). The resulting envelope
yields the most accurate capacity predictions when integer number of
elements fit into the local buckling half-wavelength Lb,cs . From Fig. 14,
it may be clearly seen that provided the beam element length is less
than Lb,cs , accurate and consistent resistance predications are achieved.

4.5. Influence of high shear forces

The presence of moment gradients inherently implies that shear is Fig. 15. Effect of shear on the bending capacity of a series of IPE 140 beams.
present. High shear has a negative influence on the bending capacity of
structural members and appropriate cross-section strength checks to low values of Ls / Lb,cs in Fig. 15). Application of the shear reduction
should be made. According to EN 1993-1-1 [1], cross-sections must factor csm leads to reduced bending capacity predictions in the mo-
satisfy a shear capacity check given by Eq. (13), where VEd is the design ment-shear interaction region and accurately replicates the shell FE
shear force and Vpl,Rd is the plastic shear capacity of the cross-section. model behaviour.
VEd The normalised moment-shear interaction diagram for beams of
1.0 varying length with IPE cross-sections is shown in Fig. 16, where Vu is
Vpl,Rd (13)
the ultimate shear capacity of the cross-section. It can be seen that the
EN 1993-1-1 [1] accounts for the interaction between shear and EN 1993-1-1 moment-shear interaction curve is conservative in
bending by reducing the cross-section bending capacity in cases where bending-dominated cases with Vu < 0.5Vpl , primarily because the ben-
the design shear force VEd exceeds half the plastic shear capacity Vpl,Rd . eficial effects of local moment gradients and strain hardening are ne-
The shear-reduced moment capacity MV,Rd is calculated using a reduced glected. Likewise, EN 1993-1-1 is conservative in the moment-shear
yield stress (1 ) fy for the shear area of the cross-section, where is a interaction region, as shown in Fig. 16. Similar conclusions were drawn
reduction factor given by Eq. (14). from experimental and numerical results on slender channels and Z-
2 sections experiencing shear buckling [53,54]. Note that in some cases,
2VEd
= 1 the shell FE models underwent large shear deformations and no peak
Vpl,Rd (14) load was reached. In these cases, failure was defined as the load at
In the proposed approach, the interaction between bending and which the tangent stiffness on the load-deformation curve of the beam
shear is accounted for through a reduction factor csm applied to the degraded to 1% of the elastic value, similar to the approach proposed
strain limit csm . The CSM shear reduction factor csm is based on the by dos Santos et al. [55].
familiar EN 1993-1-1 shear reduction factor and is defined by Eq. Cross-sections are able to resist significant shear even when the
(15). When the design shear force VEd is less than half the plastic shear applied moment is equal to or greater than the plastic moment re-
capacity, csm = 1 and no reduction is applied. For members subjected sistance of the cross-section [56], as shown by the results in Fig. 16.
to high shear (i.e. VEd > 0.5Vpl,Rd ), the shear-reduced strain limit is equal This is attributed primarily to the effective increase in yield strength
to csm csm and the strain averaging check is given by Eq. (16). It should under multi-axial stress conditions predicted by the von Mises yield
be emphasised that the strain averaging check given by Eq. (16) is criterion vM [57] in Eq. (17), where 11 is the longitudinal normal
based on the longitudinal compressive strains resulting from axial
compression and/or in-plane bending. A separate shear check given by
Eq. (13) is also still required in the proposed design approach. Simi-
larly, shear buckling must also be considered in cross-sections with
slender webs.

1 for VEd 0.5Vpl,Rd

csm = 0.5
for VEd > 0.5Vpl,Rd
0.5 + (15)

Ed,Lb
1.0 for VEd > 0.5Vpl,Rd
csm csm (16)
The influence of shear on the normalised ultimate bending capacity
Mu/ Mpl of an IPE 140 cross-section is shown Fig. 15, where beams of
varying length are considered. As the member length decreases, the
bending capacity of the shell FE models initially increases due to the
beneficial effects of moment gradients, as described in Section 4.4.
Eventually, the detrimental effects of shear begin to dominate the beam
behaviour and the bending capacity of the cross-section is reduced.
Application of the CSM strain limits to beam finite elements without Fig. 16. Moment-shear interaction for IPE beams predicted by shell FE mod-
consideration of the effects of shear leads to an over-prediction of the elling, beam FE modelling with CSM strain limits applied to strains averaged
shell FE model bending capacity in the high shear range (corresponding over Lb,cs and EN 1993-1-1.

11
A. Fieber, et al. Engineering Structures 199 (2019) 109624

Fig. 17. Explanation of how cross-sections can exceed their plastic moment resistance and resist significant shear forces when they have reached their plastic moment
resistance with reference to the midspan cross-section of an IPE 80 beam subjected to three-point bending: (a) longitudinal normal stress 11, transverse normal stress
22 and shear stress 12 distributions through the depth of the web at midspan and (b) trace of the multi-axial stress paths at monitoring points A, B and C, showing that
significant shear stresses can develop even if 11 = fy , provided that 22 > 0 . Note that no strain hardening is included in this shell FE model.

stress, 22 is the transverse normal stress and 12 is the shear stress. Si- the cross-section, thereby resulting in a redistribution of stresses. The
milar conclusions were also drawn by Gonçalves et al. [58], who red lines along the von Mises yield surface in Fig. 17(b) intersect at the
compared the results from shell finite element analyses with analytical point at which the multi-axial stress path at point B first reaches the von
solutions. Mises yield surface; this in fact occurs slightly below the point at which
the longitudinal normal stress reaches fy . As loading increases further,
2 2 2
vM = 11 11 22 + 22 +3 12 fy (17) the multi-axial stress path follows the surface of the von Mises yield
criterion (since no strain hardening is specified). The different starting
The effective increase in yield strength under multi-axial stress
locations of the stress paths in Fig. 17(b) are due to the different re-
conditions is explained in Fig. 17 with reference to an IPE 80 beam
sidual stress magnitudes at points A, B and C.
subjected to three-point bending. The longitudinal normal stress 11,
transverse normal stress 22 and shear stress 12 distributions at ultimate
load through the depth of the web at midspan are shown in Fig. 17(a). 4.6. Geometric member imperfections and residual stresses
The corresponding section moment is equal to 1.01Mpl and the section
shear force is 0.36Vpl . Considering the individual stress contributions in The load carrying capacity of columns is affected by member im-
Fig. 17(a), it can be seen that the longitudinal normal stress 11 at point perfections, particularly initial out-of-straightness and residual stresses.
B is greater than the uni-axial yield stress fy = 355 MPa ( 11 = 1.07fy ), Advanced analysis is able to directly capture the flexural buckling be-
even though no strain hardening is specified in the analysed shell FE haviour of columns provided appropriate imperfections are defined.
model. This is explained by the fact that at point B a transverse normal Member imperfections can be readily accounted for by directly mod-
stress is present, which allows additional 11 and/or 12 stress con- elling the imperfect geometry or by applying a set of equivalent hor-
tributions to develop prior to reaching the von Mises yield surface. The izontal forces. Explicit modelling of residual stresses is possible though
origin of the transverse normal stresses relates to the presence of the less practical. To overcome this issue, EN 1993-1-1 [1] allows and fa-
flanges, which provide restraint to transverse Poisson expansion in the cilitates the use of equivalent bow imperfections that implicitly account
top (longitudinally compressed) part of the beam and transverse for the combined effects of initial member out-of-straightness and re-
Poisson contraction in the bottom (longitudinally tensioned) part of the sidual stresses in a second-order analysis.
beam, inducing transverse stresses. The more substantial the flanges, Revised equivalent imperfections for inclusion in the upcoming
the greater the transverse stresses that can develop, as observed by Lee version of EN 1993-1-1 have recently been proposed by Lindner et al.
et al. [56]. Conversely, at point A, 22 and 12 are very low and the [59]. The normalised imperfection magnitude e0 / l is given by Eq. (18),
longitudinal normal stress 11 is limited to fy , while at point C, both 11 where l is the member length, is the imperfection factor for the as-
and 22 are very low and 12 is approaching the pure shear yield stress sociated Eurocode 3 column buckling curve, = 235/ fy accounts for
fy / 3 . the material strength and is the elastic reference bow imperfection
The above described behaviour can also be observed by tracing the equal to = 1/110 for major axis buckling and = 1/200 for minor axis
multi-axial stress paths at points A, B and C under increasing loading, as buckling. These imperfection values are used in the present study to
shown in Fig. 17(b). The three-dimensional von Mises yield surface is account for the detrimental effects of geometric member imperfections
also shown in Fig. 17(b) for positive 11 and 22 values. The stresses at and residual stresses on member stability. It should be re-emphasised
point A primarily develop in the 11 direction (i.e. arising from that at the cross-section level (i.e. for local buckling), the effects of
bending), with minimal shear stresses present at failure. Conversely, residual stresses and local geometric imperfections are accounted for by
shear stresses dominate the behaviour at point C. Point B experiences the CSM strain limits.
substantial normal (both 11 and 22 stresses) and shear stresses, en-
e0
abling 11 values greater than the uni-axial yield stress fy , as shown in =
(18)
l
the plan view and side elevation of the von Mises yield surface in
Fig. 17(b). Note that the change in gradient of the multi-axial stress The equivalent imperfections proposed by Lindner et al. [59] are
paths at the monitoring points are due to yielding at other points within intended to be used with second-order elastic analysis and a linear

12
A. Fieber, et al. Engineering Structures 199 (2019) 109624

Fig. 18. Cross-section interaction diagram showing

column capacity predictions using different ana-
lysis methods and different member bow im-
perfection magnitudes. Use of elastic imperfection
magnitudes set out in [59] in GMNIA may be seen
to provide accurate but safe-sided predictions of
column buckling resistance relative to the applica-
tion of EN 1993-1-1 buckling curves.

cross-section design interaction formula. Depending on the cross-sec- stresses and strains of the quad-linear material model described in
tion classification, the bending end point for the cross-section interac- Section 4.2.2 were used since in the beam element models the cross-
tion formula is either the plastic (for Class 1 or 2) or elastic (for Class 3) sectional area of the beam finite elements remained constant during the
moment capacity (Mpl and Mel respectively), and different member analyses.
imperfection magnitudes apply. The imperfections for use with the
linear plastic interaction formula are larger than those for use with the
linear elastic interaction formula in order to account for the loss in 5.1. Cross-sections subjected to bending or compression
stiffness and additional second-order deformations from material
yielding that occur in reality but are not captured in a second-order To systematically assess the proposed design approach at the cross-
elastic analysis. However, the spread of plasticity through the cross- sectional level, a series of I-sections and SHS/RHS of varying local
sections and along the members is explicitly captured during an ad- slenderness ¯ p have been analysed under pure major and minor axis
vanced analysis that features both geometric and material non- bending and pure compression, as shown in Fig. 19 and 20. Note that
linearities, as employed in the proposed approach. Consequently, the the accuracy of the continuous strength method for the prediction of
imperfection magnitudes for elastic cross-section checks are used cross-section resistances has already been verified extensively against
throughout the present study. It is possible to back-calculate the re- test and shell FE results [41,42]. The primary objective herein was to
quired equivalent imperfection magnitude from the Eurocode 3 buck- show that the same cross-section capacity can be predicted in an ad-
ling curves, but this calculation depends on the column slenderness, vanced inelastic analysis using beam finite elements by limiting the
which in turn depends on the effective length of the member. Using the peak compressive strain in the cross-section to the CSM strain limit, as
back-calculated imperfection magnitudes would thus lose one of the given by Eq. (10) for the extreme fibre or Eq. (11) for the centreline
fundamental benefits of using advanced analysis – i.e. the elimination strains. The results are shown in Fig. 19 for I-sections and Fig. 20 for
of effective lengths; back-calculated imperfection magnitudes are SHS/RHS, where Mpl,y and Mpl,z are the plastic moment capacities about
therefore not considered herein. The influence of different imperfection the major and minor axes, Mu,y and Mu,z are the ultimate bending re-
magnitudes and analysis types on the predicted major axis buckling sistances about the major and minor axes predicted by the different
resistance of an HEB 240 column is shown in Fig. 18. design approaches, Npl = Afy is the plastic compression capacity, Neff is
the effective compression capacity of Class 4 cross-sections and Nu is the
ultimate compression resistance of a cross-section predicted by the
5. Application of proposed approach different design approaches.
Overall, the proposed method of design by advanced inelastic ana-
The accuracy of the proposed method of design by advanced in- lysis is shown to accurately predict the cross-section capacity of the
elastic analysis with strain limits is assessed in this section for isolated shell FE models. The influence of strain hardening is captured in stocky
members subjected to bending, compression and combined loading. cross-sections with ¯ p < 0.4 , with an increase in bending capacity of up
The capacity predictions obtained from beam FE advanced analyses to 20% compared to Eurocode 3 when is set to 30; setting = 15
with strain limits are compared against those obtained from the yields a benefit of around 5% for Class 1 cross-sections. The gradual
benchmark shell FE models described in Section 3.3 and conventional decrease in cross-section capacity with increasing local slenderness is
EN 1993-1-1 design. The advanced analysis was performed in Abaqus also accurately reflected by the CSM strain limits.
[29] using Timoshenko (shear deformable) beam elements denoted as In contrast to the continuous capacity predictions using the CSM
B31OS and B31 in the Abaqus element library [29] for open and closed strain limits, the discrete jump in EN 1993-1-1 [1] bending capacity
cross-sections respectively. Cross-sections were discretised into 33 from Mpl to Mel between Class 2 and 3 cross-sections is an overly sim-
section points along the web and flange plates to capture the spread of plistic approximation of the continuous behaviour observed in the shell
plasticity through the cross-sections accurately. The engineering FE models. A linear transition from Mel to Mpl in Class 3 cross-sections

13
A. Fieber, et al. Engineering Structures 199 (2019) 109624

Fig. 19. Cross-section capacity predicted by shell FE models, advanced inelastic analysis using beam finite elements with CSM strain limits, and EN 1993-1-1 [1] for I-
sections in (a) pure major axis bending, (b) pure minor axis bending and (c) pure compression.

Fig. 20. Cross-section capacity predicted by shell FE models, advanced inelastic analysis using beam finite elements with CSM strain limits, and EN 1993-1-1 [1] for
SHS/RHS in (a) pure bending and (b) pure compression.

[60], as will be included in the next revision of EN 1993-1-1, provides a some Class 2 cross-sections are unable to reach their full plastic moment
better approximation to the shell FE behaviour, especially for I-sections capacity, as shown in Fig. 20(b). Similar observations were made by
as shown in Fig. 19(a) and (b). The results also indicate that the limiting Yun et al. [41]. Note that the lines in Fig. 19 and 20 for the EN 1993-1-1
slenderness for Class 2 SHS/RHS may be slightly unconservative, as resistance functions represent a typical cross-section since individual

14
A. Fieber, et al. Engineering Structures 199 (2019) 109624

normalised resistances will vary depending on their geometry, parti- gradient is kept approximately constant in each member by setting the
cularly for Class 4 sections, where the aspect ratio is important, and for length of the shear span equal to 10 times the local buckling half-wa-
sections in bending, where the shape factor is influential. For con- velength (i.e. Ls / Lb,cs = 10 ); out-of-plane failure was prevented in the
sistency, the boundaries of the EN 1993-1-1 cross-section classes in shell FE models though the provision of lateral restraints at intervals
Fig. 19 and 20 are marked at the approximate full cross-section slen- equal to the local buckling half-wavelength of the cross-section Lb,cs
derness limits. along the length of the beams. Similar to the observations made in
Modelling slender Class 4 cross-sections through one-dimensional Section 4.4, the strain averaging approach provides an accurate re-
beam finite elements is inherently problematic, since the reduction in presentation of the shell FE models across the considered range of cross-
stiffness resulting from cross-section buckling (local and/or distor- section slenderness; strain hardening is allowed for in stocky cross-
tional) is typically not accounted for. Advanced beam element for- sections with ¯ p < 0.4 , while for slender cross-sections with ¯ p > 0.68, a
mulations exist that are able to explicitly capture the loss of stiffness similar bending capacity to EN 1993-1-1 is predicted, but without the
due to local/distortional buckling in slender cross-sections [61]. Evi- need for cumbersome effective width calculations. Restricting the
dently, the loss in stiffness, and associated increase in deflections, due maximum strain limit to 15 y (i.e. = 15) conservatively limits the
to elastic local buckling cannot be captured using conventional beam bending capacity of stocky cross-sections with ¯ p 0.32 to approxi-
finite elements. Application of the CSM strain limits nonetheless still mately 1.13Mpl , but this is still significantly above the plastic moment
yields reasonably accurate strength predictions when using conven- capacity predicted by EN 1993-1-1. For semi-compact (Class 3) cross-
tional beam finite elements, since the base curve for slender cross- sections, the elastic plastic moment capacity Mep (i.e. a linear transition
sections (i.e. ¯ p > 0.68) is based on the concept of the equivalent local from Mel to Mpl ) due to be included in the next revision of EN 1993-1-1
buckling strain [43]. provides better capacity predictions than the step-wise bending re-
sistance envelope currently defined by the cross-section classification
5.2. Members subjected to bending process.
In total, 973 simply-supported I-sections and SHS/RHS beams have
The normalised bending capacities for a series of I-section beams been analysed under three-point bending, four-point bending and uni-
subjected to three-point bending are shown in Fig. 21. The moment formly distributed loading. For members with I-sections, the local
slenderness of the considered cross-sections, including common
European IPE, HEA and HEB sections, varied between 0.14 ¯ p 1.37
for major axis bending and 0.13 ¯ p 1.06 for minor axis bending. For
members with SHS/RHS, the local cross-section slenderness of the
considered cross-sections, including common European SHS and RHS
profiles, was within the range of 0.14 ¯ p 0.81. The normalised ca-
pacity predictions for the proposed method of design by advanced in-
elastic analysis and EN 1993-1-1 are shown in Fig. 22. It can be seen
that the proposed approach provides improved capacity predictions
over EN 1993-1-1, particularly in the low slenderness range and for
Class 3 cross-sections (approximately 0.45 ¯ p 0.68). The proposed
strain averaging approach leads to a reduction of the observed scatter
while also simplifying the design process. Note that the discontinuity in
the EN 1993-1-1 results shown in Fig. 22(b) is attributed to the current
cross-section classification process, while the overlap between Class 2
and 3 cross-sections is caused by the fact that, similar to Fig. 22(a), the
full cross-section slenderness ¯ p is used along the horizontal axis (as
compared to the b/ t ratio of the critical isolated plate that is used to
Fig. 21. Capacity predictions for I-sections of varying slenderness subjected to classify the cross-section according to EN 1993-1-1 [1]).
3-point bending.

Fig. 22. Predicted bending capacity Mu normalised by ultimate bending capacity from shell FE model Mshell for I-sections and SHS/RHS under three-point bending,
four-point bending and uniformly distributed loading for (a) proposed method of design by advanced inelastic analysis and (b) EN 1993-1-1.

15
A. Fieber, et al. Engineering Structures 199 (2019) 109624

5.3. Members subjected to compression Recall that in the proposed design method, failure is defined as the
lower of the peak load factor reached during the advanced analysis or
In this subsection, the proposed design approach is evaluated the load level at which the strain limit is first reached. In compact cross-
against shell FE results on pin-ended columns with I-section, H-section sections, the peak load factor tends to govern, while in slender Class 4
and SHS/RHS profiles. The ultimate major axis flexural buckling ca- cross-sections the strain limit is usually reached prior to the peak load
pacities Nu,y predicted from shell FE modelling and beam FE modelling factor, as shown in Fig. 23.
with strain limits, normalised by Npl = Afy , are compared with the EN Analogous to current EN 1993-1-1 design, in which cross-sections
1993-1-1 design resistances in Fig. 23; three member slenderness values are classified based on the first-order distribution of stresses, the cross-
¯ y = 0.5, 1.0 and 1.5, and a range of local slenderness values ¯ p are section slenderness and hence the CSM strain limits are also based on
considered. the first-order distribution of stresses, i.e. uniform compression in the
Generally, the observations made at the cross-sectional level (see case of columns. In reality, second-order moments modify the stress
Section 5.1) also apply at the member level. The predictions of the distribution of the critical cross-section at each load increment. Using a
proposed method are typically very accurate, though become con- strain limit based on a cross-section slenderness derived from a second-
servative in the case of columns with slender I-section and RHS profiles, order distribution of stresses will typically result in an increase in de-
in which the web is significantly more slender than the flanges. The formation capacity of isolated compression members and provide en-
effective width method used in Eurocode 3 is more accurate at cap- hanced capacity predictions, though is less straightforward to imple-
turing this localised reduction in strength of a single plate. For com- ment.
pression members, the accuracy of the proposed method of design by
advanced inelastic analysis is also affected by the conservative as-
5.4. Members subjected to combined compression and bending
sumption regarding the imperfection magnitude i.e. the use of the
simplified magnitudes of e0 / l , rather than the more accurate back-cal-
In this subsection, the proposed method of design by advanced in-
culated values, as discussed in Section 4.6. This is most significant for
elastic analysis is assessed for the case of beam-columns. A total of 2565
members of intermediate slenderness ( ¯ = 0.5 and 1.0); for more
beam-columns of varying member slenderness ( ¯ = 0.5, 1.0, 1.5) were
slender members with ¯ = 1.5, the buckling resistance of the column is
analysed under combined compression and major axis bending. 19
less sensitive to the assumed imperfection magnitude and the proposed
cross-section profiles of varying local slenderness were considered and
design method accurately predicts the EN 1993-1-1 buckling capacity.
the ratio of applied compression to bending was varied to cover the

Fig. 23. Normalised major axis flexural buckling capacities Nu,y / Npl of pin-ended columns of varying member slenderness ( ¯ y = 0.5, 1.0, 1.5) with (a) I-section, (b) H-
section and (c) SHS/RHS profiles with varying cross-section slenderness ¯ p .

16
A. Fieber, et al. Engineering Structures 199 (2019) 109624

Table 3
Summary of capacity predictions of the proposed method of design by advanced inelastic analysis with strain limits (for = 15), the
EN 1993-1-1 member check and second-order elastic analysis with equivalent imperfections (GNIA) and cross-section (c/s) checks
relative to shell FE results for beam-columns subjected to combined compression and major axis bending.

IPE 160, 240, 400, 500; HEA 120, 160, 240, 330, 440; HEB 100, 200, 340, 500.
Analysed cross-sections:
SHS 60 × 60 × 5, 60 × 60 × 3, 60 × 60 × 2; RHS 120 × 60 × 5, 120 × 60 × 4, 120 × 60 × 3.

Table 4
Summary of capacity predictions of the proposed method of design by advanced inelastic analysis with strain limits (for = 15), the
EN 1993-1-1 member check and second-order elastic analysis with equivalent imperfections (GNIA) and cross-section (c/s) checks
relative to shell FE results for beam-columns subjected to combined compression and minor axis bending.

Analysed cross-sections: IPE160, 240;HEA240, 330;HEB200, 340 .

complete loading range from pure compression to pure bending. Three cases and cross-sections, as well as the mean, coefficient of variation
bending moment distributions along the member length (CoV), minimum and maximum beam-column capacities obtained from
( = 1.0, 0, 0.5 ) were considered by changing the ratio of applied end the alternative design approaches relative to the shell FE results
moments = M2/ M1, where = 1 corresponds to uniform bending, ( ult / shell ), are given in Tables 3, and 4 for compression plus major and
while = 1 corresponds to double curvature bending. An additional minor axis bending respectively.
540 I-section subjected to combined compression and minor axis
bending with end moment ratios of = 1 and = 0.5 were also 5.4.1. Beam-columns: compression plus major axis bending
considered. A summary of the analysed beam-columns is given in Overall, the proposed method of design by advanced inelastic ana-
Table 2. The CSM strain limits were determined based on the first-order lysis provides consistent and accurate capacity predictions for hot-
stress distribution of the critical cross-section. The capacity predictions rolled steel beam-columns subjected to combined compression and
are compared against the results of shell FE modelling and two alter- major axis bending, with a mean of 0.965 and a CoV of 3.7% relative to
native EN 1993-1-1 [1] design approaches: (i) a member check using the shell FE results. Both EN 1993-1-1 design approaches are more
the interaction factor k yy from Annex B of EN 1993-1-1 and (ii) a linear conservative, with a mean normalised capacity prediction of 0.956 and
interaction cross-section check applied to a second-order elastic ana- CoV of 5.3% when performing a member check, and a mean normalised
lysis with equivalent imperfections. A summary of the considered load capacity prediction of 0.921 and CoV of 5.4% when applying a linear

17
A. Fieber, et al. Engineering Structures 199 (2019) 109624

interaction cross-section check to the results of GNIA (i.e. second-order limited to the elastic moment, which is seen to be overly conservative
elastic analysis with equivalent imperfections). compared to the shell FE model results. In contrast, the proposed
A key advantage of the proposed method of design by advanced method of design by advanced inelastic analysis is able to accurately
inelastic analysis compared to EN 1993-1-1 is the inclusion of the al- capture the shell FE model behaviour: for the shorter members and the
lowance for the beneficial effect of local moment gradients; this is bending dominated loading cases, the CSM strain limits govern the
highlighted by considering the three cases of first-order moment gra- load-carrying capacity, while for the more slender members and the
dient individually. For the case of uniform moment ( = 1), all three stability dominated cases the member buckling behaviour and failure is
design approaches predict similar failure loads. As the level of first- defined by the peak load factor, as highlighted in Fig. 24.
order moment gradient increases (i.e. = 0 and = 0.5), the pro- Fig. 24(b) shows similar observations for RHS 120 × 60 × 5 beam-
posed method continues to accurately predict the behaviour of the shell columns. Overall, the proposed method of design by advanced inelastic
FE model. The EN 1993-1-1 member check accounts for the influence of analysis is able to capture the nonlinear moment-axial force interaction
the moment gradient on global buckling through the equivalent mo- of the shell FE models better than the two EN 1993-1-1 design ap-
ment factor Cmy . However, the effects of the local moment gradient and proaches. Fig. 24(b) also shows how the capacity predictions of the
strain hardening on the local cross-section capacity are not accounted proposed design method improve as bending becomes increasingly
for, resulting in increased conservatism for the cases of = 0 and dominant and the influence of the assumed imperfection magnitude
= 0.5. The method of performing a second-order elastic analysis diminishes.
plus a linear interaction cross-section check is shown to be conservative Fig. 24(c) shows the normalised M-N interaction diagrams for IPE
by around 8% on average. This is primarily due to the fact that the 240 beam-columns subjected to combined compression and major axis
simplified linear interaction cross-section check does not reflect the true bending. According to EN 1993-1-1, the limiting width-to-thickness
nonlinear cross-section behaviour, particularly for the more bending ratio for the classification of the web changes as the applied (major
dominated loading cases. axis) moment to axial compression ratio changes. The class-dependent
The normalised moment-axial force interaction diagram for a series cross-section capacity limits result in discrete jumps in the Eurocode 3
of HEA 240 beam-columns subjected to combined compression and capacity predictions, as shown in Fig. 24(c). In the bending dominated
uniform major axis bending is shown in Fig. 24(a). According to EN cases, an IPE 240 in grade S355 steel is classified as Class 1/2; hence the
1993-1-1, an HEA 240 is classified as a Class 3 cross-section for both plastic moment capacity Mpl and squash load Npl define the anchor
pure compression and pure bending. Hence, the bending capacity is points of the interaction curve. Once classification switches to Class 3 as

Fig. 24. Normalised ultimate capacity of pin-ended beam-columns subjected to combined compression and major axis bending for the following cross-sections: (a)
HEA 240, (b) RHS 120 × 60 × 5, (c) IPE 240 and (d) SHS 60 × 60 × 3.

18
A. Fieber, et al. Engineering Structures 199 (2019) 109624

the proportion of compression increases, a linear interaction surface 5.5. Overview and discussion
anchored to Npl and Mel applies. In the compression dominated cases,
the cross-section is classified as Class 4 and the effective area is used to The capacity predictions obtained using the proposed method of
calculate the compression resistance of the cross-section. Fig. 24(c) also beam FE advanced inelastic analysis with CSM strain limits and = 15,
shows that the EN 1993-1-1 capacity predictions are less accurate for as well as using the EN 1993-1-1 member design checks, are compared
cases with nonlinear moment distributions, particularly for bending against the benchmark shell FE results for each analysed load case in
dominated load cases. The CSM strain limits on the other hand are able Table 5. A value of ult / shell 1.0 indicates safe-sided predictions,
to capture the effects of moment gradients and strain hardening. while lower coefficients of variation indicate less scatter in the results.
Applying a linear interaction cross-section check to the results of a The limiting cross-section slenderness values of the analysed cases are
second-order elastic analysis generally results in an under-prediction of also given. Overall, the proposed method of design by advanced in-
the shell FE model capacity for bending dominated cases when a mo- elastic analysis including the CSM strain limits is more accurate and
ment gradient is present, as shown in Fig. 24(c) and (d). While the consistent compared to EN 1993-1-1 design, particularly for bending
effects of geometric imperfections and residual stresses are accounted dominated cases. For compression members, the accuracy of the ad-
for in the analysis through equivalent imperfections, the influence of vanced analysis is limited by the accuracy of the assumed imperfection
the local moment gradient is ignored and the linear interaction cross- magnitude. This may result in slightly conservative results at an iso-
section check does not allow for appropriate spread of plasticity at the lated member level, though it is expected that this drawback is more
cross-sectional level, thereby resulting in conservative capacity pre- than outweighed by the advantages of advanced inelastic analysis for
dictions. indeterminate structural systems. Note that the proposed method has
currently only been verified for I-sections and SHS/RHS members. Ex-
tension to circular hollow section (CHS) members is currently un-
5.4.2. Beam-columns: compression plus minor axis bending derway, and requires the development of a CHS-specific base curve to
The capacity predictions for I-shaped beam-columns subjected to determine the CSM strain limits (available in [49]), the definition of
combined compression and minor axis bending are shown in Table 4. suitable local buckling half-wavelengths and comprehensive verifica-
The proposed method of design by advanced inelastic analysis with tion of the method for a range of loading scenarios, as presented herein
strain limits accurately reflects the shell FE model behaviour in all for I-sections and SHS/RHS members.
cases, with a mean predicted capacity of 0.966 relative to the shell FE The key features and advantages of the proposed method of design
ultimate capacity. The EN 1993-1-1 design predictions are overly con- by advanced inelastic analysis using beam finite elements with strain
servative in most of the considered cases of I-sections subjected to limits are summarised as follows:
combined compression and minor axis bending, as shown in Fig. 25(a)
for an IPE 240; this is because the web is typically classified as slender • Beam finite elements are used for computational efficiency
(Class 4). Similarly, when performing a second-order elastic analysis, • No cross-section classification, member checks or effective lengths
even though the linear interaction curve based on plastic section are required
properties is used for Class 1 and 2 cross-sections, the bending capacity • A standardised stress-strain model [31] captures the yield plateau
is restricted to 1.25Mel [59] – this again results in overly conservative and strain hardening behaviour of hot-rolled carbon steels
capacity predictions for bending dominated cases, as shown in • CSM strain limits mimic local buckling in beam finite elements and
Fig. 25(b) for an HEB 200. control the spread of plasticity through the cross-section and around

Fig. 25. Normalised ultimate capacity of pin-ended beam-columns subjected to combined compression and minor axis bending for the following cross-sections: (a)
IPE 240 and (b) HEB 200.

19
A. Fieber, et al. Engineering Structures 199 (2019) 109624

Table 5
Overview of all analysed cases, limiting cross-section slenderness values and performance of proposed method of design by advanced inelastic analysis and traditional
EN 1993-1-1 member design.

ult / shell

No. of simulations ¯p AA + CSM ( = 15) EN 1993-1-1

Profile Load case Min Max Mean CoV Mean CoV

I-sections Cross-section (N) 143 0.19 1.67 0.963 0.076 0.974 0.080
Cross-section (My ) 143 0.17 1.37 0.952 0.056 0.915 0.071
Cross-section (Mz ) 143 0.11 0.90 0.845 0.120 0.706 0.148
4-point bending (My ) 97 0.22 1.21 0.968 0.042 0.933 0.052
3-point bending/UDL (My ) 429 0.14 1.37 0.958 0.053 0.907 0.075
3-point bending (Mz ) 143 0.13 1.06 0.819 0.128 0.691 0.151
M V interaction 272 0.24 0.41 0.860 0.087 0.767 0.110
Column (major)a 666 0.24 1.67 0.930 0.070 0.972 0.032
Column (minor)a 210 0.30 1.67 0.904 0.091 0.980 0.015
Beam-column (N + My )a 1755 0.28 1.16 0.965 0.033 0.948 0.050
Beam-column (N + Mz )a 540 0.23 0.65 0.966 0.057 0.866 0.118

Total 4541

SHS/RHS Cross-section (N) 123 0.19 1.11 0.986 0.060 0.968 0.082
Cross-section (My and Mz ) 123 0.14 0.81 0.959 0.085 0.940 0.106
4-point bending (My and Mz ) 123 0.14 0.81 0.959 0.085 0.940 0.106
3-point bending (My and Mz ) 324 0.14 0.81 0.901 0.107 0.868 0.130
Columna 144 0.19 1.11 0.902 0.050 0.951 0.052
Beam-column (N + My )a 810 0.21 0.73 0.963 0.047 0.939 0.067

Total 1647

a
Capacity predictions for proposed method of advanced inelastic analysis using beam FE with CSM strain limits is based on simplified imperfection magnitude
defined by Eq. (18). Back-calculated imperfections would give capacity predictions close to EN 1993-1-1 values.

the structure References

• The beneficial effects of local moment gradients are accounted for
through a strain averaging approach [1] EN 1993-1-1. Eurocode 3: Design of steel structures – Part 1-1: General rules and

• Advanced analysis provides an accurate representation and visua- rules for buildings. European Committee for Standardization, Brussels; 2005.
[2] Gardner L, Yun X, Fieber A, Macorini L. Steel design by advanced analysis: material
lisation of the failure mechanisms
•
modeling and strain limits. Engineering 2019;5(2):243–9.
Compared to existing methods, design is quicker, more direct and [3] Gardner L. The continuous strength method. Proc Inst Civ Eng Struct Build
more accurate 2008;161(3):127–33.
[4] Liew JYR, Chen WF, Chen H. Advanced inelastic analysis of frame structures. J
Constr Steel Res 2000;55(1–3):245–65.
6. Conclusions [5] Chen WF. Advanced analysis for structural steel building design. Front Archit Civ
Eng China 2008;2(3):189–96.
[6] Kim SE, Chen WF. Design guide for steel frames using advanced analysis program.
A new method of structural steel design based on advanced inelastic Eng Struct 1999;21(4):352–64.
analysis, i.e. geometrically and materially nonlinear analysis with im- [7] Trahair NS, Chan SL. Out-of-plane advanced analysis of steel structures. Eng Struct
perfections (GMNIA), and strain limits is presented. The continuous 2003;25(13):1627–37.
[8] Buonopane SG, Schafer BW. Reliability of steel frames designed with advanced
strength method (CSM) strain limits are employed to simulate local analysis. J Struct Eng ASCE 2006;132(2):267–76.
buckling, thereby making it possible to use inelastic analysis methods [9] Rasmussen KJR, Zhang H, Cardoso FS, Liu W. The direct design method for cold-
with beam finite elements for all cross-section classes. The CSM strain formed steel structural frames. Eighth international conference on steel and alu-
minium structures, Hong Kong, 2016. Hong Kong. 2016.
limits provide a rational means of limiting the cross-section deforma-
[10] Surovek AE. Advanced analysis in steel frame design. Guidelines for direct second-
tion and the corresponding spread of plasticity to appropriate levels order inelastic analysis. Reston, Virginia: ASCE; 2012.
depending on the local cross-section slenderness. Furthermore, by ap- [11] Liew JYR, White DW, Chen WF. Beam-column design in steel frameworks - insights
on current methods and trends. J Constr Steel Res 1991;18(4):269–308.
plying the strain limits to compressive strains that are averaged over a
[12] Chan SL, Liu YP, Zhou ZH. Limitation of effective length method and codified
characteristic length of member, the beneficial effects of local moment second-order analysis and design. Steel Compos Struct 2005;5(2–3):181–92.
gradients can be exploited. The proposed design approach has been [13] EN 1993-1-5. Eurocode 3: Design of steel structures - Part 1–5: Plated structural
assessed against a total of 6188 FE results on hot-rolled steel members elements. European Committee for Standardization, Brussels; 2006.
[14] Driscoll Jr. GC, Beedle LS. The plastic behavior of structural members and frames.
with I-section, SHS and RHS profiles of varying local slenderness. Weld J 1957;36(6). 275–s.
Members subjected to bending, compression and combined loading [15] Lay MG. The experimental bases of plastic design. Fritz Engineering Laboratory
were considered. It is shown that the proposed design method is con- Report No. 297.3. Lehigh University; 1963.
[16] Gioncu V, Petcu D. Available rotation capacity of wide-flange beams and beam-
sistently more accurate than current EN 1993-1-1 design. Research is columns Part 2. Experimental and numerical tests. J Constr Steel Res
currently underway to extend the design method to structural systems, 1997;43(1–3):219–44.
including continuous beams and frames. [17] Wang J, Afshan S, Gkantou M, Theofanous M, Baniotopoulos C, Gardner L. Flexural
behaviour of hot-finished high strength steel square and rectangular hollow sec-
tions. J Constr Steel Res 2016;121:97–109.
Acknowledgement [18] Geschwindner LF. A practical look at frame analysis, stability and leaning columns.
Eng J AISC 2002;39:167–81.
[19] Chan SL, Liu YP, Liu SW. Structural design in the post-effective length era. Procedia
The authors would like to acknowledge the Engineering and Eng 2011;14:1005–12.
Physical Sciences Research Council (EPSRC) for the financial support [20] Byfield MP, Nethercot DA. An analysis of the true bending strength of steel beams.
(Grant EP/M5078784/1). Proc Inst Civ Eng Struct Build 1998;128:188–97.

20
A. Fieber, et al. Engineering Structures 199 (2019) 109624

[21] Lee GC, Galambos TV. The post buckling strength of wide-flange beams. Fritz loading - Part 2: Parametric studies and CSM design. J Constr Steel Res
Engineering Laboratory Report No. 205E.12. Lehigh University; 1961. 2018;148:265–74.
[22] Sawyer Jr. HA. Post-elastic behavior of wide-flange steel beams. J Struct Div ASCE [43] Zhao O, Afshan S, Gardner L. Structural response and continuous strength method
1961;87(8):43–71. design of slender stainless steel cross-sections. Eng Struct 2017;140:14–25.
[23] Lay MG, Adams PF, Galambos TV. Experiments on high strength steel members. [44] Gardner L, Yun X, Macorini L, Kucukler M. Hot-rolled steel and steel-concrete
Welding Research Council Bulletin No. 110; 1965. composite design incorporating strain hardening. Structures 2017;9:21–8.
[24] Sfintesco D. Experimental basis of European buckling curves. Constr Métallique [45] Li Z, Schafer BW. Buckling analysis of cold-formed steel members with general
1970;3:5–12. boundary conditions using CUFSM: Conventional and constrained finite strip
[25] Yun X. Material modelling and design of hot-rolled and cold-formed steel structures methods. Twentieth international speciality conference on cold-formed steel
[PhD thesis]. London: Department of Civil and Environmental Engineering, structures. Saint Louis, Missouri, USA. 2010.
Imperial College; 2018. [46] Seif M, Schafer BW. Local buckling of structural steel shapes. J Constr Steel Res
[26] van Kuren RC, Galambos TV. Beam-column experiments. Fritz Engineering 2010;66(10):1232–47.
Laboratory Report No. 205A.30. Lehigh University; 1961. [47] Gardner L, Fieber A, Macorini L. Formulae for calculating elastic local buckling
[27] Aglietti RA, Lay MG, Galambos TV. Tests on A36 and A441 steel beam-columns. stresses of full structural cross-sections. Structures 2019;17:2–20.
Fritz Engineering Laboratory Report No. 278.14. Lehigh University; 1964. [48] Lan X, Chen J, Chan TM, Young B. The continuous strength method for the design of
[28] Wang J, Gardner L. Flexural buckling of hot-finished high-strength steel SHS and high strength steel tubular sections in compression. Eng Struct 2018;162:177–87.
RHS columns. J Struct Eng ASCE 2017;143(6):1–12. [49] Buchanan C, Gardner L, Liew A. The continuous strength method for the design of
[29] Abaqus. Version 6.14. Analysis user’s guide. Dassault Systemes, Simulia Corp, circular hollow sections. J Constr Steel Res 2016;118:207–16.
Providence RI, USA; 2014. [50] Zhao O, Gardner L. The continuous strength method for the design of mono-sym-
[30] Crisfield MA. A fast incremental/iterative solution procedure that handles snap- metric and asymmetric stainless steel cross-sections in bending. J Constr Steel Res
through. Comput Struct 1981;13:55–62. 2018;150:141–52.
[31] Yun X, Gardner L. Stress-strain curves for hot-rolled steels. J Constr Steel Res [51] Lay MG, Galambos TV. The inelastic behavior of beams under moment gradient.
2017;133:36–46. Fritz Engineering Laboratory Report No. 297.12. Lehigh University; 1964.
[32] ECCS. Ultimate limit state calculations of sway frames with rigid joints. No: 33, [52] Lay MG. Some studies of flange local buckling in wide-flange shapes. Fritz
Technical Committe 8 of European Convention for Constructional Steelwork Engineering Laboratory Report No. 297.10. Lehigh University; 1964.
(ECCS), Brussels; 1984. [53] Pham CH, Davis AF, Emmett BR. Numerical investigation of cold-formed lapped Z
[33] Kucukler M. Stiffness reduction approach for structural steel design [PhD thesis]. purlins under combined bending and shear. J Constr Steel Res 2014;95:116–25.
London: Department of Civil and Environmental Engineering, Imperial College; [54] Pham CH, Hancock GJ. Experimental investigation of high strength cold-formed C-
2015. sections in combined bending and shear. J Struct Eng ASCE 2010;136(7):866–78.
[34] Yun X, Gardner L, Boissonnade N. Ultimate capacity of I-sections under combined [55] dos Santos GB, Gardner L, Kucukler M. A method for the numerical derivation of
loading - Part 1: experiments and FE model validation. J Constr Steel Res plastic collapse loads. Thin-Walled Struct 2018;124:258–77.
2018;147:408–21. [56] Lee SC, Lee DS, Yoo CH. Flexure and shear interaction in steel I-girders. J Struct Eng
[35] Fieber A, Gardner L, Macorini L. Formulae for determining elastic local buckling ASCE 2013;139(11):1882–94.
half-wavelengths of structural steel cross-sections. J Constr Steel Res [57] Mises Rv. Mechanik der festen Körper im plastisch-deformablen Zustand.
2019;159:493–506. Nachrichten von der Gesellschaft der Wissenschaften zu Göttingen, Mathematisch-
[36] Kucukler M, Gardner L, Macorini L. A stiffness reduction method for the in-plane Physikalische Klasse; 1913. p. 582–92.
design of structural steel elements. Eng Struct 2014;73:72–84. [58] Gonçalves R, Coelho T, Camotim D. On the plastic moment of I-sections subjected to
[37] Zhao O, Gardner L, Young B. Behaviour and design of stainless steel SHS and RHS moderate shear forces. Thin-Walled Struct 2014;78:138–47.
beam-columns. Thin-Walled Struct 2016;106:330–45. [59] Lindner J, Kuhlmann U, Jörg F. Initial bow imperfections e0 for the verification of
[38] Afshan S, Gardner L. The continuous strength method for structural stainless steel flexural buckling according to Eurocode 3 Part 1–1 - additional considerations. Steel
design. Thin-Walled Struct 2013;68:42–9. Constr 2018;11(1):30–41.
[39] Ashraf M, Young B. Design formulations for non-welded and welded aluminium [60] Greiner R, Lechner A. Background information to design guidelines for cross-section
columns using continuous strength method. Eng Struct 2011;33(12):3197–207. and member design according to Eurocode 3 with particular focus on semi-compact
[40] Su MN, Young B, Gardner L. The continuous strength method for the design of sections. Graz: Institute for Steel Structures and Shell Structures, Graz University of
aluminium alloy structural elements. Eng Struct 2016;122:338–48. Technology; 2012.
[41] Yun X, Gardner L, Boissonnade N. The continuous strength method for the design of [61] Rasmussen KJR, Zhang X, Zhang H. Beam-element-based analysis of locally and/or
hot-rolled steel cross-sections. Eng Struct 2018;157:179–91. distortionally buckled members: Theory. Thin-Walled Struct 2016;98:285–92.
[42] Yun X, Gardner L, Boissonnade N. Ultimate capacity of I-sections under combined

21