Lateral-Torsional Buckling Resistance

of Castellated Beams
Delphine Sonck, Ph.D. 1; and Jan Belis, Ph.D. 2

Abstract: The existing design expressions for the lateral-torsional buckling behavior of castellated beams conflict. Furthermore, they do not
take into account the detrimental effect of the residual stress modification attributable to the fabrication process, which was only recently
demonstrated by the authors. This makes these design rules possibly unsafe. In this paper, the lateral-torsional buckling behavior of doubly
Downloaded from ascelibrary.org by Tufts University on 10/19/16. Copyright ASCE. For personal use only; all rights reserved.

symmetric castellated beams loaded by a constant bending moment is investigated numerically. The numerical model, including the modified
residual stresses, was validated by comparing its results with experimental results. A preliminary design approach is proposed based on the
current European guidelines for the calculation of the lateral-torsional buckling resistance of I-section beams. According to the proposed
approach, the lateral-torsional buckling resistance of castellated beams can be determined using the cross-sectional properties calculated at the
center of the web opening. The modification of the residual stresses during the fabrication process results in resistances that lie approximately
one buckling curve lower. DOI: 10.1061/(ASCE)ST.1943-541X.0001690. © 2016 American Society of Civil Engineers.
Author keywords: Castellated beams; Lateral-torsional buckling; Residual stresses; Numerical simulations; Experiments; ABAQUS;
Metal and composite structures.

Introduction fabrication of the castellated members. A full overview of the

failure behavior of castellated beams is given by Kerdal and
Castellated members are steel I-section members with hexagonal Nethercot (1984).
openings located at fixed distances along the web. They are usually Castellated members loaded in bending can fail by lateral-
made by cutting a hot-rolled I-section member according to a torsional buckling (LTB). In this global buckling mode, the lateral
polygonal pattern, after which both obtained halves are shifted movement of the compressed flange of the beam will be restrained
and welded together again (Fig. 1). The resulting castellated by the part of the beam subjected to tension. This results in a com-
member is typically 40–60% higher than its original parent section, bined lateral and torsional movement of the cross section, as de-
which greatly increases the strong-axis bending resistance. By picted in Fig. 2. LTB of castellated beams has been investigated
varying the cutting pattern and the properties of the parent sec- by Nethercot and Kerdal (1982), as well as Gietzelt and Nethercot
tion(s), a wide variation of opening shapes and beam geometries (1983), who demonstrated that the LTB behavior of these beams is
is possible, such as cellular members or tapered members. qualitatively similar to the LTB failure of plain-webbed I-section
Castellated members are most suited for load cases where they beams. However, a quantitative modification of the lateral-torsional
are loaded in strong-axis bending because of the corresponding buckling resistance was introduced because of the modified geom-
economic material use. A further advantage of castellated beams etry. Accordingly, it was proposed to use the design approach valid
is the ability to guide service ducts through the web openings in- for plain-webbed members without web openings, but with all
stead of under the beams, which maximizes the usable height. Fur- cross-sectional properties calculated at the center of the castellation
thermore, the lighter appearance of exposed castellated beams can (2T approach).
be architecturally appealing. However, because of the presence of As will be illustrated in the next section, some ambiguity
the web openings their behavior is more complex than that of plain- still exists on how to calculate the LTB resistance using this 2T
webbed I-section members. Consequently, the number of necessary approach. Furthermore, as illustrated in recent work by the authors,
design checks for these members increases. New local failure the production process modifies the already present residual
modes will arise around the openings, and already existing failure stresses in the hot-rolled parent sections (Sonck et al. 2015). The
modes will be altered as a result of the modified geometry and experimental results described in this paper illustrate that the com-
imperfections. An important example of such an imperfection is pressive residual stresses in the flanges of the castellated members
the residual stress pattern, which will be modified during the increase, compared to the stresses in the original parent sections.
Because this is detrimental for the LTB resistance, this should also
1
Postdoctoral Assistant, Dept. of Structural Engineering-LMO, Ghent be considered during the design. However, it was never taken into
Univ., Tech Ln. Ghent Science Park—Campus A, Technologiepark- account in the currently existing design approaches for the LTB
Zwijnaarde 904, 9052 Ghent, Belgium (corresponding author). ORCID: resistance, which makes these approaches possibly unsafe.
http://orcid.org/0000-0001-8340-8797. E-mail: Delphine.Sonck@UGent.be In this work, the authors resolve the ambiguity and the possible
2
Professor, Dept. of Structural Engineering-LMO, Ghent Univ., Tech unsafety of the currently existing design approaches for LTB of
Ln. Ghent Science Park—Campus A, Technologiepark-Zwijnaarde 904,
castellated beams by performing a parametric numerical study, con-
9052 Ghent, Belgium. E-mail: Jan.Belis@UGent.be
Note. This manuscript was submitted on March 1, 2016; approved on
sidering the modified residual stresses. The numerical model used
September 8, 2016; published online on October 19, 2016. Discussion per- in this study was validated by comparing its results with LTB ex-
iod open until March 19, 2017; separate discussions must be submitted for periments on castellated members that were executed by the first
individual papers. This paper is part of the Journal of Structural Engineer- author. The currently existing design approaches will be compared
ing, © ASCE, ISSN 0733-9445. with the results of this study, after which a first design approach

© ASCE 04016197-1 J. Struct. Eng.

J. Struct. Eng., 04016197

 Fig. 1. Standard fabrication method of castellated members starting from a plain-webbed parent section

for the LTB moment will be proposed. This approach will include factor χLT can be calculated using the general method from EC3,
the effect of the modified residual stress pattern and modified according to Eqs. (2) and (3). In these equations, the influence of
geometry. imperfections such as geometric imperfections and residual stresses
The results obtained for LTB of castellated beams will be an is taken into account by the imperfection factor αLT . This factor
extension of a similar investigation of the LTB behavior of cellular increases with increasing imperfection influences, lowering the
members described in two other studies (Sonck 2014; Sonck and LTB resistance.
Belis 2015). While only circular openings were considered in For each member, the value of αLT depends on which buckling
Downloaded from ascelibrary.org by Tufts University on 10/19/16. Copyright ASCE. For personal use only; all rights reserved.

the latter study, now the effect of hexagonal opening shapes with curve is imposed by EC3. As shown in Table 1, buckling curve a
different aspect ratios on the critical LTB moment M cr and the LTB should be used for members with a low influence of imperfections,
resistance M Rd can be determined. while buckling curve d should be used for members with large im-
In the following section the existing design approaches on perfections or a large imperfection sensitivity. The buckling curve
which the proposed approach is based will be discussed. Next, choice depends on the height-to-width ratio H=b of the cross sec-
the LTB experiments used for the validation of the numerical model tion, as well as the beam’s fabrication method. In Table 2, the pre-
will be described. Subsequently, the finite-element (FE) model and scribed buckling curves for hot-rolled beams subjected to LTB
its validation will be treated, as well as the geometries used in the are given.
parametrical study. After this, the results for the critical LTB mo- The nondimensional slenderness λ̄LT should be calculated
ment M cr will be presented. Finally, the results obtained for the according to Eq. (4), using the critical buckling moment M cr . In
LTB resistance M Rd will be shown and a first preliminary proposal Eq. (5), the critical buckling moment M cr of a simply supported
will be made for the LTB resistance design approach. beam with fork supports loaded by a constant bending moment
is given. In this equation, L is the length of the beam, GI t is
the torsion stiffness, EI z the weak-axis bending stiffness, and
Proposed Design Approach for LTB of Castellated EI w the warping stiffness of the beam’s cross section
Members χLT W y fy
M Rd ¼ ð1Þ
The proposed design approach is based on the design method used γ M1
in Annex N of the European prestandard ENV 1993-1-1:1992 for 1
the calculation of steel structures (CEN 1998). This design method χLT ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ≤ 1 ð2Þ
ϕ þ ϕ2 − λ̄2LT
is based on the findings of Nethercot and Gietzelt mentioned earlier
(Nethercot and Kerdal 1982; Gietzelt and Nethercot 1983). Accord-  
ing to these findings, the 2T approach can be used, calculating the ϕ ¼ 0.5 1 þ αLT ðλ̄LT − 0.2Þ þ λ̄2LT ð3Þ
LTB resistance in a similar manner as for plain-webbed I-section
members, but with all cross-sectional properties calculated at
sffiffiffiffiffiffiffiffiffiffiffi
W y fy
the center of the castellation. Because the proposed 2T approach λ̄LT ¼ ð4Þ
M cr
will be based on the design method used for LTB of plain-webbed
I-section members, the latter method will be concisely discussed
subsequently.
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
π π2
According to the current version of the European standard for M cr ¼ GI t EI z þ 2 EI z EI w ð5Þ
L L
the calculation of steel structures EN1993-1-1:2005 (CEN 2005,
which will be referred to in the following as EC3), the buckling
resistance M Rd of a steel I-section beam should be calculated ac- Table 1. Imperfection Factors Corresponding with Different Buckling
cording to Eq. (1). In this equation, W y is the appropriate section Curves, according to EC3
modulus, f y the yield stress, and γ M1 is the partial factor for
Imperfection
the resistance of members to instability assessed by member
Buckling curve factor αLT
checks (for which EC3 recommends a value of 1.0). The reduction
a 0.21
b 0.34
c 0.49
d 0.76

Table 2. LTB Buckling Curve Selection for Hot-Rolled Sections according

to EC3
Geometry of Buckling
cross section curve
Fig. 2. Lateral-torsional buckling failure of a castellated member H=b ≤ 2 a
loaded by a bending moment H=b > 2 b

© ASCE 04016197-2 J. Struct. Eng.

J. Struct. Eng., 04016197

 Table 3. Different Preliminary Buckling Curve Proposals according to while it is not known how the wider array of hexagonal opening
Literature, in Function of the Height-to-Width Ratio H=b of the shapes in castellated members will influence the LTB critical
Completed Cellular or Castellated Member moment and LTB resistance. Using the results of the numerical
Source of proposal H=b ≤ 2.0 H=b > 2.0 parametric study, the validity of the 2T approach will be confirmed
Sonck and Belis (2015) c c
for castellated beams and a first proposal will be made for the buck-
Boissonnade et al. (2013) c c ling curve choice.
Lakusic et al. (2008) b c
Maquoi et al. (2003) a b
Description of LTB Experiments

According to the proposed approach, the section modulus W y The utilized numerical model for the study of the LTB behavior
and the critical buckling moment M cr are calculated using the of castellated members was validated by comparing its results with
cross-sectional properties at the center of the castellation. It is pro- the results of three LTB experiments on castellated beams executed
posed to use the plastic section modulus W y;pl as section modulus, at Ghent University, which are described in this section.
because none of the examined sections were prone to local buckling
Downloaded from ascelibrary.org by Tufts University on 10/19/16. Copyright ASCE. For personal use only; all rights reserved.

effects (as will be confirmed by the numerical results given later in

Specimens
this work).
While specific instructions are given for I-section beams All three tested castellated beams were made from an IPE160
(e.g., Table 2), it is less clear which buckling curve should be used parent section with steel grade S275, of which the nominal yield
for the castellated beams. In literature, different proposals have stress f y is 275 MPa. The geometry of these members was similar,
been made, depending on the height-to-with ratio H=b of the but the length differed. The mean measured dimensions of the
completed castellated or cellular beam (Table 3). Additionally, section (according to Fig. 3) and the specimen length L are given
the detrimental influence of the production process on the present in Tables 4 and 5. The yield stresses of the top flange f y;tfl , bottom
residual stresses and the corresponding LTB resistance, as illus- flange fy;bfl , and the web f y;web were measured using an uniaxial
trated in earlier work (Sonck et al. 2015), was not taken into ac- tensile test (Table 5). All members were made according to the stan-
count in certain studies (Boissonnade et al. 2013; Lakusic et al. dard production method depicted in Fig. 1.
2008; Maquoi et al. 2003), which makes these proposals possibly The proposed residual stress pattern for the completed castel-
unsafe. Additionally, only cellular members were considered for lated members is given in Fig. 4. This pattern is based on residual
other studies (Sonck and Belis 2015; Boissonnade et al. 2013), stress measurements that were executed in the flanges and web of
the castellated beam specimens. These measurements were
executed at the web post (between two openings) and at the tee
section (at the center of an opening). The sectioning method
was used to determine the residual stresses, and the relaxation
strains were measured using electrical strain gages. More details

Fig. 3. Castellated member geometry

Table 4. Mean Measured Dimensions (in mm) of Castellated Beam

Specimens
Dimension Measured value
tf 7.3
tw 5.5
H 220.0 Fig. 4. Residual stress patterns for castellated members derived
b 83.1 from the results in Sonck et al. (2014) (reprinted from Journal
a 118.9 of Constructional Steel Research, Vol. 105, Delphine Sonck and
w 73.8 Jan Belis, “Lateral–torsional buckling resistance of cellular beams,”
c 31.2 pp. 119–128, copyright 2016, with permission from Elsevier)

Table 5. Overview of Specimen Lengths L, Load Distances LF , Number of Openings n, Measured Yield Stresses f y , and Comparison of Experimentally
and Numerically Obtained Failure Loads Fexp and Fabq
L LF fy;tfl f y;bfl f y;web Fexp Fabq Fexp =Fabq − 1
Specimen (m) (m) N (MPa) (MPa) (MPa) (kN) (kN) (%)
CS1_L3 3.15 0.21 15 352 349 332 25.92 24.29 6.7
CS1_L4 3.99 1.89 19 348 342 320 23.06 24.25 −4.9
CS1_L6 6.09 1.89 29 352 349 332 6.39a — —
a
Failure load possibly not reached during experiment.

© ASCE 04016197-3 J. Struct. Eng.

J. Struct. Eng., 04016197

 from this experiment can still be used for the validation study by
comparing the experimentally and numerically obtained load-
displacement curves during the prebuckling stage.

Finite-Element Modeling and Validation

The LTB behavior of castellated members was studied by perform-

ing a numerical parametric study in ABAQUS 6.12. The utilized
numerical model was validated using the experimental results.
The validation study, numerical model, and varied parameters in
the parametric study will be described subsequently.

Validation of Numerical Model

Downloaded from ascelibrary.org by Tufts University on 10/19/16. Copyright ASCE. For personal use only; all rights reserved.

The numerical model was validated by comparing its results with

the experimental results described earlier. In the numerical model,
the flanges and the web were modeled by quadratic shell elements
Fig. 5. LTB test setup (reprinted from Journal of Constructional Steel with reduced integration (S8R). The dimensions of the cross
Research, Vol. 105, Delphine Sonck and Jan Belis, “Lateral–torsional section and openings corresponded with the measured values of
buckling resistance of cellular beams,” pp. 119–128, copyright 2016, Table 4. The boundary conditions corresponded with those used
with permission from Elsevier) in the experiments. The beam was loaded by two vertical point
loads at the locations of load application.
For each of the three specimens, a GMNIA-type (geometrical
and material nonlinear analysis with imperfections) calculation
about the measurements and the derivation of the proposed patterns was executed by performing a geometrically nonlinear arc-length
can be found in previous studies (Sonck et al. 2014; Sonck 2014; controlled analysis. The material of the beam was modeled as per-
Sonck et al. 2015). fectly elastic-plastic without hardening. The elastic behavior was
modeled by using a Young’s modulus E of 205 GPa [which is
Experimental Setup the mean value for steel structures according to Simões et al.
(2009)] and a Poisson’s modulus ν of 0.3. As yield stresses fy ,
The specimens were tested in a symmetric four-point bending the measured values from Table 5 were used.
test setup, as sketched in Fig. 5. Each beam was simply supported A geometric imperfection corresponding with an LTB
by fork supports, which correspond with the commonly used eigenmode for a similar member loaded in constant bending
LTB boundary conditions in theoretical calculations. The length L was introduced. The amplitude of the imperfection was L=1000.
between the supports is given in Table 5. Each beam was subjected The residual stresses from Fig. 4 were introduced by means of
to two point loads F=2 at a distance of LF apart, by means of a the ABAQUS user subroutine SIGINI. Using this subroutine, the
loading beam that was loaded centrally by a hydraulic jack initial stresses at the integration points of each element could be
(Table 5). The loading system allowed free lateral movement and determined as a function of its coordinates. Because these coordi-
rotation of the points of load application. Thus, the lateral movement nates are the coordinates of the imperfect beam, the geometric im-
and rotation of the specimens during the LTB deformation were not perfection needed to be deducted from these coordinates to obtain
hindered. More details about the test setup and instrumentation are the coordinates of the perfect member. Using the latter coordinates,
given in previous studies (Sonck 2014; Sonck and Belis 2015). a constant residual stress pattern was introduced for each beam,
the exact residual stress values depending on the beam’s geometry
Results (as specified in Fig. 4).
A comparison of the obtained failure loads for the two shorter
The three castellated beams all failed because of lateral-torsional CS1_L3 and CS1_L4 sections is given in Table 5. The numerical
buckling. The achieved total maximum loads Fexp are given in (marked abq) and experimental (marked exp) load-displacement
Table 5. As expected, the failure loads were the largest for the curves are given in Fig. 6. For the CS1_L6 specimen, only the pre-
shortest specimens. For each of the three specimens, the load- failure deformations can be compared, as explained earlier. As
displacement curves are given in Fig. 6 (marked exp). The displace- shown, the agreement between the numerical model and the experi-
ment w is the mean vertical displacement of the two loading points, ments is very good. Further validation of the numerical model can
while the rotation φ is the mean torsional rotation of the two be found in Sonck (2014) and Sonck and Belis (2015), where a
loading points. similar model for LTB of cellular members is validated by compar-
The plastic behavior of the shorter CS1_L3 specimens is clearly ing its results with five experimental results.
visible in Fig. 6; a considerable plastic deformation remains present
after the complete unloading of the specimen. For the two longer
Numerical Model Used for Parametric Study
specimens, CS1_L4 and CS1_L6, the plastic deformation was con-
siderably smaller, as expected for members with a larger slender- For each considered beam, two types of analyses were done in the
ness and more elastic behavior. For the longest CS1_L6 specimen, parametric study. Firstly, the elastic critical LTB moment M cr;abq
both the load and displacement were still slightly increasing at the was determined using a linear buckling analysis (LBA) calculation.
removal of the load. Because of the manual operation of the hy- Secondly, the true LTB failure moment M Rd;abq of each beam was
draulic jack, it was very difficult to further increase the load at this determined using a GMNIA calculation in an arc length controlled
point, which resulted in a premature load removal. This possibly analysis step (with a modified Riks algorithm). The LBA calcula-
caused in an incorrect value of the failure load. However, the results tions determined the eigenvalue of a perfect and linear elastic

© ASCE 04016197-4 J. Struct. Eng.

J. Struct. Eng., 04016197

Downloaded from ascelibrary.org by Tufts University on 10/19/16. Copyright ASCE. For personal use only; all rights reserved.

Fig. 6. Load-displacement curves UGent: total load F versus vertical displacement w and total load F versus torsional rotation φ (note: the experi-
mental results are indicated with subscript exp, while the numerical results are indicated with abq)

member. However, the geometrically nonlinear GMNIA calcula- a constant bending moment were considered. The beams were sup-
tions included geometric imperfections, elastic-plastic material ported by fork supports, which were modeled by preventing the
behavior, and residual stresses. Thus, the modified residual stress displacements U y and U z in y and z direction, as well as the rotation
pattern will influence only the GMNIA results, while the modified about the x-axis URx at both ends of the beam. Additionally, the
geometry will influence both the LBA and GMNIA results. longitudinal displacement U x was prevented at the central node of
All beams were modeled using S8R shell elements for the one beam end [Fig. 7(a)]. The bending moments were applied at the
flanges and the web, using the nominal dimensions of each section. beams’ ends by means of line loads on the web and the flanges
In the parametric study, only simply-supported members loaded by [Fig. 7(b)]. The magnitude of the line loads was chosen such that

Fig. 7. FE model: (a) applied boundary conditions; (b) load introduction

© ASCE 04016197-5 J. Struct. Eng.

J. Struct. Eng., 04016197

 Table 6. Dimensions of the Parent Sections (in mm) according to Fig. 3 to Table 6. The total beam height H ¼ fH · h, opening angle α,
Parent and web post width w ¼ f w · l0 ¼ fw · ðw þ 2cÞ (Fig. 3), along
section IPE300 IPE600 HE320A HE650A HE320M HE650M with the minimum length Lmin ¼ fL · H were varied (Table 7).
For each parent section, five different f L values were chosen so
h 300 600 310 640 359 668
b 150 220 300 300 309 305
that the slenderness values λ̄LT varied around 0.5, 1, 1.5, 2, or
tw 7.1 12.0 9.0 13.5 21.0 21.0 2.5, with a minimum fL value of 5.
tf 10.7 19.0 15.5 26.0 40.0 40.0 The castellated member geometries made from the HE320M
and HE320A parent sections have a height-to-width ratio
H=b ≤ 2. However, the other four castellated member geometries
(IPE300, IPE600, HE650A, HE650M) have geometries with
they would correspond with uniform and opposite longitudinal H=b > 2. This means that different LTB curves should be used
stresses in the top and bottom halves. for both geometry groups, according to some of the existing design
Only steel of grade S235 was considered. In this manner, rules for LTB of castellated members (Table 3).
the obtained numerical results would be safe for all steel grades All feasible castellated geometries made from the six different
because the most detrimental influence of the residual stresses is parent sections were considered by taking into account the geomet-
Downloaded from ascelibrary.org by Tufts University on 10/19/16. Copyright ASCE. For personal use only; all rights reserved.

expected for this steel grade. Again, the material was modeled us- ric constraints from CEN (1998) and CTICM (2006). In total, 747
ing a bilinear stress strain curve with E ¼ 210 GPa, ν ¼ 0.3, and different castellated beam geometries were considered. All of these
f y ¼ 235 MPa. Strain-hardening was not considered, as its effect geometries have Class 1 or 2 cross sections, which means that the
on the buckling resistance will be small for the considered slender- cross section can become fully plastic before local buckling effects
ness values in the parametric study (Boissonnade and Somja 2012). arise. Thus, the use of the full plastic moment in the design rule is
The applied geometric imperfection was a half sine wave lateral acceptable. More detailed information about the parametric study
imperfection with amplitude L=1000. This imperfection amplitude can be found in Sonck (2014).
corresponds with the amplitude used for the original derivation
of the European buckling curves (Beer and Schulz 1970; Maquoi
and Rondal 1978). Local imperfections were not introduced, as Results and Discussion for Elastic Critical LTB
their influence on the global buckling behavior will be limited Moment M cr
(Boissonnade and Somja 2012). The residual stresses from Fig. 4
were introduced for the GMNIA calculations, as described in the The elastic critical buckling load Mcr;abq was determined by per-
previous section. forming LBA calculations. As expected, the observed buckling
behavior is qualitatively similar to that of plain-webbed members;
all members failed in an LTB mode without local buckling effects.
Parametric Study In Fig. 8, a comparison is given between the obtained numerical
In the parametric study, a large variety of realistic castellated beam buckling loads M cr;abq and the analytically calculated values M cr;2T
geometries was made starting from six different parent sections, using the 2T approach for the critical buckling moment [Eq. (5)].
which were European steel sections with dimensions according It is shown that the analytical approach overestimates the critical
buckling load for the shorter members, making this approach
unsafe. These unsafe deviations are caused by web distortion, as
explained for plain-webbed members by Bradford (1992), Schafer
Table 7. Varied Cross-Sectional Parameters in Parametric Study
and Ádány (2005), and Trahair (1993). Once the effects of plastic-
Parameter Considered values ity are taken into account the afflicted short members will fail
fH 1.4 by plastic yielding instead of elastic buckling so that this effect
1.5 can be neglected (Sonck et al. 2011; Zirakian and Showkati
1.6 2006). A similar behavior was observed for cellular members in
α 45° previous studies (Sonck 2014; Sonck and Belis 2015).
60° For the longer lengths, an underestimation of M cr is visible,
75° which is caused by the (safe) underestimation of the torsion con-
fw 0.1 stant I t by the 2T approach. This will be more evident for the longer
0.3
lengths, as can be deduced from Eq. (5). A more correct value of the
0.5
torsion constant I t can be obtained by using a weighted average

Fig. 8. Critical LTB moment: comparison of numerically obtained values M cr;abq with analytically obtained values M cr;2T and Mcr;avg

© ASCE 04016197-6 J. Struct. Eng.

J. Struct. Eng., 04016197

 opening for which α ¼ 60° and c ¼ w · cosð60°Þ. For this hexago-
nal opening, l0;avg ¼ w þ 1.5 · w · cosð60°Þ ¼ 1.75w ¼ 0.875 · 2w ¼
0.875 · l0 . This agrees well with the results obtained for cellular
members with circular openings with diameter a, for which an
equivalent rectangular opening with height a and width 0.9a
was proposed.

Results and Discussion for LTB Resistance M Rd

The LTB failure moment M Rd;abq was determined numerically in
GMNIA calculations as the maximum of the load-deflection dia-
Fig. 9. Equivalent rectangular openings for calculation of weighted gram. However, for most of the longer geometries, this maximum
average I t;avg and corresponding M cr;avg could not be determined as the diagram kept increasing past the
critical LTB moment. This is caused by the stabilizing effects of
Downloaded from ascelibrary.org by Tufts University on 10/19/16. Copyright ASCE. For personal use only; all rights reserved.

the prebuckling deformations in the geometrically nonlinear analy-

sis (Trahair 1993). Consequently, the results for these geometries
approach for a beam with equivalent rectangular openings of length were not taken into account.
l0;avg instead of the total hexagonal opening length l0 ¼ w þ 2c The results of the numerical calculations M Rd;abq can now be
(Fig. 9). The corresponding weighted average torsion constant compared with the analytical results M Rd obtained for the different
I t;avg can be calculated using Eq. (6), in which I t;2T is calculated buckling curves, using the weighted average approach for I t and the
using the 2T approach and I t;full is the torsion constant of the gross 2T approach for the other cross-sectional properties. In Fig. 10, a
cross section without an opening. After a comparison with the direct comparison of the obtained results with the different buckling
numerical results, it was found that the equivalent opening length curves is shown. For this purpose, the slenderness λ̄LT was calcu-
l0;avg ¼ w þ 1.5c provided the best fit. As is shown in Fig. 8, this lated for each considered geometry using Eq. (4). The numerically
considerably improves the agreement between the numerical results obtained reduction factor χabq for each geometry is calculated
and the analytical calculations. In the remainder of this paper, the using Eq. (1), with M Rd ¼ MRd;abq . In both equations, the plastic
weighted average approach will be used to calculate the torsion section modulus of the 2T section is used. In Table 8, the results for
constant I t;avg for the critical moment Mcr each parent section are compared with the results obtained for
buckling curves a–d, by means of the minimum and maximum val-
n · lo;avg n · lo;avg
 
I t;avg ¼ I t;2T þ 1 − I t;full ð6Þ ues of the deviation ΔM Rd [Eq. (7)]
L L
 
M Rd;abq
The dimensions of the equivalent rectangular openings can ΔM Rd ¼ − 1 · 100 % ð7Þ
be compared with the dimensions obtained for cellular members M Rd
in Sonck and Belis (2015) by considering a regular hexagonal
Compared with the results for the other sections, very high re-
duction factors are obtained for the HE320M sections (around
buckling curve a). The results for the other sections lie around
LTB curve c. However, based on the existing design rules, in which
the buckling curve choice depends on the H=b ratio of the sections,
it was expected that similar results would be obtained for the
HE320A and HE320M sections. Thus, the results show that the
H=b criterion is ill-suited for the current lateral-torsional buckling
curve choice. This conclusion is also valid for plain-webbed mem-
bers, as already demonstrated in several studies (Taras and Greiner
2008, 2010; Taras 2010).
For the HE320M geometries, buckling curve a would be most
suited, while buckling curve c would be more suitable for the other
geometries, disregarding the small unsafe deviations for the
HE320A geometries. If only one buckling curve were to be pro-
Fig. 10. LTB resistance: comparison of numerically obtained values
posed for all parent section geometries, then buckling curve c
χabq with buckling curves
would be most suited. This corresponds with the findings obtained

Table 8. Agreement between Different Buckling Curves and Numerical Results: Deviation ΔM Rd for Each Parent Section Group
A B C Dd
ΔM Rd Minimum Maximum Minimum Maximum Minimum Maximum Minimum Maximum
HE320A −20.7 −1.9 −11.6 5.4 −6.0 13.6 3.1 27.6
HE320M −2.2 2.3 4.7 14.1 11.8 26.2 23.5 45.9
IPE300 −16.6 3.5 −10.0 8.9 −3.2 15.0 7.8 27.3
HE650A −17.3 3.7 −10.5 9.3 −3.7 15.6 6.7 27.9
HE650M −14.1 7.8 −7.3 14.2 −0.2 21.4 11.2 36.1
IPE600 −20.3 4.2 −12.2 9.0 −3.8 14.6 9.9 26.3

© ASCE 04016197-7 J. Struct. Eng.

J. Struct. Eng., 04016197

 for cellular beams in previous studies (Sonck and Belis 2015; The 2T approach can also be used for the calculation of the LTB
Boissonnade et al. 2013). resistance M Rd . Using a numerical study, a design approach that
In Maquoi et al. (2003) higher resistances were found because would fit in the current EC3 approach for the calculation the
the detrimental modification of the residual stress pattern was not LTB resistance of plain-webbed I-section beams was determined.
taken into account in this work. Instead, the more advantageous In this study, the detrimental effect of the fabrication process on the
residual stress pattern valid for the original parent sections was residual flange stresses, which was previously never considered,
used. The proposal made by Lakusic et al. (2008) seems accurate was taken into account. The results showed that buckling curve c
for all sections with H=b > 2.0, but is overly conservative com- would be most suited for all sections. Earlier obtained results
pared with the HE320M results and unsafe for the HE320A sec- illustrate that the modification of the residual stress pattern during
tions. Because it is unclear which assumptions were used in the the fabrication process of castellated beams will result in resistan-
numerical simulations by Lakusic et al. (2008), it is not possible ces that lie approximately one buckling curve lower than the resis-
to determine the cause of these deviations. tances obtained using the original residual stress pattern. This
These findings are still preliminary, and it should be checked corresponds with a maximum decrease of the buckling resistance
whether the assumed residual stress pattern, which was based of 13%.
on residual stress measurements in castellated members made from However, further research will be necessary to confirm the val-
Downloaded from ascelibrary.org by Tufts University on 10/19/16. Copyright ASCE. For personal use only; all rights reserved.

relatively light IPE160 parent sections, is also valid for heavier pa- idity of the used residual stress pattern for other sections and differ-
rent sections or castellated members made using different cutting or ent fabricators, as well as to study the effect of different bending
welding procedures. Additionally, it would be appropriate to con- moment diagram shapes. For the former, the composition of a
sider more bending moment patterns in the numerical simulations. worldwide database of residual stress measurement results for cas-
Based on these additional data, a definitive lateral-torsional buck- tellated and cellular members would prove very beneficial. In order
ling curve formulation could be proposed, using a similar statistical to obtain this database it is recommended to measure the residual
evaluation as the one performed by Rebelo et al. (2009). stresses at the web post and tee section of each tested cellular or
In previous studies (Sonck 2014; Sonck and Belis 2015; Sonck castellated member specimen.
et al. 2015) the influence on the failure behavior of the modification
of the residual stresses during the production was studied for cel-
lular members. It was found that the residual stress change resulted Acknowledgments
in a decrease of the buckling resistance of approximately one buck- The authors would like to acknowledge Huys-Liggers (Venlo,
ling curve. It is expected that this will also be the case for the cas- Netherlands) for the production of the castellated members used
tellated members, for which only the opening shape is different. for the LTB experiments performed at Ghent University. Part of
This is confirmed by earlier work (Sonck et al. 2015), where the computational resources (STEVIN Supercomputer Infrastruc-
the effect of the modification of the residual stress pattern on ture) and services used in this work were kindly provided by Ghent
the lateral-torsional and weak-axis flexural buckling resistance University, the Flemish Supercomputer Center (VSC), the Hercules
was examined numerically. In this work, GMNIA calculations were Foundation, and the Flemish Government, department EWI.
executed for cellular and castellated members. These members
were simply supported in fork supports and loaded by either a com-
pressive load or a constant bending moment. Two different residual References
stress patterns were considered in the GMNIA simulations: the
original residual stress pattern and the modified pattern from Fig. 4. ABAQUS 6.12 [Computer software]. Dassault Systèmes, Waltham, MA.
As original residual stress pattern, the standard pattern for I-section Beer, H., and Schulz, G. W. (1970). “Bases Théoriques des Courbes
Européennes de Flambement.” Construction Métallique, 3, 37–57.
members proposed by the ECCS (1984) was used. Modifying this
Boissonnade, N., and Somja, H. (2012). “Influence of imperfections in
pattern to the residual stress pattern proposed in Fig. 4 resulted in a FEM modeling of lateral torsional buckling.” Proc., Annual Stability
maximum decrease of the resistance of 13%. The cellular and cas- Conf., Structural Stability Research Council, (SSRC2012), Structural
tellated members displayed similar behavior, and the 13% decrease Stability Research Council (SSRC), Chicago.
was found for both lateral-torsional buckling and weak-axis flexu- Boissonnade, N., Nseir, J., Moussa, L., and Somja, H. (2013). “Design of
ral buckling. More details about this study can be found in Sonck cellular beams against lateral torsional buckling.” Proc. Inst. Civ. Eng.
et al. (2015). Struct. Build., 167(SB7), 436–444.
Bradford, M. A. (1992). “Lateral-distortional buckling of steel I-section
members.” J. Constr. Steel Res., 23(1–3), 97–116.
CEN (European Committee for Standardization). (1998). “Design of steel
Conclusions Structures. Part 1-1: General rules and rules for buildings. Annex N:
Openings in webs.” ENV 1993-1-1, Eurocode 3, Brussels, Belgium.
This paper investigated the LTB behavior of castellated beams CEN (European Committee for Standardization). (2005). “Design of steel
using a numerical parametric study based on a validated finite- structures. Part 1-1: General rules and rules for buildings.” ENV 1993-1-
element model. The suitability of the already existing 2T approach 1, Eurocode 3, Brussels, Belgium.
paired with the current EC3 approach for LTB of plain-webbed CTICM (Centre Technique Industriel de la Construction Métallique).
I-section beams was examined. According to the 2T approach, (2006). “Arcelor cellular beams: Detailed technical description.”
the LTB resistance is calculated in an identical manner as for Saint-Aubin, France.
plain-webbed I-section members, but all cross-sectional properties ECCS (European Convention for Constructional Steelwork). (1984).
are calculated at the location of the opening. “Ultimate limit state calculations of sway frames with rigid joints.”
Publication No. 33, Brussels, Belgium.
It was confirmed that the 2T approach would be suitable for the
Gietzelt, R., and Nethercot, D. A. (1983). “Biegedrillknicklasten von
calculation of the critical buckling moment M cr of castellated Wabenträgern.” Stahlbau, 11, 346–349.
beams. This method could be improved even more by using a Kerdal, D., and Nethercot, D. A. (1984). “Failure modes for castellated
weighted average approach for the torsional constant I t , consider- beams.” J. Constr. Steel Res., 4(4), 295–315.
ing a beam with equivalent rectangular openings with height a and Lakusic, V. T., Dzeba, I., and Androic, B. (2008). “The buckling curve
width w þ 1.5c. for lateral-torsional buckling resistance of castellated beams.” Proc.,

© ASCE 04016197-8 J. Struct. Eng.

J. Struct. Eng., 04016197

 5th European Conf. on Steel Structures (Eurosteel 2008), ECCS, Sonck, D., Van Impe, R., and Belis, J. (2014). “Experimental investigation
Brussels, Belgium, 1587–1592. of residual stresses in steel cellular and castellated members.” Constr.
Maquoi, R., et al. (2003). Lateral torsional buckling in steel and composite Build. Mater., 54, 512–519.
beams, European Commision, Brussels, Belgium. Sonck, D., Vanlaere, W., and Van Impe, R. (2011). “Influence of plasticity
Maquoi, R., and Rondal, J. (1978). “Mise en equation des nouvelles courbes on lateral-torsional buckling behaviour of cellular beams.” Mater. Res.
Européennes de flambement.” Construction Métallique, 1, 17–29. Innov., 15(sup1), s158–s161.
Nethercot, D. A., and Kerdal, D. (1982). “Lateral-torsional buckling of Sonck, D., Van Lancker, B., and Belis, J. (2015). “Influence of residual
castellated beams.” Struct Eng., 60B(3), 53–61. stresses on the global buckling resistance of cellular and castellated
Rebelo, C., Lopes, N., Simões da Silva, L., Nethercot, D. A., and Vila Real, members.” Proc., Annual Stability Conf., Structural Stability Research
P. M. M. (2009). “Statistical evaluation of the lateral-torsional buckling Council, (SSRC2015), Structural Stability Research Council (SSRC),
resistance of steel I-beams. Part 1: Variability of the Eurocode 3 resis-
Chicago.
tance model.” J. Constr. Steel Res, 65(4), 818–831.
Taras, A. (2010). “Contribution to the development of consistent stability
Schafer, B. W., and Ádány, S. (2005). “Understanding and classifying local,
design rules for steel members.” Ph.D. dissertation, Graz Univ. of Tech-
distortional and global buckling in open thin-walled members.” Proc.,
nology, Graz, Austria.
Annual Stability Conf., Structural Stability Research Council,
(SSRC2012), Structural Stability Research Council (SSRC), Chicago, Taras, A., and Greiner, R. (2008). “Development of consistent buckling
curves for torsional and lateral-torsional buckling.” Steel Constr.,
Downloaded from ascelibrary.org by Tufts University on 10/19/16. Copyright ASCE. For personal use only; all rights reserved.

1–20.
Simões da Silva, L., Rebelo, C., Nethercot, D. A., Marques, L., Simões, R., 1(1), 42–50.
and Vila Real, P. M. M. (2009). “Statistical evaluation of the lateral- Taras, A., and Greiner, R. (2010). “New design curves for lateral-torsional
torsional buckling resistance of steel I-beams. Part 2: Variability of steel buckling. Proposal based on a consistent derivation.” J. Constr. Steel
properties.” J. Constr. Steel Res., 65(4), 832–849. Res., 66(5), 648–663.
Sonck, D. (2014). “Global buckling of castellated and cellular steel beams Trahair, N. S. (1993). Flexural-torsional buckling behaviour of structures,
and columns.” Ph.D. dissertation, Ghent Univ., Ghent, Belgium. E & FN Spon, London.
Sonck, D., and Belis, J. (2015). “Lateral-torsional buckling resistance of Zirakian, T., and Showkati, H. (2006). “Distortional buckling of castellated
cellular beams.” J. Constr. Steel Res., 105, 119–128. beams.” J. Constr. Steel Res., 62(9), 863–871.

© ASCE 04016197-9 J. Struct. Eng.

J. Struct. Eng., 04016197