Journal of Constructional Steel Research 166 (2020) 105911

Contents lists available at ScienceDirect

Journal of Constructional Steel Research

Numerical and analytical analyses of high-strength steel cellular beams:

A discerning approach
Rabee Shamass a,⁎, Federico Guarracino b
a
Division of Civil and Building Services Engineering, School of Build Environment and Architecture, London South Bank University, UK
b
Department of Structural Engineering, University of Naples `Federico II', Via Claudio, 21 - 80125 Napoli, Italy

a r t i c l e i n f o a b s t r a c t

Article history: The behaviour of cellular beams made from normal and high strength steel with various geometries is
Received 9 September 2019 investigated through a large number of finite element analyses and a simple mechanical model for the
Received in revised form 4 December 2019 Web-Post Buckling (WPB) failure is developed and analysed in order to highlight the factors which influence
Accepted 13 December 2019
its occurrence and development for both normal and High-Strength (HS) steels. The performed FE analyses
Available online xxxx
and the proposed modelling, once calibrated, allow to shed some light on the characteristics of the phenomenon
Keywords:
and to provide the basis of a reliable design method to predict shear buckling of web-post of cellular beams made
Cellular beams both of mild and HS steel.
Web-post buckling © 2019 Elsevier Ltd. All rights reserved.
High strength steel

1. Introduction Many experimental tests have been conducted on cellular beams

with different geometries to investigate their structural behaviour
Cellular steel beams are widely used in modern constructions due to and failure modes. Tsavdaridis and D'Mello [4] tested closely spaced
their economic and aesthetic advantages, compared to solid I-beam. web openings of various shapes and found that all the tested cellular
They are applicable to situations where services (e.g. conduits for venti- beams failed by WPB. More recently Grilo etal. [5] tested nine cellu-
lation or air conditioning) are to be fitted within required structural lar beams and the WPB failure dominates the failure of all beams.
depths or long spans are requested along with desirable creative Erdal and Saka [6] tested 12 cellular beams and found that 6 beams
appearance. They are manufactured from hot-rolled I-beam section by failed by WPB and two beams failed by combined WPB and VB.
cutting along web length in a certain pattern to create the upper and Panedpojaman et al. [7] tested 9 cellular beams with different geom-
lower halves of the cellular beam and re-welding again to form a cellular etries and loading conditions (i.e. one or two point loads) and found
beam, as seen in the Fig. 1, where H is the total height of the cellular sec- that seven beams failed by WPB and two beam failed by combined
tion; tw is the web thickness; tf is the flange thickness; bf is the flange WPB and LTB. It can be noted from the various experiments that
width; d is the height of the parent section; do is the opening diameter; web-post buckling is typically the limiting cause, especially for
s is the centre-to-centre spacing of adjacent openings; so is the edge-to- large closely spaced openings ones.
edge spacing of adjacent opening. The opening up of rolled beam in- An early empirical design method approaching the WPB failure was
creases its section modulus and moment of inertia, resulting in greater officially published by SCI publication 100 [1] and was adopted by AISC
strength and rigidity. The reduction in the beams weight has an obvious [8] for the design of castellated and cellular beams. The design curves of
effect on material and handling cost. the web-post were obtained from a series of validated non-linear finite
Unfortunately, the flexural behaviour of cellular I-shaped steel beam element analyses of single web-post model. Another empirical
can be a complex issue on account of its susceptibility to several failure web-post buckling model proposed by Tsavdaridis and D'Mello [4]
types and instability modes: Vierendeel bending (VB) due to shear dis- was derived from the finite element analyses and depends on the
tribution across the opening (Fig. 2a), overall bending failure (BF) opening-spacing ratio and opening-web thickness ratio. Lawson et al.
(Fig. 2b), shear failure (SF) due to reduced steel section (Fig. 2c), web- [9] proposed a different design methodology to calculate the shear
post buckling (WPB) due to horizontal shear force in the web-post buckling resistance based on the design strut analogy in which an effec-
(Fig. 2d), lateral torsional buckling (LTB) (Fig. 2e), and rupture of weld tive length in the compressed diagonal strut is determined, and the
joints (Fig. 2f). compression stress resistance in this diagonal strut is calculated using
the BS 5950–1:2000 compression resistance formulation or the relevant
buckling curve of Eurocode 3 [10]. Panedpojaman et al. [7] proposed
⁎ Corresponding author. further improvement of the effective length of the strut model that
E-mail address: [email protected] (R. Shamass). takes into consideration the effect of the strut's boundary conditions.

https://doi.org/10.1016/j.jcsr.2019.105911
0143-974X/© 2019 Elsevier Ltd. All rights reserved.
2 R. Shamass, F. Guarracino / Journal of Constructional Steel Research 166 (2020) 105911

Fig. 1. (a) Parent I-beam section; (b) Cellular beam geometry.

Grilo et al. [5] proposed a new formulation to determine the shear resis- members made from HS steel can be significantly different from those
tance in cellular beams for the web-post buckling based on resistance made from conventional mild steel. The vast majority of research into
curved derived from the finite element analyses. HS steel has been focused on bare beam steel sections (e.g. [11–13]),
High strength (HS) steels whose nominal yield strength exceeds column steel sections (e.g. [14–16]) and connections (e.g. [17,18]). A
460 MPa are produced widely nowadays and there has been a strong number of research studies has been conducted on the use of HS steel
demand worldwide for their use in many buildings and bridges struc- in composite structures such as composite beams (e.g [19–21]), and
tures. The use of HS steel instead of normal strength steel in steel composite columns (e.g. [22]).
construction can reduce the member size, leading to smaller founda- However, there seems to have been little or no research, at least in
tions and can reduce the amount of coating material as well as lower the public domain, into the web-post buckling of cellular beams made
construction and transportation costs. from HS steel. Current design methods proposed in the literatures to
HS steel has been the subject of intensive research from the struc- calculate the shear strength of web-post are applicable to normal
tural engineering community in order to provide useful and efficient de- strength steel Grade S235 or S355. Hence the effect of steel strength
sign data. The structural behaviour, failure modes and ductility of steel on the WPB of cellular beams is addressed in this study.

Fig. 2. Some failure modes of cellular beams: (a) Vierendeel bending, VB [1]; (b) overall bending failure, BF [1]; (c) shear failure, SF [2]; (d) web-post buckling WPB; (e) lateral torsional
buckling, LTB [3]; and (f) rupture of weld joints [2].
 R. Shamass, F. Guarracino / Journal of Constructional Steel Research 166 (2020) 105911 3

The objective of the present study is to investigate, through finite 15%. The values of fy and fu can be obtained according to Grilo et al. [5]
element analyses, the comparative behaviour of cellular beams made for specimens A1, A2, A3, A5, A6, B1, B2, B5 and B6, and Tsavdaridis
from normal and high strength steel with various geometries. Also, and D'Mello [4] for specimen C1 and C2. The “engineering” stresses
since there is no unanimous consensus about the design procedure for and strains are converted into “true” stresses and strains in the numer-
the shear resistance determination and the results obtained by various ical model in order to consider the effects of the reduction of the cross-
authors present significant differences when compared with each section during the tensile test. The formulae for conversion are:
other, a simple mechanical model for the WPB failure is developed
and analysed on the basis of geometrical and mechanical parameters σ true ¼ σ ð1 þ εÞ
only in order to highlight the factors which influence the occurrence
and the development of the WPB both in normal and HS steels. The pro- εtrue ¼ ln ð1 þ εÞ
cedure is essentially based on the modelling proposed by Lawson et al.
[9] but, differently from other methods, does not make reference to The cellular steel beam section and stiffeners are modelled using a
code prescriptions and focuses on the mechanics of the phenomenon general-purpose three-dimensional reduced integration shell element,
only. It is shown that the proposed procedure, once calibrated with named S4R. This element is widely used for industrial applications
the aid of the performed FE analyses, can be used as a reliable design since it provides robust and accurate solutions in all loading conditions
method to predict shear buckling of web-post of cellular beams made for thin and thick shell problems and is suitable for large-strain analysis.
both of mild and HS steel and sheds some light on the characteristics The size of the mesh element plays an important role in the accuracy of
of the phenomenon. It is finally worth noticing that the finite element the results and thus a mesh sensitivity analysis has been performed for
models have been developed using commercial software ABAQUS and various element size. It was found that a balanced mesh density provid-
validated against existing experimental results conducted by Grilo ing both a good accuracy of the FE results and a reasonable computa-
et al. [5] and Tsavdaridis and D'Mello [4]. In fact, there are no tests avail- tional cost corresponded to an element size of 15 mm for all studied
able in the literature on the flexural behaviour of cellular beams made cellular beams.
from HS steel and make resort to numerical models is currently the The boundary conditions of the cellular beams are simply supported
only possibility to conduct parametric studies of cellular beams geome- at the bottom edges and the loading is applied under displacement-
tries and their material properties. The present work aims therefore to control on the top flange in the downward direction (Fig. 4).
constitute a first step towards the understanding of the buckling behav- Grilo et al. [5] stated that the vertical point load was applied with an
iour of cellular beams made of HS steel. eccentricity of 20 mm of the mid-span in order to induce the web-post
buckling, while Tsavdaridis and D'Mello [4] applied the loading at the
2. Development of the numerical modelling mid-span of the beams C1 and C2.
Tie contact was defined between the surface of steel section and the
Single web models are widely used in the literature to study the edges of the stiffeners. A point at the mid-span of the beams was later-
web-post buckling of cellular beam (e.g. [4,7]). However, the single ally restrained in order to prevent the specimens from buckling in a lat-
web model cannot capture the structural behaviour of cellular beams eral torsional mode. Two analyses have been conducted: first a linear
such as deflections along the span and post-buckling behaviour. buckling analysis in order to obtain eigenvalues and eigenmodes and,
Furthermore, there is no consensus about the ideal boundary condition successively, a nonlinear analysis using a Newton-Raphson solution
for these models. Therefore, here a full beam model has been developed method in order to obtain the buckling load and to follow the post-
using finite element software ABAQUS. The numerical model of the buckling path of the beam after failure. Geometrical imperfections due
beams is validated against the specimens tested by Tsavdaridis and to the process of cutting and fabrication of the cellular beams are inev-
D'Mello [4] and Grilo et al. [5]. itable and can cause significant drops of the value of the buckling load.
The elastic-plastic material model uses Von Mises yield criterion The imperfections contribute to initiate the bucking and their post-
with associated plastic flow and isotropic hardening. The uniaxial be- buckling behaviour is affected by the size and shape of the geometrical
haviour of steel is modelled using multi-linear stress-strain relation, as imperfections. Thus, in order to obtain reliable and accurate numerical
shown in Fig. 3. This relation is defined using Young modulus E = 210 results, initial imperfections were included in the FE model. The initial
fy imperfection shape used in the numerical analysis has been based on
Gpa, yield stress fy, yield strain εy ¼ , strain at the onset of hardening the lowest buckling mode obtained from the linear buckling analysis
E
εst =1.8%, ultimate stress fu,and corresponding ultimate strain εu = in order to apply the most critical imperfection. The imperfection ampli-
tude was assumed to be equal to δw/25 for the specimens tested by Grilo
et al. [5] (δw is the imperfection measured value) as it led to the most ac-
curate results when the models were compared to experimental results.
For the specimens C1 and C2, the imperfection size was assumed as
tw/200 = 7.6/200 = 0.038 mm [4].

Fig. 3. Stress-strain relationship of steel material in cellular beam. Fig. 4. The loading, boundary conditions and geometries of the FE model.
4 R. Shamass, F. Guarracino / Journal of Constructional Steel Research 166 (2020) 105911

Table 1 250
Comparison between experimental and numerical buckling shear load. A5 Test
Specimen Buckling shear load, V (kN) VFE/Vtest % 200 FE

Test Vtest Numerical VFE

A1 38.0 40.0 5.3 150

A6
A2 61.9 59.2 −4.3

Load (kN)
A3 70.7 65.6 −7.2
A5 99.1 98.4 −0.7 100
A2
A6 102.2 102.5 0.3
B1 54 54.9 1.6
B2 79 75.6 −4.3 50
B5 138.5 134.5 −2.9
B6 150 148.6 −0.9
C1 144.4 147.7 2.3 0
C2 127.5 120 −5.9 0 5 10 15
Deflection (mm)

350
3. Validation of the finite element model
Exp
300 B6
In order to evaluate the accuracy of the beam model described above FE
250
for analysing the structural behaviour of cellular beams, the buckling
load, the load-deflection relationship and the failure modes for the cel-

Load (kN)
200
lular beams are compared with the corresponding experimental results B2
150
obtained by Grilo et al. [5] for specimens A1, A2, A3, A5, A6, B1, B2, B5
and B6, and by Tsavdaridis and D'Mello [4] for specimen C1 and C2. 100
Due to the symmetry of the beam model, the vertical bucking shear B1
50
load of cellular beam corresponds to half of the ultimate load.
0
3.1. The web-post bucking resistance of the cellular beams 0 5 10 15 20
Deflection mm
The numerical buckling loads for all studied specimens are pre-
sented together with their experimental counterparts in Table 1. It can Fig. 5. Load-deflection relationship predicted by FE model and obtained experimentally.
be observed that the predicted loads are in very good agreement with
the corresponding experimental results in which the numerical vertical accurate prediction of the behaviour of steel cellular beams in terms of
shear loads differ from the experimental ones from −7.2% to 5.3% with buckling shear load, load-displacement response and failure mode.
an average deviation of −1.5%. Consequently, it can be concluded that There are some small differences in terms of initial bending stiffness,
the FE model adopted in the present study is adequate for predicting but this is most likely due to localised yielding in the test or due to the
the buckling shear strength of cellular beams with various geometries. machine compliance, which cannot be assessed accurately in the nu-
merical model. Therefore, in the following sections, the model is used
3.2. Load-displacement response to investigate the effect of the geometrical and material parameters on
the behaviour of cellular beams under analysis.
Fig. 5 shows the load vs mid-span displacement response in the nu-
merical simulation and in the experimental setting. For brevity, only
cases A2, A5, A6, B1, B2 and B6 are presented as they are reflective of
the results for all specimens. It is observed that the initial bending stiff-
ness obtained from FE model are in good agreement with those
obtained experimentally for all specimens, except for A5 and A6 at
which the slope of the load-displacement curves before the ultimate
load is attained are higher than those observed experimentally. This
can be attributed to the difference between the multi-linear stress-
strain relationships of Fig. 3 and the actual constitutive laws, see Yun
and Gardner [23]. Furthermore, the failure mode predicted numerically
for all specimens is web-post buckling (WPB), which agrees with the
failure mode observed experimentally. Fig. 6 illustrates the comparison
between the predicted failure mode and the experimental failure mode
for the beam B5 and A5.
The geometric parameters H/do, s/do and do/tw are those which es-
sentially affect the behaviour of cellular beams. The specimens A2 and
A5 present the ratio H/do equal to 1.26 and 1.64, respectively, and the
ratio do/tw equal to 71.4 and 51.8, respectively. The yield stress for
both specimens is 416 Mpa. It can be noted that as the ratio H/do in-
creases and do/tw decreases, the ultimate load and the initial bending
stiffness increase and the deflection at the ultimate load decrease.
Similar observations can be made for the specimens B1, A6 and B6 for
which the H/do ratio is equal to 1.26, 1.63 and 1.67, respectively.
Overall, on the basis of data presented in this and previous sections,
it can be concluded that the FE model adopted is capable of providing an Fig. 6. Comparison between FE and experimental failure mode for (a) B5 and (b) A5.
 R. Shamass, F. Guarracino / Journal of Constructional Steel Research 166 (2020) 105911 5

Table 2 A large number of numerical simulation have been conducted to inves-

Material properties of cellular beams for the parametric study [20]. tigate the effect of each of these parameters on the ultimate load or
Steel Grade fy (MPa) fu (MPa) εst (%) εu (%) buckling shear load, on the deflection corresponding to the ultimate
S355 355 510 2.5 18
load, on the initial bending stiffness, on the failure mode (WPB or VB)
S460 460 550 2.0 14 and on the post-buckling behaviour. The beams B2 and B6 with
S690 690 770 εy 8.8 H/d o = 1.25 and 1.67, respectively, are used for this parametric
S960 960 980 εy 5.5 study. The members are assumed to be simply supported with a single
point load with an eccentricity of 20 mm from the mid-span in order
to prompt the web-post buckling. Initial imperfection is introduced
in the models in the shape of the first buckling mode obtained from
the Eigenvalue analysis [24] with a scale factor of H/500 [7]. Five s/do
ratios in the range of 1.1 to 1.5 are studied with five web thicknesses
of 5.1 mm, 5.6 mm, 6.0 mm, 6.6 mm and 7.6 mm. d/tw ratio varies
from 40.13 to 59.80 and do/tw ratio varies from 32.24 to 48 for
beams with H/do = 1.67 and varies from 46.33 to 69 for beams with
H/do = 1.25. Four steel grades S355, S460, S690 and S960 are used
throughout this parametric study.
Multi-linearstress-strain relationship for the normal and high
strength steel are considered, as shown in the Fig. 3. The elastic modulus
of steel is taken as 200 GPa and the stresses and strains limits values
shown in the Fig. 3 are illustrated in Table 2. Fig. 7 shows the graphical
Fig. 7. The stress-strain relationship of steel grades used in the Table 2. representation of the stress-strain of the steel grades used in the Table 2.

4. Parametric study 4.1. Effect of geometric parameters H/do, s/do and do/tw

In this section, the validated FE model is used to study the influence Fig. 8a–d show the load-deflection relationship of cellular beams
of geometrical and material parameters on the structural behaviour of with H/do = 1.67, s/do = 1.2, 1.4 and with H/do = 1.25, s/do = 1.2,
high strength steel cellular beams. These include the steel grade, the 1.4 for S355, S460, S690 and S960 steel grades, respectively. For all
depth-to-opening diameter ratio, H/do, the opening diameter-web members the slenderness ratio d/tw is chosen to be 59.8 (the web thick-
thickness ratio, do/tw, and the spacing-to-opening diameter, s/do. ness is chosen equal to 5.1 mm).

H/do=1.67,s/do=1.2
H/do=1.67, s/do=1.2 300
250
H/do=1.25,s/do=1.2
H/do=1.25,s/do=1.2
250 H/do=1.67, s/do=1.4
200 H/do=1.67,s/do=1.4
H/do=1.25, s/do=1.4
H/do=1.25,s/do=1.4 200
150
Load (kN)

Load (kN)

150
100
100

50
50

0 0
0 5 10 15 20 25 0 5 10 15 20 25
Deflection (mm) Deflection (mm)

(a) (b)

H/do=1.67,s/do=1.2 H/do=1.67,s/do=1.2
400 450
H/do=1.25,s/do=1.2 H/do=1.25, s/do=1.2
350 400
H/do=1.67,s/do=1.4 H/do=1.67, s/do=1.4
350 H/do=1.25, s/do=1.4
300 H/do=1.25,s/do=1.4
300
250
250
Load (kN)
Load (kN)

200
200
150 150
100 100
50 50

0 0
0 5 10 15 20 25 0 5 10 15 20 25
Deflection (mm) Deflection (mm)

(c) (d)

Fig. 8. Load-deflection curves for members made from (a) S355; (b) S460; (c) S690; (d) S960.
6 R. Shamass, F. Guarracino / Journal of Constructional Steel Research 166 (2020) 105911

It can be noted from Fig. 8a to d that as the ratio H/do increases, S460, S690 and S960, respectively. It can be noticed that the
the failure load and initial bending stiffness increase, while employment of high strength steel grades leads to higher buckling
the deflection at failure decreases. For instance, for members with shear capacities and lower deflections with respect to normal steel
s/do = 1.4, when the ratio H/do increases from 1.25 to 1.67, the when the ratio do/tw decreases.
deflection at the web-post failure load decreases by 45%, 37%, 35% Figs. 10 and 11 illustrate the bucking shear load for members
and 33% for members with S355, S460, S690 and S960, respectively. with s/do varies from 1.1 to 1.5 and with do/tw varies from 32.24 to
For the same members, when the ratio H/do increases from 1.25 to 69 for normal and HS steel. The results are obtained for members
1.67, the web-post failure load increases by 53%, 60%, 66% and 70% with H/do = 1.67 and H/do = 1.25, respectively. It can be noticed
for members with S355, S460, S690 and S960, respectively. As it that the buckling shear load and the failure mode (web-post buckling
was expectable, it can be noticed that the use of high strength or Vierendeel failure, see Fig. 12) are significantly affected by the geo-
steel grades in cellular beams tends to lead to higher buckling metrical parameters H/do, s/do and do/tw, and by the steel grade. The
shear load and lower deflection with respect to those obtained bucking shear load increases with increasing ratios s/do and H/do, as
from cellular beams made of normal strength steel when the ratio well as with increasing yield stress, and with a decrease of the ratio
H/do increases. However, it also worth pointing out that, as it is ex- do/tw. The observed mode of failure for all member with H/do =
pectable for an intrinsically non-linear phenomenon, the decrease 1.67 is web-post buckling while the Vierendeel failure is observed
in deflection and the corresponding increase in the value of the fail- in some members with H/do = 1.25, particularly for the beams with
ure load cannot be related linearly. s/do = 1.5 and with high ratio of do/tw and with low yield stress
Fig. 9a–d show the load-deflection relationship of cellular beams (i.e. S355) while the use of HS steel shifts the failure mode into the
with do/tw = 32.24, 40.83 and 48.0 for S355, S460, S690 and S960 web-post buckling and provides higher failure loads. This is due to
steel grades, respectively. For all members the ratio H/do and s/do is cho- the fact that the web-post is characterised by an increment in the
sen to be 1.67 and 1.2, respectively. One can observe that as the ratio do/ width, thickness and yield stress.
tw decreases, the web-post failure load and the corresponding deflec- From Figs. 10 and 11 one can observe that HS steel provides higher
tion substantially increases while the initial bending stiffness increases web-post buckling capacities than those provided by normal strength
only slightly. For instance, as the ratio do/tw decreases from 48.0 to steel. Based on the analysis of the whole cellular beams members used
32.24, the deflection at the failure load increases by 194%, 158%, 120% in the parametric study (i.e. beams with H/do = 1.67 and 1.25) with
and 72% for steel grades S355, S460, S690, S960, respectively, and the S355, S460, S690 and S960 steel grades, it is found that the buckling
web-post failure load increase by 69%, 65%, 72% and 79% for S355, shear load for members with S460, S690 and S960 increases by 20%,

350
400
300 350
250 300

200 250
Load (kN)

Load (kN)

150 200

do/tw=32.24 150
100
do/tw=32.24
do/tw=40.83 100
50 do/tw=40.83
do/tw=48.0
50 do/tw=48.0
0
0 5 10 15 20 25 30 0
Deflection (mm) 0 5 10 15 20 25 30
Deflection (mm)
(a) (b)

600 700

500 600

500
400
400
Load (kN)
Load (kN)

300
300
200 do/tw=32.24 do/tw=32.24
200
do/tw=40.83 do/tw=40.83
100 100
do/tw=48.0 do/tw=48.0

0 0
0 5 10 15 20 25 30 0 5 10 15 20 25 30
Deflection (mm) Deflection (mm)

(c) (d)

Fig. 9. Load-deflection curves for members with different do/tw ratios and made from (a) S355; (b) S460; (c) S690; (d) S960 (H/do and s/do is 1.67 and 1.2).
 R. Shamass, F. Guarracino / Journal of Constructional Steel Research 166 (2020) 105911 7

200 240

180 220
do/tw=48.0 do/tw=48.0

buckling shear laod V(kN)

buckling shear laod V(kN)

do/tw=43.75 200 do/tw=43.75

160
do/tw=40.83 180 do/tw=40.83
140 do/tw=37.12 do/tw=37.12
160
do/tw=32.24 do/tw=32.24
120 140
120
100
100
80
80

60 60
1 1.1 1.2 1.3 1.4 1.5 1.6 1 1.1 1.2 1.3 1.4 1.5 1.6
s/do s/do

(a) (b)

360 410

310 do/tw=48.0 360 do/tw=48.0

buckling shear laod V(kN)

buckling shear laod V(kN)

do/tw=43.75 do/tw=43.75
310
260 do/tw=40.83 do/tw=40.83
do/tw=37.12 do/tw=37.12
260
210 do/tw=32.24 do/tw=32.24
210
160
160
110
110

60
60
1 1.1 1.2 1.3 1.4 1.5 1.6
1 1.1 1.2 1.3 1.4 1.5 1.6
s/do
s/do

(c) (d)

Fig. 10. Influence of s/do and do/tw geometrical parameters on the buckling shear load of cellular beams with H/do = 1.67 and made from (a) S355; (b) S460; (c) S690; (d) S960.

140 160
Vierendeel failure do/tw=69 do/tw=69
120 do/tw=62.88
buckling shear laod V(kN)

do/tw=62.88 140
buckling shear laod V(kN)

do/tw=58.68
do/tw=58.68
100 120 do/tw=53.35
do/tw=53.35 do/tw=46.33
80 do/tw=46.33
100

60
80

40
60
20
1 1.1 1.2 1.3 1.4 1.5 1.6 40
s/do 1 1.1 1.2 1.3 1.4 1.5 1.6
s/do
(a)
(b)
240 300
do/tw=69 280 do/tw=69
220
do/tw=62.88 260 do/tw=62.88
buckling shear laod V(kN)

buckling shear laod V(kN)

200 240
do/tw=58.68 do/tw=58.68
180 do/tw=53.35 220
200 do/tw=53.35
160 do/tw=46.33
180 do/tw=46.33
140 160
120 140
100 120
100
80
80
60 60
1 1.1 1.2 1.3 1.4 1.5 1.6 1 1.1 1.2 1.3 1.4 1.5 1.6
s/do s/do
(c) (d)

Fig. 11. Influence of s/do and do/tw geometrical parameters on the buckling shear load of cellular beams with H/do = 1.25 and made from (a) S355; (b) S460; (c) S690; (d) S960.
8 R. Shamass, F. Guarracino / Journal of Constructional Steel Research 166 (2020) 105911

Fig. 12. Observed failure modes of the cellular beams (a) Vierendeel failure; (b) web-post buckling.

60% and 100% on average compared to the cellular beams made from 4.2. Effect of high strength steel grade on the stress distribution at the
normal steel S355. web-post
Overall, all the parametric studies show that the results derived from
adopting HS steels cannot be linearly extrapolated from the increment Fig. 13 illustrates the Von-Mises stress distribution at the failure load
in the value of the yielding stress and for such a reason, as it will be of cellular beam with H/do = 1.67, s/do = 1.2 and do/tw = 40.83 made
shown in the Section 5, design formulae need to be based on a critical from normal and high strength steel. Only the stress zones that exceed
analysis of the failure mechanism. the yield value are shown. As anticipated in the literature, the

Fig. 13. The Von-Mises stress distribution at the web-post for (a) S355, (b) S460, (c) S690, (d) S960
 R. Shamass, F. Guarracino / Journal of Constructional Steel Research 166 (2020) 105911 9

Fig. 16. Inclination α of the tangent to the web holes.

Mörsch, which is also at the basis of the procedure proposed by Lawson

et al. [9] but, differently from other methods, does not make reference to
code prescriptions and focuses on some key parameters which have not
been highlighted before.
Thus, in the present section reference is made first to a simplified
model for web-post buckling based on the elementary model of an
Fig. 14. Typical two-waves web-post buckling mode. ideal inclined compressed strut within the web-post, in analogy of the
one suggested by Lawson et al. [9].
The difference here is that the modelling is based only on geometri-
compressive and tensile stresses tend to act across the web-post on op- cal and mechanical considerations, without making any reference to
posite diagonals in order to transmit shear forces and the compressive code prescriptions or design curves. In this manner the underlying me-
stress tends to give origin to the web-post buckling. It can be noticed chanics of the phenomenon can be highlighted and the level of approx-
that the yielding region of the web-post increases as the yield stress of imation precisely assessed with respect to the numerical models.
steel increases. For S690 or S960 HS steel, the yielding region evolves The adopted model is founded on the so-called classical truss anal-
over the whole web-post while the yielding region is localised at the ogy developed by Ritter and Mörsch. The analogy is thus based on a
ends of the web-post for the S355 normal strength steel. It worth notic- truss model with parallel chords and web members connected by
ing that no yielding region is observed at the middle of the web-post. means of pin joints, where the compressed inclined strut is inclined at
This suggests that the ideal compressed diagonal strut buckles, as it is a certain angle α' with respect to the longitudinal axis of the cellular
shown by the FE analyses, with a two-wave buckling mode, see Fig. 14. beam, see Fig. 15.
The value of the force Q is given by
5. A discerning model for web-post buckling design
ΔM
Q¼ ð1Þ
Despite a number of works in the past few years, there is still no con- H−t f
sensus about the design procedure for the shear resistance determina-
tion and the results obtained by various authors present significant where
differences when compared with each other, especially when the geo-
metrical parameters become different from the most used ones. This
ΔM
fact becomes even more evident when the existent models are applied V¼ ð2Þ
s
to HS steel cellular beams.
For this reason, a simple mechanical model for the WPB failure is
Q
developed and analysed on the basis of geometrical and mechanical pa- Qc ¼ ð3Þ
2 cosα 0
rameters in order to highlight the factors which influence the occur-
rence and the development of the WPB both in normal and HS steels.
The procedure is essentially based on the shear model by Ritter and

Fig. 15. Forces Q acting at the top and bottom flanges and decomposition into components
Qt and Qc through the web member. Fig. 17. Basic geometric relationships.
10 R. Shamass, F. Guarracino / Journal of Constructional Steel Research 166 (2020) 105911

Fig. 18. Width of the ideal strut.

Fig. 20. Horizontal shear in the web post of cellular beam.

so that
0 1
b
Rþ
Q
 0 −1 B 2 C
V ¼ 2 c H−t f cosα 0 ð4Þ α ¼ sin @ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi cosα A ð10Þ
s 1 2
b þ 4R so þ so 2
2
being M and V the bending moment and the vertical shear buckling load qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2
acting on the cellular beam, respectively. d R b þ 4R so þ s2o
l0 o ¼
o
¼   ð11Þ
The inclination α' of the ideal compressed strut depends on both sinα 0 b
Rþ cosα
geometrical and mechanical factors. In fact, the position of the web 2
holes and their dimensions define the angle α, as shown in Fig. 16.
The angle α and the indicated length lo can be easily derived with So far, all the formulae have been derived from purely geometrical
reference to Fig. 17. consideration and it is worth noticing that the length of the ideal strut,
It is

rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Table 3

 so 2 s2 Ratio of the predicted shear strength to FE shear strength V/VFE based on the effective
a¼ Rþ −R2 ¼ Rso þ o ð5Þ length model lLawson, lPanedpojaman (a)
and lPanedpojaman (b)
and new proposal for S355 steel
2 4 e e e
grade.

s/do H/do V/VFE d/tw

R
α ¼ sin−1 so ð6Þ
59.8 54.46 50.83 46.21 40.13
Rþ
2 1.1 1.25 lPanedpojaman (b)
1.07 1.08 1.07 1.06 1.03
e
lPanedpojaman
e
(a)
1.18 1.17 1.16 1.13 1.08
and, since R = do/2, lLawson
e 0.47 0.51 0.54 0.58 0.63
proposed formulation 0.93 0.87 0.83 0.78 0.72
1.67 lPanedpojaman
e
(b)
0.73 0.70 0.67 0.64 0.62
do lPanedpojaman (a)
lo ¼ ¼ do þ so ð7Þ e 0.65 0.64 0.62 0.59 0.58
sinα lLawson
e 0.37 0.39 0.40 0.41 0.43
proposed formulation 0.62 0.59 0.56 0.53 0.51
1.2 1.25 lPanedpojaman (b)
1.14 1.24 1.28 1.35 1.44
The width of the ideal strut b defines the inclination angle α' as a e
lPanedpojaman
e
(a)
1.35 1.45 1.48 1.53 1.59
function of α as shown in Fig. 18: lLawson 0.74 0.82 0.88 0.95 1.08
e
Thus, with reference to Figs. 17 and 18, it is proposed formulation 1.31 1.27 1.22 1.16 1.09
1.67 lPanedpojaman
e
(b)
1.03 1.02 1.02 1.00 0.98
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi lPanedpojaman (a)
0.87 0.88 0.89 0.89 0.90
1 2 b e
b þ 4R so þ s2o Rþ lLawson
e 0.57 0.62 0.65 0.68 0.73
2 ¼ 2 ð8Þ proposed formulation 0.97 0.94 0.92 0.89 0.86
sinβ sinα 0 1.3 1.25 lPanedpojaman (b)
0.90 0.96 1.00 1.05 1.18
e
lPanedpojaman
e
(a)
1.11 1.16 1.21 1.25 1.36
π π  lLawson
e 0.78 0.83 0.88 0.93 1.06
β ¼ −α → sinβ ¼ sin −α ¼ cosα ð9Þ proposed formulation 1.23 1.13 1.08 1.00 0.95
2 2
1.67 lPanedpojaman
e
(b)
1.12 1.14 1.15 1.16 1.18
lPanedpojaman
e
(a)
0.90 0.95 0.98 1.01 1.06
lLawson
e 0.70 0.75 0.80 0.85 0.92
proposed formulation 1.04 1.01 0.99 0.97 0.94
1.4 1.25 lPanedpojaman
e
(b)
0.72 0.78 0.83 0.93 0.93a
lPanedpojaman
e
(a)
0.91 0.97 1.03 1.13 0.93a
lLawson
e 0.78 0.84 0.90 0.99 0.93a
proposed formulation 1.12 1.04 1.00 0.96 0.93a
1.67 lPanedpojaman
e
(b)
1.09 1.13 1.17 1.21 1.26
lPanedpojaman
e
(a)
0.87 0.94 0.99 1.05 1.13
lLawson
e 0.77 0.84 0.89 0.96 1.05
proposed formulation 1.04 1.01 0.99 0.97 0.95
1.5 1.25 lPanedpojaman
e
(b)
0.62 0.69 0.96a 0.92a 0.94a
lPanedpojaman
e
(a)
0.78 0.87 0.96a 0.92a 0.94a
lLawson
e 0.78 0.87 0.96a 0.92a 0.94a
proposed formulation 1.06 1.01 0.96a 0.92a 0.94a
1.67 lPanedpojaman
e
(b)
0.97 1.04 1.10 1.16 1.26
lPanedpojaman
e
(a)
0.81 0.89 0.95 1.02 1.14
lLawson
e 0.81 0.89 0.95 1.02 1.14
proposed formulation 0.99 0.97 0.96 0.93 0.92
Fig. 19. Area of the web-post. a
Failure by Vierendeel mechanism.
 R. Shamass, F. Guarracino / Journal of Constructional Steel Research 166 (2020) 105911 11

0
l o , results function of the length lo of Fig. 16, of the inclination angle α where Et is the tangent modulus of the material at the buckling stress
and of the width of the strut, b. This is different from all the models in and n = 2 for the typical two-waves web-post buckling mode found
literature, which do not make the inclination of the ideal strut function in both the numerical and experimental analyses.
of its effective width, which is generally assumed as half the web-post The use of tangent, secant or reduced moduli for the buckling of
width at the mid-height, so/2 (e.g. [4,7]), even if it is widely accepted structures in the plastic range might turn to be rather an intricate mat-
that the band width in a narrow web-post is significantly wider than ter, depending on the particular case at hand (see, for example, [25–28])
this approximation. and it has been object of debate for over one century since the first work
From physical considerations and validation through the analysis of by Engesser back in 1895. Here reference is made to the stress-strain
the stress distribution in the web-post that results from the performed curves discussed by Yun and Gardner [23] and to the analysis of tangent
numerical analyses, it is here proposed that the width of the ideal and secant moduli proposed by Guarracino and Simonelli [29], so that
strut is proportional to the ratio between the actual area of the web- the equivalent tangent moduli of 50, 60, 80, 100 Gpa are used for
post and the total height of the section, H, see Fig. 19. S355, S460, S690 and S960, respectively.
However, the calculation of the buckling load Qc from Eq. (13) and,
h π 2i consequently, of the maximum shear load V from Eq. (4), requires also
Hðdo þ so Þ− do 0
b¼ 4 ð12Þ the definition of the coefficient ξ and of the effective length l o of the
γH
ideal Euler's strut.
0
This because both the geometrical lengths lo and l o in Eqs. (7) and
with γ = 6.
(11) have been derived from purely geometrical considerations,
At this point the buckling load Qc in Eq. (4) can be directly derived by
starting from what is indicated in Fig. 16. In reality, the actual length
the Euler's formula for the buckling of a simple strut, l0 o of the ideal Euler's strut will extend past the region delimitated by
the web holes and the strut will also be subject to some form of restrain
n2 π2 Et b t 2w
Qc ¼  0 2 ð13Þ due to the continuity of the web.
12 ξ l o

Table 4 Table 5
Ratio of the predicted shear sterngth to FE shear strength V/VFE based on the effective Ratio of the predicted shear sterngth to FE shear strength V/VFE based on the effective
length model lLawson
e , lPanedpojaman
e
(a)
and lPanedpojaman
e
(b)
and new proposal for S460 steel length model lLawson
e , lPanedpojaman
e
(a)
and lPanedpojaman
e
(b)
and new proposal for S690 steel
grade. grade.

s/do H/do V/VFE d/tw s/do H/do V/VFE d/tw

59.8 54.46 50.83 46.21 40.13 59.8 54.46 50.83 46.21 40.13

1.1 1.25 lPanedpojaman

e
(b)
1.06 1.09 1.09 1.09 1.07 1.1 1.25 lPanedpojaman
e
(b)
0.91 0.96 1.00 1.01 1.04
lPanedpojaman
e
(a)
1.20 1.22 1.21 1.18 1.14 lPanedpojaman
e
(a)
1.08 1.12 1.15 1.14 1.14
lLawson
e 0.43 0.47 0.50 0.53 0.59 lLawson
e 0.33 0.36 0.39 0.42 0.49
proposed formulation 0.96 0.91 0.86 0.80 0.73 proposed formulation 0.93 0.88 0.85 0.78 0.71
1.67 lPanedpojaman
e
(b)
0.79 0.77 0.76 0.73 0.70 1.67 lPanedpojaman
e
(b)
0.82 0.81 0.80 0.78 0.76
lPanedpojaman
e
(a)
0.69 0.68 0.68 0.67 0.65 lPanedpojaman
e
(a)
0.68 0.69 0.69 0.69 0.68
lLawson
e 0.35 0.37 0.40 0.42 0.45 lLawson
e 0.28 0.31 0.33 0.36 0.41
proposed formulation 0.64 0.61 0.60 0.57 0.54 proposed formulation 0.62 0.60 0.58 0.56 0.53
1.2 1.25 lPanedpojaman
e
(b)
1.01 1.12 1.19 1.30 1.35 1.2 1.25 lPanedpojaman
e
(b)
0.81 0.89 0.95 1.05 1.21
lPanedpojaman
e
(a)
1.23 1.34 1.41 1.51 1.52 lPanedpojaman
e
(a)
1.01 1.10 1.16 1.26 1.43
lLawson
e 0.64 0.72 0.78 0.88 0.96 lLawson
e 0.49 0.55 0.59 0.67 0.80
proposed formulation 1.29 1.26 1.22 1.19 1.06 proposed formulation 1.26 1.20 1.16 1.11 1.07
1.67 lPanedpojaman
e
(b)
1.05 1.05 1.05 1.05 1.04 1.67 lPanedpojaman
e
(b)
1.04 1.04 1.04 1.05 1.05
lPanedpojaman
e
(a)
0.84 0.87 0.89 0.91 0.93 lPanedpojaman
e
(a)
0.76 0.80 0.82 0.86 0.90
lLawson
e 0.52 0.56 0.60 0.65 0.72 lLawson
e 0.42 0.46 0.49 0.54 0.62
proposed formulation 0.96 0.93 0.91 0.89 0.86 proposed formulation 0.94 0.90 0.88 0.85 0.82
1.3 1.25 lPanedpojaman
e
(b)
0.82 0.88 0.92 0.97 1.07 1.3 1.25 lPanedpojaman
e
(b)
0.65 0.73 0.79 0.87 0.95
lPanedpojaman
e
(a)
1.02 1.09 1.14 1.18 1.27 lPanedpojaman
e
(a)
0.83 0.93 1.00 1.08 1.17
lLawson
e 0.71 0.76 0.80 0.85 0.95 lLawson
e 0.56 0.63 0.68 0.75 0.83
proposed formulation 1.28 1.18 1.12 1.02 0.93 proposed formulation 1.27 1.22 1.18 1.10 0.98
1.67 lPanedpojaman
e
(b)
1.10 1.13 1.15 1.17 1.20 1.67 lPanedpojaman
e
(b)
1.03 1.07 1.09 1.12 1.16
lPanedpojaman
e
(a)
0.84 0.90 0.93 0.98 1.05 lPanedpojaman
e
(a)
0.73 0.78 0.82 0.88 0.96
lLawson
e 0.63 0.68 0.73 0.79 0.88 lLawson
e 0.52 0.56 0.60 0.66 0.76
proposed formulation 1.03 1.00 0.98 0.95 0.93 proposed formulation 1.03 0.98 0.95 0.91 0.88
1.4 1.25 lPanedpojaman
e
(b)
0.65 0.71 0.75 0.82 1.00a 1.4 1.25 lPanedpojaman
e
(b)
0.53 0.59 0.64 0.69 0.79
lPanedpojaman
e
(a)
0.82 0.89 0.94 1.02 1.00a lPanedpojaman
e
(a)
0.68 0.76 0.81 0.88 0.98
lLawson
e 0.70 0.76 0.81 0.88 1.00a lLawson
e 0.58 0.64 0.69 0.75 0.85
proposed formulation 1.16 1.08 1.03 0.97 1.00a proposed formulation 1.21 1.14 1.09 1.01 0.91
1.67 lPanedpojaman
e
(b)
1.03 1.08 1.12 1.18 1.26 1.67 lPanedpojaman
e
(b)
0.91 0.97 1.01 1.08 1.17
lPanedpojaman
e
(a)
0.80 0.86 0.91 0.99 1.09 lPanedpojaman
e
(a)
0.67 0.73 0.77 0.85 0.96
lLawson
e 0.70 0.76 0.81 0.89 1.00 lLawson
e 0.58 0.63 0.67 0.75 0.86
proposed formulation 1.03 0.99 0.97 0.96 0.94 proposed formulation 1.04 0.99 0.95 0.92 0.89
1.5 1.25 lPanedpojaman
e
(b)
0.54 0.60 0.65 1.00a 1.02a 1.5 1.25 lPanedpojaman
e
(b)
0.45 0.49 0.53 0.56 1.02a
lPanedpojaman
e
(a)
0.69 0.76 0.83 1.00a 1.02a lPanedpojaman
e
(a)
0.57 0.63 0.67 0.72 1.02a
lLawson
e 0.69 0.76 0.83 1.00a 1.02a lLawson
e 0.57 0.63 0.67 0.72 1.02a
proposed formulation 1.07 1.01 0.98 1.00a 1.02a proposed formulation 1.13 1.05 1.00 0.91 1.02a
1.67 lPanedpojaman
e
(b)
0.89 0.96 1.01 1.10 1.22 1.67 lPanedpojaman
e
(b)
0.73 0.82 0.87 0.96 1.10
lPanedpojaman
e
(a)
0.73 0.80 0.86 0.94 1.08 lPanedpojaman
e
(a)
0.59 0.66 0.71 0.79 0.94
lLawson
e 0.73 0.80 0.86 0.94 1.08 lLawson
e 0.59 0.66 0.71 0.79 0.94
proposed formulation 1.00 0.96 0.94 0.92 0.91 proposed formulation 0.98 0.96 0.93 0.90 0.89
a a
Failure by Vierendeel mechanism. Failure by Vierendeel mechanism.
12 R. Shamass, F. Guarracino / Journal of Constructional Steel Research 166 (2020) 105911

It is worth recalling that Lawson et al. [9] suggested that the effective Furthermore, Panedpojaman et al. [7] suggested an additional coef-
length of the strut can be assumed equal to half of the diagonal distance ficient for the effective length (k) that takes into consideration the tee
across the web-post height that restrains the buckling
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Panedpojaman ðbÞ Panedpojaman ðaÞ
Lawson
le ¼ 0:5 so 2 þ do
2
ð14Þ le ¼ k le ð18Þ

The vertical shear buckling force is given by with

 2  
ðH−2yt Þ s do do
V ¼ Vh ð15Þ k ¼ 0:9 ≤ min 1:15 ; 1:15 ð19Þ
s do d d

where yt is the depth of the elastic neutral axis of tee from the outer face
Here, without loss of generality and once again on the bases of the
of the flange (Fig. 20), and Vh is horizontal shear force,
analyses of the results from the performed numerical analyses, it is
V h ¼ σ c so t w ð16Þ proposed that

According to Lawson and Hicks [30] σc is the buckling strength ob- l0 o ¼ H þ do 1 ð20Þ
tained from the relevant buckling curve c of Eurocode 3 [10]. 2 sinα 0
However, Lawson's effective length model underestimates the pre-
dicted shear strength for narrow web-posts, therefore, Panedpojaman and
et al. [7] suggested an alternative effective length model as the length
tw
between the mid-height and the point of tangency ξ¼ψ ð21Þ
H
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Panedpojaman ðaÞ 2
le ¼ 0:5 s2 −do ð17Þ By setting ψ = 20 , the proposed formulation yields the results for the
performed analyses which are illustrated in the following Tables.
The predicted buckling shear strength of web-post are based on
Table 6 three effective length models lLawson
e , lPanedpojaman
e
(a)
and lPanedpojaman
e
(b)

Ratio of the predicted shear sterngth to FE shear strength V/VFE based on the effective for the models by Lawson et al. [9] and by Panedpojaman et al. [7],
length model lLawson
e , lPanedpojaman
e
(a)
and lPanedpojaman
e
(b)
and new proposal for S960 steel using the Eq. (15), and on the proposed procedure. All the calculations
grade. are made for normal and high strength steel grades. The analytical
s/do H/do V/VFE d/tw buckling shear strength of web-post is compared with the numerical
59.8 54.46 50.83 46.21 40.13
buckling shear strength to validate the effective length model and to
assess the capability of the design provision explained above. Again
1.1 1.25 lPanedpojaman
e
(b)
0.81 0.87 0.92 0.97 1.00
the beams B2 and B6 are chosen for this comparison where the ratios
lPanedpojaman
e
(a)
0.99 1.05 1.09 1.13 1.13
lLawson
e 0.27 0.31 0.33 0.37 0.41 s/do varies in the range of 1.1 to 1.5 and the ratio d/tw varies from
proposed formulation 0.94 0.88 0.85 0.80 0.71 40.13 to 59.80. Four steel grades S355, S460, S690 and S960 are
1.67 lPanedpojaman
e
(b)
0.86 0.85 0.85 0.84 0.83 taken into consideration.
lPanedpojaman
e
(a)
0.67 0.69 0.70 0.71 0.73 Tables 3 and 4 illustrate the ratio between the calculated shear
lLawson 0.24 0.26 0.29 0.32 0.38
e
proposed formulation 0.62 0.59 0.58 0.56 0.54
strength V using three effective length models, lLawson e , lPanedpojaman
e
(a)

1.2 1.25 lPanedpojaman

e
(b)
0.69 0.76 0.82 0.90 1.04 and lPanedpojaman
e
(b)
and the proposed model, and the FE shear
lPanedpojaman
e
(a)
0.87 0.95 1.02 1.11 1.26 strength VFE for S355 and S460 steel. For small values of H/do and s/do
lLawson
e 0.41 0.46 0.50 0.56 0.66 (i.e. H/do = 1.25 and s/do b 1.3), lPanedpojaman (b)
overestimates the shear
e
proposed formulation 1.28 1.20 1.16 1.10 1.03
buckling results by up to 44% and 35% for steels S355 and S460, respec-
1.67 lPanedpojaman
e
(b)
1.06 1.06 1.06 1.07 1.09
lPanedpojaman
e
(a)
0.71 0.75 0.78 0.82 0.89 tively, while it overestimates the shear buckling results by up to 26% for
lLawson
e 0.37 0.40 0.43 0.47 0.55 both S355 and S460 and for high values of H/do and s/do (i.e. H/do =
proposed formulation 0.97 0.91 0.88 0.84 0.81 1.67 and s/do ≥ 1.3). The overestimation yielded by lPanedpojaman
e
(b)
in-
1.3 1.25 lPanedpojaman
e
(b)
0.56 0.49 0.67 0.74 0.86 creases with larger values of slenderness d/tw. Similarly, le Panedpojaman (a)
lPanedpojaman
e
(a)
0.72 0.63 0.85 0.94 1.08
lLawson 0.48 0.42 0.57 0.64 0.75
overestimates the shear buckling results by up to 59% and 59% and 52%
e
proposed formulation 1.32 0.98 1.18 1.12 1.04 for S355 and S460, respectively, for beams with small values of H/do
1.67 lPanedpojaman
e
(b)
0.97 1.01 1.04 1.09 1.16 and s/do (i.e. H/do = 1.25 and s/do ≤ 1.3). On the other hand, lLawson e
lPanedpojaman
e
(a)
0.65 0.70 0.74 0.80 0.90 provides predictions which underestimate the load capacity by up to
lLawson 0.45 0.49 0.52 0.58 0.67
e
63% and 65% for smaller values of s/do (i.e. s/do = 1.1, 1.2) and by up to
proposed formulation 1.06 0.99 0.95 0.91 0.87
1.4 1.25 lPanedpojaman
e
(b)
0.46 0.51 0.54 0.60 0.69 30% and 37% for high values of s/do (i.e. s/do ≥ 1.3) except for some
lPanedpojaman
e
(a)
0.59 0.65 0.70 0.77 0.88 cases, i.e. H/do = 1.25, d/tw = 40.13 and s/do ≥ 1.2 for S355 and s/do =
lLawson
e 0.50 0.55 0.59 0.65 0.75 1.5 for S460 where only a slight overestimation is found.
proposed formulation 1.26 1.18 1.12 1.05 0.95 When H/do = 1.67 and s/do = 1.1, the proposed formulation under-
1.67 lPanedpojaman
e
(b)
0.81 0.85 0.92 1.00 1.12
lPanedpojaman (a)
0.58 0.62 0.68 0.75 0.87
estimates the shear buckling by up to 49% and 46% for steel S355 and
e
lLawson
e 0.50 0.53 0.59 0.65 0.77 S460, respectively. However, the proposed formulation results in
proposed formulation 1.07 0.97 0.96 0.92 0.89 good agreements with FE shear buckling for beams with H/do = 1.67
1.5 1.25 lPanedpojaman
e
(b)
0.39 0.43 0.45 0.47 0.57 and s/do N 1.1. For beams with H/do = 1.25 and s/do = 1.2, the proposed
lPanedpojaman (a)
0.51 0.55 0.58 0.61 0.73
e
formulation overestimates the predictions by up to 31% and 29% for
lLawson
e 0.51 0.55 0.58 0.61 0.73
proposed formulation 1.21 1.12 1.05 0.92 0.87 steel S355 and S460, respectively. Similarly, the proposed formulation
1.67 lPanedpojaman
e
(b)
0.66 0.72 0.76 0.85 1.02 overestimates the predictions for beams with H/do = 1.25, s/do = 1.3
lPanedpojaman
e
(a)
0.52 0.58 0.61 0.69 0.84 and d/tw = 59.8, 54.46 and 50.83 by up to 23% and 28% for steel S355
lLawson
e 0.52 0.58 0.61 0.69 0.84 and S460, respectively. However, the it provides good agreements
proposed formulation 1.04 0.98 0.93 0.91 0.90
with FE shear buckling for H/do = 1.25 and s/do = 1.1, 1.4, 1.5.
 R. Shamass, F. Guarracino / Journal of Constructional Steel Research 166 (2020) 105911 13

Fig. 21. Comparison of the predicted shear strength with the FE shear strength for beams made from (a) S355; (b) S460; (c) S690; and (d) S960 steel grades.

Tables 5 and 6 show the ratio between the calculated shear Fig. 21shows a graphical representation of all the analytically pre-
strength V and the FE shear strength VFE using three effective length dicted and FE calculated shear strength for beams made from mild and
models lLawson
e , lPanedpojaman
e
(a)
and lPanedpojaman
e
(b)
and the new proposal HS steel. It can be seen that the use of lPanedpojaman
e
(a)
or lPanedpojaman
e
(b)

for S690 and S960 high strength steel. lPanedpojaman

e
(b)
overestimates the as effective buckling length in the calculation of shear buckling using
shear buckling in some cases by up to 21% and 26% when d/tw = 40.13 nominal buckling equations of BS EN 1993-1-1 generally tends to overes-
and up to 10% and 9% when H/do = 1.67 and s/do = 1.2, 1.3. timate the predicted buckling shear strength of cellular beams for both
On the other hand, the predicted shear buckling for S690 and S960 normal and high strength steel. The use of lLawson
e tends to provide accept-
obtained using lLawson
e is underestimated by up to 72% and 76% for able predictions for the normal strength steel and very conservative pre-
smaller values of s/do (i.e. s/do = 1.1, 1.2) and for larger s/do (i.e. s/do = dictions for HS steel, especially for the S960.
1.3, 1.4, 1.5) the results are underestimated by up to 48% and 55% for
S690 and S960, respectively.
When H/do = 1.67 and s/do = 1.1, the proposed formulation under-
estimates the shear buckling by up to 47% and 46% for steel S690 and Table 7
Summary of RMS errors.
S960, respectively. However, the proposed procedure results in good
agreements with FE shear buckling for beams with H/do = 1.67 and s/ Steel Grade Analytical model RMS error
do N 1.1. For beams with H/do = 1.25 and s/do = 1.2, the proposed for- S355 lPanedpojaman
e
(b)
21.8
mulation overestimates the predictions by up to 31% and 29% for steel lPanedpojaman
e
(a)
21.8
S690 and S960, respectively. Similarly, the proposed formulation over- lLawson
e 29.3
estimates the predictions for beams with H/do = 1.25, s/do = 1.3 and proposed formulation 18.5
S460 lPanedpojaman
e
(b)
25.1
d/tw = 59.8, 54.46 and 50.83 by up to 23% and 28% for steel S690 and lPanedpojaman (a)
24.5
e
S960, respectively. However, the it provides good agreements with FE lLawson
e 38.3
shear buckling for H/do = 1.25 and s/do = 1.1, 1.4, 1.5. proposed formulation 22.5
Overall, the proposed analytical formulation results in better agree- S690 lPanedpojaman
e
(b)
33.9
lPanedpojaman (a)
38.9
ment with the FE results for the whole range of investigated geometries e
lLawson
e 67.3
and both for normal and HS steel. This can be essentially attributed to proposed formulation 29.8
the ability to adapt both the width and the inclination of the compressd S960 lPanedpojaman
e
(b)
49.1
ideal strut to the geometry of the case, see Eqs. (12) and (20), lPanedpojaman
e
(a)
59.7
respectively. lLawson
e 99.5
proposed formulation 37.1
14 R. Shamass, F. Guarracino / Journal of Constructional Steel Research 166 (2020) 105911

Table 8
Summary of the reliability analysis for the proposed formulation (s/do = 1.1 cases
included).

n b kd, n Vr γM0

S355 46 1.05 3.04 0.23 1.44

S460 45 1.11 3.04 0.22 1.55
S690 49 1.07 3.04 0.22 1.55
S960 49 1.09 3.04 0.22 1.56

Table 7 illustrate the calculated root-mean-square (RMS) errors

for the predicted shear strength V using three effective length models,
lLawson
e , lPanedpojaman
e
(a)
and lPanedpojaman
e
(b)
and the proposed model, for
both normal and high-strength steel. The RMS error is defined as
rffiffiffiffiffiffiffiffi
SSE
RMS error ¼ ð22Þ
n

where n is the number of data points over which the sum of the squared
Fig. 22. Failure modes different from web-post buckling.
error SSE between the finite-element shear strength VFE i and the pre-
dicted shear strength Vpred.
i , given as:

N 
X 2
pred: The material over-strength of high strength steel was taken equal to
SSE ¼ VFE
i −Vi ð23Þ
1
1.135 with a coefficient of variation COV of 0.055, while the COV of geo-
metric properties was assumed equal to 0.02 [32]. For normal strength
steel, the material over-strength of high strength steel was taken
It can be pointed out that the RMS error ranges between 18.5 and
equal to 1.25 with a coefficient of variation COV of 0.055 [33]. The
37.7 for the shear strength predicted by the proposed formulation
COV between the experimental and the numerical results, which was
while it ranges between 21.8 and 49.1 for the shear strength predicted
found equal to 0.025, was also considered. Performing a First Order
by lPanedpojaman
e
(b)
, between 21.8 and 59.7 for the shear strength pre-
Panedpojaman (a) Reliability Method (FORM) in accordance with the Eurocode target reli-
dicted by le and between 29.3 and 99.5 for shear strength
ability requirements, the partial factors γM0 were evaluated.
predicted by lLawson
e .
It has to be mentioned again that for s/do = 1.1 the failure, see
Thus, the RMS error for the predicted shear strength V according
Fig. 22, is not in the fashion of the typical two waves post buckling
to the proposed model results on average lower than the ones
shown in Fig. 14. Therefore, the reliability analysis has been here per-
relative to the use of the effective length models lLawson
e , lPanedpojaman
e
(a)
Panedpojaman (b) formed both considering and not considering the cases s/do = 1.1. The
and le .
results are collected in Tables 1 and 2, respectively, and it is evident
In the end, it is evident that the proposed formulation seems to pro-
that the elimination of the failure modes not modelled by the proposed
vide shear buckling results that are in much more reasonable agreement
formulation leads to much more consistent results. It goes without say-
with those obtained by FE analysis for both mild and HS steel.
ing that none of the other formulations would be able to cover these
atypical cases, either.
6. A statistical evaluation in the fashion of Annex D EN 1990
Overall, the performed statistical analysis seems to confirm the reli-
ability of the proposed formulation.
Even if, as stated in the Introduction, the proposed procedure does
not make reference to code prescriptions and focuses on the mechanics
of the phenomenon only, a statistical analysis in the fashion of the pro- 7. Conclusions
visions of Annex D EN 1990 (2002) [31] has been carried out in order to
assess the reliability of the proposed formulation. However, in the A comprehensive numerical analysis of the behaviour of cellular
framework of the present study the statistical evaluation of the pro- beams made from normal and high strength steel with various geome-
posed prediction model is done here against the numerical results. tries has been performed and discussed. The finite element models have
Tables 8 and 9 below summarise the following key statistical param- been developed using commercial software ABAQUS and validated
eters: the number of tests and FE simulations n, the design fractile factor against existing experimental results conducted by Grilo et al. [5] and
(ultimate limit state), kd, n, the average ratio of FE to model resistance Tsavdaridis and D'Mello [4].
based on a least squares fit to all the data, b, the combined coefficient Since there is currently no unanimous consensus about the de-
of variation incorporating both model and basic variable uncertainties, sign procedure for the shear resistance determination, on the basis
Vr, and the partial safety factor for cross-section resistance γM0. of the obtained numerical results a simple mechanical model for
the WPB failure has been proposed in order to highlight the factors
which influence the occurrence and the development of the WPB
Table 9
Summary of the reliability analysis for the proposed formulation (s/do = 1.1 cases
both in normal and HS steels.
excluded). The resulting analytical formulation results in better agreement with
the FE results for the whole range of investigated geometries and both
n b kd, n Vr γM0
for normal and HS steel, on account of its capability to adapt both the
S355 41 1.02 3.04 0.137 1.16 width and the inclination of the compressd ideal strut to the geometry
S460 40 1.03 3.04 0.14 1.29 of the case.
S690 44 1.05 3.04 0.15 1.31
As a consequence, it is found that the proposed analytical procedure
S960 44 1.04 3.04 0.152 1.32
can be used as basis for the development of a reliable design method to
 R. Shamass, F. Guarracino / Journal of Constructional Steel Research 166 (2020) 105911 15

predict shear buckling of web-post of cellular beams made both of mild [13] J.L. Ma, T.M. Chan, B. Young, Experimental investigation of cold-formed high
strength steel tubular beams, Eng. Struct. 126 (2016) 200–209.
and HS steel. [14] K.J.R. Rasmussen, G.J. Hancock, Tests of high strength steel columns, J. Constr. Steel
Res. 34 (1) (1995) 27–52.
[15] H. Ban, G. Shi, Y. Shi, M.A. Bradford, Experimental investigation of the overall buck-
Declaration of Competing Interest ling behaviour of 960 MPa high strength steel columns, J. Constr. Steel Res. 88
(2013) 256–266.
The authors declare that they have no known competing financial [16] Y.B. Wang, G.Q. Li, S.W. Chen, F.F. Sun, Experimental and numerical study on the be-
havior of axially compressed high strength steel columns with H-section, Eng.
interests or personal relationships that could have appeared to influ- Struct. 43 (2012) 149–159.
ence the work reported in this paper. [17] R. Puthli, O. Fleischer, Investigations on bolted connections for high strength steel
members, J. Constr. Steel Res. 57 (3) (2001) 313–326.
[18] P. Može, D. Beg, J. Lopatič, Net cross-section design resistance and local ductility of
Acknowledgements elements made of high strength steel, J. Constr. Steel Res. 63 (11) (2007)
1431–1441.
The first author would like to acknowledge the Centre for Civil and [19] B. Uy, R.J. Sloane, Behaviour of composite tee beams constructed with high strength
steel, J. Constr. Steel Res. 1 (46) (1998) 203–204.
Building Services Engineering at London South Bank University for the
[20] R. Shamass, K.A. Cashell, Behaviour of composite beams made using high strength
encouragement and provision of technical supports for this research. steel, Structures, vol. 12, Elsevier 2017, pp. 88–101.
[21] H. Ban, M.A. Bradford, B. Uy, X. Liu, Available rotation capacity of composite beams
References with high-strength materials under sagging moment, J. Constr. Steel Res. 118 (2016)
156–168.
[1] J. Ward, Design of composite and non-composite cellular beams, The Steel Construc- [22] B. Uy, M. Khan, Z. Tao, F. Mashiri, Behaviour and design of high strength steel-
tion Institute, SCI Publication, 1990 100. concrete filled columns, Proceedings of the 2013 World Congress on Advances in
[2] K. Kuchta, M. Maślak, Failure modes determining the resistance and the stability of Structural Engineering and Mechanics (ASEM13), Jeju, Korea 2013, pp. 150–167.
steel cellular beams, Czasopismo Inżynierii Lądowej, Środowiska i Architektury 62 (4) [23] X. Yun, L. Gardner, Stress-strain curves for hot-rolled steels, J. Constr. Steel Res. 133
(2015) 263–280. (2017) 36–46.
[3] J. Nseir, M. Lo, D. Sonck, H. Somja, O. Vassart, N. Boissonnade, Lateral torsional buck- [24] Abaqus, Reference Manual, Version 6, vol. 11, Dassault Systèmes, France: Simulia,
ling of cellular steel beams, Proceedings of the Annual Stability Conference Struc- 2011.
tural Stability Research Council 2012, April, pp. 18–21. [25] R. Shamass, G. Alfano, F. Guarracino, A numerical investigation into the plastic buck-
[4] K.D. Tsavdaridis, C. D'Mello, Web buckling study of the behaviour and strength of ling paradox for circular cylindrical shells under axial compression, Eng. Struct. 75
perforated steel beams with different novel web opening shapes, J. Constr. Steel (2014) 429–447.
Res. 67 (10) (2011) 1605–1620. [26] R. Shamass, G. Alfano, F. Guarracino, An investigation into the plastic buckling par-
[5] L.F. Grilo, R.H. Fakury, G. de Souza Veríssimo, Design procedure for the web-post adox for circular cylindrical shells under non-proportional loading, Thin-Walled
buckling of steel cellular beams, J. Constr. Steel Res. 148 (2018) 525–541. Struct. 95 (2015) 347–362.
[6] F. Erdal, M.P. Saka, Ultimate load carrying capacity of optimally designed steel cellu- [27] R. Shamass, G. Alfano, F. Guarracino, An analytical insight into the buckling paradox
lar beams, J. Constr. Steel Res. 80 (2013) 355–368. for circular cylindrical shells under axial and lateral loading, Math. Probl. Eng.
[7] P. Panedpojaman, T. Thepchatri, S. Limkatanyu, Novel design equations for shear (2015), 514267. https://doi.org/10.1155/2015/514267.
strength of local web-post buckling in cellular beams, Thin-Walled Struct. 76 [28] R. Shamass, G. Alfano, F. Guarracino, On elastoplastic buckling analysis of cylinders
(2014) 92–104. under nonproportional loading by differential quadrature method, Int. J. Struct.
[8] S. Fares, J. Coulson, D. Dinehart, Steel Design Guide 31: Castellated and Cellular Beam Stab. Dyn. 17 (7) (2017), 1750072. https://doi.org/10.1142/S0219455417500729.
Design – AISC, 2016 2016. [29] F. Guarracino, M.G. Simonelli, The torsional instability of a cruciform column in the
[9] R.M. Lawson, J. Lim, S.J. Hicks, W.I. Simms, Design of composite asymmetric cellular plastic range: analysis of an old conundrum, Thin-Walled Struct. 113 (2017)
beams and beams with large web openings, J. Constr. Steel Res. 62 (6) (2006) 273–286.
614–629. [30] R.M. Lawson, S.J. Hicks, Design of Composite Beams with Large Web Openings: in
[10] C.E.N. Eurocode, 3: Design of Steel Structures, Part 1–1 “General Rules and Rules for Accordance with Eurocodes and the UK National Annexes, Steel Construction Insti-
Buildings”, European Committee for Standardization, CEN, Brussels, 2005. tute, 2011.
[11] J.M. Ricles, R. Sause, P.S. Green, High-strength steel: implications of material and [31] EN 1990, Eurocode - Basis of Structural Design (English), 2002.
geometric characteristics on inelastic flexural behavior, Eng. Struct. 20 (4–6) [32] M. Gkantou, M. Theofanous, N. Antoniou, C. Baniotopoulos, Compressive behaviour
(1998) 323–335. of high-strength steel cross-sections, Proceed. Inst. Civil Eng. Struct. Build. 170 (11)
[12] C.H. Lee, K.H. Han, C.M. Uang, D.K. Kim, C.H. Park, J.H. Kim, Flexural strength and ro- (2017) 813–824.
tation capacity of I-shaped beams fabricated from 800-MPa steel, J. Struct. Eng. 139 [33] F. McCann, L. Gardner, Numerical analysis and design of slender elliptical hollow
(6) (2012) 1043–1058. sections in bending, Thin-Walled Struct. 139 (2019) 196–208.