Journal of Constructional Steel Research 162 (2019) 105728

Contents lists available at ScienceDirect

Journal of Constructional Steel Research

Web compression buckling capacity prediction for unstiffened I-sections

with opposite patch loading
Fatmir Menkulasi a,⁎, Nahid Farzana b
a
Department of Civil and Environmental Engineering, Wayne State University, Detroit, MI 48202, USA
b
Department of Civil Engineering, Louisiana Tech University, Ruston, LA, USA

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

Article history: Two web compression buckling capacity prediction methods are introduced for unstiffened steel I-sections sub-
Received 12 March 2019 ject to opposite patch loading applied to the flanges. The methods are generally posed as a function of loaded
Received in revised form 8 June 2019 width to web depth ratio, and are applicable for opposite patch loading applied at the interior of a wide flange
Accepted 8 August 2019
section or at the end of it, where the web has a free edge. The proposed methods include three parts: 1) an ex-
Available online xxxx
pression for predicting the squash load, 2) an expression for predicting the elastic buckling load, and 3) a resis-
Keywords:
tance function. The squash load is calculated using an empirically derived effective width concept based on
Opposite patch loading observations at the ultimate load from an extensive experimental database and validated numerical simulations.
Web compression buckling Web slenderness is defined as the square root of the ratio of the web squash load to the web critical elastic buck-
Critical elastic buckling ling load. The critical elastic buckling load is defined consistently with that obtained with a plate buckling energy
Finite element analysis solution for patch loading on infinitely long strips and considers the shortened web buckling half-wavelength
Capacity prediction resulting from flange rotational restraint provided to the web. The methods are validated with existing experi-
mental data and shell finite element collapse simulations, and are shown to be more accurate and more widely
applicable than current American Institute for Steel Construction (AISC) Specification provisions.
© 2019 Elsevier Ltd. All rights reserved.

1. Introduction
The research presented in this paper deals with the behavior of
unstiffened steel I-sections subject to opposite patch loading away or
at member ends (Fig. 1a and b). Fig. 1c provides some practical exam-
ples of steel I-sections subject to opposite patch loading. One example
is a through-girder in which the column above and the column below
align but the girder needs to cantilever over the column below for var-
ious detailing reasons. Another example is a beam-column moment
connection under gravity loads, which features beams framing on
both sides of the column.
Research on opposite patch loading resistance of steel plates dates
back to the early 1900s with some researchers dealing primarily with
the elastic stability of steel plates and some others with their ultimate
resistance. The problem of elastic stability of a simply supported rectan-
gular plate subject to opposite patch loading was first attempted by
Sommerfield [1] and later by Timoshenko [2] who used a strain energy
approach to develop an approximate solution by neglecting the exten-
sional deformation of the middle surface during buckling. Leggett [3]
developed an accurate solution for the elastic stability of infinitely
⁎ Corresponding author.
E-mail addresses: fatmir.menkulasi@wayne.edu (F. Menkulasi), nfa004@latech.edu
(N. Farzana).
long plates subject to equal and opposite concentrated forces, and Hop-
kins [4] specialized the general problem of opposite patch loading to
that of sets of concentrated normal forces, and to that of a normal
force distributed uniformly over a finite length. Subsequently, Yamaki
[5] investigated the elastic stability of a rectangular plate under opposite
patch loading for the case when all edges are simply supported, and
when loaded edges are simply supported while the other edges are
clamped. Khan and Walker [6] and Lagerqvist and Johansson [7] pro-
posed a value for the elastic buckling coefficient, k, which resulted in
good capacity predictions.
Winter and Pian [8] developed empirical formulas for predicting the
crushing strength of steel thin webs in cold-formed steel members sub-
ject to opposite concentrated loads. While they admit that such a purely
empirical approach is likely to produce results of somewhat limited va-
lidity, they also state that the range of variations of dimensions and
shapes of test specimens considered is wide enough to cover the practi-
cally important range of elements.
Section J10 of AISC Specifications [9] addresses conditions in which
flanges and webs are subject to concentrated forces and provides two
equations for calculating the web compression buckling capacity of
members subject to equal and opposite concentrated forces. However,
in these equations, the loaded width is not a variable since these equa-
tions were originally developed for directly welded beam-column mo-
ment connections in which the concentrated loads come from the

https://doi.org/10.1016/j.jcsr.2019.105728
0143-974X/© 2019 Elsevier Ltd. All rights reserved.
2 F. Menkulasi, N. Farzana / Journal of Constructional Steel Research 162 (2019) 105728
Fig. 1. a) Opposite patch loading away from member ends, b) opposite patch loading at member ends, c) practical examples of opposite patch loading.
flanges of the beams and therefore are applied over small widths. As a
result, currently, the equations provided in AISC Specifications [9] result
in identical capacities for members subject to opposite patch loading re-
gardless of what the width of the applied load is.
While the studies conducted on the elastic stability of plates subject
to opposite patch loading are useful for calculating elastic buckling loads
for plates subject to certain boundary conditions, there is a need to use
these theoretical formulations to develop a generalized methodology
for predicting the elastic buckling capacity of I-sections that considers
the shortened web buckling half-wavelength resulting from flange rota-
tional restraint provided to the web. With respect to capacity prediction,
the most relevant and comprehensive work is the study conducted by
Lagerqvist and Johansson [7], who developed a methodology for
predicting the web compression buckling capacity of I-sections subject
to opposite patch loading away from the member ends. However, in
this study the case of I-sections subject to opposite patch loadings at
member ends was not evaluated. As a result, the goals of this paper
are to:
1) Determine the applicability of current AISC Specification [9] equa-
tions for checking the limit state of web compression buckling in
wide flange sections for a variety of opposite patch loading condi-
tions away or at member ends featuring various loaded widths;
2) Evaluate the methodology proposed by Lagerqvist and Johansson [7]
for I-beams subject to opposite patch loadings away from member
ends using an expanded experimental database and data from vali-
dated numerical simulations that systematically capture the influ-
ence of the loaded width;
3) Present two generalized methodologies for predicting web compres-
sion buckling capacity of steel I–sections subject to opposite patch
loading that consider the loaded width and position of load with re-
spect to the end of the member.
The two proposed methods are generally posed as a function of
loaded width to web depth ratio, and are applicable for opposite
patch loading applied at the interior of a wide flange section or at
the end of it, where the web has a free edge. The proposed methods
are based on a von Karman approach, are consistent with the formu-
lation of buckling related problems provided in most modern steel
design specifications, and include three parts: 1) an expression for
predicting the squash load, 2) an expression for predicting the elastic
buckling load, and 3) a resistance function. The squash load is calcu-
lated using an empirically derived effective width concept based on
ultimate load capacities obtained from an extensive experimental
database and validated numerical simulations. Web slenderness is
defined as the square root of the ratio of the web squash load to
the web critical elastic buckling load. The critical elastic buckling
load is defined consistently with that obtained from a plate buckling
energy solution for patch loadings on infinitely long strips and con-
siders the shortened web buckling half-wavelength resulting from
flange rotational restraint provided to the web. Resistance is calcu-
lated using the column global buckling curve equations provided in
the AISC Specifications [9] and the resistance function developed
by Lagerqvist and Johansson [7] for the first and second proposed
methods, respectively. The proposed methods are validated with
existing experimental data and shell finite element collapse simula-
tions, and are compared with current American Institute for Steel
Construction (AISC) [9] provisions and the method developed by
Lagerqvist and Johansson [7] in terms of accuracy, range of applica-
bility, and appropriate safety.
2. AISC 360–16 specifications prediction method for web compres-
sion buckling
Section J10 of AISC Specifications [9] addresses conditions in which
flanges and webs are subject to concentrated forces. The limit states ad-
dressed in this section are: flange local bending, web local yielding, web
crippling, web sidesway buckling, web compression buckling, and web
panel zone shear. The provisions for web compression buckling apply
to a pair of compressive single-concentrated forces or the compressive
components in a pair of double-concentrated forces, applied at both
flanges of a member at the same location.
Eq. (1) is used to check the limit state of web compression buck-
ling when the pair of concentrated compressive forces is applied at
 F. Menkulasi, N. Farzana / Journal of Constructional Steel Research 162 (2019) 105728
a distance from the member end that is greater than d/2, where d is
the overall depth of the member. When this distance is less than
d/2, a 50% reduction in capacity is recommended, which results in
Eq. (2). Both Eqs. (1) and (2) are used when the ratio between the
loaded width and overall member depth (N/d) is b1. When N/d is
not small, it is recommended that the member web be designed as
a compression member in accordance with Chapter E of AISC Speci-
fications [9]. The web compression buckling capacity, Rn, is a func-
tion of web thickness, t w , clear distance between flanges less the
fillet or corner radius for rolled shapes, h, modulus of elasticity, E,
and web yield stress, Fyw. Loaded width, N, is not a variable.
Eqs. (1) and (2) were developed by Chen and Newlin [10,11] and
Chen and Oppenheim [12,13] to quantify the column web strength in
directly welded beam-column moment connections. Fig. 2 illustrates
the investigated beam-column moment connection configurations,
as well as the test setup used to emulate the compressive forces in
the column web. Because the test setup was devised to emulate com-
pressive loads coming from the bottom flanges of the beams,
Eqs. (1) and (2) may not be applicable to other conditions such as
those illustrated in Fig. 1. The difference in the case of the through-
girder is clear because the loaded width provided by the top bearing
plate and the bottom cap plate is much larger than the thickness of a
beam flange in a directly welded beam-column moment connection.
Additionally, the applicability of Eqs. (1) and (2) to other types of
beam-column moment connections, such as the stiffened end plate
moment connection illustrated in Fig. 1 is also questionable because
of the ability of the stiffened end plate to further distribute the load
coming from the bottom flanges of the beam. Also, the 50% reduction
included in Eq. (2) was introduced in absence of applicable research
[9]. The main shortcoming of Eqs. (1) and (2) is the fact that the
loaded width is not a variable. Clearly, the greater the loaded width
the greater the portion of the web that can be engaged to resist the
Fig. 2. a) Schematic of typical interior beam to column connection and simulation of the compression region (adapted from [10]), b) test setup used by Chen and Newlin [10] to investigate
web buckling strength (adapted from [10]).
3
applied pair of concentrated compressive forces.
pffiffiffiffiffiffiffiffiffiffiffi
24t 3w EF yw
Rn ¼ away from the ends ð1Þ
h
pffiffiffiffiffiffiffiffiffiffiffi
12t 3w EF yw
Rn ¼ at member ends ð2Þ
h
The origin of Eq. (1) is the classical elastic buckling equation for a
plate subject to equal and opposite uniformly distributed loading. In
the derivation used by Chen and Newlin [10,11] and Chen and Oppen-
heim [12,13] it is assumed that the column flange acts as a bearing
plate and it distributes the load caused by the beam compression flange
of thickness, tb, to some larger length tb + 5 k at the edge of the column
web. Chen and Newlin [10,11] and Chen and Oppenheim [12,13] use the
variables N and tb interchangeably to denote the thickness of the beam
compression flange. In Fig. 2 the variable N is used and it represents the
thickness of the beam compression flange only and does not account for
the thickness of the weld. The distance from the column flange to the
edge of the column web is defined by k, which is the distance from
the outer face of the flange to the web toe of the fillet. This k is different
from the elastic plate buckling coefficient. It is further assumed that the
concentrated beam-flange load acts on a square panel whose dimen-
sions are h × h, where h is the clear distance between flanges less the fil-
let or corner radius. The theoretical elastic buckling stress for a plate
subject to equal and opposite uniformly distributed loading is provided
by Eq. (3), where k is the elastic plate buckling coefficient, E is the mod-
ulus of elasticity, υ is the Poisson's ratio, b is the plate height (the equiv-
alent of h used in Eqs. (1) and (2)), and t is the plate thickness (the
equivalent of tw used in Eqs. (1) and (2)). In the case of a square plate
simply supported on all four sides, k = 4 (elastic plate buckling coeffi-
cient). Additionally, when E = 29,000 ksi (200 GPa) and ν = 0.3,
4 F. Menkulasi, N. Farzana / Journal of Constructional Steel Research 162 (2019) 105728

Eq. (3) reduces to Eq. (4). To obtain the critical elastic buckling load for procedure into an iterative process because λ is a function of Fy, which
the plate, Eq. (4) is multiplied by the thickness, t, and the loaded width, in turn is function of ly, which in turn is a function of m2, and m2 is a
b, of the plate. This results in Eq. (5). Chen and Newlin [10,11] proposed function of λ.
Eq. (6) to check the limit state of web compression buckling, in which A resistance function, Х, is used to relate yield resistance to the ca-
they adjusted the coefficient in Eq. (5) to fit the results of the most crit- pacity of I-sections subject to the patch loading configurations men-
ical test and introduced the yield stress as a variable. When the yield tioned above (Eq. (13)). This resistance function is expressed as a
stress is 50 ksi (345 MPa), Eq. (6) reduces to Eq. (7). When the modulus function of the slenderness ratio, λ, which in turn is expressed as the
of elasticity and yield stress in Eq. (1) are taken equal to 29,000 ksi square root of the ratio of yield resistance, Fy, to the critical elastic buck-
(200 GPa), and 50 ksi (345 MPa), respectively, the coefficient becomes ling load, Fcr (Eq. (14)). The critical elastic buckling load, Fcr, is obtained
28,900, which is the rounded down version of the coefficient provided by multiplying the critical elastic buckling stress for a steel plate with
in Eq. (7). the thickness of the plate and height of the plate (Eq. (15)). The critical
elastic buckling load is a function of the buckling coefficient, kF
π2 E (Eq. (16)), which was formulated empirically to capture the influence
F cr ¼ k  2 ð3Þ
b of several parameters such as the loaded width to web height ratio, ss/
12ð1−υ2 Þ
t hw, web height to web panel length ratio, hw/a, and the coefficient β,
which captures the influence of flange and web geometry (Eq. (17)).
104842 It should be noted that in the design recommendations for inclusion in
F cr ¼  2 ð4Þ
b Eurocode Part 2 chapter 5.7, Lagerqvist and Johansson [7] proposed a
t simplified equation for determining the buckling coefficient kF. How-
ever, in the evaluation conducted in this paper Eq. (16) is used as op-
104842t 3 posed to the simplified equation. Additionally, the formulation of the
P cr ¼ ð5Þ critical elastic buckling load and the buckling coefficient are different
b
from that presented in this paper as part of the proposed methods.
pffiffiffiffiffiffi These differences will be discussed further in detail later in this paper.
4100t 3 σ y
P cr ¼ ð6Þ
b
F R ¼ F y X ðλÞ ð8Þ
3
28991 t
P cr ¼ ð7Þ F y ¼ f yw t w ly ð9Þ
b
 pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ly ¼ ss þ 2t f 1 þ m1 þ m2 ð10Þ
3. Lagerqvist and Johansson prediction method
f yf b f
m1 ¼ ð11Þ
Lagerqvist and Johansson [7] developed a design procedure for the f yw t w
resistance of I-girders to concentrated loads, which covers three differ-
 2
ent loading conditions namely patch loading, opposite patch loading hw
m2 ¼ 0:02 if λ ≥0:5; otherwise m2 ¼ 0 ð12Þ
and end patch loading. Patch loading and end patch loading represented tf
cases when the load was applied only on one flange away from the end
of the member and near the end of the member, respectively. The case 0:47
X ðλÞ ¼ 0:06 þ ≤1:0 ð13Þ
of opposite patch loading covered only cases when the opposing loads λ
were applied away from member ends. The case of opposite patch load- sffiffiffiffiffiffiffi
ing at member ends was not addressed. While many of the equations Fy
λ¼ ð14Þ
developed by Lagerqvist and Johansson [7] apply to all three loading F cr
conditions mentioned above, in this section only the design procedure
that pertains to opposite patch loading applied away from member π2 E t 3w
ends is discussed. F cr ¼ k F ð15Þ
12ð1−υ2 Þ hw
The proposed design methodology for opposite patch loading away
from member ends is summarized herein and consists of three parts:    2 ! pffiffiffiffi
!
ss hw
1) an expression for yield resistance, 2) an expression for predicting kF ¼ 1þ 3:4 þ 1:8 þ 0:1 4 β ð16Þ
2hw a
the critical elastic buckling load, and 3) an expression for predicting ca-
pacity. Eq. (8) gives the resistance of I-sections subject to all three load-
ing conditions mentioned above including opposite patch loading away b f t 3f
β¼ ð17Þ
from member ends. Eq. (9) represents the model for yield resistance, hw t 3w
which is expressed as a function of ly, which represents the length of
the region that undergoes complete yielding (both the web in trans-
verse compression and the flange in bending assuming that no buckling 4. Experiments on I-section webs with opposite patch loading
of any sort can take place. In this way, ly sets the upper limit of the resis-
tance. Additionally, ly is expressed primarily as a function of the loaded Experimental research on opposite patch loading on I-girder webs
width, ss, and flange thickness, tf, although it also captures the influences dates back to the 1960s with Graham et al. [14] conducting tests on col-
of web thickness, tw, flange width, bf, and material yield stress, fy. Also, ly, umn webs to emulate the compressive forces coming from the beam
can be calculated using Eq. (10), which was obtained using a combina- compression flanges in directly welded beam-column moment connec-
tion of phenomenological models for yield resistance based on collapse tions. Chen and Oppenheim [10,11] and Chen and Newlin [12,13] con-
mechanisms developed using the principles of concentrated plasticity tinued the research on this topic and proposed equations for
and empirically obtained coefficients (Eq. (10) through (12)). The em- estimating the strength of column webs subject to equal and opposite
pirical adjustment is reflected in the coefficient 0.02 in Eq. (12) for concentrated forces that are now part of the AISC Specifications [9].
cases that feature a slenderness ratio, λ, greater than or equal to 0.5. It Other experimental works on I-sections subject to opposite patch load-
should be noted, that the dependence of m2 on λ turns the design ing include studies conducted by Zoetmeijer [15], Aribert et al. [16,17],
 F. Menkulasi, N. Farzana / Journal of Constructional Steel Research 162 (2019) 105728
Table 1
Scope of parametric study.
Beam sections h/w h/w
1 2 3 4 5 1 2 3 4 5
W 200 × 15
W 310 × 23.8
W 410 × 46.1
Interior bearing condition End bearing condition
W 530 × 66
W 690 × 125
W 760 × 134
Lagerqvist and Johansson [7], and most recently Dos Santos et al. [18].
An experimental database of 84 tests on opposite patch loading was
compiled with the purpose of evaluating the AISC Specifications [9] for
opposite patch loading, evaluating the method proposed by Lagerqvist
and Johansson [7], as well as for evaluating the two methods proposed
in this paper. Out of the 84 tests, 76 tests were taken from the database
compiled by Lagerqvist [19] and 8 additional tests were added from the
study conducted by Dos Santos et al. [18]. The database compiled by
Lagerqvist [19] was further filtered to exclude tests that either featured
an extra loading plate or included bending moments in addition to op-
posite patch loading. This resulted in a total of 74 tests which were
used for evaluation purposes.
Details about the geometry and mechanical properties of the I-
sections tested in opposite patch loading configuration can be found
in Lagerqvist [19] and Dos Santos et al. [18]. The I-beams tested by
Dos Santos et al. [18] consisted of stainless steel sections, whereas the
database compiled by Lagerqvist [19] consisted of welded girders
made of high strength, quenched and tempered steel as well as girders
with ordinary steel grade. The main goal of these experimental pro-
grams was to observe the behavior of the girders near the ultimate
limit state and record the ultimate load capacity. The parameters that
varied between the tests were primarily the depth and thickness of
the web and the length of the applied load.
5. Simulations to collapse of I-section webs with opposite patch
loading
The compiled experimental database featured I-sections subject to
opposite patch loading when the load was applied away from the mem-
ber ends. Additionally, the goal of these experiments was to obtain the
Fig. 3. Loading cases and modeling approach considered in the parametric study.
5
Loaded width range (w) (mm) Column sections h/w Loaded width range (w) (mm)
10 15 20
41–201 W 250 × 73 13–25
61–305 W 310 × 97 Moment connection 15–30
81–404 W 360 × 91 18–36
104–526
135–678
150–749
ultimate load capacity of the investigated sections. Because the results
from physical tests already include the influence of geometric imperfec-
tions, yielding, and residual stresses, experimental setups do not pro-
vide an avenue for investigating the elastic stability of these sections
and for obtaining an accurate estimation of the critical elastic buckling
load, which is typically a key parameter in most modern steel design
codes for buckling related problems. Therefore, to complement
the information obtained from the experimental database, several
simulations to collapse of I-sections with opposite patch loading were
conducted to: 1) systematically study the influence of the loaded
width on elastic buckling loads and ultimate load capacity when the
load is applied away or at member ends, and 2) to develop an accurate
formulation for the critical elastic buckling load of I-sections subject to
opposite patch loading.
The wide flange sections and depth (h) over loaded width (w) ratios
that were considered are provided in Table 1 and feature both beam and
column sections. The beam sections are selected to represent various
bearing type loading conditions for which the limit state of web com-
pression buckling applies. The column sections are selected to represent
the loading condition present in beam-column moment connections
under gravity loading. A total of six beam sections with approximate
depths ranging from 203 mm to 762 mm are investigated. This range
of beam depths covers the majority of wide flange section depths used
as beams in steel structures. For each beam section, five overall depth
over loaded width ratios are considered, which range from 1.0 to 5.0.
Additionally, for each beam section two loading cases are examined:
1) interior bearing (opposite patch loading away from member ends),
and 2) end bearing (opposite patch loading at member ends) (Fig. 3).
The loaded width ranges for the beam sections are shown in Table 1.
The h/w ratios for the beam sections are selected such that they cover
6 F. Menkulasi, N. Farzana / Journal of Constructional Steel Research 162 (2019) 105728
a variety of bearing conditions. A lower bound of 1.0 was used for the h/
w ratio to void cases for which the loading condition may no longer rep-
resent opposite patch loading. This is consistent with AISC Specifications
in which it is recommended that for cases when h/w b 1.0 the member
web be designed as a compression member in accordance with
Chapter E of AISC Specifications [9].
Also, three column sections are investigated with approximate
depths ranging from 254 mm to 356 mm. The column sections are se-
lected such that they represent the majority of wide flange sections
used in column applications. Three overall depth to loaded width ratios
are considered for the columns and they ranged from 10 to 20 (Table 1).
The h/w ratios for the column sections were selected such that they sim-
ulate compressive loads coming from beam flanges in moment resisting
connections. Each wide flange section is subject to compressive loads at
both flanges. The compression load is applied in the form of a uniformly
distributed load over the loaded width (w) defined in Table 1 (Fig. 3).
5.1. Description of finite element modeling protocol
A total of 69 nonlinear finite element analyses were performed to
obtain failure loads for the investigated specimens, to evaluate existing
prediction methods, and to develop generalized mathematical models
for checking the limit state of web compression buckling that take
into account the influence of the loaded width and location of load
with respect to member end. The numerical simulations were per-
formed using the commercially available finite element analysis
software Abaqus [20]. The wide flange section was modeled as a deform-
able body with shell elements. The top and bottom plates shown in
Fig. 3 were modeled as rigid bodies and were connected to the top
and bottom flanges with a tie constraint such that the plates and the
corresponding portions of the flanges moved together. The length of
the top and bottom plates matched the loaded width, w, under consid-
eration. It was assumed that at the point of load application, the flanges
are held in position in the lateral direction either by their own stiffness
or by lateral bracing. A similar assumption was made in the study con-
ducted by Lagerqvist and Johansson [7]. Additionally, the top plate
was restrained against translations in directions 1 and 3 and against ro-
tations about all three axis to simulate out-of-plane lateral bracing, the
restraint provided by the rest of the beam, and the restraint provided
by the slab or any other supported member. The top plate was free to
translate in the vertical direction to accommodate the application of
the load. The bottom plate was restrained against all translations and
rotations.
The details shown in Fig. 1c justify the utilization of rigid top and
bottom plates in the numerical simulations. For example, in the detail
that features a through-girder (Fig. 1c left), the assembly of the stub col-
umn base plate at the top of the girder together with the composite con-
crete slab is rather rigid. Also, typical bracing details (such as knee
braces) used at the top of the lower column in Fig. 1c (left) justify the
assumption to restrain rotations as well as translations. Additionally,
the presence of concrete slab at the top of the through girder in Fig. 1c
(left) justifies the decision to restrain rotations and translations at the
top of the girder web. Similarly, in Fig. 1c (left), the assembly of the
floor column footprint together with the column cap plate at the bottom
of the girder is also rather rigid. Although, in this case there could be
some bending in the column cap plate. In this case, the engineer needs
to exercise judgment as to what the actual loaded width, w, is. In any
case, the loaded width should be taken as the distance over which the
applied force is effectively distributed to the supporting member [7].
Lagerqvist and Johansson [7] provide excellent guidance on this topic
by illustrating various loading conditions and by presenting recommen-
dations for how to determine the loaded width. Finally, in the beam-
column moment connection shown in Fig. 1c (right) the assembly of
the beam compression flange, beam web and beam end plate together
with the stiffener is rather rigid further justifying the use of rigid top
and bottom plates in the simulations. Also, the use of rigid top and
bottom plates is consistent with the majority of the tests conducted
on I-sections (Fig. 2), many of which featured bars welded to the top
and bottom flanges.
The length of the wide flange sections was selected such that it was
equal to three times the overall depth of the section to allow for a suffi-
cient distribution of the applied load in the web of the section. For ex-
ample, if the applied load was distributed to the web at a 45o angle,
and the ratio between the overall section depth (h) and loaded width
(w) is one, then the minimum section length necessary to allow for
this distribution is 2 h. Accordingly, a section length equal to 3 h was
chosen in case the distribution of load to the web occurs at smaller
angle than 45o with the horizontal axis. The restraint provided by the
continuation of the wide flange section to the vertical edges of the
webs was ignored.
Both flanges and the web were modeled using S8R5 shell elements.
The S8R5 element is a doubly-curved thin shell element with eight
nodes and it employs quadratic shape functions. The “5” in S8R5 de-
notes that each element has five degrees of freedom (three transla-
tional, two rotational) instead of six (three translational, three
rotational). The rotation of a node about the axis normal to the element
mid-surface is removed from the element formulation to improve com-
putational efficiency [21]. The “R” in the S8R5 designation denotes that
the calculation of the element stiffness is not exact; the number of
Gaussian integration points is reduced to improve computational effi-
ciency and avoid shear locking [21]. This element is designed to capture
the large deformations and through-thickness yielding expected to
occur during the out-plane buckling of the web to failure. The height
of the web was determined such that it extended from the line that rep-
resents the mid-thickness of the top flange to the line that represents
the mid-thickness of the bottom flange. Tie constraints were used to
connect the top and bottom of the web to the corresponding flanges.
The size of the mesh was selected such that each element side did not
exceed 25 mm in length and was determined based on results from con-
vergence studies to provide a balance between accuracy and computa-
tional expense. It was assumed that the self-weight of the specimens
was negligible compared to the applied loads. Although the cross-
sections were symmetrical about the major axis, it was necessary to
model the full cross-section because the buckled shape could be non-
symmetrical.
The finite element model takes into account both material and geo-
metric nonlinearities. The structural steel material was modeled using a
bilinear stress strain relationship based on coupon test data provided by
Arasaratnam et al. [22]. The true stress versus true strain relationship is
shown in Fig. 4 and was input into Abaqus to define the limits of the Von
Mises yield surface. Modulus of elasticity, E, was set to 200 GPa and
Poisson's ratio ν, was set to 0.3. The magnitude of the yield stress and
the slope of the strain hardening branch were later varied as part of a
parametric study to investigate their influence on section capacity.
Fig. 4. True stress-strain curve based on data from Arasaratnam et al. [22].
 F. Menkulasi, N. Farzana / Journal of Constructional Steel Research 162 (2019) 105728 7

To initiate buckling, an initial small out-of-plane geometric im- Table 2

perfection, in the form of the first mode shape obtained from an ei- Beam sections and material properties used in FEA of tested beams.

genvalue buckling analysis, was imposed to the model. Fig. 5 shows Test Section E (GPa) ν σy (measured) (MPa) σu (true) ϵu (true)
the first buckled mode shapes for W530 × 66 and W310 × 97, respec- 1a W 250 × 44.8 200 0.3 287 344 0.16
tively, for various overall depth (h) to loaded width (w) ratios. The 2a W 250 × 58 200 0.3 841 1009 0.16
corresponding elastic buckling loads in terms of uniformly distrib- 3b W 310 × 40 200 0.3 281 337 0.16
uted loads obtained from an eigenvalue buckling analysis are also 4a W 310 × 44.5 200 0.3 274 329 0.16
5b W 310 × 52 200 0.3 763 915 0.16
illustrated.
6b W 310 × 67 200 0.3 372 447 0.16
An Abaqus.fil file is created for each eigenbuckling analysis, 7b W 310 × 67 200 0.3 392 470 0.16
which is then called from the nonlinear.inp file with the *IMPERFEC- 8a W 310 × 67 200 0.3 815 978 0.16
TION command. During the design phase the imperfections are typ- a
Chen and Oppenheim [10,11].
ically unknown and are accounted for in the design equations used b
Chen and Newlin [12,13].
to estimate the capacity of the members. They are usually used as
general random quantities that can be rigorously treated by stochas- magnitude of the residual thermal and cold-work stresses resulting
tic techniques [23]. The magnitude of the initial imperfection consid- from gagging or rotary straightening rolled or welded built up
ered in this study is h/100, where h is the overall depth of the shapes to introduce or remove camber is far greater in the longitudi-
member. Initial imperfections larger than this magnitude were con- nal direction of the shape than in any transverse direction except for
sidered too large to be acceptable. Material nonlinearity is simulated surface defects [24]. Finally, the variation of residual stresses
in Abaqus with classical metal plasticity theory, including the as- through the thickness of the element is usually not important except
sumption of a Von Mises yield surface. The influence of residual for very thick elements and for walls of cold-formed tubular mem-
stresses on the web compression buckling capacity of steel bers neither of which were the subject of this study [24].
I-sections was considered to be inconsequential because residual The modified Riks method was used to determine the nonlinear
stresses due to non-uniform cooling of the metal typically form in response of the wide flange section. The modified Riks method (i.e.,
the longitudinal direction of the member and this study deals with *STATIC,RIKS in Abaqus), was developed in the early 1980's and
the buckling of the I-section webs due to opposite patch loading ap- enforces an arc length constraint on the Newton-Raphson incre-
plied in the transverse direction of the member. Additionally, the mental solution to assist in the identification of the equilibrium

Fig. 5. First buckled mode shapes for a beam and a column section, a) W530 × 66, b) W310 × 97.
8 F. Menkulasi, N. Farzana / Journal of Constructional Steel Research 162 (2019) 105728

Table 3 load along the length of the web at the top. All simulations were
Comparison of failure loads obtained from physical tests and FEA. conducted in a load controlled mode. Two reference nodes were
Test Section PultTest (kN) PultFEA (kN) Ratio = PultTest/ PultFEA assigned; one at the centroid of the top plate, and the other at
1a W 250 × 44.8 400 360 1.11
the centroid of the bottom plate (Fig. 3). For each case, the vertical
2a W 250 × 58 1125 1030 1.09 displacement of the reference node at the top plate and the reac-
3b W 310 × 40 285 262 1.09 tion at the reference node of the bottom plate were recorded. The
4a W 310 × 44.5 271 284 0.95 maximum vertical displacement at the reference node of the top
5b W 310 × 52 1045 853 1.22
flange was typically limited to 6 mm, because such a vertical dis-
6b W 310 × 67 738 650 1.14
7b W 310 × 67 1157 1221 0.95 placement corresponded with loads that were lower than the
8a W 310 × 67 747 680 1.10 peak load and were typically well into the descending branch of
Avg. 1.08
the load displacement curve.
COV (%) 8.45
a 5.2. Verification of modeling protocol with existing test data
Chen and Oppenheim [10,11].
b
Chen and Newlin [12,13].
To validate the modeling approach the failure loads for eight beam
path at highly nonlinear points along the load-deflection curve tests performed by Chen and Oppenheim [10,11] and Chen and Newlin
[25]. The load was applied as a uniformly distributed shell edge [12,13] were compared to the failure loads obtained from finite element

Fig. 6. Total load versus vertical displacement at the top of the web (interior bearing-beams).
 F. Menkulasi, N. Farzana / Journal of Constructional Steel Research 162 (2019) 105728
analyses. The tests were performed on various wide flange sections,
which were compressed on both flanges until the web buckled using
the test setup illustrated in Fig. 2. Table 2 summarizes the wide flange
sections and materials properties used in the finite element analyses
of tested beams. Modulus of elasticity, E, and Poisson's ratio, υ, were as-
sumed to be equal to 200 GPa and 0.3, respectively, since these proper-
ties were not reported by Chen and Oppenheim [10,11] and Chen and
Newlin [12,13]. The yield stress matched that measured from coupon
tests. The ultimate stress and strain were not reported in the studies
by Chen and Oppenheim [10,11] and Chen and Newlin [12,13]. The ulti-
mate true stress was assumed to be 20% greater than the measured yield
stress. The ultimate true strain was assumed to be 0.16.
A summary of the experimentally obtained failure loads and
those computed using finite element analyses is provided in
Table 3. The average ratio between the peak load obtained from the
tests and that obtained from finite element analyses is 1.08. The co-
efficient of variation is 8.45%. These statistics confirm that the
modeling approach used in this study provides reliable capacity
predictions.
Fig. 7. Total load versus vertical displacement at the top of the web (end bearing-beams).
9
5.3. Load-deformation response trends of I-section webs with opposite
patch loading
Fig. 6 shows the applied load versus vertical displacement curves for
all beam sections, which were loaded to simulate an interior bearing
condition (opposite patch loading away from member ends). Table 8
provides a summary of the peak loads obtained from finite element
analyses. As expected, when the loaded width is larger (i.e. h/w ratio
lower) the peak load is also larger. This is due to the fact that a larger
loaded width engages a greater portion of the web in resisting the ap-
plied load, thus resulting in a higher peak load. The difference between
the peak loads for h/w ratios equal to 1.0 and 5.0 varies from 50% to 58%
for a given section. As can be seen, the magnitude of the loaded width
influences the buckling capacity of the web. This change in capacity as
it relates to the limit state of web compression buckling would have
not been captured using the current AISC Specification [9] equations.
Fig. 7 illustrates the applied load versus vertical displacement curves
for all beam sections, which were loaded to simulate an end bearing
condition (opposite patch loading at member ends). Also, in this case
10 F. Menkulasi, N. Farzana / Journal of Constructional Steel Research 162 (2019) 105728

Fig. 8. Total load versus vertical displacement at the top of the web (interior bearing-columns).

the peak loads increase as the loaded width increases. As expected, the
peak loads for bearing conditions at the end of the beam are lower than
those obtained for a bearing condition at the interior of the beam. This
difference becomes more pronounced for higher h/w ratios. For exam-
ple, the difference between the peak loads obtained for interior bearing
and end bearing conditions for a W410 × 46.1, when the h/w ratio is
equal to 1.0, is 24%. However, when the h/w ratio for the same section
is 5.0, the difference in the peak loads is 213%. This is due to the fact
that the difference between the portions of the web that are effective
in resisting the applied loads is relatively smaller when the loaded
width is large. The effective width used in providing resistance to the
applied loads consists of the loaded width plus an additional portion
of the web, which is engaged in resisting the load due to the lateral dis-
tribution of the load in the web. When the loaded width is small, then
the majority of the effective web width comprises of the portion of the
web that is engaged due to the lateral distribution of the load.
Fig. 8 shows the load versus vertical displacement curves for the col-
umn sections. For the column sections the distinction between the peak
loads for various h/w ratios is not as pronounced. This is due to the fact
that the difference between the loaded widths is not as significant as the
one considered for the beam sections. For example, in the case of W310
× 97 an h/w ratio equal to 20 corresponds to a loaded width equal to
15 mm. When the h/w ratio is 10 then the loaded width is 31 mm.
Accordingly, a difference between a 15 mm and 31 mm loaded width
did not result in a marked difference in the peak loads obtained from fi-
nite element analysis.
5.4. Influence of member length and boundary conditions on section
capacity
As stated earlier, the wide flange section lengths (L) used in the non-
linear finite element analyses were taken equal to three times the over-
all depth of the section (3 h) to allow for a sufficient distribution of the
applied load to the web of the section. Additionally, the edges of the
webs for all wide flange sections were conservatively modeled as free.
The influence of greater section lengths and alternative boundary condi-
tions for the edges of the web on the buckling capacity of the web for in-
terior loading conditions was investigated by considering three
additional section lengths: 4 h, 5 h, and 6 h. W410 × 46.1 and an h/w
= 3.0 were chosen as the wide flange section and height over loaded
width ratio to investigate the influence of the section length and web
edge boundary conditions on the computed web compression buckling
capacity. Each case was analyzed assuming that the boundary condi-
tions for the edges of the web were free or restrained against transla-
tions in directions 1 and 3 (Fig. 3) while all other degrees of freedom

Table 4
Effect of section length and web boundary conditions.
Table 5
Section Boundary condition Computed web compression Effect of the slope of the strain-hardening branch.
buckling capacity (kN) (h/w = 3.0)
Section and aspect ratio Boundary conditions Fy (MPa) Fu (MPa) Rn (kN)
3h 4h 5h 6h
407 414 485
W 410 × 46.1 Free Web 489 512 512 516 407 552 489
Pinned Weba 507 516 516 516 W 410 × 46.1 (h/w = 3.0) Free Web 407 621 489
a 407 690 489
Restrained against translations in directions 1 and 3 (Fig. 3) while all other degrees of
407 703 489
freedom are free.
 F. Menkulasi, N. Farzana / Journal of Constructional Steel Research 162 (2019) 105728 11

Table 6
Effect of the yield stress on capacity.

Beam Section Peak load (interior)

h/w = 1.0 h/w = 5.0

Fy (MPa) Fy (MPa)

345 414 483 552 621 690 345 414 483 552 621 690

W 200 × 15 352 400 434 469 496 531 228 255 283 303 324 345
W 530 × 66 1496 1641 1786 1903 1993 2082 945 1048 1131 1214 1276 1338
W 760 × 134 2710 2986 3192 3413 3599 3730 1744 1931 2082 2220 2337 2441

Peak load (end)

W 200 × 15 296 331 359 386 414 427 110 124 138 145 152 159
W 530 × 66 1227 1358 1448 1517 1607 1689 455 503 545 572 607 641
W 760 × 134 2227 2379 2593 2703 2806 2944 827 910 986 1048 1096 1151

Column section Peak load (column)

h/w = 10 h/w = 20

W 250 × 73 848 965 1089 1200 1303 1413 821 938 1055 1165 1269 1372
W 310 × 97 1103 1262 1413 1551 1689 1841 1069 1220 1365 1503 1634 1744
W 360 × 91 1076 1234 1379 1510 1641 1751 1034 1186 1324 1455 1579 1689

were free. Table 4 provides a summary of the computed web compres-
sion buckling capacity for these additional cases. The difference be-
tween the buckling capacities for different section lengths and a given
web edge boundary condition is negligible. Similarly, the difference be-
tween buckling capacities for different web edge boundary conditions
and a given section length is also negligible. These results suggest that
neither the section length nor the assumption for the boundary condi-
tions at the edges of the web results in a significant difference in the
computed web compression buckling capacities.
5.5. Influence of strain hardening on section capacity
The effect that the slope of the strain hardening branch has on the
peak loads was examined because the peak loads obtained from the
nonlinear finite element analyses were achieved after the wide flange
sections were compressed past their linear range. The yield stress was
assumed to be 407 MPa and the ultimate stress was varied from
414 MPa to 703 MPa. W410 × 46.1 and free web edge boundary condi-
tions were chosen to study the influence of the slope of the strain hard-
ening branch. Table 5 provides the computed peak loads for each
investigated case. The results suggest that the slope of the strain hard-
ening branch has a negligible effect on the computed peak loads. This
is because after a certain point during the loading process geometric
nonlinearities take over regardless of what the slope of the strain
hardening branch is and result in computed capacities that are similar
in all cases.
5.6. Influence of yield stress magnitude on section capacity
Because the peak loads obtained from the nonlinear finite element
analyses were achieved after the wide flange sections were compressed
past their linear range, the effect of the yield stress on the computed ca-
pacities was examined. Three beam sections and three column sections
were considered to study the influence of the yield stress. Also, for each
section the two most extreme h/w ratios were investigated. The yield
stress was varied from 345 MPa to 690 MPa in 69 MPa increments.
Table 6 provides a summary of the peak loads for all considered cases.
The magnitude of the yield stress appears to have a significant influence
on the computed capacities. The higher the yield stress the higher the
computed capacity. As a result, the yield stress was included as a vari-
able in both proposed prediction methodologies.
6. Proposed web compression buckling strength limit state predic-
tion method for opposite patch loading
Two methods are presented for predicting the capacity of I-sections
to opposite patch loading. Each prediction method consists of three
parts: 1) an expression for predicting capacity, 2) another for predicting
yield resistance, and 3) an expression that relates yield resistance to sec-
tion capacity, which is a function of the square root of the ratio between
yield resistance and critical elastic buckling load. In the first method, the
expression that relates yield resistance to section capacity is based on
the AISC Specification [9] column global buckling curve, while in the
second method this expression is equal to the resistance function devel-
oped empirically by Lagerqvist and Johansson [7] . Both methods adopt
the same formulation for the critical elastic buckling load, and are based
on a von Karman approach for the formulation of the yield resistance.

6.1. Critical elastic buckling load

Fig. 9. Relationship between ratio of loaded width to web depth, w/hw, and elastic buckling
coefficient, k.
As stated earlier, classical plate solutions exist for the critical elastic
buckling load, Pcrl, of a plate subjected to opposite patch loading [2–6].
The solution selected for both prediction methods presented in this
paper is based on Hopkins' [4] approach, which employs a variational
energy treatment considering a prebuckling reference stress distribu-
tion pattern that includes St. Venant force spreading from the load ap-
plication points. The critical elastic buckling stress for a rectangular
12 F. Menkulasi, N. Farzana / Journal of Constructional Steel Research 162 (2019) 105728

plate subject to opposite uniformly distributed compressive loads can thickness, tw, and loaded width, w (Eq. (19)). This formulation is
be calculated using Eq. (18). The critical elastic buckling load, Pcrl, is cal- adopted to predict critical elastic buckling loads for both prediction
culated as the product of the critical elastic buckling stress, σcrl, web methods presented in this paper and is different from that used by

Fig. 10. a) Comparison of computed and predicted critical elastic buckling loads, b) comparison of computed and predicted capacities, c), d) comparison of computed and predicted
capacities normalized by the squash load as a function of slenderness ratio, e) ratio of computed and predicted capacities as a function of slenderness ratio.
 F. Menkulasi, N. Farzana / Journal of Constructional Steel Research 162 (2019) 105728 13

Lagqerqvist and Johansson [7], Graham et al. [14], Chen and Newlin
[10,11], and Chen and Oppenheim [12,13], who obtained the critical
elastic buckling load by multiplying the critical elastic buckling stress,
σcrl, with the width of the web, tw, and height of the web, hw, as opposed
to the loaded width, w.
The equation relating loaded width, w, and plate width (web depth,
hw) to the plate buckling coefficient k, is solved and reported for several
cases in Hopkins [4]. The solution is derived for infinitely long, rectangu-
lar, isotropic plates, simply supported at the long edges and is illustrated
in Fig. 9. The critical elastic buckling loads obtained from the numerical
simulations were used to back calculate the buckling coefficient, k, for
all investigated cases using Eqs. (18) and (19). As can be seen, from
Fig. 9 the shape of the curve that relates the buckling coefficient, k, to
the ratio of the loaded width over web depth, w/hw, is similar with
that obtained analytically by Hopkins [4] for infinitely long, rectangular,
isotropic plates, simply supported at the long edges. Higher values of k
are obtained for lower values of w/hw. The curves for opposite patch
loading away from and at member ends are both offset from the curve
developed using Hopkins' [4] solution due to the difference in boundary
conditions. As stated above, the solution derived by Hopkins [4] applied
to infinitely long plates simply supported at the long edges, whereas the
numerical solutions obtained during this study apply to I-sections sub-
ject to opposite patch loading in which all rotations and translations at
the web-flange connection were restrained with the exception of the
vertical translation at the top of the web which was set free to accom-
modate the application of the load. As a result, a modification to the as-
sumed buckling half-wavelength that forms vertically along the web in
Hopkins' [4] solution is required to account for the rotational restraint
provided by the flanges to the web. The reason why the curve for end
loading is closer to Hopkins' [4] curve is because the combination of
boundary conditions in this case results in solutions that are closer to
Hopkins' solutions compared to those obtained for interior loading.
By looking at the shape of the curves, it becomes evident that the re-
lationship between k and w/hw can be best approximated by a power

Table 7
Comparison of predicted and computed elastic buckling loads (kN).

Section h/w Interior Loading End Loading

Lagerqvist Johannsson [7] Proposed FEA Ratio = Pcrl-FEA/Pcrl-predicted Proposed FEA Pcrl FEA/
(Pcrl_LJ) (Pcrl proposed) (Pcrl FEA) (Pcrl proposed) (Pcrl FEA) Pcrl proposed
PcrlFEA/ Pcrl FEA/
Pcrl LJ Pcrl proposed

W 200 × 15 416 419 446 1.07 1.06 358 353 0.99

W 310 × 23.8 588 594 626 1.06 1.05 506 496 0.98
W 410 × 46.1 1 879 883 961 1.09 1.09 754 760 1.01
W 530 × 66 1374 1387 1488 1.08 1.07 1181 1184 1.00
W 690 × 125 2431 2449 2649 1.09 1.08 2087 2104 1.01
W 760 × 134 2333 2353 2537 1.09 1.08 2002 2019 1.01

W 200 × 15 345 343 344 1.00 1.00 216 213 0.99

W 310 × 23.8 489 487 483 0.99 0.99 305 300 0.98
W 410 × 46.1 2 729 723 749 1.03 1.04 454 463 1.02
W 530 × 66 1141 1136 1145 1.00 1.01 712 715 1.00
W 690 × 125 2019 2006 2054 1.02 1.02 1257 1277 1.02
W 760 × 134 1939 1929 1956 1.01 1.01 1207 1220 1.01

W 200 × 15 322 322 317 0.98 0.99 171 159 0.93

W 310 × 23.8 456 456 441 0.97 0.97 242 237 0.98
W 410 × 46.1 3 680 677 691 1.02 1.02 360 369 1.03
W 530 × 66 1064 1064 1046 0.98 0.98 565 566 1.00
W 690 × 125 1882 1878 1888 1.00 1.01 997 1014 1.02
W 760 × 134 1807 1806 1791 0.99 0.99 958 967 1.01

W 200 × 15 310 312 304 0.98 0.97 150 148 0.99

W 310 × 23.8 439 443 421 0.96 0.95 213 208 0.98
W 410 × 46.1 4 655 657 664 1.01 1.01 316 326 1.03
W 530 × 66 1025 1034 998 0.97 0.97 496 495 1.00
W 690 × 125 1813 1824 1810 1.00 0.99 876 890 1.02
W 760 × 134 1742 1755 1713 0.98 0.98 841 847 1.01

W 200 × 15 303 308 297 0.98 0.96 138 137 0.99

W 310 × 23.8 429 437 410 0.96 0.94 196 191 0.97
W 410 × 46.1 5 640 648 649 1.01 1.00 291 299 1.03
W 530 × 66 1002 1019 971 0.97 0.95 458 454 0.99
W 690 × 125 1772 1798 1767 1.00 0.98 808 818 1.01
W 760 × 134 1702 1730 1669 0.98 0.96 777 776 1.00

W 250 × 73 1904 1980 2050 1.08 1.04 Avg 1.00

10
W 310 × 97 2357 2454 2510 1.07 1.02 COV 0.02
W 360 × 91 1817 1892 1967 1.08 1.04 Min. 0.93

W 250 × 73 1873 1995 2029 1.08 1.02 Max. 1.03

15
W 310 × 97 2317 2473 2473 1.07 1.00 % cons. 57
W 360 × 91 1787 1907 1940 1.09 1.02

W 250 × 73 1857 2015 2018 1.09 1.00

20
W 310 × 97 2298 2499 2450 1.07 0.98
W 360 × 91 1772 1927 1922 1.08 1.00

Avg. 1.03 1.01

COV 0.05 0.04
Min. 0.96 0.94
Max. 1.09 1.09
% cons. 59 54
14 F. Menkulasi, N. Farzana / Journal of Constructional Steel Research 162 (2019) 105728
function. As a result, an equation of the form y = axb + c was used to fit a
curve to the data points obtained numerically for opposite patch loading
away from member ends and at member ends. In general, the values of k
obtained for a given w/hw ratio for opposite patch loading at member
ends were lower than those obtained for opposite patch loading away
from member ends, which is logical. The obtained expressions for k for
opposite patch loading away from member ends (interior bearing) and
at member ends (end bearing) are given by Eqs. (20) and (21).
Fig. 9 also shows the relationship between w/hw ratio and k if k were
to be back calculated using the formulation for the critical elastic buck-
ling load used by Lagerqvist and Johansson [7] (Eq. (15)). In this case,
the relationship is linear; higher k values are obtained for higher w/hw
ratios.
The difference between these k-values and those obtained from
Eqs. (20) and (21) is because in Eq. (15) hw is used when deter-
mining Fcr as opposed to w in Eq. (19) when determining Pcrl. As
a result, the way the critical elastic buckling load is formulated de-
termines the form of the function adopted for the buckling coeffi-
cient, k.
kπ2 Et 2w
σ crl ¼
12ð1−υ2 Þhw
2 between the two resistance functions is the relationship between ca-
P crl ¼ σ crl t w w
 −1:072
w for low slenderness ratios. Whereas in the formulation developed by
k ¼ 3:261 þ 2:311 ðinterior loading Þ
hw
 −1:205
w ceeds the squash load. The influence of these differences on model pre-
k ¼ 0:7569 þ 3:942 ðend loading Þ
hw
6.2. Capacity prediction method 1
Strength prediction of I-section webs subject to opposite patch
loading is considered by assuming the existing AISC Specification [9]
column global buckling curve represents an accurate relationship be-
tween web slenderness and capacity. Eqs. (22) and (23) are used to cal-
culate the web compression buckling capacity, Rn, and take into
account the presence of initial imperfections and yielding. Web slen-
derness, λc, is defined as (Py/Pcrl)0.5 (Eq. (24)), where Py is the squash
load and Pcrl is the critical elastic buckling load. The squash load, Py, is
determined by multiplying the web thickness, tw, with an effective
web width, weff, and the yield stress, Fy (Eq. (6)). The effective web
width, weff, was obtained by considering a spreading yield zone, and
it can be calculated using Eqs. (26) and (27). As can be seen, the effec-
tive width is a function of the loaded width and flange thickness. The
coefficients for the loaded width and flange thickness were obtained it-
eratively to arrive at a reasonable balance between accuracy and appro-
priate safety.
 2
 conservatively 59% of the cases considered, while the proposed method
Rn ¼ 0:658λc P y for λc ≤1:5
!
0:877
Rn ¼ P y for λc N1:5
λ2c
sffiffiffiffiffiffiffi
Py
λc ¼
P crl
P y ¼ weff t w F y
weff ¼ w þ 12t f for interior loading
weff ¼ w þ t f for end loading ð27Þ
6.3. Capacity prediction method 2
0:877
In this method, the reduction factors 0.658λc and
2
used in
λ2c
Eqs. (22) and (23), respectively, are replaced by the resistance function
X(λ) developed by Lagerqvist and Johansson [7]. This function is given
in Eq. (13) and is based on a von Karman approach in which the squash
load is related to capacity using the following equation Х = 1/λ, where
Х is the resistance function and λ is the slenderness ratio. Lagerqvist and
Johansson [7] modified this expression to Х = a + b/λ, where a and b
are constants developed to better fit experimental data. One difference
between the AISC column curve and the resistance function developed
by Lagerqvist and Johansson [7] is that in AISC Specifications [9] the in-
fluence of the squash load disappears for cases when the slenderness
ratio is N1.5 (elastic buckling region), whereas the formulation devel-
oped by Lagerqvist and Johansson [7] captures the influence of possible
post buckling resistance for large slenderness ratios. Another difference
ð18Þ
pacity and squash load in regions of low slenderness ratios. In the
AISC Specifications [9], the column global buckling curve expressed as
ð19Þ
the relationship between the ratio of capacity to squash load, Rn/Py,
and slenderness ratio, λ, transitions gradually to a ratio of Rn/Py = 1.0
ð20Þ
Lagerqvist and Johansson [7] there is a sudden cut in the curve for
those cases that feature low slenderness ratios in which capacity ex-
ð21Þ diction accuracy is investigated in the subsequent sections.
7. Comparison of predicted and computed elastic buckling loads
and capacities
Critical elastic buckling loads obtained from eigenvalue buckling
analyses were compared with those obtained based on the equation
proposed by Lagerqvist and Johansson [7] (Eq. (15)) and that proposed
in this paper (Eq. (19)). The results are illustrated graphically in Fig. 10a
and in tabular format in Table 7. Table 7 provides a summary of the crit-
ical elastic buckling loads and the ratio between computed and pre-
dicted values for opposite patch loading away from and at member
ends. It should be noted that the method proposed by Lagerqvist and
Johansson [7] applies only to opposite patch loading away from member
ends and is included in comparison only for this loading condition.
Both methods provide rather accurate estimations of elastic buckling
loads despite the fact that the prediction for this load was formulated
differently. The average ratio between the computed and predicted
elastic buckling load and the corresponding coefficient of variation
(COV) is 1.03 and 0.05 for the method proposed by Lagerqvist and
Johansson [7], and 1.01 and 0.04 for the method proposed in this
paper. The method proposed by Lagerqvist and Johansson [7] predicts
ð22Þ predicts conservatively 54% of the cases. However, it should be noted
that the minimum and maximum computed to predicted ratios for the
elastic buckling loads for the method proposed by Lagerqvist and
ð23Þ Johansson [7] and the method proposed in this paper are 0.96 and
1.09, and 0.94 and 1.09, respectively. As a result, both methods provide
rather accurate predictions of the critical elastic buckling load, albeit the
proposed method is more general because it covers opposite patch load-
ð24Þ
ing away from and at member ends whereas the method proposed by
Lagerqvist and Johansson [7] covers only loading conditions where the
opposing loads are applied away from member ends.
ð25Þ
Table 8 provides a summary of computed and predicted capacities
based on AISC Specification section J10 [9], the method proposed by
ð26Þ
Lagerqvist and Johansson [7], the first proposed method, and the second
 F. Menkulasi, N. Farzana / Journal of Constructional Steel Research 162 (2019) 105728 15

Table 8
Comparison of predicted and computed web buckling capacities (kN).

Section h Interior Loading End Loading

w
AISC Lager. and Prop.1 Prop.2 FEA Ratio = Rn-FEA/Rn-predicted AISC Prop.1 Prop.2 FEA Ratio = Rn-FEA/
360–16 Johan. [7] (Rn_P_1) (Rn_P_2) (Rn_FEA) 360–16 (Rn_P_1) (Rn_P_2) (Rn_FEA) Rn-predicted
(Rn_AISC) (Rn_LJ) (Rn_AISC)
Rn FEA Rn FEA Rn FEA Rn FEA Rn FEA Rn FEA Rn FEA
Rn AISC Rn LJ Rn P 1 Rn P 2 Rn AISC Rn P 1 Rn P 2

W 200 × 15 89 245 291 235 254 2.85 1.04 0.87 1.08 45 237 191 210 4.70 0.89 1.10
W 310 × 23.8 127 410 473 392 411 3.24 1.00 0.87 1.05 63 394 324 340 5.37 0.86 1.05
W 410 × 46.1 1 188 662 741 638 681 3.63 1.03 0.92 1.07 94 613 514 547 5.82 0.89 1.06
W 530 × 66 296 1049 1163 1001 1050 3.55 1.00 0.90 1.05 148 976 828 867 5.86 0.89 1.05
W 690 × 125 522 1830 2042 1747 1868 3.58 1.02 0.91 1.07 261 1703 1432 1496 5.74 0.88 1.05
W 760 × 134 502 1910 2025 1809 1909 3.80 1.00 0.94 1.06 251 1709 1505 1521 6.06 0.89 1.01

W 200 × 15 89 177 202 164 196 2.20 1.11 0.97 1.19 45 129 105 125 2.80 0.97 1.19
W 310 × 23.8 127 293 336 271 316 2.49 1.08 0.94 1.17 63 220 178 205 3.24 0.93 1.15
W 410 × 46.1 2 188 477 549 448 534 2.84 1.12 0.97 1.19 94 346 283 338 3.60 0.98 1.20
W 530 × 66 296 749 849 689 808 2.73 1.08 0.95 1.17 148 554 454 525 3.55 0.95 1.16
W 690 × 125 522 1314 1495 1213 1454 2.79 1.11 0.97 1.20 261 963 786 930 3.57 0.97 1.18
W 760 × 134 502 1363 1508 1240 1472 2.93 1.08 0.98 1.19 251 982 825 951 3.79 0.97 1.15

W 200 × 15 89 154 169 141 178 1.99 1.16 1.05 1.27 45 93 77 90 2.02 0.97 1.17
W 310 × 23.8 127 252 283 229 285 2.25 1.13 1.01 1.24 63 161 130 160 2.53 0.99 1.24
W 410 × 46.1 3 188 414 476 384 489 2.60 1.18 1.03 1.27 94 256 206 262 2.79 1.02 1.27
W 530 × 66 296 645 725 583 729 2.47 1.13 1.01 1.25 148 409 331 409 2.77 1.00 1.24
W 690 × 125 522 1136 1281 1031 1329 2.55 1.17 1.04 1.29 261 710 573 724 2.78 1.02 1.26
W 760 × 134 502 1174 1296 1046 1342 2.67 1.14 1.04 1.28 251 734 599 734 2.93 1.00 1.22

W 200 × 15 89 142 152 129 169 1.89 1.19 1.11 1.31 45 74 63 86 1.93 1.16 1.37
W 310 × 23.8 127 232 255 208 271 2.14 1.17 1.06 1.30 63 130 105 138 2.18 1.06 1.31
W 410 × 46.1 4 188 381 437 351 469 2.50 1.23 1.07 1.33 94 209 168 231 2.46 1.11 1.37
W 530 × 66 296 591 656 529 694 2.35 1.17 1.06 1.31 148 334 269 350 2.37 1.05 1.30
W 690 × 125 522 1045 1164 940 1263 2.42 1.21 1.08 1.34 261 579 466 627 2.40 1.08 1.34
W 760 × 134 502 1077 1176 947 1276 2.54 1.19 1.08 1.35 251 603 487 639 2.55 1.06 1.31

W 200 × 15 89 134 141 121 162 1.81 1.20 1.15 1.33 45 63 54 80 1.79 1.28 1.48
W 310 × 23.8 127 219 237 196 261 2.06 1.19 1.10 1.33 63 111 91 126 1.99 1.14 1.39
W 410 × 46.1 5 188 362 412 332 454 2.42 1.26 1.10 1.37 94 180 145 212 2.26 1.18 1.46
W 530 × 66 296 559 611 497 670 2.27 1.20 1.10 1.35 148 287 232 320 2.16 1.12 1.38
W 690 × 125 522 989 1090 884 1222 2.34 1.24 1.12 1.38 261 497 402 577 2.21 1.16 1.43
W 760 × 134 502 1017 1099 887 1235 2.46 1.21 1.12 1.39 251 521 419 581 2.32 1.12 1.39

W 250 × 73 581 620 596 590 618 1.06 1.00 1.04 1.05 Avg. 3.22 1.02 1.24
10
W 310 × 97 721 786 748 738 804 1.12 1.02 1.08 1.09 COV 0.41 0.10 0.11
W 360 × 91 556 700 737 667 786 1.41 1.12 1.07 1.18 Min. 1.79 0.86 1.01

W 250 × 73 581 603 574 579 605 1.04 1.00 1.05 1.05 Max. 6.06 1.28 1.48
15
W 310 × 97 721 763 718 721 787 1.09 1.03 1.10 1.09 %Con. 100 50 100
W 360 × 91 556 677 707 651 765 1.38 1.13 1.08 1.18

W 250 × 73 581 594 564 574 600 1.03 1.01 1.06 1.04
20
W 310 × 97 721 751 703 715 778 1.08 1.04 1.11 1.09
W 360 × 91 556 666 693 644 756 1.36 1.14 1.09 1.17

Avg. 2.28 1.12 1.03 1.21

COV 0.34 0.07 0.07 0.10
Min. 1.03 1.00 0.87 1.04
Max. 3.80 1.26 1.15 1.39
% Con. 100 95 69 100

proposed method. The method proposed by Lagerqvist and Johansson
[7] is included in comparison with computed capacities only for loading
conditions that feature opposite patch loading away from member
ends.
The prediction method provided in the AISC Specification section J10
[9] provides average ratios between computed and predicted capacities
equal to 2.28 and 3.22 for opposite patch loading away from and at
member ends, respectively. The corresponding coefficients of variation
are 0.34 and 0.41, respectively. The minimum and maximum ratios be-
tween computed and predicted capacities for each case are 1.03 and
3.80 for opposite patch loading away from member ends, and 1.79
and 6.06 for opposite patch loading at member ends. While the AISC
Specification [9] method predicts conservatively 100% of the cases in-
vestigated numerically, it is rather inaccurate. This is due to the fact
that most of the I-sections considered in the numerical study featured
hw/w ratios that were relatively small, while the equations provided in
the AISC Specifications were developed for large hw/w ratios (column
webs subject to concentrated forces coming from beam flanges) and
do not account for the variation in the loaded width.
Additionally, the ratios show that this underestimation becomes
more pronounced as the h/w ratios become smaller. This is expected be-
cause higher h/w ratios are closer to the assumption of a simply sup-
ported square panel used in the derivation of Eq. (1). However, even
for h/w ratios equal to 5.0 the AISC equation still significantly underesti-
mates the buckling capacity of the web. Depending on which beam is
considered the underestimation of the buckling capacity for an interior
bearing condition and an h/w ratio equal to 5.0 varies from 45% to 59%.
The underestimation of the buckling capacity becomes more pro-
nounced as the section depth gets larger. For h/w ratios higher than or
equal to 3.0 the ratios between the predicted capacity and computed ca-
pacity for interior and end loading conditions are similar. For these cases
this justifies the 50% reduction for end bearing condition included in
Eq. (2), however the coefficients 24 and 12 in Eqs. (1) and (2), respec-
tively, are approximately half of what they should be if the predicted
16 F. Menkulasi, N. Farzana / Journal of Constructional Steel Research 162 (2019) 105728
load were to match the computed one. For h/w ratios equal to 1 and 2
the ratios between the predicted and computed capacities for end bear-
ing conditions are lower than those calculated for interior bearing con-
ditions. This suggests that the 50% reduction for these h/w ratios is
significantly conservative.
Eq. (1) was used to predict the web compression buckling capacity
of the investigated column webs. The ratios between predicted capaci-
ties and computed capacities suggest that Eq. (1) does a reasonably
good job at predicting the buckling capacity of the column web for sec-
tions W250 × 73 and W310 × 97. For section W360 × 91, the prediction
of Eq. (1) errs on the conservative side by 26–30%. Because the h/w ra-
tios considered for the column sections result in loaded widths that do
not vary as much as those considered in the beam sections, the pre-
dicted versus computed ratios for a given section are similar. This is
due to the fact that Eqs. (1) and (2) were originally developed specifi-
cally to quantify the capacity of column webs in beam-column moment
connections.
The variable h in Eqs. (1) and (2) is defined as the clear distance be-
tween flanges less the fillet or corner radius for rolled shapes. Because
the wide flange sections in this study were modeled using shell ele-
ments for the top and bottom flanges as well as for the web, the variable
h was taken equal to the distance between the centerlines of top and
bottom flanges to make a consistent comparison with the peak loads
obtained from finite element analyses.
The prediction method proposed by Lagerqvist and Johansson [7]
provides an average ratio between computed and predicted capacities
equal to 1.12 and a COV equal to 0.07 for opposite patch loading applied
away from member ends. The minimum and maximum ratios between
predicted and computed capacities are 1.00 and 1.26, respectively, and
95% of the cases are predicted conservatively. These statistics suggest
that the prediction method proposed by Lagerqvist and Johansson [7]
for opposite patch loading away from member ends provides fairly ac-
curate predictions of capacities and is rather ideal for design purposes.
The first proposed prediction method provides an average ratio be-
tween computed and predicted capacities equal to 1.03 and a COV
equal to 0.07 for opposite patch loading applied away from member
ends. The minimum and maximum ratios between computed and pre-
dicted capacities are 0.87 and 1.15, respectively, and the number of
cases that are predicted conservatively are 69% of the total cases consid-
ered. For load cases that feature I-sections with opposite patch loading
at member ends, the average ratio between computed and predicted ca-
pacities is 1.02, the COV is 0.10, the minimum and maximum ratios
between computed and predicted capacities are 0.86 and 1.28, respec-
tively, and the percentage of I-sections whose capacities are predicted
conservatively are 50%. As a result, this method is more general and accu-
rate than the method proposed by Lagerqvist and Johansson [7],
although the number of I-sections whose capacities are predicted
unconservatively is greater. Although the minimum ratio between com-
puted and predicted capacities based on the proposed method is
contained to 0.86 for all cases considered, for design purposes the method
proposed by Lagerqvist and Johansson [7] appears to be more appropriate
for cases that feature opposite patch loading away from member ends.
The second proposed method provides an average ratio between
computed and predicted capacities equal to 1.21 and a COV equal to
0.10 for opposite patch loading applied away from member ends. The
minimum and maximum ratios between computed and predicted ca-
pacities are 1.04 and 1.39, respectively, and the capacity of all consid-
ered I-sections is predicted conservatively. For load cases that feature
I-sections with opposite patch loading at member ends, the average
ratio between computed and predicted capacities is 1.24, the COV is
0.11, the minimum and maximum ratios between computed and pre-
dicted capacities are 1.01 and 1.48, respectively, and again, the capacity
of all considered I-sections is predicted conservatively. In contrast to the
first proposed prediction method, the second method is slightly less ac-
curate although it is more appropriate for design since all predictions
yield capacities that are higher than the computed ones.
A graphical comparison of all considered prediction methods is pro-
vided in Fig. 10b. The differences between the two proposed methods
and the one proposed by Lagerqvist and Johansson [7] is barely notice-
able since they all do a good job at predicting capacity. However, the un-
derestimation of capacity provided by the method in the AISC
Specifications is evident because most of the points are below the per-
fect fit line.
In Fig. 10c the difference between the two proposed prediction
methods is illustrated graphically in terms of a web compression buck-
ling curve expressed as the ratio between predicted capacity and squash
load and the slenderness ratio. It is clear that the curve generated based
on the AISC Specifications [9] global buckling curve leaves a balanced
number of points below and above it offering improved average accu-
racy, but leaving a considerable number of points on the unconservative
side. Although, as stated earlier the minimum ratio between computed
and predicted capacity is contained to 0.86. The second curve, which
represent the second proposed prediction method, does a very good
job at keeping all data points in the conservative range without being
too conservative (maximum ratio between computed and predicted ca-
pacity is no N1.48). The curve that corresponds exactly to a capacity
being equal to the critical elastic buckling load, Pcrl, is also shown and
denoted with 1/λ2. This curve serves as an upper bound for all cases in-
vestigated numerically. It should be noted that the slenderness ratio, λ,
in all numerical simulations is no N1.5.
Fig. 10d provides a similar illustration with that provided in Fig. 10c
and illustrates how the prediction curve proposed by Lagerqvist and
Johansson [7] is able to keep most data points on the safe zone while
not compromising accuracy. This comparison is provided as a separate
figure to enhance clarity and since this comparison includes only the
numerically obtained data points for opposite patch loading away
from member ends. The curve that corresponds exactly to a capacity
being equal to the critical elastic buckling load, Pcrl, again serves as an
upper bound for all numerically investigated cases.
Finally, Fig. 10e provides a comparison between the two proposed
prediction methods and that proposed by Lagerqvist and Johansson
[7] in terms of the ratio between computed and predicted capacity
and slenderness ratio. This figure illustrates one more time the differ-
ence between the three prediction methods in terms of accuracy and
appropriate safety for design. In summary, the first proposed method
offers improved accuracy and generality compared to the method pro-
posed by Lagerqvist and Johansson [7] and the second proposed
method. The second proposed method offers improved generality com-
pared to the method proposed by Lagerqvist and Johansson [7] and
more appropriate safety for design compared to the first proposed
method. Finally, the method proposed by Lagerqvist and Johansson [7]
offers the best balance between accuracy and appropriate safety for de-
sign for opposite patch loading away from member ends, but lacks the
generality that the other two methods offer in terms of opposite patch
loading location. In general, the presented statistics suggest that the
proposed prediction models as well as that proposed by Lagerqvist
and Johansson [7] predict well the elastic and inelastic behavior of I-
sections subject to opposite patch loading.
Table 9
Comparison of predicted and tested web buckling capacities (kN).
Ratio = Rn_tested/Rn_predicted
AISC Lagerqvist and Johannsson Proposed_1 Proposed_2
360–16 [7]
Average 1.30 1.28 1.32 1.24
Std. dev. 1.03 0.25 0.32 0.19
COV 0.79 0.19 0.25 0.15
Min. 0.38 0.95 0.87 0.94
Max. 5.96 2.20 2.85 1.78
% conserv.a 54 96 93 95
a
% predicted conservatively.
 F. Menkulasi, N. Farzana / Journal of Constructional Steel Research 162 (2019) 105728
8. Comparison of predicted and tested web buckling capacities
To complement the comparison between computed and predicted
capacities, an experimental database of 74 tests on opposite patch load-
ing was used to evaluate all considered prediction models. As stated
earlier, details about the geometry and mechanical properties of the I-
sections tested in opposite patch loading configuration can be found
in Lagerqvist [19] and Dos Santos et al. [18]. It should be noted that all
these tests featured opposite patch loading away from member ends.
Table 9 provides a summary of the comparison of tested and pre-
dicted capacities for the 74 I-sections considered in the experimental
database. This comparison is expressed in terms of the ratio between
tested over predicted capacities using various statistical indicators. The
average ratios between tested and predicted capacities vary from 1.24
to 1.32 and the COV vary from 0.15 to 0.79. The second proposed
method provides the best balance between accuracy and appropriate-
ness for design in addition to offering generality because it covers oppo-
site patch loading conditions away from and at member ends. The
average ratio between tested and predicted capacities and the corre-
sponding COV for the second proposed method are 1.24 and 0.15, re-
spectively. The corresponding statistics for the method proposed by
Lagerqvist and Johansson [7] are similar, 1.28 and 0.19, respectively.
The second proposed method and the method proposed by Lagerqvist
and Johansson [7] offer minimum and maximum ratios between tested
and predicted capacities that are equal to 0.94 and 1.78, and 0.95 and
2.20, respectively. The second proposed method predicts conservatively
the capacity of 95% of the tested I-sections and the method proposed by
Lagerqvsit and Johansson [7] predicts conservatively 96% of the tested
specimens. These statistics suggest that both of these methods provide
Fig. 11. a) Comparison of tested and predicted capacities, b), c), comparison of tested and predicted capacities normalized by the squash load as a function of slenderness ratio, d) ratio of
tested and predicted capacities as a function of slenderness ratio.
17
a good balance between accuracy and appropriateness for design with
the second proposed method offering also generality in terms of the po-
sition of the opposite patch loading.
The first proposed method ranks third because it offers an aver-
age ratio between tested and predicted capacities equal to 1.32 and
a COV of 0.25. In addition, the minimum and maximum ratios
between tested and predicted capacities are 0.87 and 2.85, respec-
tively, and the capacity of 93% of the specimens is predicted
conservatively.
The prediction method provided in section J10 of the AISC Specifica-
tions offers the least accurate prediction and is also the least appropriate
for design because it offers an average ratio between tested and pre-
dicted capacities equal to 1.30 while the corresponding coefficient of
variation is 0.79. Also, the minimum and maximum ratios of tested
and predicted capacities are 0.38 and 5.96, respectively, and the capacity
of only 54% of the specimens is predicted conservatively. This is in con-
trast to the statistics obtained for the numerically investigated cases for
which the AISC Specification prediction methods predicted conserva-
tively 100% of the investigated specimens and minimum and maximum
ratios between computed and predicted capacities varied from 1.03 to
3.80 for interior loading cases. It should be noted that there are some
differences in the characteristics of the specimens investigated numeri-
cally versus those investigated experimentally. For examples, the h/w
ratio for the specimens investigated numerically ranged from 1.0 to 20
with the average ratio being 5.77, whereas for those specimens investi-
gated experimentally it varied from 1.02 to 49.03 with the average ratio
being 10.40. Nonetheless, the results obtained from the experimental
dataset reveal that the prediction method provided in the AISC Specifi-
cations does not have the ability or versatility to capture with good
18 F. Menkulasi, N. Farzana / Journal of Constructional Steel Research 162 (2019) 105728
accuracy the web compression buckling capacity of I-sections subject to
a variety of loading conditions featuring various characteristics and var-
ious h/w ratios.
This relative comparison between all methods is also shown graph-
ically in Fig. 11a, which highlights once again the relative accuracy and
suitability for design of the proposed methods and that proposed by
Lagerqvist and Johansson [7], as well as the spread in prediction pro-
vided by the method in the AISC Specifications [9]. The two proposed
methods and the method proposed by Lagerqvist and Johansson [7]
were also evaluated by comparing the corresponding prediction curves
and the tested data in terms of the ratio between tested or predicted ca-
pacities divided by the squash load versus the slenderness ratio. As
stated earlier, to enhance clarity the curve for the prediction method
proposed by Lagerqvist and Johansson [7] is shown separately
(Fig. 11b) from the curves for the two proposed prediction methods
(Fig. 11c). Additionally, the curve that corresponds exactly to a capacity
being equal to the critical elastic buckling load is also shown and de-
noted with 1/λ2. This curve serves as un upper bound for all tests that
feature a slenderness ratio, λ, smaller than 1.5. The rest of the capacity
over squash load ratios for tests that feature slenderness ratios N1.5
are above the elastic buckling curve implying that in these tests the
postbuckling capacity of these I-sections was mobilized.
Fig. 11c shows how the curve generated based on the second predic-
tion method does a better job in aligning with tested data compared to
the curve that was generated based on the first prediction method in-
cluding cases that feature high slenderness ratios (N1.5). This is because
in prediction method 2, the function that relates capacity to squash load,
includes yield stress as a parameter even for sections that feature high
slenderness ratios thus accounting for the postbuckling capacity of the
section.
Finally, the strength of each prediction model was evaluated by plot-
ting the ratio of tested over predicted capacities as a function of the slen-
derness ratio. Fig. 11d shows the consistency, accuracy, and suitability
for design in prediction offered by the second proposed method and
the method proposed by Lagerqvist and Johansson [7]. The first pro-
posed method also offers good consistency, accuracy and appropriate-
ness for design but is slightly less consistent and accurate compared to
the other two methods due to the presence of some outlier tests.
9. Summary and conclusions
Two web compression buckling capacity prediction methods were
introduced and validated for steel wide flange sections considering op-
posite patch loading applied to the flanges. The methods are generally
posed as a function of loaded width to web depth ratio, and are applica-
ble for patch loading applied at the interior of a wide flange section or at
the end of it, where the web has a free edge. Web slenderness is defined
as the square root of the ratio of the web squash load to the web critical
elastic buckling load. The squash load is calculated considering a spread-
ing yield zone. The critical elastic buckling load is formulated consis-
tently with the approach used by Hopkins [4] for opposite patch
loading on infinitely long strips and considers the shortened web buck-
ling half-wavelength resulting from flange rotational restraint provided
to the web. The first proposed method adopts the global column buck-
ling curve provided in the AISC Specifications [9], while the second pro-
posed method adopts a resistance function developed empirically by
Lagerqvist and Johansson [7]. The proposed methods were validated
with existing experimental data and shell finite element collapse simu-
lations. Since the compiled experimental database included only
I-sections with opposite patch loading away from member ends, a com-
putational study was undertaken to systematically study the influence
of various loaded width to web height ratios away from member ends
as well as at member ends. The computational study was also used to
help develop a methodology for accurately predicting the elastic buck-
ling load of I-sections subject to opposite patch loading. The following
conclusions are drawn:
• The method provided in AISC Specifications [9] is not able to capture
the influence of the loaded width when estimating the web compres-
sion buckling capacity of I-sections subject to opposite patch loading
and therefore provides inaccurate and inconsistent capacity predic-
tions. Additionally, its formulation does not have the ability or versa-
tility to capture with good accuracy the web compression buckling
capacity of I-sections subject to a variety of loading conditions featur-
ing various characteristics and various h/w ratios.
• Both proposed methods were shown to be more accurate and more
widely applicable than current AISC Specification [9] method;
• While the method proposed by Lagerqvist and Johansson [7] provides
a rather good combination of accuracy and appropriateness for design
for opposite patch loading away from member ends, it lacks the gen-
erality provided by the two proposed methods in terms of addressing
opposite patch loading conditions at the ends of I-sections;
• The method proposed by Lagerqvist and Johansson [7] as well as the
two proposed methods offer rather accurate predictions of elastic
buckling loads despite the fact that the formulation for calculating
such load is different;
• The influence of the loaded width on the beam sections was more pro-
nounced than on the column sections, because the range of loaded
widths in the column sections was not as large as the range for the
beam sections;
• Section length and web boundary conditions did not have a significant
influence on the buckling capacity of the web provided that the sec-
tion length is at least three times the overall depth of the member;
• The slope of the strain-hardening branch had a negligible effect on the
buckling capacity of the web.
• Yield stress had a pronounced effect on the buckling capacity of the
web due to the fact that the computed peak load was obtained
when the sections were compressed past their linear range and is in-
cluded in both proposed prediction methods.
References
[1] A.Z. Sommerfield, Z. Math. Phys. 54 (1906).
[2] S.P. Timoshenko, Z. Math. Phys. 58 (1910) 357.
[3] D.M.A. Leggett, The effect of two isolated forces on the elastic stability of a flat rect-
angular plate, Proc. Camb. Philos. Soc. 33 (1937) 325–339.
[4] H.G. Hopkins, Elastic stability of infinite strips, VIIth International Congress of Ap-
plied Mechanics, London, UK, 1949.
[5] N. Yamaki, Buckling of a rectangular plate under locally distributed forces applied on
the two opposite edges, Reports of the Institute of High Speed Mechanics, Tohoku
University, vol. 34, 1954, pp. 55–71.
[6] M.Z. Khan, A.C. Walker, Buckling of plates subject to localized edge loading, Struct.
Eng. 50 (6) (1972) 225–232.
[7] O. Lagerqvist, B. Johansson, Resistance of I-girders to concentrated loads, J. Constr.
Steel Res. 39 (2) (1996) 87–119.
[8] G. Winter, R.H.J. Pian, Crushing strength of thin steel webs, Engineering Experiment
Station, Bulletin No. 35, Part 1, Cornell University, Ithaca, New York, 1946.
[9] AISC, Specification for Structural Steel Buildings, AISC, Chicago, IL, 2016.
[10] F.W. Chen, D.E. Newlin, Column web strength in steel beam-to-column connections,
ASCE Annual and National Environmental Engineering Meeting October 18–22, 1971 ,
St. Louis, Missouri.
[11] W.F. Chen, D.E. Newlin, Column web strength in beam-to-column connection, J.
Struct. Div. 99 (9) (1973) 1978–1984.
[12] W.F. Chen, I.J. Oppenheim, Web buckling strength of beam-to-column connections,
Fritz Laboratory Reports, 1970 , Paper 299, September 1970 (74-1) PB 234 621/AS.
[13] W.F. Chen, I.J. Oppenheim, Web buckling strength of beam-to-column connections,
J. Struct. Div. 100 (1) (1974) 279–285.
[14] J.D. Graham, A.N. Sherbourne, R.N. Khabbaz, C.D. Jensen, Welded interior beam-to-
column connections, WRC Bull. (63) (1960).
[15] P. Zoetemeijer, The Influence of Normal-, Bending- and Shear Stresses on the Ulti-
mate Compression Force Exerted Laterally to European Rolled Sections, Facultiet
der Civiele Techniek, Technische Universitiet Delft, 1980 Report 6-80-5, Delft.
[16] J.M. Aribert, A. Lachal, M. Moheissen, Investigation of the Local Compressive
Strength of the Web of Metallic H or I Sections. Final Report (Draft), Laboratoire
Structures, I.N.S.A, Rennes, France, 1989 July.
[17] J.M. Aribert, A. Lachal, M. Moheissen, Interaction du voilement et de la resistance
plastique de l'ame d'un profile lamine soumis a une double compression locale (nu-
ance d'acier allant jusqu'a FeE 460), Constr. Metallique 27 (1990) 3–23.
[18] G.B. Dos Santos, L. Gardner, M. Kucukler, Experimental and numerical study of stain-
less steel I-sections under concentrated internal one-flange and internal two-flange
loading, Eng. Struct. 175 (2018) 355–370.
 F. Menkulasi, N. Farzana / Journal of Constructional Steel Research 162 (2019) 105728 19

[19] O. Lagerqvist, Patch loading–resistance of steel girders subjected to concentrated
forces, Doctoral thesis, Luleå University of Technology, Div. of Steel Structures,
1994 , 159D, Luleå, 1995.
[20] Abaqus User's Manual Version 6.14–2. , Dassault Systemes Simulia Corp, 2017.
[21] D.C. Moen, Direct strength design of cold-formed steel members with perforations,
PhD Dissertation, John Hopkins University, Baltimore, MD, 2008.
[22] P. Arasaratnam, K.S. Sivakumaran, M.J. Tait, True Stress-True Strain Models for Struc-
tural Steel Elements, International Scholarly Research Notices, 2011.
[23] M.R. Soltani, A. Bouchaïr, M. Mimoune, Nonlinear FE analysis of the ultimate behav-
ior of steel castellated beams, J. Constr. Steel Res. 70 (2012) 101–114.
[24] R.D. Zieman, Guide to Stability Design Criteria for Metal Structures, John Wiley &
Sons, Inc, Hoboken, New Jersey, 2010.
[25] M.A. Crisfield, A fast incremental\iteration solution procedure that handles
snapthrough″, Comput. Struct. 13 (1981) 55–62.