Sidesway Web Buckling of Steel Beams

G. Y. GRONDIN and J. J. R. CHENG

ABSTRACT
The lateral stability of the tension flange of doubly symmetric
beams with the compression flange braced against lateral
movement was investigated experimentally and numerically.
Steel beams of W360x33 (W14x22) and W360x39
(W14x26) cross-section were tested to failure. Lateral supports were provided to the compression flange to prevent
lateral-torsional buckling while the tension flange was free to
move laterally over its entire length. Failure took the form of
sidesway web buckling. A numerical model incorporating the
effect of residual stresses, initial imperfections, large displacements and material yielding was implemented using the
finite element method. The model was validated by comparison with the test results. The current AISC design specification for sidesway web buckling was evaluated by comparing
predicted capacities with the test results and the results of a
limited parametric study performed with the numerical
model. It was found that the sidesway web buckling capacity
predicted using the AISC model is very conservative. Less
conservative, but safe, prediction equations are proposed as
possible alternatives to the current AISC equation.
Keywords: Steel beam, buckling, tension flange, inelastic,
residual stresses, initial imperfections
INTRODUCTION
The first observation of instability of the tension flange of a
doubly symmetric beam section loaded in bending was made
in the early 1970s (Costley, 1970; Bansal, 1971). During test
programs designed to investigate the lateral stability of continuous beams, unexpected failure by tension flange movement accompanied by a sudden decrease in load carrying
capacity was observed. In these early tests, the failure of the
beams by tension flange lateral movement was attributed to
a misalignment of the loads. A closer examination of the
problem later showed that the failure was not a result of
second order out-of-plane effects. It could be attributed to the
presence of a critical compressive stress field in the web,
below the load point (Summers and Yura, 1982). Expanding
on the model of local web buckling under a point load
G. Y. Grondln, Department of Civil and Environmental Engineering, University of Alberta, Edmonton, Alberta, Canada.
J. J. R. Cheng, Department of Civil and Environmental Engineering, University of Alberta, Edmonton, Alberta, Canada.

proposed by Basler, Yen, Mueller, and Thurlimann (1960),

Summers and Yura (1982) proposed that the compression
field in the web is analogous to a column and that the tension
flange acts as a lateral spring restraint. The proposed model
of a column supported at the top by a pin and at the base by
a lateral spring illustrated the potential for lateral instability
of a point loaded beam, braced against lateral compression
flange movement. This simple model was later adopted in the
AISC design specification (AISC, 1994) and is presented in
some detail in the following section.
This paper presents the results of an experimental and a
numerical investigation of sidesway web buckling. The experimental program and the numerical analysis were designed to assess the level of conservatism in the current AISC
design standard. Modified versions of the AISC design equation are developed based on the results of the finite element
analysis. These simple equations are then evaluated by a
comparison with the results of a parametric study.

BACKGROUND
For a beam where the compression flange is not restrained
against rotation, a simple model of a column pin supported at
the top and at the bottom, and restrained laterally at the base
with a spring is assumed. This model is illustrated in Figure
1. It is also assumed that the vertical load on the web is linearly
distributed. This is consistent with Basler et al.'s (1960)
assumption of a triangular vertical stress distribution in the
web below a point load. Replacing this triangular load distribution by a constant load of 50 percent of the maximum load,
the critical load, Pcn can be expressed as:
Pcr = 2kbh

(1)

where kh is the lateral stiffness of the tension flange, and h is

the height of the web. The lateral stiffness kh of the tension
flange can be expressed as:

where E is the modulus of elasticity, /yis the moment of inertia

of the tension flange about its strong axis, and / is the distance
between the points of lateral support on the tension flange.
The constant C reflects the type of restraint at the points of
lateral support. For simple lateral supports, C takes a value
of 48 and for fixed lateral supports, the value of C is
192. Assuming an intermediate value of 80, using

ENGINEERING JOURNAL / FOURTH QUARTER /1999 169

E = 200,000 MPa (29,000 ksi), and substituting Equation (2)

into (1), the following expression for Pcr is obtained,
Pcr = 2,670 tfh

(3)

I
\ J

where fyis the tension flange thickness, &yis the flange width
and h and / are as defined above. The unit of force in Equation
3 is kilonewton (kN) and the unit of length is millimetre (mm).
For the force expressed in kips and the length in inches, the
constant term in Equation (3) becomes 387,000. Multiplying
both sides of Equation (3) by 0.4 (h tw)3 and solving for Pcr,
the equation can be rewritten in the following form:
Pcr = 6,670 -

0.4

fh/tw
l/bf

(4)

For units of force in kips and length in inches, the constant

in Equation (4) becomes 967,000, which is almost identical
to Equation Kl-7 presented in the AISC standard with the
exception of the constant term which has been taken as
960,000 in the standard. In order to account for inelastic
effects, the critical load predicted by Equation (4) is reduced
by 50 percent if the moment at the point of load application
exceeds the yield moment capacity of the section (AISC,
1994). In addition, the magnitude of the critical load given by
Equation (4) cannot exceed the load that would cause buckling of the web as a column. For the case where the flanges
do not offer rotational restraint to the web, assuming that a
portion of the web h wide is effective, the buckling load, PE,
of the web would be:
2

n EL
PF = 2-

(5)

Braced
Top Flange

y<wwv|
Vertical Load
Distribution

Unbraced
Bottom Flange

Fig. 1. Column model used by Summers and Yura (1982).

170 ENGINEERING JOURNAL/ FOURTH QUARTER /1999

Table 1. Test Specimens Dimensions

Test

Span
Length
(mm)

Flange
Flange
Height
Web
Width Thickness of Web Thickness
(mm)
(mm)
(mm)
(mm)

6,261

127

8.18

333

5.67

7,482

127

10.57

332

6.44

6,261

128

10.57

332

6.44

5,039

128

10.30

332

6.57

where the constant 2 is used to account for the assumed

linearly varying axial load applied on the web, Iw is the
moment of inertia of the effective section of the web given as
(h tw3)/12, and Le is the effective height of the web taken as h
for the case where the bottom and top flanges do not offer any
rotational restraint to the web.
The condition for sidesway web buckling can therefore be
obtained from Equation (4) and Equation (5) as:
h/tw
(h V3
<0.5
l/bf

(6)

V J

If it is assumed that h/tf = 40, an assumption that is generally true for economy beams, Equation (6) reverts to AISC's
condition for sidesway web buckling:
h/tw
<1.7
l/bf

(7)

The following presents an assessment of the above model

using test results obtained at the University of Alberta (Mullin
and Cheng, 1994) and a finite element model validated with
the test results. The finite element model was developed to
expand the database of test results so that the above model
could be tested for a wider range of geometric properties.
TEST PROGRAM
A full-scale test program was designed to study the lateral
tension flange movement behaviour of rolled steel I-beams
(Mullin and Cheng, 1994). Four full-scale steel beam specimens were tested using two different beam sizes and three
different spans. The cross-sections used for the test specimens
were W360x33 (W14x22) and W360x39 (W14x26). The
W360x33 beam specimens were tested for span lengths of
6.260 mm and 5,040 mm, whereas the W360x39 beam specimens were tested for span lengths of 7,480 mm and
6.261 mm. The material properties were characterized using
six tension coupons from each section. The average yield
strength and average tensile strength of the W360x33 section
were found to be 375 MPa and 538 MPa, respectively. The
average yield and tensile strengths of the W360x39 were
355 MPa and 555 MPa, respectively.

Test Setup and Instrumentation

In order to create a structural system particularly susceptible
to tension flange movement buckling, an end condition that
provided some negative moment resistance in the plane of the
beam while preventing the development of warping stresses
in the ends of the bottom flange was constructed. The introduction of negative moments at the beam ends has the effect
of reducing the bottom flange lateral stiffness by the introduction of compression stresses over part of the bottom flange.
A sketch of the overall test setup showing the location of
the load and lateral supports is presented in Figure 2. Figure
3 shows a cross-section of the end supports. The top flange
attachment was designed to resist only flexural forces. The
bottom flange assembly was designed to resist both the flange
flexural force and the shear force at the support. The bottom
flange support assembly allowed rotation about the axis of the
connection (the vertical axis), thus eliminating warping
stresses in the bottom flange. The vertical shear force was
carried through a thrust bearing to the reaction column while
the bottom flange flexural force was transmitted to the reaction column through a tapered roller bearing. Stiffeners were
placed between the top and bottom flange restraint assemblies
to prevent buckling of the beam web at the supports.

A point load was applied at midspan through a sliding

frame that provided a guided vertical movement while restraining the loaded (top) flange laterally (see Figure 4). The
load was applied through rockers and rollers to allow rotation
of the flange about the beam web. Displacement along the
longitudinal axis of the beam was prevented at midspan. The
lateral supports at quarter points allowed free displacements
along the longitudinal axis and the strong axis of the member.
Each brace point restrained the lateral displacement of the top
flange and rotation of the beam about the vertical axis.
Instrumentation of the test specimens consisted of strain
gauging five sections along the span, displacement measurement of the bottom flange along the span length, and measurement of the out-of-plane deformation of the web near the
applied load. The location of the electrical resistance strain
gauges and out-of-plane displacement measurements are outlined in Figure 5. Three strain gauge rosettes were placed on
the web at midspan to determine the strain distribution in the
web under the concentrated load. The strain measurements at
FHAA&t M 0 RCHMT
OmwFrwr*

+-M

imwFfflira
BflM#lQ

ttgia Rtefingt

Restraint Frances
End Support
Assembly

Loading
Frame

Test
Specfrrten

E3

LCMMTQ Rod* I

^pa&a^***-^::*:*:|^

r F=

M^

WSkSSH

Fig. 4. Loading frame.

HydrewftcJack

Fig. 2. Schematic of full-scale test setup.

* i&iirtKMfte
- strain QeuQt*

Top Moment Rate

..
_.
- - X
/
Wg>
Stiffener

Test Specimen (
1
\
._.
- 4

815

ms

915

mm

btMuit Uttncjfe duflwiflun

Ball Thrust Bearing

l ...
610

610

610

810

mm

Tapered Roller
Bearing
Fig. 3. Section through end support assembly.

Fig. 5. Test specimen instrumentation.

ENGINEERING JOURNAL / FOURTH QUARTER /1999 171

Table 2. Summary of Test Results.

Test

Section

Span (mm)

a*

Test Peak Load

(kN)

W360x33

6,261

0.21

W360x39

7,482

W360x39

W360x39

Predicted Peak Load (kN)

(Test/Predicted)
AISC

FEM

112

61
(1.84)

125
(0.90)

0.44

135

46
(2.93)

128
(1.05)

6,261

0.08

137

80
(1.71)

143
(0.96)

5,039

0.11

170

150
(1.13)

184
(0.92)

* a: Ratio of end moment to simply supported midspan moment

the other sections along the span were used to determine the
points of inflection, which, in turn, were used to evaluate the
degree of fixity of the end supports. Each instrumented section, away from the section at midspan, had five strain gauges
(one at mid-height and one at each flange tip). Out-of-plane
deflections of the web under the load were measured at 20
locations as indicated in Figure 5(b). Bottom flange lateral
deflections were measured at five locations along the beam.
FINITE ELEMENT ANALYSIS
The test specimens were modeled and analyzed using the
commercial finite element code ABAQUS. The geometry of
the wide flange beams was modeled with 1024 plate bending
S4R elements (see Figure 6). The S4R element is a four node,
doubly curved, shell element that allows for changes in the
thickness as well as finite membrane strains. The model
involved large displacement using a Total Lagrangian formulation. The plate material behavior was modeled by an elastic-plastic-hardening material model. Von Mises yield criterion and a kinematic strain-hardening constitutive model
were implemented.

In order to model the full range of behaviour of the beam,

including the pre-buckling and the post-buckling regimes, the
solution strategy started with a load control standard NewtonRaphson iterative procedure in the initial stage of loading,
then shifted to a modified Riks procedure as the peak load
was approached. The Riks procedure, also referred to as the
arc length method, permits tracing the behaviour of the softening post-buckling regime.
Spring elements were used at the points of lateral restraint
to simulate the flexibility of the tension rods and restraint
frame assemblies were used to provide the lateral restraint in
the test specimens. The end support assemblies were more
difficult to model since their exact stiffness was difficult to
evaluate. Referring to the end assembly shown in Figure 3, it
is expected that some rotational and lateral restraints are
provided by the flange connection. The lateral stiffness of the
flange connection plates was evaluated based on the actual
plate dimensions. They were modeled with springs placed at
the top and bottom flanges in the lateral direction (direction
2 in the model shown in Figure 6). The rotational restraint
contribution from the various components of the end connections was modeled with a single rotational spring placed at
the centroid of the beam section at both supports. The stiffness
of the rotational spring was calculated based on the location
of the inflection point measured during testing. The stiffness
of the end rotational spring was evaluated using the following
expression:
k=

2EI

a
1-a

(8)

where E is the modulus of elasticity, / is the moment of inertia

of the beam section about the strong axis, L is the span length
of the beam, and a is the ratio of the end moment to the simple
span maximum moment, which is a function of the measured
Fig. 6. Finite element model of steel beam.

172 ENGINEERING JOURNAL / FOURTH QUARTER /1999

location of the point of inflection. The factor a obtained for

each test specimen is presented in Table 2.
Residual Stresses
The effect of residual stresses was also included in the finite
element model. The longitudinal residual stresses arising
from the differential cooling of the cross-section during the
manufacturing process were incorporated directly into the
model. Since the residual stresses were not measured in the
test specimens, the residual stress pattern shown in Figure 7
was assumed with a maximum residual stress of 30 percent
of the yield strength of the material. The residual stresses were
introduced in the model by imposing initial strains in the form
of a temperature distribution. In order to avoid introducing
transverse residual stresses in the model, an orthotropic temperature material property was used that had zero thermal
expansion coefficients in directions 2 and 3, the directions
perpendicular to the axis of the member. The initial strains
introduce initial stresses, upon which iteration is carried out
to establish equilibrium. The first load step in all of the
analyses consisted of the application of the residual stresses.
The following load step consisted of applying the point load
at midspan to obtain the load versus deformation response of
the beam both in the pre- and post-buckling regimes.

+0.3 0 y

-0.3 o

-0.3 Oy

Initial Imperfections
The initial imperfections in the test specimens were not
measured. To account for the presence of initial imperfections
in the analysis, initial imperfections in the form of lateral
bending of the top flange and the bottom flange were introduced in the model. A sinusoidal wave was adopted for the
shape of the initial imperfections with a maximum magnitude
at midspan of 0.1 percent of the span length. The sinusoidal
wave in the top flange was in the opposite direction to the one
in the bottom flange. The resulting distorted shape of the
beam cross-section at midspan is shown in Figure 8.
A residual stress free mesh, with the distortions described
above, was first generated. The residual stresses were then
applied to the beam. At the end of the first load step the
deformed shape of the beam consisted of the initial imperfections described in this section plus the superimposed deformations created by the application of the residual stresses. In
the second load step of the analysis the beam was loaded to
failure with a point load applied at midspan.
TEST RESULTS AND
COMPARISON WITH FEA RESULTS
Curves of the load versus the measured out-of-plane deflection of the tension flange at midspan are presented in Figure
9 for each of the four test specimens. The initial portion of the
load versus lateral displacement plots is nearly linear. As the
load approached 75 to 80 percent of the peak load, out-ofplane deflections increased at a greater rate and the out-ofplane and in-plane deflection response became distinctly
non-linear. Test specimen 1 experienced a sudden lateral
deflection of the tension flange of about 25 mm accompanied
by a decrease of load capacity of about 10 percent of the peak
load. Test specimens 2, 3, and 4 displayed stable behavior
with a gradual increase in lateral deflections accompanied by
a slow drop of vertical load. Since tests 2, 3, and 4 were
conducted with the same beam, the tests were stopped shortly

+0.3 oy

Fig. 7. Residual stress pattern used in the finite element model

8b = 0.001 L

5 0 - 0.0011

Fig. 8. Beam distortion at midspan due to initial imperfections.

1D

15

2D

25

Lateral Displacement of Tension flange at Mdspan(mrn)

Fig. 9. Beam test results.

ENGINEERING JOURNAL/ FOURTH QUARTER /1999 173

after a drop of vertical load was observed so that damage to

the beam section would be avoided.
Strain gauges spaced at 916 mm along the length of the
beams were used to determine the location of the points of
inflection on the beam, from which end moments could be
evaluated. The ratio of end moment to maximum simple span
moment obtained for each of the test specimens is presented
in Table 2. A value of 0.5 would indicate a fully rigid moment
connection while a simple support would have a value of 0.0.
As can be seen, the degree of fixity of the end connections
varied significantly between the tests.
The finite element model described above was used to
predict the test results. Figure 10 shows a comparison between the loads versus lateral displacement response obtained
from the tests and the finite element analysis. A comparison
of the test and predicted peak load level is presented in Table
2. With the exception of test specimen 2, Figure 10 indicates
that, at the initial stage of loading, the slope of the load versus
deflection curves measured for all of the specimens is lower
than predicted by the finite element model. Also, with the
exception of test specimen 2, the predicted load carrying
capacity was higher than the measured capacity. However, the
predicted capacity of the test specimens was all predicted to
within 10 percent of the measured peak load. A study of the
effect of initial imperfections and residual stresses, two variables that were not specifically measured in the test specimens, indicated that the difference between the measured and
predicted behavior can be associated with the effect of initial
imperfections and residual stresses. Considering the variability of initial imperfections and residual stresses (Tall and
Alpsten, 1969) that can be expected in rolled members, the
finite element model presented in the previous section is
believed to be adequate to predict reliably the sidesway web
buckling behavior of steel beams.
The sidesway web buckling capacity of the test specimens
was also predicted using AISC's model described in Equation
(4). Table 2 presents a comparison between the experimental
200

results and the predicted capacities. AISC's model is found to

underestimate significantly the tension flange buckling capacity with values of test to predicted ratios varying from 1.13
to 2.93. This is despite the fact that the test specimens were
subjected to end moments that are not accounted for in AISC's
equation. It is expected that the test to predicted ratios would
be even greater if the beam specimens were tested without
end moments. It is therefore apparent that there is room for
improvement in the existing model.
MODIFIED AISC MODEL
The finite element model was used to expand the database of
test results from which modifications of the AISC model were
derived. The distribution of vertical stresses in the web below
the point load is shown in Figure 11. This stress distribution
is typical for all four specimens from the test program. Figure
12 shows a plot of the normalized vertical stresses in the web
along a section directly below the point of application of the
load when the stresses were still elastic. It is evident from the
figure that the linear stress assumption can be improved to
obtain a better representation of the actual stress distribution
in the web. In an attempt to improve upon the linear stress
distribution assumed by Summers and Yura (1982), various
non-linear load distributions were investigated, namely, a
I P*2ftfcN

/ '
/ '
/

| 1
J
j

If
1

fl

0*

"in

f*5
.fa1

N4r%;

\ 10

/ 1

>\

>\
>

ft

11
J

&

s fcV

Fig. 11. Elastic vertical stress distribution under a point load.

350

180
160
140

i
1

..<%**^ X
^ >"
X

ys
I.

60

40

JJr
m

,5

10

FEA,Beem1
Experimental, Beam 1
FEA, town 2
Experimental, Beam 2
FEA,Beam3
Experimental, Beam 3
FEA,Beun4
Experimental, Beam 4

\X

<

20

25

Z^
/ <*
AV

30

Bo(lottRijfliliiilP!tjriatfff>wt(nHn)

Fig. 10. Comparison of FEA predictions with the test results.

174 ENGINEERING JOURNAL / FOURTH QUARTER /1999

A <'

\x

0,0
15

1
1
t

1 100

0,2

XX

* FEA
* Linear
--

04

Cubic
ExponwTfteJ

0,8

14)

NOffMffiXMl Wfcfe S M l f t

Fig. 12. Vertical force distributions in beam web

under a concentrated load.

1J2

Table 3. Prediction of Test Results Using Equation (9).

Test Specimen

Test Peak Load

(kN)

0.21

Predicted Peak Load (kN)

(Test/Predicted)
Linear

Quadratic

Cubic

Exponential

112

70
(1.6)

105
(1.1)

111
(1.0)

121
(0.93)

0.44

135

63
(2.1)

95
(1.4)

99
(1.4)

109
(1.2)

0.08

137

101
(1.4)

151
(0.91)

159
(0.86)

173
(0.79)

0.11

170

173
(0.98)

259
(0.66)

272
(0.62)

298
(0-57)

quadratic stress distribution, a cubic stress distribution and an

exponential stress distribution (see Figure 12). Using these
non-linear stress distributions and a web model similar to that
proposed by Summers and Yura (1982), simple web buckling
expressions can be obtained.
In addition to the change made to the vertical load distribution, modifications were also made to the web model.
Rotational springs were added at the top and bottom of the
rigid bar to simulate the rotational restraint provided to the
web by the flanges. A translation spring was also added to the
top flange to model the flexible lateral restraint provided in
the test specimens. The critical load, Pcr, was calculated for
each web load distribution discussed above using the principle of stationary potential energy. Appendix A presents a
detailed derivation of the model for a quadratic web load
distribution. The critical load can be expressed as:
Prr = Akhh 1
Kf ~r Itjj

Kh2

(9)

where kb is the lateral stiffness of the bottom flange and can

be determined using Equation (2), kt is the lateral stiffness
provided at the top of the beam and reflects both the stiffness
of the flange and the stiffness of the lateral support provided
to the top flange. Since the top flange is in compression, thus
reducing the lateral bending stiffness of the top flange, the
stiffness kt can be taken as the stiffness of the lateral brace.
The terms kx and k2 are the stiffness of rotational springs at
the top and the bottom of the web, respectively. The constant
A takes a value of 2.0 for an assumed linear force distribution
along the web height, 3.0 for a quadratic load distribution,
3.15 for a cubic load distribution, and 3.45 for an exponential
load distribution.
If one assumes that the rotational restraint offered by the
flanges to the web is negligible, that the stiffness of the lateral
brace on the compression flange is very large, and that the
stiffness of the bottom spring, kb9 can be approximated by

Equation (2), then Equation (9) reverts to Equation (4) for a

linearly varying axial load on the rigid bar.
To account for the effect of yielding on the buckling capacity
of the tension flange, it is assumed that the sidesway web
buckling capacity is reduced by 50 percent if the moment at the
point of loading exceeds the yield moment. This is consistent
with the assumption made in AISC's design procedure.
ASSESSMENT OF THE PREDICTION MODELS
Equation (9) was used to predict the test results presented
previously. The bottom flange lateral stiffness, kb, was obtained from Equation (2) and the lateral stiffness at the top
flange, kt, was taken as 45 kN/mm. This is representative of
the stiffness of the lateral bracing system used in the test
program. The stiffness terms kx and k2 were calculated assuming that the flanges are partially restrained from twisting at
the end supports and free to twist along the length of the beam.
The torsional stiffness of the top and bottom flanges, assuming that the ends are fully restrained, is given by:
,b tl
kh2 = 0.5128 'ff f

(10)

where bf and tf are the flange width and thickness, respectively. In order to account for some flexibility in the end
connections the stiffness predicted by Equation (10) was
reduced by 30 percent.
Table 3 presents a comparison between the test results and
the peak loads predicted using Equation (9) for a linear,
quadratic, cubic, and exponential web load distribution. Except for the linear load model, all of the models overestimate
the capacity of some of the test specimens. However, it should
be recalled that the test specimens were partially fixed at their
ends to decrease the sidesway web buckling capacity. The
simple model given in Equation (9) does not account for end
moments. Consequently, it is expected that Equation (9)
would have the tendency to overestimate the capacity of
beams that are subjected to negative end moments. As was

ENGINEERING JOURNAL/ FOURTH QUARTER/1999 175

the case for the non-linear web load models, the linear web
load model was not developed to account for beams with
negative end moments. It is therefore possible that the linear
web load model will overestimate the capacity of some beams
with fixed end boundary conditions.
A limited parametric study, using a modification of the
above finite element model, was conducted to test the ability
of the above simple models to predict the sidesway web
buckling capacity of beams. The modified model used for the
parametric study consisted of a beam with simple end supports and perfectly rigid lateral supports at the top flange that
prevented lateral displacement but allowed free rotation
about the web-to-flange junction. The yield strength of the
material was taken as 300 MPa. Initial imperfections, as
described above, were used for this investigation. Table 4
presents the results of this parametric study where the beam
length, L, the web height, h, and thickness, tw, and the flange
width, bf, and thickness, tfi were varied. All the cases presented in Table 4 satisfy the sidesway web buckling requirement of Equation (4). Assuming that the flanges offer negligible rotational restraint to the web (i.e. kl = k2 = 0), Equation
(9) can be expressed in a form similar to Equation (4) for
various vertical load distributions. The critical load for the
quadratic web force distribution can be obtained from:
1 0 , 0 0 0 ^ 0.4

'h/t ^
L/bf
V

(10)

however, leads to a decrease in capacity. In general, the

simplified models also indicate this type of behavior.
Some of the beams used in the parametric study were
proportioned so that buckling of the tension flange would
occur only after significant yielding of the beam. As indicated
in Table 4, the predicted buckling capacity obtained from the
simplified models was decreased by 50 percent to account for
yielding. A comparison of the buckling capacity obtained
using the simple equations with the finite element results
indicates that Equation (4), the most conservative of the four
equations investigated, underestimates significantly the sidesway web buckling capacity. The quadratic and cubic stress
models provide intermediate predicted capacities between
those predicted by the linear stress model presented in Equation (4) and the exponential stress distribution model presented in Equation (12). It can also be observed that although
Equation (12) is the least conservative in most cases, it
provides a safe estimate of the sidesway web buckling capacity. One exception, however, is beam R15 for which Equation
(12) over-estimates the capacity. In this case, however, the
capacity based on web crippling is lower than the capacity
based on sidesway web buckling (325 kN for web crippling
compared to 431 kN for sidesway web buckling). In this case
the predicted web crippling capacity of the web is a good
estimate of the capacity predicted from the finite element
model. Therefore, of all the simple equations investigated
here, Equation (12) gives the best prediction of the sidesway
web buckling capacity of simply supported beams.

The corresponding equation for a cubic distribution is:

Pcr= 10,500

_w ii 0.4 (h/tj\

(11)

L/bf

and the critical load predicted for an exponential stress distribution is given as:
Pcr= 11,500-

0.4

h/t
L/b

(12)

To be consistent with Equation (4) the units in Equations

(10) to (12) are kN and mm. For units of force and length
expressed in kips and inches, the constant term in Equations
(10), (11), and (12) is 1,451,000, 1,525,000, and 1,667,000,
respectively. The results of the numerical analysis and the
buckling capacity predicted using Equations (4), (10), (11)
and (12) are presented in Table 4. The finite element results
indicate clearly that a change in web thickness leads to a
significant change in buckling capacity. This, however, cannot be accounted for with the simplified models presented
here since the beam web is replaced by a rigid bar in these
models. The numerical analysis also indicates that the capacity of the beam increases with increasing web height, flange
width, and flange thickness. An increase in span length,

176 ENGINEERING JOURNAL / FOURTH QUARTER /1999

SUMMARY AND CONCLUSIONS

An experimental and analytical investigation of sidesway
web buckling of steel beams was conducted. Four full-size
steel beams were tested to obtain test results from which a
numerical model was validated. The numerical model incorporated the effect of residual stresses, initial imperfections,
inelastic material response, partial lateral restraint of the
compression flange, and the rotational restraint provided at
the end supports. The numerical model predicted the test
results with good accuracy. The finite element model was
used to determine the stress distribution in the beam web
under the applied load. This information was used to develop
modified versions of the Summers and Yura model to derive
simple prediction models. The simple models include the
effect of the top flange translational stiffness and both the top
flange and the bottom flange rotational stiffness. The validated numerical model was used to perform a limited parametric study to provide a database for comparison with the
prediction models, including the one adopted by AISC. Prediction models were derived for four web load distributions,
namely, a linear, quadratic, cubic, and exponential distribution. The effect of web thickness was not incorporated in these
simplified models. The parametric study, however, indicated
that the web thickness has a significant effect on the sidesway

Table 4. Comparison of Simplified Models with FEA Results.

Beam Dimensions

Sidesway Web Buckling Capacity (kN)

Simplified Model

L(mm)

h(mm)

bf (mm)

tf (mm)

t w (mm)

FEA

Eg. (4)

Eg. (10)

Eg. (11)

Eg. (12)

6,000

300

100

10

91

37

56

58

64

R2

6,000

300

100

10

10

117

37

56

58

64

R3

6,000

300

100

20

162

74

111

117

128

R4

6,000

300

100

20

10

191

74

111

117

128

R5

6,000

300

150

10

124

62*

94*

99*

108*

R6

6,000

300

150

10

10

150

62*

94*

99*

108*

R7

6,000

300

150

20

228

124*

188*

197*

216*

R8

6,000

300

150

20

10

259

124*

188*

197*

216*

R9

6,000

600

100

10

180

74

111

117

128

R10

6,000

600

100

10

10

315

74

111

117

128

R11

6,000

600

100

20

283

147

222

234

255

R12

6,000

600

100

20

10

459

147

222

234

255

R13

6,000

600

150

10

220

124*

188*

197*

216*

R14

6,000

600

150

10

10

387

248

188*

197*

216*

R15

6,000

600

150

20

336

248*

375*

394*

431*

R16

6,000

600

150

20

10

594

248*

375*

394*

431*

R17

8,000

300

100

10

68

16

23

25

27

R18

8,000

300

100

10

10

86

16

23

25

27

R19

8,000

300

100

20

122

31

47

49

54

R20

8,000

300

100

20

10

142

31

47

49

54

R21

8,000

300

150

10

93

52

40*

42*

45*

R22

8,000

300

150

10

10

112

52

79

83

45*

R23

8,000

300

150

20

172

105

79*

83*

91*

R24

8,000

300

150

20

10

194

105

79*

83*

91*

R25

8,000

600

100

10

148

31

47

49

54

R26

8,000

600

100

10

10

219

31

47

49

54

R27

8,000

600

100

20

242

62

94

99

108

R28

8,000

600

100

20

10

339

62

94

99

108

R29

8,000

600

150

10

188

105

158

166

91*

R30

8,000

600

150

10

10

289

105

158

166

182

R31

8,000

600

150

20

309

209

158*

166*

182*

R32

8,000

600

150

20

10

447

209

317

333

182*

Beam

* Capacity adjusted for the effect of yielding

ENGINEERING JOURNAL / FOURTH QUARTER /1999 177

web buckling capacity. Of the four models investigated, the

model with an exponential load distribution provides a better
description of the vertical stress distribution in the web under
the applied point load and leads to a better prediction of the
beam capacity. The linear load distribution used in AISC leads
to very conservative estimates of the beam capacity.
The present prediction equations have two shortcomings: 1)
they do not account for the web flexibility and, consequently,
cannot reflect the effect of the web thickness; and 2) the models
are strictly applicable to simply supported beams. However, the
parametric study performed using the finite element method
demonstrated that, even with a web thickness as small as 5 mm,
the simplified models gave conservative predictions of the buckling capacity. Therefore, although the simplified models are
unable to account for changes in web thickness, they will predict
the capacity of beams of various web thickness with some
conservatism. Beams that develop negative end moments are
more susceptible to sidesway web buckling since part of the
bottom flange is in compression. Use of the equations reviewed
in this report for beams with end restraint could lead to an
overestimation of their sidesway web buckling capacity. This
was demonstrated in a comparison with the test results. The
designer should therefore use the proposed equations with caution for these cases. Further investigation of beams with end
rotational restraint is required.
REFERENCES
AISC (1994), Load and Resistance Factor Design Specification for Structural Steel Buildings, American Institute of
Steel Construction, Inc., Chicago, II.
Bansal, J. P. (1971), The Lateral Instability of Continuous
Beams, AISI Project 157, Department of Civil Engineering, University of Texas, Austin, Texas, 126 pp.

List of Symbols
A
bf
C
E
h
/
If
k
kb
kt
kx
k2
/
L
Le
PE
Pcr
tf
tw
a

= constant used in Equation (9)

= width of the tension flange
= constant
- modulus of elasticity
- web height
= moment of inertia of the beam
= moment of inertia of tension flange
= stiffness of the rotational springs at the beam ends
= translational stiffness of the bottom flange
= translational stiffness of the top flange
= rotational stiffness of the top flange
= rotational stiffness of the bottom flange
= length of tension flange between points of lateral
support
= span length
= effective height of the web
= load that causes buckling of the web as a column
= sidesway web buckling load
= tension flange thickness
= web thickness
= ratio of the end moment to the simple span maximum
moment
Appendix A - Derivation of Buckling Load

The simplified two degrees of freedom model shown in the

Figure A-l consists of a rigid bar restrained at the top and
bottom by springs of stiffness kt and kb, respectively. At
buckling, the rigid bar adopts the configuration shown in
dashed lines in the figure. The two degrees of freedom are
taken as the translation at the top of the rigid element, 8 r , and
its rotation, 9.

Basler, K.,Yen, B. T , Mueller, J. A., and Thurlimann, B.

(1960), Web Buckling Tests on Welded Plate Girders, Bulletin No. 64, Welding Research Council, New York.
Costley, H. E. (1970), Lateral and Local Instability of Continuous Beams, ACE Project 157, Department of Civil
Engineering, University of Texas, Austin, Texas, 74 pp.
Mullin, D, and Cheng, J. J. R. (1994), The Effect of Tension
Flange Movement on the Strength of Point Loaded Beams,
Structural Engineering Report No. 208, Department of
Civil and Environmental Engineering, University of Alberta, Edmonton, Alberta.
Summers, P. A. and Yura, J. A. (1982), The Behavior of Beams
Subjected to Concentrated Loads, Phil M. Ferguson Structural Engineering Laboratory Report No. 82-5, University
of Texas, Austin, TX, August.
Tall, L. and Alpsten, G. A. (1969), On the Scatter in Yield
Strength and Residual Stress in Steel Members, IABSE
Symposium on Concepts of Safety of Structures and Methods of Design, London.

178 ENGINEERING JOURNAL / FOURTH QUARTER /1999

8t+9h
Figure A-l Modified buckling model.

The lateral displacement of the bar at buckling, y, is given

as
(A-l)

y = 8 , + 6(/i-Jc)

The internal strain energy stored in the top and bottom

translational and rotational springs can be expressed in terms
of the degrees of freedom as follows:
2

u=^[(kt+kb)S*

+ (kbh +kl + k2)e +

Using the principle of stationary potential energy, equilibrium is satisfied if:

= 0andr = 0
30
38,
In matrix form Equation A-5 can be expressed as
kh h2 + L+ k?

Ph

2kbh8teyA-2)

dx
o

-\i

fxQ
dx =
h

/z02
-P-r(A-3)
o

where the first derivative of y with respect to x was obtained

from differentiation of Equation (A-l). Equation A-3 was
obtained assuming a quadratic variation of the axial load, P,
over the height, h, of the rigid bar. From Equations (A-2) and
(A-3) the total potential energy is

/zG 2

kbhbtQ-P-zo

khh
= 0

khh
where the variables are as defined in Figure A-1. The potential
energy of the external load P is given as:

(A-5)

(A-6)

Kt + kh

The buckling condition can be satisfied by setting the

determinant of the coefficient matrix to zero, and solving for
the axial load magnitude P. This results in the following
expression for the buckling capacity of a rigid bar restrained
by translational and rotational springs at the top and bottom
and loaded axially with a load varying as a second order
polynomial:
Prr=3khh

I"

/Cj i K2

kh + k,

kb h

(A-l)

(A-4)

ENGINEERING JOURNAL/ FOURTH QUARTER /1999 179

180 ENGINEERING JOURNAL/ FOURTH QUARTER/1999