Unified Flexural Resistance Equations for Stability Design of

Steel I-Section Members: Uniform Bending Tests

Donald W. White, M.ASCE1; and Se-Kwon Jung, A.M.ASCE2

Abstract: The 2004 AASHTO and 2005 AISC provisions for flexural design of steel I-section members have been revised in their
entirety relative to previous specifications to simplify their logic, organization, and application, while also improving their accuracy and
Downloaded from ascelibrary.org by New York University on 05/16/15. Copyright ASCE. For personal use only; all rights reserved.

generality. This paper evaluates the lateral-torsional and flange local buckling 共LTB and FLB兲 predictions from these and previous
specifications versus uniform bending experimental test results. A total of 154 rolled and 123 welded I-section member LTB tests, and 11
rolled and 36 welded I-section member FLB tests are considered. Reliability indices are estimated for load and resistance factor design
共LRFD兲 of buildings based on the test statistics combined with established statistics for material and fabrication bias factors and the ASCE
7 load model. The notional reliability for LTB is found to be reasonably constant and consistent with the targeted level in the first AISC
LRFD specification of 1986. The unified equations, combined with a design-oriented procedure for calculation of elastic LTB K factors,
are shown to capture the test results accurately throughout the inelastic and elastic LTB ranges, leading to substantial liberalization of the
calculated resistances in certain cases. The mean resistances for inelastic LTB and FLB are captured accurately by a linear equation in the
corresponding slenderness parameters. The reliability for FLB is found to be slightly higher than that for LTB.
DOI: 10.1061/共ASCE兲0733-9445共2008兲134:9共1450兲
CE Database subject headings: Synthesis; Steel beams; High strength steel; Flexural strength; I beams; Structural stability; Experi-
mental data; Bending.

Introduction Overview of Unified Flexural Resistance Provisions

The calibration of the original AISC 共1986兲 load and resistance The unified flexural resistance equations in AASHTO 共2004兲 and
factor design 共LRFD兲 flexural resistance provisions was based on AISC 共2005兲 are closest in form and function to those of AISC
limited experimental testing of the many different types of 共1999兲. White 共2008b兲 provides a detailed technical overview of
I-section members 共Yura et al. 1978; Cooper et al. 1978兲. A num- the updated provisions. The new AISC and AASHTO provisions
ber of important additional experimental studies have been con- are largely equivalent, but with an emphasis on streamlined de-
ducted in subsequent years. As a part of the process of developing sign of slender-web I-section members in AASHTO 共2004兲 and
updated AASHTO 共2004兲 and AISC 共2005兲 provisions for flex- compact doubly symmetric I-section members in AISC 共2005兲.
ural design of steel I-section members, an effort was made to All the flange local buckling 共FLB兲 and lateral-torsional buckling
collect and analyze the results from a large set of relevant experi- 共LTB兲 equations in the new specifications are based consistently
mental tests. Also, updated material and fabrication bias factors on the two anchor points shown in Fig. 1. Anchor Point 1 is
have been published subsequent to the original LRFD effort. This located at the length KLb = L p for LTB, or the flange slenderness
paper presents an analysis of the data from statically determinate 关␭fc = bfc / 2tfc兴 = ␭pf for FLB, corresponding to development of the
uniform bending tests. The statistics obtained for the ratio of the maximum potential flexural resistance. This resistance is labeled
test to predicted strengths 共M test / M n兲 are utilized along with es- in Fig. 1 as M max 共in terms of the bending moment兲 or Fmax 共in
tablished statistics for material and fabrication bias factors and the terms of the compression flange flexural stress兲. The resistance
ASCE 7 load model to assess the reliability associated with the M max⫽plastic moment M p for I sections having a compact web.
updated provisions in the context of building design. The com- However, it is generally less than M p for other cross sections.
panion paper 共White and Kim 2008兲 presents the results for Anchor Point 2 is located at the largest compression flange stress
moment-gradient tests. or member bending moment for which the buckling behavior is
nominally elastic. The ordinate of Anchor Point 2 is taken 共in
1
Professor, School of Civil and Environmental Engineering, Georgia terms of moment兲 as RbFyrSxc = Rb共0.7Fy兲Sxc = Rb共0.7M yc兲 for
Institute of Technology, Atlanta, GA 30332-0355. most I shapes, where M yc⫽yield moment associated with the
2
Civil Structural Engineer, Bechtel Power Corp., Frederick, MD compression flange, FycSxc, and Rb⫽web bend buckling strength
21705; formerly, Graduate Research Assistant, School of Civil and Envi- reduction factor, equal to 1.0 for sections with a compact or non-
ronmental Engineering, Georgia Institute of Technology, Atlanta, GA. compact web, but equal to a value less than 1.0 for sections with
Note. Associate Editor: Scott A. Civjan. Discussion open until a slender web. The abscissa of Anchor Point 2 is denoted by the
February 1, 2009. Separate discussions must be submitted for individual
length Lr, or the flange slenderness ␭rf. The inelastic LTB and
papers. The manuscript for this paper was submitted for review and pos-
sible publication on August 24, 2005; approved on May 16, 2006. This FLB resistances are expressed simply as straight lines between
paper is part of the Journal of Structural Engineering, Vol. 134, No. 9, the anchor points. For KLb ⬎ Lr or ␭ f ⬎ ␭rf, the nominal resistance
September 1, 2008. ©ASCE, ISSN 0733-9445/2008/9-1450–1470/ is defined explicitly as the theoretical elastic buckling moment or
$25.00. flange stress. The format shown in Fig. 1, adapted from AISC

1450 / JOURNAL OF STRUCTURAL ENGINEERING © ASCE / SEPTEMBER 2008

J. Struct. Eng. 2008.134:1450-1470.

 many practical situations, the use of a LTB effective length
factor K ⬍ 1 in critical unbraced segments more than offsets
this decrease in the value of L p. For unbraced lengths with
substantial end-restraint, more accurate and more liberal esti-
mates are obtained when the effective length KLb is used,
where K is determined using a design-oriented procedure rec-
ommended by Nethercot and Trahair 共1976兲 and by Galambos
共1998兲.
• Web local buckling is no longer handled as a separate limit
state for noncompact-web I-section members. Rather,
noncompact-web effects are handled in a fashion similar to the
way that slender-web effects are addressed in Appendix G of
AISC 共1999兲. That is, coefficients similar to the web bend-
buckling strength reduction factor for slender-web members,
Downloaded from ascelibrary.org by New York University on 05/16/15. Copyright ASCE. For personal use only; all rights reserved.

Fig. 1. Basic form of unified FLB and LTB resistance equations R PG in AISC 共1999兲 and Rb in AASHTO 共1998兲, are used to
quantify the influence of a noncompact web on the M max val-
ues for FLB and LTB in Fig. 1. These coefficients are termed
共1999兲, greatly facilitates the definition of simple yet comprehen- the web plastification factors Rpc and Rpt, as they account for
sive flexural resistance equations. the development of web yielding necessary to achieve M max
There are a number of minor differences between AASHTO ⬎ M yc or M yt. The parameter Rpc applies to the FLB and LTB
共2004兲 and AISC 共2005兲. To facilitate the understanding of both limit states whereas Rpt applies to the limit state associated
specifications, White 共2008b兲 presents the base unified flexural with tension flange yielding 共TFY兲, which governs only for
resistance provisions and explains the minor deviations from highly monosymmetric sections with a larger flange in com-
these provisions in the places they occur. Correspondingly, the pression. For compact-web sections, Rpc and
base unified provisions are utilized in this paper as a point of Rpt⫽cross-section shape factors M p / M yc and M p / M yt.These
departure, followed by consideration of the effects of deviations improvements are based on observations, discussed in this
from these provisions. One exception is made relative to the uni- paper, that the shape of the FLB and LTB strength curves is
fied provisions in this paper. The unified provisions require the influenced by the web slenderness, not just the maximum re-
use of the slender-web member equations when Fyf ⬎ 485 MPa sistance M max. The new provisions define the governing resis-
共70 ksi兲. This restriction, which is implemented in AASHTO tance as the smaller of the values determined from the FLB,
共2004兲, is due to the more limited testing of girders with these LTB, and TFY calculations, similar to the approach in Appen-
steels, the implications of the use of these steels with conven- dix G of AISC 共1999兲 but applicable to compact-,
tional strength concrete slabs in composite bridge construction, noncompact-, and slender-web members.
and the larger potential reduction in the LTB resistance due to • The web compactness limit ␭pw has been updated for singly
web distortional flexibility for large Fyf. For the noncomposite symmetric I sections to provide an improved characterization
tests considered in this paper, this restriction is waived, generally of the demands on the web for development of the plastic
producing a more liberal and more accurate estimate of the test moment resistance M p. This is important particularly for mem-
strengths. In a few cases, the notation is different in AISC 共2005兲 bers with a smaller flange in compression, typical of compos-
and AASHTO 共2004兲. In these situations, the notation from ite bridge I girders in positive bending prior to their being
AASHTO 共2004兲 is utilized for purposes of discussion. The cor- made composite, or for negative moment regions of composite
responding AISC 共2005兲 terms are provided in the notation list members, where the depth of the web in compression may be
and are defined along with the AASHTO terms where the terms large due to the large effective tension flange.
first appear. • The elastic LTB resistance is based on a single equation appli-
Some of the key improvements in the unified provisions rela- cable for all types of I-section members. This equation gives
tive to AISC 共1999兲 and AASHTO 共1998兲 are as follows: the exact beam-theory solution for LTB of doubly symmetric
• The flange stress corresponding to the nominal onset of yield- members, and it gives an accurate to somewhat conservative
ing, Fyr 关FL in AISC 共1999, 2005兴, has been updated. The approximation for singly symmetric noncomposite members
equation for Fyr in the new provisions is essentially the same and composite members in negative bending 共White and Jung
for all types of I-section members. For most I shapes, Fyr 2003a,b; White 2008b兲. This equation can be written in terms
= 0.7Fyc. This represents a substantial enhancement in the FLB of the compression flange flexural stress as

冑
and LTB resistances for slender-web I girders having unbraced
lengths or flange slenderness values near the maximum limits ␲ 2E 0.078
Fcr = Cb 1+ 共KLb/rt兲2 共1兲
for inelastic buckling 共Anchor Point 2兲, justified by the results 共KLb/rt兲2 X2
presented in this paper as well as other studies. The previous
where
AISC 共1999兲 Appendix G and AASHTO 共1998兲 slender-web
member provisions assumed Fyr = 0.5Fyc at Anchor Point 2 Sxch
共Fig. 1兲. X2 = 共2兲
J
• A single equation is employed for the compact bracing limit L p
for all types of I-section members. This equation provides a rt⫽radius of gyration of the compression flange plus one-third
better fit to the experimental results summarized in this paper of the area of the web in compression; Sxc⫽elastic section
than prior specification equations for L p. The recommended L p modulus to the compression flange; h⫽distance between the
equation is generally more restrictive than the corresponding centroids of the flange elements; and J⫽St. Venant torsion
equations in AISC 共1999兲 and AASHTO 共1998兲. However, in constant of the steel I section. Eq. 共2兲 is a simple ratio of the

JOURNAL OF STRUCTURAL ENGINEERING © ASCE / SEPTEMBER 2008 / 1451

J. Struct. Eng. 2008.134:1450-1470.

 bending and torsional efficiencies of the cross section. For Interrogation of Test Data and Prediction of Test
doubly symmetric sections, Eq. 共2兲 is approximately equal to Flexural Resistances
2Ix / J.
For column-type I sections with d / bfc ⬵ 1 and compact web The flexural resistances are calculated in this paper using the
and/or flange elements, the second term under the radical in measured geometry and material properties for each component
Eq. 共1兲 tends to be significantly larger than 1. As such, it plate, where available, and using the specific design-oriented pro-
would be quite uneconomical to discount this major contribu- cedure detailed by Nethercot and Trahair 共1976兲 and Galambos
tion to the resistance to obtain a simpler form for Eq. 共1兲. 共1998兲 for determining the LTB effective length factors, K. White
However, in many situations involving beam- or girder-type I 共2008a兲 discusses the application of this approach to singly-
sections with D / bfc greater than about 2 and b f / 2t f near the symmetric I-section members and composite I-section members
compact-flange limit ␭pf or larger, the contribution from the in negative bending. For rolled members, separate flange and web
second term in Eq. 共1兲 is relatively small 共White and Jung yield strengths are utilized in this research, when this information
2003a兲. For slender-web members, the contribution from this is available, and also the web-to-flange fillet areas are included in
term is neglected altogether, due to the reduction in the effec-
Downloaded from ascelibrary.org by New York University on 05/16/15. Copyright ASCE. For personal use only; all rights reserved.

the calculations.
tive St. Venant torsional stiffness associated with web distor- These details typically affect the predicted strengths by only a
tional flexibility 共i.e., the deformation of the web into an S few percent, but prior research demonstrates that their inclusion
shape upon twisting of the cross section, and the correspond- gives the best correlation with physical test results, e.g., see Fu-
ing reduction in the twist rotation of the flanges兲 共White and kumoto et al. 共1980兲. For welded members, separate web and
Jung 2007兲. In this case, Eq. 共1兲 reduces to the form flange yield strengths are utilized whenever the data are available,
but web–flange fillets are not included in the calculation of the
cross-section properties as in common practice. The use of an
␲ 2E effective length factor K is essential for accurate prediction of test
Fcr = Cb 共3兲
共KLb/rt兲2 resistances in the inelastic and elastic LTB ranges of the response.
This paper evaluates the accuracy associated with the use of the
used traditionally by AISC and AASHTO for slender-web design-oriented procedure originally proposed by Nethercot and
members. Trahair 共1976兲. Using this procedure, a consistent set of K values
Eq. 共1兲 results in a more succinct expression for the noncom- is obtained for all of the considered tests. Somewhat different K
pact lateral brace spacing than in AISC 共1999兲, values are reported in a number of the experimental studies using
other methods of calculation.

冑 冑 冉 冊
A large number of test results have been reported in the litera-
1.95rt E Fyr 2 4 ture in which only the nominal section geometry is specified, the
Lr = 1+ 1 + 6.76 X 共4兲
X Fyr E geometry measurements are provided to a small number of sig-
nificant digits, and/or only a single yield strength is listed for the
applicable for all types of compact- and noncompact-web I entire cross section. In this research, cases in which the flange
sections, whereas Eq. 共3兲 gives dimensions and/or the web or section depth are specified as nomi-
nal values or reported to less than three significant digits are

冑
handled as a separate group, labeled as “tests with nominal or
E approximate measured geometry.” A single exception to this rule
Lr = ␲rt 共5兲
Fyr is made for deep girders in which the web depth is reported as an
integer value in inches, but based on the assumption that the
AISC 共1999兲 does not provide an Lr equation for singly actual dimensions are rounded to the reported integer values, the
symmetric I-section members in its Appendix F. Engineers actual dimensions differ from the reported values by less than 1%.
often assume that this noncompact bracing limit must be cal- Also, cases in which only the web thickness is nominal or ap-
culated iteratively. White and Jung 共2003b兲 give a closed-form proximate and tests where only a single yield strength is listed are
expression for Lr based on the rigorous application of open- still classified as “tests with accurate measured geometry.” This is
section thin-walled beam theory. However, Eq. 共4兲 is simpler because deviations in the web thickness and yield strength from
to evaluate and provides good accuracy for most practical sin- reported values tend to influence the predicted resistance to a
gly symmetric members. Also, due to the larger influence of minor degree compared to deviations in the flange geometry and
web distortion in singly symmetric members, the rigorous the web depth. For rolled members that do not have cross-section
beam theory equation does not necessarily give a better esti- areas reported including a contribution from the web-to-flange
mate of the physical buckling resistance 共White and Jung fillets, circular fillet areas are included in this work based on the
2007兲. difference between the nominal area listed in section property
White 共2008b兲 discusses many other improvements in the uni- tables and the area calculated using only the nominal plate dimen-
fied provisions. The reader is referred to this companion paper for sions. The average cross-section area including the web–flange
a complete overview of the developments. When specialized to fillets can be determined by weighing the member and dividing
apply only to compact doubly symmetric I-section members, the by the steel weight density and the member length. However, this
unified provisions are dramatically simplified. AISC 共2005兲 gives information is rarely reported. The flange yield strength is used
these simplified provisions explicitly at the beginning of its Chap- for the web–flange fillets in this work. Specific geometric dimen-
ter F. However, welded bridge I girders typically have a larger sions, material yield strengths, test configurations, and observa-
web slenderness, are not doubly symmetric, and are composite in tions about the test behavior are detailed by White and Jung
their final constructed condition. As such, AASHTO 共2004兲 em- 共2004兲 for all of the tests.
phasizes the simplified design of general slender-web I-section In concept, measured static yield strengths from tension cou-
members. pon tests are utilized for all the resistance calculations. Galambos

1452 / JOURNAL OF STRUCTURAL ENGINEERING © ASCE / SEPTEMBER 2008

J. Struct. Eng. 2008.134:1450-1470.

 共1998兲 in Appendix B8 specifies a recommended procedure for • The reported M p or M y deviate substantially from values cal-
determining these static yield strengths. This procedure involves a culated using the specified dimensions and yield strengths.
modification of the ASTM E8 共2000兲 method for tension testing. • The tests were stopped prior to achieving a significant reduc-
It is well recognized that the rate of loading in tension tests has a tion in the slope of the load-deflection curve associated with
significant impact on the measured yield properties. The proce- the limit load, typically such that the members could be modi-
dure recommended by Galambos 共1998兲 requires that the cross- fied and used in subsequent tests.
head motion be stopped for approximately 5 min at several • The predominant failure occurred in a component other than
intervals within the yield plateau. At these stages, the plot of the the test segment subjected to uniform bending.
load–strain curve drops from its original course, indicating a de- • Significant lateral movement of a brace point or failure of one
crease in load. The magnitude of the load is recorded at the end of or more of the braces occurred prior to reaching the limit load
this time, or when the load has stabilized. Five minutes is sug- of the test.
gested as a maximum time limit as, in the case of hydraulic ma- Also, a number of investigators have tested annealed beams in
chines, strain reversal and a lower equilibrium load may be their research programs, to facilitate assessment of the effect of
recorded due to leakage of oil in the system. Nevertheless, Adams residual stresses in other tests. No annealed beam tests are in-
Downloaded from ascelibrary.org by New York University on 05/16/15. Copyright ASCE. For personal use only; all rights reserved.

et al. 共1964兲 show test data in which the load does not stabilize cluded in the data collected in this research. Further, members
completely until a period of roughly 24 h has elapsed. They con- with longitudinally stiffened webs are not addressed in this work.
clude that the static yield strength problem is not one of deter- Lastly, tests in which the adjacent unbraced lengths are subjected
mining a difference in load, but of determining a value of the load to moment gradient and are more critical for LTB 共using calcu-
共stress兲 corresponding to the desired duration of loading on the lated KLb values and including Cb ⬎ 1 for these segments兲 are not
structure. Thus the wait time in determining the static yield stress, included here. These tests are classified as moment-gradient cases
as well as the duration at which the displacements are held con- and are addressed by White and Kim 共2008兲.
stant on the structure during static testing at inelastic load levels, White and Jung 共2004兲 list a number of tests in which the
are sources of variability in M test / M n. In some cases, the tension specimens were repaired from a failed 共i.e., damaged兲 condition
coupon and the flexural member tests were conducted at specifi- and then retested as well as other tests in which the specimens
cally reported slow rates that correspond to near static yield had D / b f ⬎ 7.5. The results from these tests are not considered in
strengths. These tests are still included in the collected data. The the statistical analyses of this research. Obviously, prior damage
tests conducted by Dibley 共1969, 1970兲 are examples of this sort. may influence the member resistance. Also, the predictions are
Another common source of error in the calculation of M test / M n highly variable for the limited number of tests with D / b f ⬎ 7.5:
is the determination of the steel elastic modulus. It is well known two tests have M test / M n = 0.77 and 0.82 whereas another two tests
that measured E values are sensitive to numerous attributes of test have M test / M n = 1.11 and 1.28. The low M test / M n values are con-
procedures 共ASTM 1997; Galambos and Ravindra 1978; Adams sistent with the results from a number of high-shear high-moment
et al. 1964兲. Baker and Kennedy 共1984兲 report a bias factor for tests with D / b f ⬎ 6 considered by White et al. 共2008兲. AASHTO
the elastic modulus, equal to the ratio of the mean from test data 共2004兲 restricts the design of I-section members to D / b f 艋 6.
to E = 200 GPa 共29,000 ksi兲, of ¯␳E = 1.02 with a coefficient of For some slender-web members, the predominant failure mode
variation VE = 0.012 based on a reanalysis of the data from Galam- involves vertical flange buckling 共VFB兲, i.e., a folding of the
bos and Ravindra 共1978兲. This is consistent with recent unpub- compression flange into the web. As reported in prior experimen-
lished data from extensive elastic modulus testing of bridge steels tal studies, VFB failures occur generally only after significant
by the Federal Highway Administration. Elastic moduli listed in yielding of the compression flange 共e.g., see Cooper 1967; Lew
experimental reports often are not based on procedures such as and Toprac 1968; Owen et al. 1970兲. Correspondingly, the reduc-
ASTM E111 共1997兲 necessary for accuracy of the measurements. tion in the resistance due to VFB is small. Therefore, the test data
For this reason, the elastic modulus is taken uniformly as the involving VFB are included in the statistical analyses of this
nominal value E = 200 GPa in all the calculations of this work. study.
Variations in the actual E from its nominal value are reflected
automatically in the M test / M n ratios and need not be considered
further. Representative Comparisons of Predictions to Test
For the slender-web members considered in this research, the Strengths: LTB Tests
web bend-buckling strength reduction factor Rb is calculated
based on the compression flange stress M n共Rb=1兲 / Sxc, where In this study, experiments in which the LTB equations govern the
M n共Rb=1兲⫽resistance obtained using Rb = 1. The use of this stress calculated resistance are categorized as LTB tests. The LTB test
rather than Fyc gives a more precise, more liberal value for Rb. category includes members with compact, noncompact, and slen-
This is similar to the calculation of R PG in AISC 共1999兲 and is der lateral brace spacing 共see Fig. 1兲. The accuracy of the predic-
specifically allowed in AASHTO 共2004兲. AISC 共2005兲 requires tions is evaluated for specific ranges of the LTB slenderness in the
the simpler use of just Fyc in its corresponding web strength re- subsequent developments. Members in which the calculated resis-
duction factor. The effect of M n共Rb=1兲 / Sxc versus Fyc in the Rb tance is governed by the FLB equations are categorized as FLB
equation is minor 共less than a few percent兲 in the majority of the tests.
tests, but is significant in some cases with extremely large 2Dc / tw Figs. 2 and 3 show comparisons of predicted resistances to
and in which M n共Rb=1兲 is significantly smaller than M yc. measured test strengths using the above-outlined procedures for
Of more than 300 uniform bending tests reviewed in this work, two sets of representative LTB tests. The test results are compared
a number of tests are discarded for the following reasons: to both the unified equations 共White 2008b兲 as well as the AISC
• Measured static yield strengths or yield strengths from coupon 共1999兲 equations. Fig. 2 illustrates the predictions for a number of
tests conducted at specified slow rates are not reported. compact rolled universal beams 共250UB37兲 tested by Dux and
• The reported maximum load and moment do not satisfy equi- Kitipornchai 共1983兲 and Wong-Chung and Kitipornchai 共1987兲,
librium on the reported geometry. whereas Fig. 3 compares the predictions to measured strengths for

JOURNAL OF STRUCTURAL ENGINEERING © ASCE / SEPTEMBER 2008 / 1453

J. Struct. Eng. 2008.134:1450-1470.

 that the unified predictions are quite good for both K = 1 as well as
for a computed K covering a reasonably large range of practical
values. Also, these examples have KLb values covering the com-
plete inelastic LTB range. This suggests that the accuracy of the
elastic K values determined using the Nethercot and Trahair
共1976兲 procedure is sufficient for all ranges of inelastic and elastic
LTB. Additional comparisions of predicted and measured test re-
sults for 23 other welded members studied by Richter 共1998兲, 28
high-strength rolled beam tests conducted by Dibley 共1969兲, and
14 rolled beam tests conducted by Galambos and his colleagues
between 1962 and 1964 are provided by White and Jung 共2004兲.
The predictions are similar in quality and characteristics to those
shown in Figs. 2 and 3 in all cases.
It is important to note that the use of K = 1 for the tests in Fig.
Downloaded from ascelibrary.org by New York University on 05/16/15. Copyright ASCE. For personal use only; all rights reserved.

Fig. 2. Comparison of 250UB37 rolled beam test results from Dux 2 results in predictions that are from 2% conservative for one of
and Kitipornchai 共1983兲 and Wong-Chung and Kitipornchai 共1987兲 to the tests with a shorter unbraced length and K = 0.91 to as much as
unified and AISC 共1999兲 resistance equations 共b f / 2t f = 6.6 to 6.8, 49% conservative for one of the tests with a longer unbraced
D / tw = 35 to 37, D / b f = 1.6, Fyf = 285 and 293 MPa兲 length and K = 0.66. Generally, the use of K = 1 results in a sub-
stantial increase in the mean M test / M n for the LTB tests consid-
ered in this work, as well as a substantial increase in the
several doubly symmetric compact-flange noncompact-web dispersion of the M test / M n values. In practice, the writers recom-
welded members tested by Richter 共1998兲. The rolled member mend that designs can be conducted initially using K = 1 共or a
tests all have D / tw = 34– 37, a cross-section aspect ratio of D / b f value of K ⬍ 1 estimated based on judgment兲. However, the engi-
= 1.6, and Fyc = 285– 293 MPa. The above-presented welded tests neer may wish to check critical unbraced lengths that have Lb
all have D / tw = 110, a cross-section aspect ratio of D / b f = 3.6, and ⬎ L p in subsequent design stages by calculating a K factor using
Fyc = 334 MPa. the Nethercot and Trahair 共1976兲 approach. One useful property
The above-presented plots show the predicted and test M n / M p of the Nethercot and Trahair 共1976兲 procedure is that it utilizes
and M test / M p values versus the normalized LTB slenderness the elastic LTB strengths based on K = 1 as a first step.
KLb共Fyc / E兲0.5 / rt. Anchor Point 1 共see Fig. 1兲 corresponds to M n Another important characteristic illustrated in Figs. 2 and 3 is
= M p and KLb共Fyc / E兲0.5 / rt = 1.1 in the unified provisions for the the accuracy of the linear equation for inelastic LTB of both
rolled beam tests and to M n = 0.92– 0.93M p and KLb共Fyc / E兲0.5 / rt rolled and welded members. The subsequent summary statistics
= 1.1 for the welded member tests. Anchor Point 2 corresponds to presented in this paper show that the simple linear equation be-
M n = FytSxc = 0.7FycSxc in all of the above-presented tests, and to tween Anchor Points 1 and 2 in Fig. 1 provides an accurate esti-
KLb共Fyc / E兲0.5 / rt = 4.76– 4.80 in the rolled beam tests versus mate of the mean test strengths over the complete inelastic LTB
KLb共Fyc / E兲0.5 / rt = 3.87 in Richter’s tests. range.
One important attribute of the above-presented tests is that the Figs. 2 and 3 also illustrate the result that the AISC 共1999兲
rolled beams have an elastic LTB K factor ranging from 0.66 to equations tend to slightly overpredict the mean test strengths for
0.91 due to the restraint provided by less critically loaded un- both compact- and noncompact-web members in the inelastic
braced lengths on each side of the critical test segment共s兲. Con- LTB range. As illustrated in Figs. 2 and 3, this is mainly due to
versely, in the welded member tests, the unbraced lengths on each the use of
side of the critical segment共s兲 are the same length and are loaded
equally in uniform bending. These conditions give K = 1 accord-
ing to the recommended calculation procedure. One can observe L p = 1.76ry 冑 Fyc
E
共6兲

versus

L p = 1.1rt 冑 Fyc
E
共7兲

in the unified provisions. Given that rt / ry varies from approxi-

mately 1.1 to 1.3 for typical I sections, the coefficient of 1.76 in
Eq. 共6兲 corresponds to a coefficient of 1.4–1.6 in Eq. 共7兲. The
rt / ry ranges from 1.15 to 1.17 for the rolled members in Fig. 2,
giving a value for the AISC 共1999兲 Anchor Point 1 of
KLb共Fyc / E兲0.5 / rt = 1.76/ 1.15= 1.53 to 1.51. The rt / ry for the
welded members in Fig. 3 is 1.23–1.24, giving a value for the
AISC 共1999兲 Anchor Point 1 of Lb共Fyc / E兲0.5 / rt = 1.43– 1.42. In-
terestingly, the relatively large percentage difference between the
L p coefficient of 1.1 in Eq. 共7兲 versus the above-mentioned values
of 1.42–1.53 generally translates into a maximum difference of
less than 6% in the calculated M n. However, the statistical corre-
Fig. 3. Comparison of welded member test results from Richter lation of the predictions with the test results is improved, and the
共1998兲 to unified and AISC 共1999兲 resistance equations 共Tests 5 – 10, corresponding reliability indices are more uniform, when L p is
b f / 2t f = 8.0– 8.1, D / tw = 110, D / b f = 3.6, Fyf = 334 MPa兲 based on Eq. 共7兲. White and Jung 共2003a兲 and White 共2008b兲

1454 / JOURNAL OF STRUCTURAL ENGINEERING © ASCE / SEPTEMBER 2008

J. Struct. Eng. 2008.134:1450-1470.

Downloaded from ascelibrary.org by New York University on 05/16/15. Copyright ASCE. For personal use only; all rights reserved.

Fig. 4. M test / M n versus KLb共Fyc / E兲0.5 / rt for 154 rolled beam tests 共Sources: Driscoll and Beedle 1957; Lee and Galambos 1962; Prasad and
Galambos 1963; Lee et al. 1964; Adams et al. 1964; Janss and Massonnet 1967; Iyengar et al. 1969; McDermott 1969; Dibley 1969, 1970; Suzuki
and Ono 1970; Wakabayashi et al. 1970; Fukumoto and Kubo 1972; Suzuki and Kubodera 1973; Udagawa et al. 1973; Dux and Kitipornchai
1983; Wong-Chung and Kitipornchai 1987兲

address why rt is conceptually and practically a more appropriate less critical adjacent spans provide some restraint to the critical
radius of gyration for the L p equation compared to the multiple unbraced length. The unified equations are based on the philoso-
parameters ry, ryc, rT, and rt in AISC 共1999兲 and AASHTO phy that, if such restraint exists, it can be quantified accurately
共1998兲. using the Nethercot and Trahair 共1976兲 K factor calculation. This
Fig. 3 illustrates a second source of some discrepancy between gives benefits not only with respect to checking compactness of
the AISC 共1999兲 equations and the test strengths for noncompact- the lateral brace spacing, but also it can result in a substantial
web members. Fig. 3 shows two strength curves from the AISC increase in the calculated resistances for larger unsupported
共1999兲 calculations, one corresponding to the AISC 共1999兲 LTB lengths 关see White 共2008b兲兴. The development of the unified
prediction and the other corresponding to the AISC 共1999兲 web equations for Rpc, Rpt, and L p has been possible due to improved
local buckling 共WLB兲 check. In AISC 共1999兲, WLB is handled as understanding of the limit states behavior from experimental tests
a separate and independent resistance calculation. As a result, the as well as refined finite element analyses not available at the time
WLB resistance acts in effect as a cap on the resistance associated of the original LRFD research.
with the LTB limit state. The AISC 共1999兲 WLB resistance for the
members considered in Fig. 3 is approximately 0.94M p. The in-
tersection of this resistance with the LTB strength curve, labeled
as the Effective Anchor Point 1 for AISC 共1999兲 in Fig. 3, is Assessment of Data: LTB Tests
approximately Lb共Fyc / E兲0.5 / rt = 1.8. Fig. 3 illustrates the general
result that improved LTB predictions are obtained by using Eq. After discarding tests based on the criteria discussed previously,
共7兲 for L p along with an ordinate of Anchor Point 1 共M max 154 rolled and 123 welded member LTB tests are retained for
⬍ M p兲 that accounts for web slenderness effects, via the param- statistical analysis. Figs. 4 and 5 show the M test / M n values versus
eters Rb, Rpc, and Rpt discussed previously. White 共2008b兲 shows KLb共Fyc / E兲0.5 / rt for these two sets of tests. There is some notice-
an example that indicates the above-mentioned discrepancies are able dispersion in M test / M n about the ideal value of 1.0 in both of
larger for singly symmetric members with the smaller flange in these plots. The data points shown in Figs. 4 and 5 include the
compression. results from the tests that have accurate geometry measurements
The AISC 共1999兲 L p and WLB resistance equations were a as well as from the tests in which only the approximate or nomi-
part of the original AISC 共1986兲 LRFD specifications. Interest- nal geometry is available. The dispersion is reduced significantly
ingly, the original research in Galambos and Ravindra 共1976兲 when only the tests with accurate measured geometry are consid-
used Eq. 共6兲 with a coefficient of 1.4 共when expressed in normal- ered.
ized form using the ratio E / Fy兲. This corresponds to a coefficient A number of the tests with approximate/nominal geometry
of 1.1– 1.3 in Eq. 共7兲. However, the summary paper by Yura et al. have attributes, other than just the approximate flange geometry
共1978兲 used Eq. 共6兲 for L p. Yura et al. 共1978兲 justify the more and/or web depth, that may influence the accuracy of the predic-
liberal L p equation in part based on the fact that, in many cases, tions:

JOURNAL OF STRUCTURAL ENGINEERING © ASCE / SEPTEMBER 2008 / 1455

J. Struct. Eng. 2008.134:1450-1470.

Downloaded from ascelibrary.org by New York University on 05/16/15. Copyright ASCE. For personal use only; all rights reserved.

Fig. 5. M test / M n versus KLb共Fyc / E兲0.5 / rt for 123 welded member tests 共Sources: Green 2000; Richter 1998; Fukumoto et al. 1971; Fukumoto
1976; Suzuki and Ono 1973, 1976; Suzuki and Kubodera 1976; Fukumoto and Kubo 1971; Hisamitu and Okuto 1971; Morikawi and Fujino 1971;
Dimitri and Ostapenko 1970; McDermott 1969; Lew and Toprac 1968; D’Apice et al. 1966; Frost and Schilling 1964; Basler et al. 1960兲

• Twenty-nine welded and two rolled member tests have ideally conditions for the tests with fixed and cantilever-span end condi-
fully fixed end conditions. These are the tests conducted by tions. However, a large fraction of the tests with fixed–fixed end
Fukumoto and Kubo 共1972兲 in Fig. 4, and by Fukumoto et al. conditions or cantilever side spans, as well as other tests with
共1971兲, Fukumoto 共1976兲, and Fukumoto and Kubo 共1971兲 in approximate/nominal geometry, are at larger values of the LTB
Fig. 5. It is well known that fully fixed end conditions are slenderness KLb共Fyc / E兲0.5 / rt. Therefore, the extent to which the
exceedingly difficult to achieve. A value of K = 0.52 is assumed different mean and higher dispersion in M test / M n for these tests
for these tests in this study. This gives a mean M test / M n of 1.00 are due to errors in the boundary conditions and geometric dimen-
for the above-mentioned 31 tests, which is the mean M test / M n sions, versus errors in the prediction model for larger LTB slen-
for the other tests with a theoretical K ⬎ 0.5. The mean derness, is uncertain. For this reason, two broad sets of tests are
M test / M n for the 31 fixed–fixed tests is 0.97 when K = 0.50 is considered for further statistical analysis and for estimation of
used. The coefficient of variation on M test / M n for these 31 reliability indices in this research: 共1兲 tests with accurate mea-
tests is V = 10.4% using K = 0.52 and 9.6% with K = 0.50. sured geometry 共86 rolled and 50 welded members兲 and 共2兲 all of
• Twelve welded and 14 rolled member tests have cantilever the above-mentioned 154 rolled and 123 welded member tests,
spans for the unbraced lengths adjacent to the critical test seg- including the tests with fixed–fixed and cantilever side-span end
ment. All of these tests were conducted by Suzuki and Ku- conditions.
bodera 共1973兲 and are documented by Fukumoto and Kubo Table 1 summarizes the statistics for M test / M n using the uni-
共1977兲. Fukumoto and Kubo 共1977兲 specify K = 1.0 for these fied equations for the tests with accurate measured geometry.
tests, and correspondingly K = 1.0 is used for these tests in the Table 2 does the same for the larger data set obtained when the
current research. The specific effects of the cantilever end tests with nominal or approximate measured geometry are in-
spans on the stability of the critical unbraced length in these cluded with the data from Table 1. Tables 1 and 2 further subdi-
tests is uncertain. The mean M test / M n for the above-mentioned vide the rolled and welded member results into five ranges of the
26 tests is 0.98 with a coefficient of variation of 10.0%. normalized LTB slenderness, denoted by c = KLb共Fyc / E兲0.5 / rt : c
The mean and median M test / M n for the 220 combined rolled 艋 1, 1 ⬍ c 艋 2, 2 ⬍ c 艋 3, 3 ⬍ c 艋 4, and c ⬎ 4. The statistics for
and welded tests obtained by excluding the above-mentioned M test / M n are evaluated separately for each of these ranges to as-
groups is 1.00 with a coefficient of variation of 5.8% 共a mean and sess the quality of the predictions over the full range of compact
median of 1.00 and V = 5.2% for 138 rolled tests and a mean of and noncompact 共inelastic LTB兲 slenderness values. The range c
1.000, median of 1.007, and V = 6.7% for 82 welded member 艋 1 corresponds to compactly braced members as defined by
tests兲. Further, the mean and median M test / M n for the 136 com- AASHTO 共2004兲. That is, AASHTO 共2004兲 uses a coefficient of
bined rolled and welded members having accurate measured ge- 1.0 rather than 1.1 in Eq. 共7兲. Conversely, the range c ⬎ 4 corre-
ometry are 1.013 and 1.003, respectively, with V = 5.2% 共a mean sponds to members that flirt with the noncompact unbraced length
and median of 1.012 and 1.007 and V = 5.0% for 86 rolled mem- limit Lr. Some of the tests in this range have KLb ⬎ Lr; however,
bers versus 1.015 and 0.999 and V = 5.7% for 50 welded mem- the largest ratio of KLb to Lr considered is 1.21.
bers兲. These groups do not include any of the above-mentioned Experimental verification of the elastic LTB equations is ad-
tests with fixed–fixed end conditions or cantilever side spans. One dressed quite effectively in the seminal research by Galambos and
might surmise from the above-presented statistics that there is Ravindra 共1976兲, where the statistics of 185 tests from four re-
some discernible error associated with the idealization of the end search studies were considered. Therefore, the focus in the current

1456 / JOURNAL OF STRUCTURAL ENGINEERING © ASCE / SEPTEMBER 2008

J. Struct. Eng. 2008.134:1450-1470.

 Table 1. Summary of M test / M n Statistics, Unified Flexural Resistance Equations, Tests with Accurate Measured Geometry, c = KLb共Fyc / E兲0.5 / rt
Quantity All c艋1 1⬍c艋2 2⬍c艋3 3⬍c艋4 c⬎4
共a兲 Rolled members
N 86 26 33 12 8 7
Minimum 0.90 0.93 0.95 0.97 0.90 0.90
Median 1.01 1.02 1.00 1.01 1.00 0.98
Maximum 1.17 1.15 1.17 1.13 1.05 1.08
Mean 1.01 1.03 1.01 1.03 0.99 0.98
V 共%兲 4.95 4.34 4.35 5.48 5.55 5.99
共b兲 Welded members
N 50 6 20 16 8 0
Minimum 0.93 0.93 0.95 0.94 0.95
Downloaded from ascelibrary.org by New York University on 05/16/15. Copyright ASCE. For personal use only; all rights reserved.

Median 1.00 1.03 1.00 0.98 1.02

Maximum 1.18 1.11 1.18 1.09 1.11
Mean 1.02 1.03 1.02 0.99 1.03
V 共%兲 5.73 6.11 7.05 3.93 4.47
共c兲 Rolled and welded members combined
N 136 32 53 28 16 7
Minimum 0.90 0.93 0.95 0.94 0.90 0.90
Median 1.00 1.02 1.00 0.99 1.01 0.98
Maximum 1.18 1.15 1.18 1.13 1.11 1.08
Mean 1.01 1.03 1.01 1.01 1.01 0.98
V 共%兲 5.23 4.60 5.55 4.88 5.17 5.99
共d兲 Welded members with compact webs
N 6 1 4 1 0 0
Minimum 0.99 0.99
Median 1.02 1.01
Maximum 1.17 1.17
Mean 1.05 1.11 1.05 1.02
V 共%兲 6.66 7.86
共e兲 Welded members with noncompact webs
N 16 0 9 5 2 0
Minimum 0.95 0.95 0.97 1.03
Median 0.99 1.00 0.98
Maximum 1.14 1.14 1.00 1.04
Mean 1.01 1.02 0.98 1.03
V 共%兲 4.98 6.21 1.38
共f兲 Welded members with slender webs
N 28 5 7 10 6 0
Minimum 0.93 0.93 0.96 0.94 0.95
Median 0.99 1.01 0.98 0.98 1.01
Maximum 1.18 1.07 1.18 1.09 1.11
Mean 1.01 1.01 1.02 1.00 1.02
V 共%兲 5.85 5.50 8.39 4.83 5.25

study is predominantly on tests in the compact and noncompact tained for several of the column-type I-section members tested by
ranges shown in Fig. 1. Eq. 共5兲 gives cr, the coefficient associated Dibley 共1969兲 关this emphasizes the severe loss of economy that
with the Lr limit, equal to 3.75 for most slender-web members would occur for these types of members if Eqs. 共1兲 and 共4兲 were
共based on Fyr = 0.7Fy兲. Therefore, slender-web members with c replaced uniformly with Eqs. 共3兲 and 共5兲兴. However, for beam-
⬎ 4 are governed typically by Eq. 共3兲. No slender-web member type rolled sections cr is usually smaller than 5. With the excep-
tests with this characteristic were identified in this research, with tion of one case with cr = 7.19, the cr values for the welded
the exception of three I girders that have fully fixed end condi- members considered in this work are less than or equal to 5.61.
tions 共these members have M test / M n values between 1.03 and Table 1 shows that the predictions based on the unified provi-
1.10兲. For compact- and noncompact-web members, cr can be sions are generally quite good for the tests with accurate mea-
substantially larger than 4 due to the effect of the St. Venant sured geometry, both for rolled and welded members. The largest
torsional stiffness in Eq. 共4兲. Values as large as cr = 7.64 are ob- deviation from a mean of M test / M n = 1 occurs for rolled members

JOURNAL OF STRUCTURAL ENGINEERING © ASCE / SEPTEMBER 2008 / 1457

J. Struct. Eng. 2008.134:1450-1470.

 Table 2. Summary of M test / M n Statistics, Unified Flexural Resistance Equations, All Tests 共Accurate and Nominal/Approximate Geometry兲, c
= KLb共Fyc / E兲0.5 / rt
Quantity All c艋1 1⬍c艋2 2⬍c艋3 3⬍c艋4 c⬎4
共a兲 Rolled members
N 154 36 73 19 12 14
Minimum 0.85 0.90 0.89 0.90 0.85 0.90
Median 1.00 1.02 0.98 1.00 1.01 1.01
Maximum 1.17 1.15 1.17 1.13 1.07 1.15
Mean 1.00 1.02 0.99 1.01 0.99 1.04
V 共%兲 5.68 4.75 4.81 5.74 6.69 8.30
共b兲 Welded members
N 123 13 42 34 25 9
Downloaded from ascelibrary.org by New York University on 05/16/15. Copyright ASCE. For personal use only; all rights reserved.

Minimum 0.78 0.90 0.92 0.83 0.78 0.82

Median 1.00 0.93 1.00 0.98 1.03 1.03
Maximum 1.22 1.11 1.18 1.18 1.11 1.22
Mean 1.00 0.97 1.02 0.98 1.00 1.03
V 共%兲 8.16 7.15 5.74 8.36 9.96 11.15
共c兲 Rolled and welded members combined
N 277 49 115 53 37 23
Minimum 0.78 0.90 0.89 0.83 0.78 0.82
Median 1.00 1.01 0.99 0.99 1.01 1.01
Maximum 1.22 1.15 1.18 1.18 1.11 1.22
Mean 1.00 1.01 1.00 0.99 0.99 1.03
V 共%兲 6.88 5.78 5.34 7.64 8.96 9.27
共d兲 Welded members with compact webs
N 70 8 24 19 13 6
Minimum 0.78 0.90 0.92 0.83 0.78 0.82
Median 0.99 0.92 1.01 0.94 0.95 1.00
Maximum 1.22 1.11 1.17 1.18 1.10 1.22
Mean 0.98 0.94 1.01 0.96 0.96 1.01
V 共%兲 9.50 7.02 5.12 10.65 12.44 13.70
共e兲 Welded members with noncompact webs
N 16 0 9 5 2 0
Minimum 0.95 0.95 0.97 1.03
Median 0.99 1.00 0.98
Maximum 1.14 1.14 1.00 1.04
Mean 1.01 1.02 0.98 1.03
V 共%兲 4.98 6.21 1.38
共f兲 Welded members with slender webs
N 37 5 9 10 10 3
Minimum 0.93 0.93 0.96 0.94 0.94 1.03
Median 1.01 1.01 0.99 0.98 1.03 1.08
Maximum 1.18 1.07 1.18 1.09 1.11 1.10
Mean 1.02 1.01 1.02 1.00 1.03 1.07
V 共%兲 5.86 5.50 7.30 4.83 5.95 3.50

with c ⬎ 4, where the mean M test / M n = 0.98. However, there are trends for different web and LTB slenderness ranges, and to indi-
no welded members with accurate measured geometry and c ⬎ 4. cate the number of tests conducted in each of these ranges.
Also, for c ⬎ 2, the dispersion in M test / M n is smaller for the Table 2 indicates that if the tests with nominal or approximate
welded than for the rolled members. Based on the similarity of measured geometry are included, the mean strength predictions
the means and coefficients of variation for the rolled and welded are still quite good for the rolled member tests, with the largest
members, one can conclude that these data sets can be combined.
deviation in the mean M test / M n from a value of 1.0 occurring for
Therefore, Table 1 also shows the summary statistics for the com-
c ⬎ 4 共M test / M n = 1.04兲 and the smallest mean M test / M n value
bined sets of rolled and welded members. Further, the statistics
for three different welded categories—compact-, noncompact-, 共M test / M n = 0.99兲 occurring for 1 ⬍ c 艋 2 and 3 ⬍ c 艋 4. However,
and slender-web members—are shown in Table 1 to illustrate the the dispersion in the M test / M n values for the rolled members is

1458 / JOURNAL OF STRUCTURAL ENGINEERING © ASCE / SEPTEMBER 2008

J. Struct. Eng. 2008.134:1450-1470.

 larger in Table 2 than in Table 1 for all of the ranges of the end spans. In addition, the large V in the 10 hybrid member tests
parameter c. is due in part to three replicate tests of hybrid girders with a more
For the welded members, the M test / M n for c 艋 1 drops to 0.97 typical Fyw / Fyf ⬵ 0.64 conducted by Hisamitu and Okuto 共1971兲
in Table 2, and the coefficient of variation of M test / M n increases in which M test / M n ranges from only 0.84 to 0.86. It should be
to V = 9.96% for 3 ⬍ c 艋 4. The drop in the mean M test / M n for c noted that another two replicate hybrid girders tested by Hisamitu
艋 1 is due to the inclusion of seven compact welded beam tests and Okuto 共1971兲 have M test / M n = 1.02 and 1.00. Only four of the
conducted by Suzuki and Ono 共1973兲 in which all of the test ten hybrid tests in Table 2 have accurate measured geometry. The
strengths are between 0.90 and 0.93M p. The source of the smaller M test / M n for these experiments is 1.17 关the test by Frost and
M test / M n for these tests is uncertain. It may be due in part to the Schilling 共1964兲兴, 1.09, 1.08, and 1.00.
influence of residual stresses due to cutting and welding. It should
be noted that four other tests conducted in the same study, but
with c slightly greater than 1.0, exhibited M test / M p values be- Estimation of the Reliability Index „␤…: LTB Tests
tween 1.00 and 1.02. Also, Table 2 includes six compact-web
welded members with c ⬎ 4. The mean M test / M n for these tests is The structural reliability is quantified in this paper by the reliabil-
Downloaded from ascelibrary.org by New York University on 05/16/15. Copyright ASCE. For personal use only; all rights reserved.

1.01. Two of these tests have fixed-end conditions, c = 4.28 and ity index ␤ 共Ellingwood et al. 1980, 1982; Galambos et al. 1982;
4.39 versus cr = 4.12 and 4.00, and M test / M n = 1.10 and 1.22, re- Galambos 2004兲 in the context of LRFD of steel building mem-
spectively. The other four of these tests have cantilever side bers. The general format of the LRFD equations is
spans, c = 4.40, 5.21, 5.50, and 5.87 versus cr = 4.88, 5.22, 5.06,
and 4.88, and M test / M n = 0.82, 0.92, 0.99, and 1.01, respectively. ␾Rn 艌 兺 ␥ iQ i 共8兲
Based on the above-noted test characteristics and results, one
must conclude that additional uniform bending tests would be where Rn represents the nominal resistance defined in the design
provisions; Qi represents the nominal load effects from ASCE 7
useful, particularly for compact-web welded members with c
共ASCE 2006兲; the resistance factor ␾ accounts for the uncertainty
艋 1 and c ⬎ 4. Also, there are no noncompact-web member tests
in the resistance; and the load factors ␥i account for the probabili-
listed in Table 2 for c 艋 1 and c ⬎ 4. Therefore, it would be useful
ties associated with the various loads acting in combination on the
to conduct a number of additional noncompact-web member tests
structure. This format provides the most effective design-oriented
in these ranges of the LTB slenderness. Finally, there are only a
means of accounting for the separate and distinctly different sta-
few compact-web member tests with accurate measured geometry
tistical properties of various resistance limit states and design load
in the intermediate range of the LTB slenderness 关see Table 1共d兲兴.
combinations. If the resistance R and the load effect Q are as-
For this reason, a number of additional tests would also be useful
sumed to be lognormally distributed, then
for compact-web members with 1 ⬍ c 艋 4. It is important to note
that for the six compact-web member tests with accurate mea-
sured geometry considered in Table 1共d兲, M test / M n ranges from
0.99 to 1.17. A substantial number of compact-, noncompact-, and
␤=
ln 冉冊 R̄
Q̄
共9兲
slender-web welded members have been tested for various ranges 冑V2R + VQ2
of c under moment-gradient loading. These tests are addressed by
White and Kim 共2008兲. where R̄ and Q̄⫽mean values of the resistance and load effects
The means and coefficients of variation for the rolled and and VR and VQ⫽corresponding coefficients of variation. The re-
welded tests in Table 2 are sufficiently close such that, as in Table sistances R are expressed by
1, one might consider combining these data sets. Also, it is im-
portant to understand the trends in the welded member data for R = R n␳ M ␳ G␳ P 共10兲
different web and LTB slenderness ranges. Therefore, all the sta- where ␳ M , ␳G, and ␳ P⫽material, geometry, and professional bias
tistical summaries from Table 1 are again repeated in Table 2. coefficients. These coefficients are assumed to be independent
Most of the slender-web and all of the noncompact-web member random parameters that characterize the variability of the resis-
uniform bending tests have accurate measured geometry. tance due to uncertainties in the material properties, the cross-
Prior to considering estimates of the reliability index 共␤兲, it is section geometry, and the prediction model used for the
important to consider several other categories of the above- calculation of Rn. The mean resistance and its coefficient of varia-
mentioned welded tests: hybrid members and singly symmetric tion are thus given by
members. There are only eight singly symmetric member tests in
Table 2. The mean M test / M n for these tests is 0.97 and V = 0.069. R̄ = Rn¯␳ M¯␳G¯␳ P = Rn¯␳R, VR = 冑V2M + VG2 + V2P 共11兲
Six of these tests have only nominal/approximate measured ge-
ometry as well as fully fixed end conditions. Only two have ac- For flexural members, the parameter ¯␳ P⫽mean M test / M n and
curate measured geometry and less than fully fixed end conditions V P⫽coefficient of variation of M test / M n, where M n is to be cal-
共M test / M n = 1.04 for both of these tests兲. There are a total of ten culated strictly using the actual 共measured兲 geometric and static
hybrid member tests. Their mean M test / M n is 1.00 and V = 0.115. material properties of the test specimens. Using a similar first-
The large V in this case is due in part to one four-point bending order reliability approach, the mean and the coefficient of varia-
test conducted by Frost and Schilling 共1964兲 with M test / M n tion of the load effects are
= 1.17. The web is A7 and the flanges are USS T-1 Type A steel
共Fyf = 820 MPa兲 in this test. Also, the adjacent unbraced lengths, 冑兺 共␳¯ QiV QiQ i兲
2

which are nearly equally critical with respect to elastic LTB by Q̄ = 兺 ¯␳Q Qi,
i
VQ =
Q̄
共12兲
the Nethercot and Trahair 共1976兲 calculations, are subjected to
high moment gradient and have reinforced webs to prevent shear where Qi⫽nominal load effects; ¯␳Qi⫽ratios of the respective
failure. This test reaches M test = 1.05M p, although its value of c mean values to the nominal values; and VQi⫽corresponding coef-
= 1.92, possibly due to end restraint from the less heavily yielded ficients of variation.

JOURNAL OF STRUCTURAL ENGINEERING © ASCE / SEPTEMBER 2008 / 1459

J. Struct. Eng. 2008.134:1450-1470.

 ASCE 7 共ASCE 2006兲 specifies dead and live load factors of Kennedy and Baker also recommended the use of two inde-
1.2 and 1.6 for the combination of dead plus live load. Galambos pendent values for ¯␳Fy and VFy for welded members, one based on
共2004兲 shows that this load combination gives the most critical the variation in the yield strength expected from heat to heat, and
共smallest兲 ␤ values for all but small live-to-dead load ratios 共L / D兲 the other based on the variation between coupons taken at differ-
less than about 0.2. As a simplification, the reliability is consid- ent locations across the width and along the length of steel plates.
ered only in the context of the ASCE 7 dead plus live load com- The first set of these parameters was developed from the data
bination in this paper. Given the above-presented expressions, Eq. collected by Kennedy and Gad Aly 共1980兲 from Canadian mills
共9兲 can be rearranged into the form for CSA G40.20-1974 grade 44W and 44T steels, whereas the

冋冉 冊册
second set of parameters was based on data from the American
1 ¯␳R 1.2 + 1.6共L/D兲 Iron and Steel Institute 共AISI 1974兲. The recommended values
␤= 共13兲
冑V2R + VQ2 ln ␾ ¯␳D + ¯␳L共L/D兲 were ¯␳Fy1 = 1.10, VFy1 = 0.092, ¯␳Fy2 = 1.00, and VFy2 = 0.061. These
parameters give a combined net ¯␳Fy of 1.10 and VFy = 0.110. The
for a given L / D and ␾, and the coefficient of variation of the load writers are not aware of published values for other plate steels
Downloaded from ascelibrary.org by New York University on 05/16/15. Copyright ASCE. For personal use only; all rights reserved.

effects may be written as such as A572 Grade 345 commonly used in the United States.
At the present time, most rolled shapes are produced to the
冑共␳¯ DVD兲2 + 共␳¯ LVL共L/D兲兲2 single material grade specification ASTM A992. Bartlett et al.
VQ = 共14兲 共2003兲 assess the properties of this steel based on 207 flat-strap
¯␳D + ¯␳L共L/D兲 tensile test specimens from 38 heats of steel and eight different
shapes provided by three producers. Galambos 共2004兲 recom-
The following statistical parameters are obtained from Elling-
mends the uniform application of ¯␳ M = 1.06 and V M = 0.06 for both
wood et al. 共1980兲 for the effects of dead and live load due to use
rolled and welded members based on the data from this study.
and occupancy: ¯␳D = 1.05, VD = 0.10, ¯␳L = 1.00, and VL = 0.25. It
Given the above-presented knowns and unknowns, it was de-
should be noted that the AASHTO 共1998, 2004兲 load models and
cided to adopt the recommendations of Galambos 共2004兲 to ob-
their corresponding statistical properties are different than these
tain ␤ estimates for LTB slenderness values of c 艋 4 in this work.
value, resulting in a base target reliability index of ␤ = 3.5 for
For assessment of the data for c ⬎ 4, the response is assumed to
bridge design. The ␤ values are estimated only in the context of
be affected predominantly by the variability associated with the
building design in this paper.
elastic modulus. In this regard, the values ¯␳E = 1.02 and VE
= 0.012 are adopted for calculation of ␤ using the results from
prior elastic LTB tests collected by Galambos and Ravindra
Material Bias and Coefficient of Variation 共1976兲, where the measured elastic modulus was used in deter-
mining ¯␳ P and V P. For the test data with c ⬎ 4 collected in the
The parameters ¯␳ M and V M account for the differences between
present work, ¯␳ M is taken equal to 1.00 and V M is taken equal to
the relevant measured static material properties and the nominal
zero as E = 200 GPa 共29,000 ksi兲 is used in all of the calculations.
material properties used in design. These parameters were taken
Therefore, the variability associated with the elastic modulus is in
as representative values of 1.05 and 0.10, respectively, for all
essence folded into the calculation of the professional bias and
I-section member flexural limit states in the seminal LRFD re-
coefficient of variation in these cases.
search 共Yura et al. 1978; Cooper et al. 1978兲 based on the con-
siderations discussed by Galambos and Ravindra 共1978兲. In
subsequent research, Kennedy and Gad Aly 共1980兲 used a more
refined approach to determine the effective material bias for the Geometry Bias and Coefficient of Variation
different flexural resistance limit states in Canadian Standards As-
sociation 共CSA兲 S16.1 共1984兲. Subsequently, Baker and Kennedy The parameters ¯␳G and VG account for the differences between the
共1984兲 and Kennedy and Baker 共1984兲 used a simpler approach actual 共measured兲 geometry and the nominal values used for the
for determining ¯␳ M and V M in the context of the design of steel geometry in design. The effective values of ¯␳G and VG vary as a
highway bridges and in a separate assessment of CSA resistance function of the flexural limit state 共e.g., full plastification of com-
factors for laterally unsupported steel beams. In their approach, pactly braced compact sections, compression flange yielding of
the bias and coefficient of variation of Fy are used for limit states compact-flange slender-web I girders, inelastic LTB of compact-,
involving full cross-section plastification, and the effective ¯␳ M noncompact-, or slender-web members, elastic LTB of compact-,
and V M vary from ¯␳Fy and VFy to ¯␳E and VE as the unbraced length noncompact-, or slender-web members, etc.兲, similar to variations
increases from the CSA equivalent of L p to the CSA equivalent of in the effective ¯␳ M and V M . Kennedy and Baker 共1984兲 concluded,
Lr. As noted previously, Baker and Kennedy 共1984兲 determined based on a review of the more rigorous studies by Kennedy and
¯␳E and VE as 1.02 and 0.012, respectively, based on a reanalysis Gad Aly 共1980兲, that it was appropriate to base ¯␳G and VG on the
of the data from Galambos and Ravindra 共1976兲. Kennedy and statistics for the plastic section modulus throughout the entire
Baker 共1984兲 also determined ¯␳Fy = 1.06 and VFy = 0.051 for rolled range of inelastic buckling for all types of rolled and welded
beams based largely on data collected from Canadian mills by I-section members in their work. Also, they concluded that essen-
Kennedy and Gad Aly 共1980兲. Kennedy and Gad Aly 共1980兲 de- tially the same final ¯␳R and VR values are obtained for elastic LTB
termined ¯␳Fy values of 1.11 and 1.05, respectively, from web and as in the more rigorous study by Kennedy and Gad Aly 共1980兲 if
flange coupon data, and recommended the use of ¯␳Fy = 1.05 for ¯␳G and VG are taken as ¯␳Iy and VIy along with ¯␳ M = ¯␳E and V M
flexure, where the strength of the flanges predominates. The val- = VE. Further, Kennedy and Baker 共1984兲 recommended ¯␳Zx
ues suggested for ¯␳E, VE, ¯␳Fy and VFy by Kennedy and Baker = 0.99, VZx = 0.038, ¯␳Iy = 1.00, and VIy = 0.058 for rolled I sections,
共1984兲 include consideration of other than normal distributions of and ¯␳Zx = 1.02, VZx = 0.015, ¯␳Iy = 1.030, and VIy = 0.020 for welded I
the data as well as an adjustment for errors in measurement 共using sections based on the data collected by Kennedy and Gad Aly
the method described subsequently in this paper兲. 共1980兲. Bartlett et al. 共2003兲 report ¯␳Zx = 1.00 and VZx = 0.034 in

1460 / JOURNAL OF STRUCTURAL ENGINEERING © ASCE / SEPTEMBER 2008

J. Struct. Eng. 2008.134:1450-1470.

 their recent study of rolled members. They do not report statistics geometry information. The above-presented reductions might also
for other geometric properties. be considered to account to some extent for reporting of the same
The original LRFD research adopted ¯␳G = 1.00 and VG = 0.05 yield strength for the flanges and web in some of the tests.
uniformly for all flexural limit states as a reasonable estimate,
indicating good control in fabrication 共Yura et al. 1978; Cooper et
al. 1978兲. These values are retained by Galambos 共2004兲 and also Reliability for Elastic LTB
are selected as reasonable representative values for this work.
The focus of this research is primarily on the collection and
analysis of the data from tests in which KLb is less than Lr.
Accounting for Errors in Measurement Galambos and Ravindra 共1976兲 collected results from 185 elastic
LTB tests by various investigators in their research. The resulting
Kennedy and Gad Aly 共1980兲, Kennedy and Baker 共1984兲, and ¯␳ P was 1.03 and V P was 0.093. These results are considered to be
Baker and Kennedy 共1984兲 also recognized that variations in representative in this range of the response. Upon adjusting for
Downloaded from ascelibrary.org by New York University on 05/16/15. Copyright ASCE. For personal use only; all rights reserved.

measured values are due in part to errors in the measurements measurement error using the above-discussed procedure, V P is
themselves. They concluded that since measurement errors are modified to 共0 . 0932 − 0 . 042兲1/2 = 0.084. The bias factor and coef-
not part of the variability of the test data per se, the corresponding ficient of variation for the resistance are then obtained as ¯␳R
VR = 冑V2M + VG2 + V2P
coefficients of variation for the test data should be determined on
= ¯␳ M¯␳G¯␳ P = 共1.02兲共1.00兲共1.03兲 = 1.05 and
a net basis. Kennedy and Baker explain their approach to account 冑 2 2 2
= 0 . 012 + 0 . 05 + 0 . 084 = 0.098. The use of these values in Eq.
for measurement errors as follows:
共13兲 along with a resistance factor ␾ = 0.9 and an assumed live-
“MacGregor, in Ellingwood et al. 共1980兲, shows, in consider-
to-dead load ratio of L / D = 3 gives ␤ = 2.6. This value is identical
ing the results of tests, that errors in the measurements of tests
to the targeted reliability index for statically determinate beams
themselves contribute to the variation in the results and that,
subjected to uniform bending moment in the first 1986 AISC
therefore, such variation should be removed from the variation of
LRFD specification 共Galambos 2004兲. For statically determinate
the measured property. Mirza and MacGregor 共1982兲 suggested a
beams under moment gradient, and for statically indeterminate
coefficient of variation equal to 0.040 to account for errors in
beams and simple frames, the estimated reliability is higher 共Yura
measurement when assessing the strength of both prestressed and
et al. 1978; White and Kim 2008兲.
reinforced concrete beams. This coefficient in itself is obtained as
共0 . 0272 + 0 . 0302兲1/2. The first term represents errors due to varia-
tions in actual specimen dimensions, and differences in material
strengths at the failure section and those measured in control Reliability for Inelastic LTB
specimens. The second term represents uncertainties in the load-
ing and measuring procedures. Fig. 6 shows the estimated ␤ values obtained using the above-
For this study, when the coefficient of variation on the mea- mentioned approach and the statistics from Tables 1 and 2. Each
sured property was equal to or less than about 0.06 the coefficient of the plots shows the estimated ␤ values for live-to-dead load
of variation for the errors in measurement was considered to vary ratios L / D of 1 to 5 in increments of 1 for each range of c
linearly from a value of 0.00 to a value of 0.04 when the coeffi- = KLb共Fyc / E兲0.5 / rt considered in the previous tables, i.e., c 艋 1,
cient of variation on the measured property varied from 0.00 to 1 ⬍ c 艋 2, 2 ⬍ c 艋 3, 3 ⬍ c 艋 4, and c ⬎ 4. Also, the reliability in-
0.06. Above a value of 0.06 for the coefficient of variation of the dices for the elastic LTB assessment summarized in the previous
measured property, the coefficient of variation for the errors in section are shown for L / D from 1 to 5 as the rightmost set of data
measurement was taken to be a constant value of 0.04. Most points in each of the plots. A resistance factor of ␾ = 0.9 is as-
measurements fell in the latter category. Using this method, the sumed in all cases. The first three plots in Fig. 6 correspond to the
maximum reduction in the coefficient of variation, taking into data including tests with nominal or approximate measured ge-
account the errors in measurement, is 25%. This procedure was ometry from Table 2, whereas the last three plots correspond to
not applied to the measurements of geometric properties, which the data for the tests with accurate measured geometry from Table
had relatively small coefficients of variation.” 1. Figs. 6共a and d兲 show the estimated ␤ values for rolled mem-
The above-mentioned procedure is adopted in this research to bers, Figs. 6共b and e兲 summarize the results for welded members,
adjust the previously reported coefficients of variation on and Figs. 6共c and f兲 show the reliability indices based on the
M test / M n, prior to their use in Eqs. 共11兲 and 共13兲 to estimate the combined statistics for rolled and welded members.
reliability index ␤. The values of V P reported in Tables 1 and 2 One can observe that most of the ␤ values are above about 2.5.
are unadjusted values. It should be noted that no attempt is made This is in the range of the reliability index of 2.6 associated with
in this work to account for the dispersion in the M test / M n values the first 1986 AISC LRFD specification 共Galambos 2004兲. The ␤
due the lack of precise geometric dimensions in the tests with values for larger c tend to be slightly less than 2.5 at L / D = 3 in
nominal or approximate measured geometry. The data for these several of the plots. For example, in Fig. 6共d兲, corresponding to
tests could be adjusted potentially in a manner similar to the the rolled members with accurate measured geometry, ␤ = 2.4 for
above-presented adjustment for errors in measurement, and a c ⬎ 4 and L / D = 3. This estimate is based on seven tests 关see Table
pooled standard deviation could be calculated using the separate 1共a兲兴.
statistics for the data sets with accurate and approximate geom- There are no welded member tests with accurate measured
etry. However, it was felt that such a process would be too exces- geometry and c ⬎ 4 关see Table 1共b兲兴. Therefore, no data points are
sive of a manipulation of the data given the other approximations shown for c ⬎ 4 in Fig. 6共e兲. There are only nine welded member
invoked in estimating the reliability indices. One might consider tests with nominal or approximate measured geometry in this
that the above-presented reductions in the coefficients of varia- LTB slenderness range 关Table 2共b兲兴. Six of these are compact
tion, accounting for measurement error, compensate to some ex- welded I-section members, the specifics of which have been dis-
tent for the additional dispersion in M test / M n due to imprecise cussed previously. For the welded members with c ⬎ 4 关Table

JOURNAL OF STRUCTURAL ENGINEERING © ASCE / SEPTEMBER 2008 / 1461

J. Struct. Eng. 2008.134:1450-1470.

 welded members with accurate measured geometry 关Table 1共c兲兴,
the estimated ␤ values are approximately constant for c 艋 4 共be-
tween 2.7 and 2.9 for L / D = 3兲 关Fig. 6共f兲兴. A smaller ␤ = 2.4 is
obtained for c ⬎ 4 in Fig. 6共d兲 关Fig. 6共f兲 shows the same ␤ as Fig.
6共d兲 for c ⬎ 4 as there are no welded member tests with accurate
measured geometry and c ⬎ 4兴 largely due to two compact welded
member tests with low M test / M n discussed previously. Therefore,
the majority of the test data supports the conclusion that the uni-
fied provisions provide a reliability commensurate with the level
targeted for determinate beams in uniform bending in the original
LRFD specification. As noted previously, more tests of compact-
and noncompact-web welded members would be useful.
Fig. 6共b兲 shows a decrease in the estimated reliability, index
for welded members with short unbraced lengths 共c 艋 1兲. This is
Downloaded from ascelibrary.org by New York University on 05/16/15. Copyright ASCE. For personal use only; all rights reserved.

due to seven compact-web welded beam tests from Suzuki and

Ono 共1973兲, noted previously, which had M test values between
0.90 and 0.93M p. However, four additional tests with c slightly
larger than 1.0 in this same study had M test values between 1.00
and 1.02. The low ␤ values for c 艋 1 in Fig. 6共b兲 are influenced
somewhat by the subdivision of the data into the discrete ranges
of c. A better estimate of ␤ for small c values in Fig. 6共b兲 can be
determined as a weighted average of the values for c 艋 1 and 1
艋 c ⬍ 2. This gives a reliability index between 2.5 and 2.8, which
is consistent with the ␤ values estimated using just the welded
member data with accurate measured geometry in Fig. 6共e兲. This
also reduces the estimated ␤ value for welded members with 1
艋 c ⬍ 2 关Fig. 6共b兲兴 to a value comparable to that shown for rolled
members with 1 艋 c ⬍ 2 in Fig. 6共a兲.
Given the above-noted qualities and limitations of the col-
lected test data, possibly Fig. 6共c兲 represents the best overall es-
timate of the reliability index for statically determinant beams
subjected to uniform bending moment for different ranges of c
= 共Fyc / E兲0.5 KLb / rt and values of L / D. For L / D = 3, this plot in-
dicates that ␤ decreases slightly from a value of 2.7 for c 艋 1 to
␤ = 2.5 for tests with c slightly larger than 4. These values are
reasonably consistent with the estimate of ␤ = 2.6 for elastic LTB
using the data from the original studies by Galambos and Ravin-
dra 共1976兲.

Influence of Lp and Fyr Equations on the Reliability

for Inelastic LTB

Fig. 7 illustrates the influence of several important variations of

the L p and Fyr equations from those associated with the base
unified provisions. These plots have the same format as the plots
in Fig. 6 except that the results are only for L / D = 3. They are
based on the larger data sets from Table 2, as the number of tests
in several of the categories considered are rather sparse in Table
1. Figs. 7共a and b兲 show the influence of using a coefficient of
either 1.1 or 1.4 in Eq. 共7兲 as well as the influence of using Eq. 共6兲
for L p 关such that the corresponding coefficient in Eq. 共7兲 is c p
Fig. 6. Reliability indices for various ranges of c = KLb共Fyc / E兲0.5 / rt, = 1.76ry / rt as shown in the legends for these plots兴 for rolled and
and from elastic LTB studies by Galambos and Ravindra 共1976兲, for compact-web welded beams. One can observe that Eq. 共7兲 with a
several live-to-dead load ratios 共L / D兲 and ␾ = 0.9 coefficient of either 1.1 共as specified in the base unified provi-
sions兲, or 1.0 关as specified in AASHTO 共2004兲兴 gives essentially
the same level of reliability. The ␤ values are slightly larger for
2共b兲兴 and L / D = 3, ␤ = 2.4 is obtained as shown in Fig. 6共b兲. A the intermediate LTB slenderness values when Eq. 共7兲 is used
reliability index of approximately 2.5 is obtained for the ranges with the AASHTO 共2004兲 coefficient of 1.0. Conversely, the use
2 ⬍ c 艋 3 and 3 ⬍ c 艋 4. However, if one considers the welded of Eq. 共6兲 for rolled and compact-web welded members in AISC
member tests with accurate measured geometry and 3 ⬍ c 艋 4 共2005兲 results in a small dip in the estimated reliability, relative to
关eight tests, Table 1共b兲兴, a ␤ of 2.9 is obtained at L / D = 3 关Fig. the use of Eq. 共7兲, for intermediate slenderness values. The small-
6共e兲兴. Also, if one considers the combined data sets of rolled and est estimated ␤ for rolled shapes is 2.5 关Fig. 7共a兲兴 based on 12

1462 / JOURNAL OF STRUCTURAL ENGINEERING © ASCE / SEPTEMBER 2008

J. Struct. Eng. 2008.134:1450-1470.

 关see Table 2共f兲兴. Fig. 7共c兲 shows a significant drop in the reliabil-
ity index at intermediate, lengths in the inelastic LTB region if the
AISC 共1999兲 L p equation is used in the context of the unified
equations 关the reliability of the specific AISC 共1999兲 Appendix G
LTB equations is discussed subsequently兴. However, use of either
of the coefficients 1.0 as in AASHTO 共2004兲 or 1.1 as in AISC
共2005兲 gives a nearly constant ␤ of 2.7 to 2.8 for L / D = 3 through-
out the inelastic LTB range. For c ⬎ 4, the estimated ␤ shown in
Fig. 7共c兲 is 2.9, which is larger than the ␤ = 2.6 determined using
the Galambos and Ravindra 共1976兲 elastic LTB test data 共illus-
trated by the rightmost data point in this figure兲. This greater
conservatism for slender-web members is expected, as Eq. 共3兲 is
based on J = 0 whereas generally there is some contribution from
St. Venant torsion to the resistance 共White and Jung 2007兲. Un-
Downloaded from ascelibrary.org by New York University on 05/16/15. Copyright ASCE. For personal use only; all rights reserved.

fortunately, due to web distortion, the full contribution from the

St. Venant torsional stiffness predicted by beam theory is, in gen-
eral, not realized for these types of members.
The AASHTO 共2004兲 L p coefficient of 1.0 in Eq. 共7兲 is based
on early results from this research as well as independent linear
regression analyses conducted by Yu and Sause 共2002兲 using the
data from statically determinate tests with L p 艋 Lb 艋 Lr. Yu and
Sause 共2002兲 determined that the L p coefficient in Eq. 共7兲 should
be approximately 1.1 from a linear regression analysis of the data
for compact I-section members, whereas slightly smaller coeffi-
cients were obtained from their linear regression analyses for
noncompact- and slender-web members. The L p coefficient of 1.1
was selected in the development of the unified provisions as, at
Fyc = 350 MPa 共50 ksi兲, this coefficient gives an unbraced length
approximately equal to: 共1兲 the requirement for plastic design
共Lpd兲 in AISC 共1999, 2005兲 and 共2兲 the requirement for use of
M n = M p and the enhanced resistance based on the Q-factor for-
mulas in AASHTO 共1998兲. For larger Fyc, Eq. 共7兲 is somewhat
less conservative than the AISC 共1999, 2005兲 equation for Lpd.
The AISC Lpd equation, which is identical to the requirement for
the use of M n = M p or the use of the Q-factor formulas in
Fig. 7. Reliability indices for various ranges of c = KLb共Fyc / E兲0.5 / rt AASHTO 共1998兲, is based on development of the plastic mecha-
and from elastic LTB studies by Galambos and Ravindra 共1976兲 for nism strength in three-span continuous beams having short end
L / D = 3 and ␾ = 0.9—effect of equations for L p and Fyr spans 共Yura et al. 1978; Bansal 1971兲. These tests exhibited sub-
stantial deformation capacity, expressed in terms of the total load
versus the middle-span transverse displacement. This definition of
tests in the range of 3 艋 c ⬍ 4 关Table 2共a兲兴. Smaller ␤ values are overall deformation capacity is somewhat different than the defi-
obtained using Eq. 共6兲 compared to Eq. 共7兲 for compact-web nition of inelastic rotation capacity often cited in discussions of
welded members, as shown in Fig. 7共b兲. Similar to the previous the Lpd equation.
plots in Fig. 6共b兲, the ␤ estimates vary significantly between the Fig. 7共d兲 shows the effect of using the AISC 共1999兲 equation
different ranges c 艋 1, 1 ⬍ c 艋 2, and 2 ⬍ c 艋 3 in this figure. For FL = Fyc − 69 MPa 共or FL = Fyc − 10 ksi兲 for rolled I-section mem-
2 ⬍ c 艋 3, the estimated ␤ is reduced to 2.1 based on the use of bers versus the unified equations. The reliability index drops to
Eq. 共6兲 versus 2.3 based on the use of Eq. 共7兲. The unified LTB ␤ = 2.3 in the range 3 艋 c ⬍ 4 for L / D = 3. This estimate is based
equations result in an estimated ␤ of 2.2 for compact-web welded on 12 tests 关see Table 2共a兲兴. Lastly, Fig. 7共e兲 shows the effect of
shapes with c ⬎ 4, using either Eq. 共6兲 or Eq. 共7兲 for L p. Unfortu- using the AISC 共1999兲 equation Fyc − 114 MPa 共or Fyc − 16.5 ksi兲
nately, there are only six LTB tests of compact-web welded mem- as the base equation for the Fyr of all types of welded members,
bers with accurate measured geometry 关see Table 1共d兲兴. The including slender web girders, rather than Fyr = 0.7Fy 关for hybrid
results for noncompact- and slender-web members are signifi- and/or for highly monosymmetric I sections with the larger flange
cantly better. As noted previously, additional tests of compact- in compression, both the unified as well as the AISC 共1999兲 Ap-
and noncompact-web welded members should be conducted. pendix F provisions specify other equations that may govern the
Fig. 7共c兲 shows the effect of coefficients of 1.0 共AASHTO calculation of FL or Fyr; these equations are employed where they
2004兲, 1.1 关unified provisions and AISC 共2005兲兴, and 1.76 关AISC apply兴. One can observe that the reliability is essentially the same
共1999兲 Appendix G兴 in Eq. 共7兲 for slender-web members, with all for welded members based on either of these equations. This is
the other calculations conducted as specified in the unified provi- not surprising, since for Fyc = 345 MPa, which is a representative
sions. The range c ⬎ 4 corresponds to elastic LTB for all the midrange value for the welded member tests, Fyr = 0.7Fyc
slender-web members considered in this research, with the excep- = 242 MPa, whereas FL = Fyc − 114 MPa= 231 MPa. Both values
tion of two hybrid girders with A514 flanges and A36 webs tested are nearly the same. Plots provided by White and Jung 共2004兲
by Lew and Toprac 共1968兲, where cr = 4.44. Only three of the indicate that for Fyc ⬎ 690 MPa, the M test / M n values are consis-
slender-web member tests collected in this research have c ⬎ 4 tently larger 共i.e., the predictions are more conservative兲 when the

JOURNAL OF STRUCTURAL ENGINEERING © ASCE / SEPTEMBER 2008 / 1463

J. Struct. Eng. 2008.134:1450-1470.

 FL causes the estimated reliability index to dip to values some-
what smaller than associated with the unified provisions in the
intermediate ranges of the parameter c. However, both of these
provisions tend to give approximately the same level of reliability
for unbraced lengths approaching the elastic LTB range. Fig. 8共c兲
shows that the use of FL = 0.5Fyc 共or Fyr = 0.5Fyc in the context of
the unified equations兲 results in designs that are significantly more
conservative than those associated with the unified equations for
slender-web members with Lb values close to Lr. This conserva-
tism offsets the effect of the more liberal AISC 共1999兲 Appendix
G L p equation to some extent, leading to a larger ␤ for 2 艋 c
⬍ 3 than the corresponding curve in Fig. 7共c兲. As noted previ-
ously, the unified equations give a ␤ of 2.7– 2.8 for slender-web
members for all values of c 艋 4 and L / D = 3.
Downloaded from ascelibrary.org by New York University on 05/16/15. Copyright ASCE. For personal use only; all rights reserved.

Assessment of Data: FLB Tests

Fig. 9 shows the M test / M n values versus the normalized FLB

slenderness 共Fyc / E兲0.5bfc / 2tfc from 11 rolled and 36 welded mem-
ber tests retained after discarding certain tests using the criteria
previously described. All but ten of these tests have accurate ge-
ometry measurements; also, all the tests with approximate/
nominal geometric dimensions happen to have relatively small
values of the normalized slenderness parameter. Table 3 provides
Fig. 8. Reliability indices for various ranges of c = KLb共Fyc / E兲0.5 / rt the summary statistics for the FLB tests. The mean and coefficient
and from elastic LTB studies by Galambos and Ravindra 共1976兲 for of variation of M test / M n are essentially the same whether the tests
L / D = 3 and ␾ = 0.9—comparison of unified and AISC 共1999兲 equa- with approximate/nominal geometry measurements are included
tions or not. Therefore, the following discussions focus on the full set
of FLB tests.
Table 3 further subdivides the FLB tests into the following
unified provisions are employed. There is no significant trend in categories: noncompact-flange, ultracompact-web, rolled 共NCF,
the M test / M n values as a function of Fyc for the more common UCW, Rolled兲, slender-flange, ultracompact-web, rolled 共SF,
tests with smaller Fyc values. UCW, Rolled兲, noncompact-flange, compact-web, nonhybrid
welded 共NCF, CW, NHW兲, noncompact-flange, noncompact-web,
nonhybrid welded 共NCF, NCW, NHW兲, noncompact-flange,
Reliability of Unified Provisions Relative to the AISC slender-web, nonhybrid welded 共NCF, SW, NHW兲, slender-
„1999… Provisions flange, slender-web, nonhybrid welded 共SF, SW, NHW兲,
noncompact-flange, slender-web, hybrid 共NCF, SW, Hybrid兲 and
Fig. 8 illustrates the effect of using the specific L p and FL equa- slender-flange, slendered-web, hybrid 共SF, SW, Hybrid兲. Ultra-
tions from AISC 共1999兲 on the estimated values of the reliability compact webs satisfy the traditional AISC 共1989兲 limits for plas-
index. Fig. 8共a兲 shows the influence of using Eq. 共6兲 for L p and tic design. One can observe that the mean and coefficient of
Fyc − 69 MPa 共Fyc − 10 ksi兲 for Fyr on the ␤ values for rolled variation are quite similar for all the above-mentioned categories
members. One can observe that the dip in the estimated reliability with the exception of the 共SF, SW, NHW兲 and 共NCF, SW, Hybrid兲
for intermediate LTB slenderness values is slightly larger in this groups. The mean and coefficient of variation are significantly
case compared to the corresponding curves in Figs. 7共a and d兲, larger for these exceptions. Therefore, for estimating the reliabil-
where the above-mentioned alternative equations for L p and Fyr ity index ␤, all the noncompact-flange rolled and nonhybrid
were considered separately. The estimated reliability index is only welded members are combined into one group, whereas the 共SF,
2.3 for 3 艋 c ⬍ 4 when these equations are employed 关based on 12 SW, NHW兲 and 共NCF, SW, Hybrid兲 groups are addressed sepa-
tests, Table 2共a兲兴. The use of Eq. 共7兲 for L p and 0.7Fyc for Fyr rately. The single 共SF, UCW, Rolled兲 and 共SF, SW, Hybrid兲 tests
clearly produces a more consistent ␤ for all values of LTB slen- are not included in calculating the statistics.
derness based on the calculations performed in this research. All the hybrid members in Fig. 9 and Table 3 have A514
Fig. 8共b兲 shows the effect of using the L p equations of AISC flanges and slender A36 webs. The larger mean and coefficient of
共1999兲 Appendix F for compact- and noncompact-web welded variation for M test / M n in these tests appears to be due largely to
I-section members and the L p equation of AISC 共1999兲 Appendix the extreme difference between the flange and web yield
G for slender-web welded I girders, combined with the AISC strengths, as well as the multiplicative combination of the bend
共1999兲 Appendix F equations for FL for all I-section member buckling and hybrid strength reduction factors, Rb and Rh, for
types. This is slightly different than in AISC 共1999兲, as AISC these members. The hybrid girders with larger normalized web
共1999兲 Appendix G effectively uses FL 共or Fyr兲 = 0.5RhFyc. Fig. slenderness 共Fyc / E兲0.52Dc / tw and smaller Rb values tend to have
8共c兲 investigates the effect of using FL 共or Fyr兲 = 0.5Fyc along with the most conservative predictions. As can be observed from Fig.
the AISC 共1999兲 Appendix G equation for L p for slender-web 9, with two exceptions, the normalized flange slenderness falls
members. Fig. 8共b兲 shows that the use of the AISC 共1999兲 L p between 0.45 and 0.62 for the girders having the largest M test / M n
equations along with the AISC 共1999兲 Appendix F equations for values 共values greater than 1.20兲. These are all hybrid girders.

1464 / JOURNAL OF STRUCTURAL ENGINEERING © ASCE / SEPTEMBER 2008

J. Struct. Eng. 2008.134:1450-1470.

Downloaded from ascelibrary.org by New York University on 05/16/15. Copyright ASCE. For personal use only; all rights reserved.

Fig. 9. M test / M n versus 共Fyc / E兲0.5 bfc / 2tfc for 11 rolled and 36 welded I-section members, flexural resistance governed by FLB 共Sources: Johnson
1985; Holtz and Kulak 1973, 1975; Abe and Mizukoshi 1973; Dibley 1970; McDermott 1969; Lew and Toprac 1968; Carskaddan 1968; Janss and
Massonnet 1967; Basler et al. 1960兲

The limited number of slender-flange member results in Table 0.38艋 共Fyc / E兲0.5 bfc / 2tfc 艋 0.62 and 1 艋 共Fyc / E兲0.5 2Dc / tw 艋 12.
3 and Fig. 9 indicates that the unified provisions are somewhat There are a number of tests beyond these ranges, but the corre-
more conservative for these member types compared to sponding combinations of web and flange slenderness are some-
noncompact-flange members governed by FLB. The majority of what sparse and do not include combinations of the bounding
the slender-flange member tests have flanges that are only mar- values of the web and flange slendernesses.
ginally slender, i.e., 共Fyc / E兲0.5 bfc / 2tfc is close to the noncompact The unified FLB equations are based on the following expres-
flange limit. Also, the yield strength of the flanges Fyc is between sion for the elastic flange local buckling coefficient, retained from
375 and 437 MPa in these tests. These are Johnson’s 共1985兲 stud- AISC 共1999兲:
ies with 共Fyc / E兲0.5 bfc / 2tfc between 0.69 and 0.73 共see Fig. 9兲.
There are two tests with significantly larger 共Fyc / E兲0.5 bfc / 2tfc in kc = 4/冑D/tw, 0.35 艋 kc 艋 0.76 共15兲
Fig. 9, one with a slender nonhybrid web tested by Basler et al.
共1960兲 and one with a slender hybrid web tested by Lew and Eq. 共15兲 is used with the exact analytical expression for elastic
Toprac 共1968兲. The flexural strength predictions for these tests are FLB 共for a given kc兲 to describe the elastic FLB resistance and to
quite conservative. However, McDermott 共1969兲 tested one A514 locate the abscissa ␭rf for Anchor Point 2 in Fig. 1 共White 2008a兲.
rolled member whose flange classifies as slender based on the Eq. 共15兲 defines a transition from a constant maximum kc value of
unified provisions. The FLB strength of this test is predicted rea- 0.76 for D / tw 艋 28 to a constant minimum value for kc of 0.35 for
sonably well 共M test / M n = 1.04兲. D / tw 艌 131. For rolled members, kc is taken equal to 0.76 in all
White and Jung 共2004兲 show that the FLB tests collected in cases. This is consistent with the elastic FLB coefficient used
this work fill a matrix of flange and web slenderness values rea- implicitly for rolled I sections in the original 1986 AISC LRFD
sonably well for 7.5艋 bfc / 2tfc 艋 15 and 70艋 2Dc / tw 艋 250, or specifications. The assumed kc = 0.76 is believed to be an adequate

Table 3. Statistical Summary, 11 Rolled and 36 Welded I-Section Members, Flexural Resistance Governed by FLB
Tests
with NCF
accurate NCF SF NCF NCF NCF SF NCF SF Rolled
measured UCW UCW CW NCW SW SW SW SW and
Quantity All geometry rolled rolled NHW NHW NHW NHW hybrid hybrid NHW
N 47 37 10 1 6 7 8 5 9 1 31
Minimum 0.91 0.91 0.91 1.00 0.97 1.01 1.09 0.96 0.91
Median 1.09 1.09 1.08 1.07 1.02 1.02 1.12 1.23 1.05
Maximum 1.46 1.46 1.17 1.18 1.11 1.26 1.39 1.46 1.26
Mean 1.11 1.12 1.06 1.04 1.08 1.04 1.08 1.17 1.20 1.37 1.06
V 共%兲 10.59 11.39 7.16 6.18 4.83 8.88 10.55 12.42 6.92
Note: UCW⫽ultracompact web; NCF⫽noncompact flange; SF⫽slender flange; CW⫽compact web; NHW⫽nonhybrid welded; NCW⫽noncompact web;
SW⫽slender web.

JOURNAL OF STRUCTURAL ENGINEERING © ASCE / SEPTEMBER 2008 / 1465

J. Struct. Eng. 2008.134:1450-1470.

 conservative value for atypical I sections having compact webs
combined with noncompact or slender flanges with bfc / 2tfc or
共Fyc / E兲0.5bfc / 2tfc larger than the above-mentioned ranges 共the the-
oretical kc for ideal full fixity at the web–flange juncture is 1.28兲.
In addition to the result from the single test by McDermott 共1969兲
discussed earlier, Hasham and Rasmussen 共2002兲 present finite
element analysis results that indicate the AISC 共1999兲 and unified
provisions are adequate for such members. The transition and
lower-bound values for kc in Eq. 共15兲 also are believed to provide Fig. 10. Reliability index for FLB versus the live-to-dead load ratio
an adequate conservative characterization of the flexural resis- 共L / D兲, 31 noncompact-flange rolled and nonhybrid welded members
tance for I sections having larger bfc / 2tfc and 共Fyc / E兲0.5bfc / 2tfc 共NCF Rolled and NHW兲, nine noncompact-flange slender-web hybrid
than the above-mentioned ranges within the limits 2Dc / tw 艋 250 members 共NCF, SW, Hybrid兲 and five slender-flange slender-web
and 共Fyc / E兲0.52Dc / tw 艋 12. This is due to the fact that the kc for nonhybrid welded members 共SF, SW, NHW兲, ␾ = 0.9
elastic FLB tends to be larger at greater flange slenderness values
Downloaded from ascelibrary.org by New York University on 05/16/15. Copyright ASCE. For personal use only; all rights reserved.

for a given web slenderness.

It is important to recognize that a slender web is theoretically Basler et al. 共1960兲 has a significantly larger M test / M n than the
in a postbuckled state at the limit of the resistance in the other tests in this category. If this test is removed as an outlier, the
AASHTO 共1998兲, AISC 共1999兲, and unified equations. Also, for mean M test / M n = 1.12 and V = 2.24% for the four remaining tests
most practical I sections, the limit of the flexural resistance in- conducted by Johnson 共1985兲. This mean and coefficient of varia-
volves significant inelastic actions. Therefore, elastic FLB coeffi- tion for these tests gives an estimated ␤ of 3.7.
cients determined from precise analytical or numerical buckling White and Kim 共2008兲 discuss additional FLB test results for
solutions do not, in general, provide an accurate characterization members subjected to moment gradient loadings. Their estimates
of the true FLB limit state. of the reliability for nonhybrid members subjected to moment
As noted previously, the flexural resistance is defined as the gradient is essentially the same as the above. The hybrid members
smaller of the independently calculated LTB and FLB resistances tested under moment gradient conditions have a mean M test / M n
in the unified provisions. For I-section beams, Bradford and Han- close to 1.0 as well as a small coefficient of variation comparable
cock 共1984兲 show theoretically, and Cherry 共1960兲 and Richter to the values determined for the LTB tests in this study.
共1998兲 show experimentally, that the interaction between FLB
and LTB may be neglected except for members with slender
flanges and thick webs 共Cherry’s tests are not included in this Summary and Conclusions
research as they are small scale specimens fabricated from an
alloy material兲. It is believed that the FLB coefficient of kc This paper evaluates the flexural resistances predicted by the base
= 0.76 is sufficiently conservative such that FLB–LTB interaction unified provisions implemented in AASHTO 共2004兲 and AISC
is not an issue for these atypical cases. Further study of these 共2005兲 versus a large set of uniform bending test results. The test
types of members would be useful to provide additional confir- data are subdivided into two main groups, one in which the flange
mation of the FLB and LTB equations at these limits. However, dimensions and web depth are measured to three or more signifi-
members with b f / 2t f ⬎ 12 can be prone to welding distortion and cant digits, and one in which the geometry is measured with less
other handling problems. AASHTO 共2004, 1998兲 limits b f / 2t f to accuracy or only nominal geometric properties are reported.
a maximum of 12 to facilitate handling and to alleviate welding Within these groups, the predictions for members in which the
distortion issues. resistance is governed by FLB and the predictions for all other
members, referred to in this study as LTB tests, are considered
separately. Key summary statistics for the professional bias factor
Estimation of Reliability Index „␤…: FLB Tests ␳ P = M test / M n are determined for these subgroups, as well as for
various categorizations of the members and ranges of slenderness
The procedures for estimating ␤ values from the FLB tests are parameters in these subgroups. The reliability index for combined
essentially the same as those discussed previously for LTB. Based dead-and-live load in building LRFD is determined based on the
on the characteristics shown in Table 3, it was decided to calcu- statistics from the collected test data, established values for the
late the reliability indices for noncompact-flange rolled and non- bias and coefficient of variation associated with the fabricated
hybrid welded members 共the last column of Table 3兲 as one data geometry and the material properties, and established building
set, and to assess the reliability of noncompact-flange slender-web design load statistics. The reliability index associated with lateral
hybrid members and slender-flange slender-web nonhybrid torsional buckling is shown to be more uniform using the unified
welded members separately. The number of FLB tests is more provisions, both for various ranges of the normalized slenderness
limited than the number of LTB tests considered in the previous KLb 共Fyc / E兲0.5 / rt and for various types of rolled and welded
sections. This number is judged to be insufficient to support any members. Also, the paper emphasizes that the calculation of an
finer subdivision of the data into separate flange and web slender- effective length factor K ⬍ 1 provides substantially larger pre-
ness ranges than shown in Table 3. Fig. 10 gives the estimated ␤ dicted resistances and substantial improvements in the accuracy
values versus the live-to-dead load ratio L / D for the above- in many cases for members in which the resistance is governed by
mentioned member categories. At L / D = 3, the reliability index is LTB. The use of K = 1 for members with noncompact- or slender
estimated as 2.9 for noncompact-flange rolled and nonhybrid brace spacing results in substantially greater conservatism and
welded members. For the extreme hybrid members collected in larger implicit values for the reliability index than those dis-
this study, the estimated ␤ is 3.2 at L / D = 3, and for the slender- cussed. A design-oriented procedure for calculation of elastic LTB
flange slender-web nonhybrid members, the comparable esti- K factors, presented originally by Nethercot and Trahair 共1976兲, is
mated ␤ is also 3.2. The one SF, SW, NHW test conducted by used for all the LTB calculations in this study. The reliability

1466 / JOURNAL OF STRUCTURAL ENGINEERING © ASCE / SEPTEMBER 2008

J. Struct. Eng. 2008.134:1450-1470.

 indices for FLB are shown to be generally larger than those for the inside face of the compression flange; for
LTB. nonwelded built-up shapes, the nearest line of
fasteners at the compression flange; denoted
by 2h p in AISC 共2005兲;
Acknowledgments d ⫽ total depth between the extreme fibers of the
flange elements;
The writers express their sincere thanks to the members of AISC E ⫽ elastic modulus of steel, 200 GPa 共29 000 ksi兲;
TC4 共Louis Geschwindner, Chair兲, the AISI LRFD Specification Fmax ⫽ maximum potential flexural resistance
Task Force 共Dennis Mertz, Chair兲, and the AASHTO T14 Tech- expressed in terms of a flange bending stress;
nical Committee for Steel Design 共Ed Wasserman, Chair兲 for their Fn ⫽ flexural resistance expressed in terms of an
efforts in the updating of the AISC and AASHTO flexural elastically computed flange stress;
strength provisions. Also, the NCHRP Project 12-38 共Dann Hall Fy ⫽ specified minimum yield stress;
and Chai Yoo, co-PIs兲 and 12-52 共John Kulicki, PI兲 teams are Fyc ⫽ specified minimum yield stress for the
thanked for providing substantial contributions. Special thanks compression flange;
Downloaded from ascelibrary.org by New York University on 05/16/15. Copyright ASCE. For personal use only; all rights reserved.

are extended to Michael Grubb of Bridge Software Development Fyf ⫽ specified minimum yield stress for the flange
International, Ltd. for extensive input on all attributes of the de- under consideration, or where Fyc = Fyt, of both
velopments. Professors Ted Galambos of the University of Min- flanges;
nesota and Bruce Ellingwood of Georgia Institute of Technology Fyr ⫽ compression flange flexural stress at the
provided valuable input. Professors Yuhshi Fukumoto and Masa- nominal onset of yielding, including
hiro Kubo of Nagoya University, Japan, catalogued a large num- compression flange residual stress effects;
ber of LTB tests originally in Fukumoto and Kubo 共1977兲. The denoted by FL in AISC 共1999, 2005兲;
research by these investigators greatly facilitated the data collec- Fyw ⫽ specified minimum yield stress for the web;
tion and analyses conducted in this study. Professor Richard h ⫽ distance between the centroids of the flange
Sause and Mr. Daming Yu of Lehigh University secured a number elements, denoted by h0 in AISC 共2005兲;
Lehigh reports for the writers. This research was funded by Pro- Ix ⫽ moment of inertia of the cross section about
fessional Services Industries, Inc. and the Federal Highway Ad- the centroidal major axis of bending;
ministration, and by the ASCE Structural Engineering Institute. J ⫽ St. Venant torsion constant;
The financial support from these organizations is gratefully ac- K ⫽ effective length factor for lateral torsional
knowledged. The opinions, findings and conclusions expressed in buckling;
this paper are the writers and do not necessarily reflect the views kc ⫽ flange local buckling coefficient;
of the above-mentioned individuals, groups, and organizations. L ⫽ live load;
Lb ⫽ laterally unbraced length;
Lp ⫽ limiting unbraced length to achieve the
Notation maximum potential flexural resistance of the
section, uniform moment case 共Cb = 1.0兲;
The following symbols are used in this paper: Lr ⫽ limiting unbraced length to achieve the onset
b f ⫽ flange width; of yielding in uniform bending 共Cb = 1.0兲;
bfc ⫽ width of a rectangular compression flange; M max ⫽ maximum potential flexural resistance
bft ⫽ width of a rectangular tension flange; expressed in terms of bending moment;
Cb ⫽ moment-gradient factor for lateral-torsional Mn ⫽ flexural resistance expressed in terms of
buckling; bending moment, calculated for the tests
c ⫽ normalized slenderness parameter for LTB considered in this research using the measured
equal to KLb共Fyc / E兲0.5 / rt; geometry and static material properties;
cr ⫽ value of c corresponding to Lr, equal to M n共Rb=1兲 ⫽ M n calculated assuming Rb = 1;
Lr共Fyc / E兲0.5 / rt; Mp ⫽ section plastic bending resistance;
D ⫽ for rolled shapes, clear distance between M test ⫽ measured static moment capacity from an
flanges less the fillet or corner radius; for welded experimental test;
shapes, the clear distance between the flanges; My ⫽ yield moment, FySxc = FySxt for doubly
for nonwelded built-up shapes, the distance symmetric sections; smaller of M yc and M yt for
between adjacent lines of fasteners; singly symmetric sections;
D ⫽ dead load; M yc ⫽ yield moment at the nominal onset of
Dc ⫽ distance from the cross-sectional centroid to yielding of the compression flange, FySxc;
the following: for rolled shapes, the inside M yr ⫽ moment corresponding to the onset of
face of the compression flange less the fillet or yielding, including compression flange residual
corner radius; for welded shapes, the inside stress effects, FyrSxc;
face of the compression flange; for nonwelded M yt ⫽ yield moment at the nominal onset of
built-up shapes, the nearest line of fasteners yielding of the tension flange, FySxt;
at the compression flange; denoted by 2hc in N ⫽ size of population;
AISC 共2005兲; Q̄ ⫽ mean load effect;
Dcp ⫽ distance from the cross-sectional plastic Qi ⫽ nominal load effect;
neutral axis to the following: for rolled shapes, R ⫽ resistance;
the inside face of the compression flange R̄ ⫽ mean resistance;
less the fillet or corner radius; for welded shapes,

JOURNAL OF STRUCTURAL ENGINEERING © ASCE / SEPTEMBER 2008 / 1467

J. Struct. Eng. 2008.134:1450-1470.

 Rb ⫽ slender-web bend buckling flange-strength ¯␳R ⫽ mean bias factor for the resistance, equal to
reduction factor, denoted by R PG in AISC R̄ / Rn; and
共1999兲 and R pg in AISC 共2005兲; ␾ ⫽ resistance factor.
Rh ⫽ hybrid web strength reduction factor;
Rn ⫽ resistance calculated using nominal design
values for the geometry and material
properties; References
Rpc ⫽ web plastification factor associated with the
AASHTO. 共1998兲. AASHTO LRFD bridge design specifications, 2nd Ed.
lateral-torsional buckling and flange local
with 1999, 2000, 2001, 2002 and 2003 interim provisions, Washing-
buckling resistance;
ton, D.C.
Rpt ⫽ web plastification factor associated with the AASHTO. 共2004兲. AASHTO LRFD bridge design specifications, 3rd Ed.
tension flange yielding resistance;
with 2005 interim provisions, Washington, D.C.
rt ⫽ effective radius of gyration for lateral-torsional Abe, H., and Mizukoshi, N. 共1973兲. “Buckling tests on plate girders under
buckling, represented approximately by the
Downloaded from ascelibrary.org by New York University on 05/16/15. Copyright ASCE. For personal use only; all rights reserved.

bending.” Annual Meeting of the Japanese Society of Civil Engineers,

radius of gyration of the compression flange preprint, Japan Society of Civil Engineers, Tokyo.
elements plus one-third of the depth of the Adams, P. F., Lay, M. G., and Galambos, T. V. 共1964兲. “Experiments on
web in compression; denoted by rT in AISC high strength steel members.” Fritz Engineering Laboratory Rep. No.
共1999兲; 297.8, Lehigh Univ., Bethlehem, Pa.
ry ⫽ radius of gyration of a noncomposite I section AISC. 共1986兲. Load and resistance factor design specification for struc-
about the plane of the web; tural steel buildings, Chicago.
ryc ⫽ radius of gyration of the compression flange AISC. 共1989兲. Specification for structural steel buildings, allowable
about the plane of the web; stress design and plastic design, Chicago.
Sxc ⫽ elastic section modulus corresponding to the AISC. 共1999兲. Load and resistance factor design specification for struc-
extreme compression fiber; tural steel buildings, Chicago.
Sxt ⫽ elastic section modulus corresponding to the AISC. 共2005兲. “Specification for structural steel buildings.” ANSI/AISC
360-05, Chicago.
extreme tension fiber;
American Iron and Steel Institute 共AISI兲. 共1974兲. The variation of product
t f ⫽ flange thickness;
analysis and tensile properties of carbon steel plates and wide flange
tfc ⫽ thickness of a rectangular compression flange;
shapes, Washington, D.C.
tft ⫽ thickness of a rectangular tension flange; ASCE. 共2006兲. “Minimum design loads for buildings and other struc-
tw ⫽ thickness of the web; tures.” ASCE/SEI 7-05, Reston, Va.
V ⫽ coefficient of variation, subscripts indicate the ASTM. 共1997兲. “Standard method for Young’s modulus, tangent modulus
associated quantity; and chord modulus.” ASTM E111-97, Philadelphia.
X2 ⫽ ratio of the bending and torsional efficiencies ASTM. 共2000兲. “Standard test methods for tension testing of metallic
of the cross section, Sxch / J; materials.” ASTM E8-00b, Philadelphia.
␤ ⫽ reliability index; Baker, K. A., and Kennedy, D. J. L. 共1984兲. “Resistance factors for lat-
␥i ⫽ load factor; erally unsupported steel beams and biaxially loaded steel beam col-
␭ ⫽ width-thickness ratio of a compression umns.” Can. J. Civ. Eng., 11共4兲, 1008–1019.
Bansal, J. P. 共1971兲. “The lateral instability of continuous steel beams.”
element;
Doctoral dissertation, Univ. of Texas, Austin, Tex.
␭fc ⫽ compression flange slenderness ratio bfc / 2tfc; Bartlett, F. M., Dexter, R. J., Graeser, M. D., Jelinek, J. J., Schmidt, B. J.,
␭pf ⫽ limiting slenderness parameter for a compact and Galambos, T. V. 共2003兲. “Updating standard shape material prop-
flange; erties database for design and reliability.” Eng. J., 40共1兲, 2–14.
␭pw ⫽ limiting slenderness parameter for a compact Basler, K., Yen, B. T., Mueller, J. A., and Thurlimann, B. 共1960兲. “Web
web corresponding to 2Dcp / tw; buckling tests on welded plate girders.” WRC Bulletin No. 64, Weld-
␭rf ⫽ limiting slenderness parameter for a ing Research Council, New York.
noncompact flange; Bradford, M. A., and Hancock, G. J. 共1984兲. “Elastic interaction of local
␳ ⫽ bias factor, subscripts indicate the associated and lateral buckling in beams.” Thin-Walled Struct., 2共1兲, 1–25.
quantity; Canadian National Standards Association 共CSA兲. 共1984兲. “Steel struc-
tures for buildings—Limit states design.” National Standard of
¯␳ ⫽ mean bias factor, subscripts indicate the
CanadaCAN3-S16.1-M84, Rexdale, Ont., Canada.
associated quantity; Carskaddan, P. S. 共1968兲. “The bending behavior of slender-web girders
¯␳G ⫽ mean geometry bias factor, equal to the mean with A514 steel flanges.” Applied Research, Rep. No. 57.019-904 (5)
effect on the calculated resistance of the (AS-EA-23-ps), U.S. Steel Corporation, Pittsburgh.
differences between the actual geometry and Cherry, S. 共1960兲. “The stability of beams with buckled compression
the nominal values used in design; flanges.” Struct. Eng., 38共9兲, 277–285.
¯␳ M ⫽ mean material bias factor, equal to the mean Cooper, P. B. 共1967兲. “Strength of longitudinally stiffened plate girders.”
effect of the differences between the relevant J. Struct. Div., 93共2兲, 419–451.
static material properties and the nominal values Cooper, P. B., Galambos, T. V., and Ravindra, M. K. 共1978兲. “LRFD
criteria for plate girders.” J. Struct. Div., 104共9兲, 1389–1407.
used in design;
D’Apice, M. A., Fielding, D. J., and Cooper, P. B. 共1966兲. “Static tests on
¯␳ P ⫽ mean professional bias factor, equal to
longitudinally stiffened plate girders.” WRC Bulletin No. 117, Welding
M test / M n, where M n is calculated strictly using Research Council, New York.
measured values for the geometric and Dibley, J. E. 共1969兲. “Lateral torsional buckling of I-sections in Grade 55
static material properties; steel.” Proc., Institution of Civil Engineers, Vol. 43, 599–627, Institu-
tion of Civil Engineers, London.
Dibley, J. E. 共1970兲. “A preliminary investigation into the use of high-

1468 / JOURNAL OF STRUCTURAL ENGINEERING © ASCE / SEPTEMBER 2008

J. Struct. Eng. 2008.134:1450-1470.

 strength structural steel in structures designed plastically.” BIRSA A512 Grade 65 steel beams.” Fritz Engineering Laboratory Rep. No.
Open Rep. No. EG/A/13/70, British Steel Corporation, London. 343.4, Lehigh Univ., Bethlehem, Pa.
Dimitri, J. R., and Ostapenko, A. 共1970兲. “Pilot tests on the static strength Janss, J., and Massonnet, C. 共1967兲. “The extension of plastic design to
of unsymmetrical plate girders.” WRC Bulletin No. 156, Welding Re- steel A52.” Publications of the IABSE, 27, 15–30.
search Council, New York. Johnson, D. L. 共1985兲. “An investigation into the interaction of flanges
Driscoll, G. C., and Beedle, L. S. 共1957兲. “The plastic behavior of struc- and webs in wide flange shapes.” Proc., Annual Technical Session,
tural members and frames.” Weld. J. (Miami, FL, U.S.), 36共6兲, 275- Structural Stability Research Council, Univ. of Missouri-Rolla, Rolla,
s–286-s. Mo., 395–405.
Dux, P. F., and Kitipornchai, S. 共1983兲. “Inelastic beam buckling experi- Kennedy, D. J. L., and Baker, K. A. 共1984兲. “Resistance factors for steel
ments.” J. Constr. Steel Res., 3共1兲, 3–9. highway bridges.” Can. J. Civ. Eng., 11共2兲, 324–334.
Ellingwood, B. E., Galambos, T. V., MacGregor, J. G., and Cornell, C. A. Kennedy, D. J. L., and Gad Aly, M. 共1980兲. “Limit states design of steel
共1980兲. “Development of a probability based load criterion for Ameri- structures—Performance factors.” Can. J. Civ. Eng., 7共1兲, 45–77.
can National Standard A58.” NBS Publication No. 577, Washington, Lee, G. C., Ferrara, A. T., and Galambos, T. V. 共1964兲. “Experiments on
D.C. braced wide-flange beams.” Welding Research Council Bulletin No.
Ellingwood, B. E., MacGregor, J. G., Galambos, T. V., and Cornell, C. A. 99, Welding Research Council, New York.
Downloaded from ascelibrary.org by New York University on 05/16/15. Copyright ASCE. For personal use only; all rights reserved.

共1982兲. “Probability-based load criteria: Load factors and load com- Lee, G. C., and Galambos, T. V. 共1962兲. “Post-buckling strength of wide-
binations.” J. Struct. Div., 108共5兲, 978–997. flange beams.” J. Engrg. Mech. Div., 88共1兲, 59–75.
Frost, R. W., and Schilling, C. G. 共1964兲. “Behavior of hybrid beams Lew, H. S., and Toprac, A. A. 共1968兲. “The static strength of hybrid plate
subjected to static loads.” J. Struct. Div., 90共3兲, 55–88. girders.” S.F.R.L. Technical Rep. No. P550-11, Structures Fatigue Re-
Fukumoto, Y. 共1976兲. “Lateral buckling of welded beams and girders in search Laboratory, Dept. of Civil Engineering, Univ. of Texas, Austin,
HT 80 steel.” Preliminary Rep. of 10th Congress, International Asso- Tex.
ciation for Bridge and Structural Engineering, Tokyo, 403–408. McDermott, J. F. 共1969兲. “Plastic bending of A514 steel beams.” J.
Fukumoto, Y., Fujiwara, M., and Watanebe, N. 共1971兲. “Inelastic lateral Struct. Div., 95共9兲, 1851–1871.
buckling tests on welded beams and girders.” Proc., Japan Society of Mirza, S. A., and MacGregor, J. G. 共1982兲. “Probabilistic study of
strength of reinforced concrete members.” Can. J. Civ. Eng., 9共3兲,
Civil Engineers, Tokyo, Vol. 189, 39–51 共in Japanese兲.
431–448.
Fukumoto, Y., Itoh, Y., and Kubo, M. 共1980兲. “Strength variation of lat-
Morikawi, N., and Fujino, S. 共1971兲. “Ultimate bending strength of plate
erally unsupported beams.” J. Struct. Div., 106共1兲, 65–181.
girders.” Annual Meeting of the Japanese Society of Civil Engineers,
Fukumoto, Y., and Kubo, M. 共1971兲. “Inelastic lateral buckling strength
preprint 共in Japanese兲.
of monosymmetrical I-beams.” Annual Meeting of the Japanese Soci-
Nethercot, D. A., and Trahair, N. S. 共1976兲. “Lateral buckling approxi-
ety of Civil Engineers, preprint 共in Japanese兲.
mations for elastic beams.” Struct. Eng., 54共6兲, 197–204.
Fukumoto, Y., and Kubo, M. 共1972兲. “Lateral buckling strength of girders
Owen, D. R. J., Rockey, K. C., and Skaloud, M. 共1970兲. “Ultimate load
with bracing systems.” Preliminary Rep. of 9th Congress, Interna-
behavior of longitudinally reinforced webplates subjected to pure
tional Association for Bridge and Structural Engineering, Amsterdam, bending.” IABSE Publ., 30共1兲, 113–148.
The Netherlands, 299–304. Prasad, J., and Galambos, T. V. 共1963兲. “The influence of the adjacent
Fukumoto, Y., and Kubo, M. 共1977兲. “An experimental review of lateral spans on the rotation capacity of beams.” Fritz Engineering Labora-
buckling of beams and girders.” International Colloquium on Stability tory Rep. No. 205H.12, Lehigh Univ., Bethlehem, Pa.
of Structures Under Static and Dynamic Loads, ASCE, New York, Richter, J. F. 共1998兲. “Flexural capacity of splender web plate girders.”
541–562. M.S. thesis, Univ. of Texas, Austin, Tex.
Galambos, T. V. 共1998兲. Guide to stability design criteria for metal struc- Suzuki, T., and Kubodera, M. 共1973兲. “Inelastic lateral buckling of steel
tures, T. V. Galambos, ed., Structural Stability Research Council, beams.” Annual Meeting of Architectural Institute of Japan, Architec-
Wiley Interscience, New York. tural Institute of Japan, Tokyo, preprint 共in Japanese兲.
Galambos, T. V. 共2004兲. “Reliability of the member stability criteria in Suzuki, T., and Ono, T. 共1970兲. “Experimental study of inelastic beams.
the 2005 AISC Specification.” Int. J. Steel Struct., 4共4兲, 223–230. 1: Beam under uniform moment.” Transactions of the Architectural
Galambos, T. V., Ellingwood, B. E., MacGregor, J. G., and Cornell, C. A. Institute of Japan, Architectural Institute of Japan, Tokyo, Vol. 168,
共1982兲. “Probability-based load criteria: Assessment of current design 77–84 共in Japanese兲.
practice.” J. Struct. Div., 108共5兲, 959–977. Suzuki, T., and Ono, T. 共1973兲. “Inelastic lateral buckling of steel
Galambos, T. V., and Ravindra, M. K. 共1976兲. “Load and resistance factor beams.” Annual Meeting of Architectural Institute of Japan, Architec-
design criteria for steel beams.” Research Rep. No. 27, Structural tural Institute of Japan, Tokyo, preprint 共in Japanese兲.
Division, Civil and Environmental Engineering Dept., Washington Suzuki, T., and Ono, T. 共1976兲. “Deformation capacity of high-strength
Univ., St. Louis. steel members.” Preliminary Rep., 10th Congress of the International
Galambos, T. V., and Ravindra, M. K. 共1978兲. “Properties of steel for use Association for Bridge and Structural Engineering, preprint.
in LRFD.” J. Struct. Div., 104共9兲, 1459–1468. Udagawa, K., Saisho, M., Takanashi, K., and Tanaka, H. 共1973兲. “Experi-
Green, P. S. 共2000兲. “The inelastic behavior of flexural members fabri- ments on lateral buckling of H-shaped beams subjected to monotonic
cated from high performance steel.” Doctoral dissertation, Lehigh loadings.” Transactions of the Architectural Institute of Japan, Vol.
Univ., Bethlehem, Pa. 48共212兲, 22–33.
Hasham, A. S., and Rasmussen, K. J. R. 共2002兲. “Interaction curves for Wakabayashi, M., Nakamura, T., Okamura, N. 共1970兲. “Studies on lateral
locally buckled I-section beam-columns.” J. Constr. Steel Res., 58共2兲, buckling of wide flange beams 共1兲.” Disaster Prevention Research
213–241. Institute Annuals, Vol. 14A, Kyoto Univ., Kyoto, Japan 共in Japanese兲.
Hisamitu, S., and Okuto, K. 共1971兲. “Lateral buckling tests on beams White, D. W. 共2008a兲. “Structural behavior of steel.” Chap. 6, Steel
with residual stresses.” Annual Meeting of AIJ, Architectural Institute bridge design handbook, National Steel Bridge Alliance, Chicago.
of Japan, Tokyo, preprint 共in Japanese兲. White, D. W. 共2008b兲. “Unified flexural resistance equations for stability
Holtz, N. M., and Kulak, G. L. 共1973兲. “Web slenderness limits for com- design of steel I-section members—Overview.” J. Struct. Eng.,
pact beams.” Structural Engineering Rep. No. 43, Univ. of Alberta, 134共9兲, 1405–1424.
Alta., Canada. White, D. W., Barker, M., and Azizinamini, A. 共2008兲. “Shear strength
Holtz, N. M., and Kulak, G. L. 共1975兲. “Web slenderness limits for non- and moment-shear interaction transversely-stiffened steel I-girders.”
compact beams.” Structural Engineering Rep. No. 51, Univ. of Al- J. Struct. Eng., 134共9兲, 1437–1449.
berta, Alta., Canada. White, D. W., and Jung, S.-K. 共2003a兲. “Simplified lateral-torsional buck-
Iyengar, S. N. S., Lu, L.-W., and Beedle, L. S. 共1969兲. “Experiments on ling equations for I-and channel-section members.” Structural Engi-

JOURNAL OF STRUCTURAL ENGINEERING © ASCE / SEPTEMBER 2008 / 1469

J. Struct. Eng. 2008.134:1450-1470.

 neering, Mechanics and Materials Rep. No. 24a, School of Civil and buckling strength of noncomposite discretely-braced I-beams.” Eng.
Environmental Engineering, Georgia Institute of Technology, Atlanta. Struct., 29共8兲, 1872–1888.
White, D. W., and Jung, S.-K. 共2003b兲. “Simplified lateral-torsional buck- White, D. W., and Kim, Y. D. 共2008兲. “Unified resistance equations for
ling equations for singly-symmetric I-section members.” Structural stability design of steel I-section members—Moment gradient tests.”
Engineering, Mechanics and Materials Rep. No. 24b, School of Civil J. Struct. Eng., 134共9兲, 1471–1486.
and Environmental Engineering, Georgia Institute of Technology, At- Wong-Chung, A. D., and Kitipornchai, S. 共1987兲. “Partially braced in-
lanta. elastic beam buckling experiments.” J. Constr. Steel Res., 7共3兲, 189–
White, D. W., and Jung, S.-K. 共2004兲. “Unified flexural resistance equa- 211.
tions for stability design of steel I-section members—Uniform bend- Yu, D., and Sause, R. 共2002兲. “Regression study on the inelastic lateral
ing tests.” Structural Engineering, Mechanics and Materials Rep. No. torsional buckling strength of I-beams.” ATLSS Center, Lehigh Univ.,
28, School of Civil and Environmental Engineering, Georgia Institute Bethlehem, Pa.
of Technology, Atlanta. Yura, J. A., Galambos, T. V., and Ravindra, M. K. 共1978兲. “The bending
White, D. W., and Jung, S.-K. 共2007兲. “Effect of web distortion on the resistance of steel beams.” J. Struct. Div., 104共9兲, 1355–1369.
Downloaded from ascelibrary.org by New York University on 05/16/15. Copyright ASCE. For personal use only; all rights reserved.

1470 / JOURNAL OF STRUCTURAL ENGINEERING © ASCE / SEPTEMBER 2008

J. Struct. Eng. 2008.134:1450-1470.