You are on page 1of 11

Journal of Constructional Steel Research 63 (2007) 10661076

www.elsevier.com/locate/jcsr

Proposed design methods for lateral torsional buckling of unrestrained


steel beams in fire
Ronny Budi Dharma, Kang-Hai Tan
School of Civil and Environmental Engineering, Nanyang Technological University, 50 Nanyang Avenue, Singapore 639798, Singapore

Received 27 June 2006; accepted 18 September 2006

Abstract

The behaviour of unrestrained steel I-beams has been studied by means of numerical analysis and published experimental results. The numerical
model was developed using a commercial finite element program, MSC.MARC Mentat. A series of different UB and UC sections and different
spans, subjected to both uniform moment and midspan point loads, are considered. The numerical predictions of the buckling moments are then
compared with published experimental results. Consequently, a new approach is proposed to provide more accurate and safe predictions of the
fire resistance of unrestrained beams, as well as to overcome certain weaknesses in the EC3:1.2 [European committee for standardization (CEN).
Eurocode 3: Design of steel structures, Part 1.2: General rules structural fire design, EN 1993-1-2. Brussels (Belgium); 2005] design formula.
In addition, to provide a quick and simple design approach for engineers, a straightforward and rational method known as the Rankine method
is introduced to predict the LTB failure load of steel beams in fire. It is shown that the Rankine approach generally provides a good lower bound
value for numerical predictions.
c 2007 Published by Elsevier Ltd

Keywords: Steel; Lateral torsional buckling; Fire resistance; Numerical analysis; Steel beams; Rankine approach

1. Introduction such as residual stress and lateral continuity between adjacent


segments, should also be considered. Due to uncertainties
Beams are structural members that resist applied loads arising from material and geometric imperfections, design
primarily by their bending and shearing action. When a short codes often approximate various imperfections in the form of
beam is loaded, the ultimate load-carrying capacity is governed sinusoidal curves. Despite some simplifications, the LTB design
by its full plastic moment of resistance. Its failure is due procedure is still overly complicated and lengthy, because
to excessive plastic yielding of a cross-section, unless local there is usually an attempt to provide continuity between
buckling of the compression flange or web occurs prematurely. old code formulae and new requirements when changes are
On the other hand, an unrestrained long beam may fail by made. Although the LTB problem of steel beams at ambient
elastic lateral torsional buckling at a load much lower than the temperature has been widely studied, there are relatively fewer
corresponding load for its full plastic moment capacity. For a publications addressing LTB under fire conditions. Among
beam of intermediate span between these two limits, failure the works in this field, the three main papers are Bailey
may occur by inelastic lateral torsional buckling. et al. [3], Vila Real et al. [4] and Yin and Wang [5]. Bailey
Much research has been conducted on lateral torsional et al. [3] used a three-dimensional computer model to predict
buckling at ambient temperature over the past decades [2], the behaviour of uniformly heated unrestrained beams, and
and some results from this have been incorporated into design compared them with BS5950 Part 8 [6] and ENV 1993
codes. The considerable scatter in these test results shows that
EC3:1.2 [7]. Vila Real et al. [4] conducted full scale tests
not only the effects of initial crookedness, but also those of twist
on 120 IPE beams to study the LTB behaviour of beams at
and load eccentricity, are important for LTB. Other factors,
elevated temperatures. They also proposed a modification to
the ENV 1993 EC3:1.2 [7]. Their proposals were subsequently
Corresponding author. Tel.: +65 6790 5297; fax: +65 6791 0676. incorporated into EN1993 EC3:1.2 [1]. Yin and Wang [5]
E-mail address: CKHTAN@ntu.edu.sg (K.-H. Tan). used a commercial finite element analysis package, ABAQUS,

c 2007 Published by Elsevier Ltd


0143-974X/$ - see front matter
doi:10.1016/j.jcsr.2006.09.008
R.B. Dharma, K.-H. Tan / Journal of Constructional Steel Research 63 (2007) 10661076 1067

Nomenclature o limiting slenderness


p plastic collapse load factor
C1 loading factor
R Rankine load factor
E elastic modulus
LT normalised slenderness ratio
f max maximum longitudinal stress
LT Perry imperfection factor for LTB
fp limit of proportionality
LT
T Perry imperfection factor for LTB at elevated
f pT elevated temperature limit of proportionality temperature
G shear modulus LT,fi strength reduction factor in fire
Iw warping constant non-linearity factor
Ix major axis moment of inertia
Iy minor axis moment of inertia
J torsion constant to study the effect of non-uniform temperature distributions on
kE elastic modulus reduction factor in fire LTB. Thus, compared to many other well-researched aspects
kp limit of proportionality reduction factor in fire of structural steel behaviour under fire conditions, there are
ky yield strength reduction factor in fire relatively fewer studies on LTB under fire conditions.
L beam length This paper presents a quick review of the design approach
M applied moment to provide insight into the current design basis of EC3 and
Mb,fi ultimate buckling moment capacity in fire BS5950 for LTB under fire conditions. Subsequently, the
Mcx in-plane bending moment capacity authors propose an accurate but more involved approach
Mcx,fi in-plane bending moment capacity at elevated (called the Alternative approach) to predict the buckling
temperature load of unrestrained beams. This approach is different from
ME elastic buckling moment capacity considering EC3:1.2 [1], and provides an accurate, safe and smooth
major axis curvature transition from ambient temperature to elevated temperature
M E,fi elastic buckling moment capacity considering designs. For quick calculations and more conservative
major axis curvature at elevated temperature predictions, the authors also introduce a simple interaction
ML limiting moment formula, widely known as Rankine approach, to predict the
MR Rankine ultimate moment LTB moment under fire conditions. It is worth noting that
My yield moment this approach has been successfully applied to column and
Mcr elastic buckling moment capacity frame problems under both ambient and elevated temperature
pb buckling stress conditions by the second author [812].
pb,fi buckling stress at elevated temperature The Alternative approach provides a rigorous and accurate
pE elastic buckling stress considering major axis design procedure, whereas the Rankine approach offers a
curvature quick and lower bound estimate for the design engineer. Both
p E,column elastic buckling stress of column methods are validated using published experimental results by
p TE elastic buckling stress considering major axis Vila Real et al. [4], and by finite element analyses. In the
curvature at elevated temperature modelling, a maximum initial out-of-plane bow of L/1000,
py yield strength in the form of a half sine curve, is assumed. This value
has also been commonly used by other researchers for code
p Ty , p yT elevated temperature yield strength
calibration [5]. A rectangular thick shell element is used to
Sx plastic section modulus about the major axis
model the UB and UC sections. In accordance with BS476-
T temperature
20:1987 [13], failure is assumed to have occurred when the
u lateral deflection
maximum in-plane deflection exceeds L/20, or when the
Z ex effective section modulus about major axis
deflection exceeds L/30, the beam reaches a maximum rate
Zx elastic section modulus about the major axis
of deflection L 2 /(9000 depth) mm/min (L and depth are in
Zy elastic section modulus about the minor axis
mm). As most unrestrained beams do not support a slab on top,
rotation
the temperature distribution across the section is fairly uniform.
twist
Thus, only a uniform temperature distribution is considered in
imperfection factor
this paper.
severity factor
M,fi material partial safety factor in fire 2. Review of design approaches
slenderness ratio
e elastic critical load factor BS5950 Part 8 [6] deals with unrestrained beams using
LT lateral torsional buckling slenderness ratio the limiting temperature method. Firstly, the load ratio is
LT lateral torsional buckling modified slenderness calculated based on the buckling failure at ambient temperature.
ratio Subsequently, the limiting temperature, which is a function of
LT , fi modified slenderness ratio for LTB in fire load ratio, can be found from the corresponding table [14]. A
similar method is also used by EC3:1.2, based on the degree
1068 R.B. Dharma, K.-H. Tan / Journal of Constructional Steel Research 63 (2007) 10661076

of utilisation. There is a slight difference in the definitions of approach) provides safe and more accurate buckling moment
load ratio and degree of utilisation, as pointed out by Bailey predictions compared to EC3:1.2 [1]. In addition, it overcomes
et al. [3]. The load ratio is defined as the ratio of applied a few limitations of the EC3:1.2 design formula, which will
moment at the fire limit state to the cold design moment of become apparent through the ensuing discussions. The only
resistance, whereas the degree of utilisation refers to the ratio drawback of this approach is that it requires a somewhat rather
of the applied moment at the fire limit state to the fire design lengthy procedure. Thus, the Rankine approach is introduced to
moment of resistance at time zero. A more accurate method provide a quick, but more conservative, method for calculating
is given in ENV 1993 EC3:1.2 [7] based, on bending moment the buckling stress pb under fire conditions.
capacity method. This method uses the same calculation as
at ambient temperature, but incorporates strength and stiffness 3.1. The alternative approach for LTB at elevated temperature
reduction factors in fire, and a factor of 1.2 to consider the non-
liner stressstrain relationship for steel. This method has since Vila Real et al. [4] made a proposal to modify the ENV
been modified based on Vila Reals et al. [4] proposal when 1993 EC3:1.2 [7] approach to eliminate the overestimation
Eurocode 3 status was converted from ENV to EN. Using the and the step change at LT,fi = 0.4. This has subsequently
modified slenderness for fire, the strength reduction factor for been accepted and incorporated into the latest version of EC3
LTB is calculated as: (EN 1993 EC3:1.2 [1]). However, this latest EC3 approach is
1 too conservative, especially for temperatures less than 500 C,
LT,fi = q (1) as can be seen from Fig. 1(a)(d). The horizontal axis is the
EC2 2
LT,fi
EC + LT,fi LT,fi modified slenderness ratio for LTB in fire, whereas the vertical
  axis is the non-dimensional buckling moment normalised to
2
in which, LT,fi
EC = 0.5 1 +
LT,fi + LT,fi and = is full plastic moment capacity in fire. The reason for this
p severity factor = 0.65 obtained
an imperfection factor, with conservatism may be some very low test results obtained by
from curve-fitting ( = 235/ p y ; p y = yield strength). Vila Real et al. that occurred at 500 C (Fig. 1(d)) and 600 C
The beam slenderness at elevated temperature is given by: (Fig. 1(e)). There were large differences in the failure loads
even for the same beam with the same slenderness. This
LT,fi = LT k y /k E
p
(2) discrepancy may be due to the large degradation of material
properties at that temperature range. A slight error in the
where:
measurement of temperature can cause a large difference in
LT,fi = modified slenderness ratio for LTB in fire; the test results. Besides being conservative for the medium
LT= lateral torsional buckling modified slenderness ratio slenderness range, there are two other weaknesses of the EC3
(= Mcx /Mcr ); approach. Firstly, the slope of the buckling strength curve
Mcx = in-plane bending moment capacity; for very low slenderness is too steep (Fig. 1), implying that
Mcr = elastic buckling moment capacity; even very short beams will not be able to achieve their
k y = yield strength reduction factor in fire; plastic moment capacity unless the compression flange is fully
k E = elastic modulus reduction factor in fire. restrained. That is to say, a slight increase in the unrestrained
It should be noted that EC3:1.2 does not consider the effect portion of a short beam will substantially reduce its load-
of major axis curvature in determining the elastic buckling bearing capacity below its plastic moment capacity. This seems
moment capacity. Finally, the beams LTB in fire based on EN unreasonable, and FEA of very short beams shows that they
1993 EC3:1.2 [1] is calculated using: can achieve buckling moments very close to their plastic
moment capacities, as can be seen in Fig. 1(c)(e). Secondly,
Mb,fi = LT,fi w Sx k y p y / M,fi (3)
the EC3:1.2 approach is only for elevated temperatures, and
where: a slightly different form of equation is used at ambient
Mb,fi = ultimate buckling moment capacity in fire; temperature (EC3:1.1). Thus, there exists a discontinuity
LT,fi = strength reduction factor in fire; between the approach at ambient and at elevated temperatures.
w = Z ex /Sx ; Ideally, EC3:1.2 should give the same form as EC3:1.1 at 20 C.
Z ex = effective section modulus about major axis; Based on the above arguments, the authors propose a new
Sx = plastic section modulus about the major axis; approach called the Alternative approach that gives more
M,fi = material partial safety factor in fire. accurate results and is able to overcome the weaknesses of
Clearly, in the EC3 approach to LTB design, there is a current EC3:1.2. This Alternative approach will provide a
discontinuity when going from ambient temperature [15] to logical link with the ambient temperature approach, because it
elevated temperatures [1]. The details are discussed in the can also be used at ambient temperature and will give the same
following section. result as in BS5950 Part1 [16]. It is worth noting that the limit
slenderness o below which a beam can attain its full plastic
3. Proposed approaches for design under fire conditions moment capacity, and the slope of the curve above this limit,
can vary with temperature.
In this section, two design approaches proposed by the A straight elastic beam has no out-of-plane deformation until
authors are discussed. The first approach (called the Alternative the applied moment reaches its elastic buckling moment. The
R.B. Dharma, K.-H. Tan / Journal of Constructional Steel Research 63 (2007) 10661076 1069

Fig. 1. Comparisons of test results [4], EC3 approach [1] and FEA at (a) 200 C; (b) 300 C; (c) 400 C; (d) 500 C; (e) 600 C.

elastic buckling moment of a straight beam loaded with equal


and opposite end moments at ambient temperature (Fig. 2) is
given by:
s s
2 E Iy 2 E Iw

Mcr = 2
GJ + (4)
L L2
where:
Mcr = elastic buckling moment capacity;
E = elastic modulus;
I y = minor axis moment of inertia;
L = beam length; Fig. 2. Buckling of a simply supported I-beam.
1070 R.B. Dharma, K.-H. Tan / Journal of Constructional Steel Research 63 (2007) 10661076

G = shear modulus; Z x = section modulus about the major axis;


J = torsion constant; Z y = section modulus about the minor axis;
Iw = warping constant. u = lateral deflection;
The above equation ignores the effect of major axis = twist.
curvature, and thus produces conservative estimates of the Subsequently, by taking the maximum stress as yield
elastic buckling moment. An approximate solution that includes strength, the limiting moment M L at which a crooked beam
the effect of major axis curvature is given by Kirby and attains its first yield at ambient temperature is:
Nethercot [17] as:
1.25 + Mcr /M y
 
ML
Mcr =
ME = q (5) My 2
 = p E Z ex
1 E I y /E I x
s
1.25 + Mcr /M y 2 Mcr

(10)
where: 2 My
M E = elastic buckling moment capacity taking into in which M y = yield moment.
consideration major axis curvature; By introducing the section capacity Mcx in place of the
I x = major axis moment of inertia; yield moment M y , M E is used to calculate the elastic buckling
p E = elastic buckling stress taking into consideration major moment instead of Mcr , and the factor 1.25 from Eq. (10) is
axis curvature; replaced by (1 + LT M E /Mcx ), whereby the initial curvature
Z ex = effective section modulus about major axis. and twist term LT is defined by:
The effective section modulus Z ex is taken as the plastic sec- s s !
tion modulus about the major axis Sx for class 1 and 2 sections, 2E Mcx
and as the elastic section modulus Z x for class 3 sections. LT = 0.007 0.4 0 (11)
py ME
Rearranging Eq. (5), one obtains:
one can obtain:
2E
pE = . (6) py pE
2 E Mcx pb = q (12)
py ME
LT + LT 2 p p
y E
The above expression for elastic buckling stress has the same
form as the column buckling stress ( p E,column = 2 E/2 ), in which:
with the LTB slenderness ratio defined as: p y + (LT + 1) p E
LT = . (13)
2
s s
2 E Mcx
LT = (7) The corresponding initial curvature and twist term at an
py ME
elevated temperature is therefore:
where: s s s !
p E,column = elastic buckling stress of column; 2E Mcx kE
LT = 0.007
T
0.4 0. (14)
= slenderness ratio; py ME ky
LT = lateral torsional buckling slenderness ratio.
Finally, the corresponding buckling stress at elevated
Incorporating the strength and stiffness reduction factors
temperature pb,fi can be easily obtained as:
at elevated temperatures, the corresponding elastic critical
buckling stress at elevated temperature p TE can be obtained as: k y p y p TE
pb,fi = q (15)
2k E E T + T 2 k p pT
LT y y E
p TE = = k E pE . (8) LT
2LT
in which:
In reality, beams are seldom straight, due to the presence of
k y p y + LT
T + 1 pT

initial curvature and twist. The derivations of crooked beams LT
T
= E
. (16)
can be found in [2]. The maximum longitudinal stress f max in 2
the beam can be determined from the summation of stresses It should be noted that the slenderness ratio LT in Eq. (7)
due to major axis bending, minor axis bending and warping as is a function of the geometry of the cross section and beam
defined in Eq. (9): length only. Therefore LT is the same at ambient and elevated
! temperatures, unlike LT , which is a function of both the elastic
M E I y d2 u + d f /2 modulus and yield strength for a given temperature (Eq. (2)).
f max = (9)
Zx Zy dz 2 Simply applying the material reduction factors will result in
L/2
unsafe predictions of the buckling load in the inelastic range of
where: the slenderness ratio, because the stressstrain relationship at
M = applied moment; elevated temperatures is no longer bilinear elasticplastic, but
R.B. Dharma, K.-H. Tan / Journal of Constructional Steel Research 63 (2007) 10661076 1071

Fig. 4. Variation of the non-linearity factor with temperature.

Fig. 3. Different rates of degradation of the tangent modulus with temperature.

highly non-linear, resulting in tangent stiffness at a particular


temperature much lower than the initial tangent, as shown in
Fig. 3. This affects the behaviour of an unrestrained beam
significantly, because the buckling load is highly dependent
on tangent stiffness (Eq. (8)). With a greater slenderness, the
buckling stress is generally lower, and the tangent stiffness is
closer to the initial value. For a beam of low slenderness, where
material yielding governs design instead of buckling, the beam Fig. 5. Comparison of various methods for S275 at 400 C.
bending resistance is not so sensitive to its tangent stiffness. and the limiting slenderness are temperature dependent. As a
However, for a beam of medium slenderness, where the applied result, the new form of the Perry imperfection factor for this
stress is substantial and above the limit of proportionality, Alternative approach is:
simply applying the material reduction factors can lead to
s s
a gross overestimation of LTB resistance in fire. Thus, the
s !
2E Mcx 1.5 k E
LT = 0.007
T
p
elastic modulus should be based on tangent stiffness instead of 0.4 0. (18)
py ME ky
the initial tangent. However, applying the concept of tangent
modulus in the calculation of buckling capacity requires a The above imperfection factor at ambient temperature is the
knowledge of the stress level; therefore, an iterative process same as in BS 5950, because = 1 at ambient temperature
is needed. A practical and viable option is to consider it as a (Fig. 4). Fig. 5 shows the difference in the LTB design curve
form of imperfection, and to incorporate it through the Perry of the Alternative approach, the EC3:1.2 [1] approach and FEA
imperfection factor. analyses for steel grade S275 at 400 C. It can be observed that
This is possible because the effect of inelastic buckling the Alternative approach agrees better with the trend of FEA.
and yielding is analogous to the effect of initial imperfections
that make the curve deviate from the ideal elastic buckling 3.2. Rankine approach for LTB at elevated temperatures
curve. The authors use the term f pT / p Ty to consider the highly
nonlinear stressstrain relationship at elevated temperatures. The complexity of the LTB of beams at ambient temperature
This ratio is expressed in terms of the material reduction factor increases significantly at elevated temperatures. Hence, the
as in Eq. (17), and is termed as the non-linearity factor : design approach tends to be cumbersome for hand calculations.
To provide an easy and quick design approach, the authors also
f pT f pT / p y kp introduce a simple interaction formula known as the Rankine
= = = (17) approach to solve for the LTB of beams at elevated temperature.
p Ty p Ty / p y ky
The Rankine formula is an empirical formula initially proposed
where: in the nineteenth century [18]. It was later modified in the mid-
f pT = elevated temperature limit of proportionality; twentieth century [19], and has since been adopted into various
p Ty = elevated temperature yield strength; design codes [15,16] for steel frames subjected to increasing
k p = limit of proportionality reduction factor in fire. load. The same formula is also applicable to steel columns [20].
Apparently, there has been no attempt to apply the Rankine
It is noted that varies with temperature as shown in Fig. 4.
formula to the LTB of beams at elevated temperature. The
The shape of the curve that incorporates the initial
formula has the following form:
crookedness and the limiting slenderness o are adjusted based
on the experimental results [4] and current FEA predictions 1 1 1
= + (19)
at elevated temperatures. Clearly, both the shape of the curve R p e
1072 R.B. Dharma, K.-H. Tan / Journal of Constructional Steel Research 63 (2007) 10661076

Fig. 7. Different types of node in end cross-section.


Fig. 6. LTB design curves at various temperatures using the Rankine approach.

in which, R , p and e are the Rankine, plastic collapse and Table 1


elastic critical load factors, respectively. Different types of end node boundary conditions
Applying the formula to the beam LTB, the three load factors Type of nodes restraint ux uy uz x y z
relate to the Rankine ultimate moment M R , plastic moment
Left mid-web node 0 0 0 1 1 0
capacity Mcx and elastic buckling moment M E . It can be Right mid-web node 0 0 1 1 1 0
expressed as: All other end nodes 0 1 1 1 1 0
1 1 1 Note: 0 = restraint; 1 = free.
+ = . (20)
Mcx ME MR
4. Finite element analysis
The above equation can be rearranged into:
Mcx M E In this section, numerical analyses using the commercial
MR = (21)
M E + Mcx finite element program, MSC.MARC Mentat, are described.
p y Z ex Firstly, the ability of the numerical model to simulate
pb Z ex = . (22) LTB failure is validated against the theoretical bifurcation
1 + Mcx /M E
elastic solution and published experimental results at ambient
q temperature. The validated numerical models are then used to
Substituting LT = 2 E/ p y Mcx /M E to Eq. (22), one
simulate various section sizes and beam lengths with mid-span
obtains:
point load or uniform bending at elevated temperatures.
py
pb = . (23)
2LT 4.1. Validation of numerical model
1 + 2 E/ p y

Subsequently, incorporating the respective strength and Rectangular thick shell elements from MSC.MARC Mentat
stiffness reduction factors in fire into Eq. (23), one obtains: are used to model the beam. This element has three global
displacements and three global rotations as degrees of freedom.
k y py
pb,fi = (24) The membrane strains and curvatures are obtained from the
2LT
1+ displacement and rotation fields, respectively; the transverse
2 k E E/(k y p y )
shear strains are calculated at the median plane and interpolated
in which, pb,fi = buckling stress at elevated temperature. to the integration points. Integration through the shell thickness
2
LT
Defining 2 E/ py
= LT (normalized slenderness ratio), Eq. is performed using Simpsons rule. The default number of
integration points is 11. The beam is simply supported at
(24) can be written as:
both ends, so that lateral deflection and twist rotation at the
pb,fi ky supports are prevented, while the flange ends are free to rotate
= . (25)
py k
1 + k Ey LT in horizontal planes so that the beam ends are free to warp.
This support system can be achieved using three types of end
The normalised slenderness ratio (LT ) is actually a node restraint, as shown in Table 1. The x-axis is perpendicular
ratio of plastic moment capacity to elastic buckling moment to the web, the y-axis is parallel to the web, whereas the
(Mcx /M E ). Eq. (25) can be plotted as pb,fi / p y against LT , as z-axis is along the beam, as shown in Fig. 7, where u and
shown in Fig. 6, since both are non-dimensional parameters. are displacement and rotation respectively.
By plotting the graph in this manner, an iterative procedure FEA results are firstly compared with theoretical bifurcation
requiring the guessing of the failure temperature as required solutions for elastic beams [21]. Two loading conditions are
in EC3:1.2 can be avoided, because two parameters that are considered, namely, uniform moment and mid-span point
needed to obtain the failure temperature ( pb,fi / p y and LT ), loading acting at the shear centre. It should be noted that other
are independent of temperature. Interpolation can be used for loading conditions can be considered, based on the concept of
a point that lies between the two available curves. equivalent moment factor, which takes into account different
R.B. Dharma, K.-H. Tan / Journal of Constructional Steel Research 63 (2007) 10661076 1073

Table 2
Comparison between FE predictions and their elastic theoretical bifurcation solutions for (a) Mid-span point load case; (b) Uniform moment case

(a)

Section size L (m) LT Theoretical (kN) Predicted (kN) Error (%)


838 292 176UB 12 126.6 280.90 282 0.39
533 210 82UB 8 114.8 154.20 154.7 0.32
406 178 54UB 6 100.6 136.77 136 0.56
305 102 28UB 6 150.1 23.53 23.66 0.55
254 102 28UB 6 130.6 26.78 26.7 0.30

(b)

Section size L (m) LT Theoretical (kNm) Predicted (kNm) Error (%)

838 292 176UB 12 147.2 624.22 623 0.20


533 210 82UB 8 133.5 228.44 227.5 0.41
406 178 54UB 6 117.0 151.97 151.1 0.57
305 102 28UB 6 174.6 26.14 26.04 0.38
254 102 28UB 6 151.8 29.75 29.52 0.77

Table 3 Table 4
Comparisons between FE predictions and experimental results by Kitipornchai Comparisons of test results [4] from various methods
and Trahair [22]
Test/Rankine Test/EC3 Test/Alt.approach
Section L (m) Experimental Current FEA (kN)
Mean 1.637 1.719 1.495
size (kN)
L/1000 L/2000 L/3000 Straight T = 200 C SD 0.291 0.286 0.266
COV 0.177 0.166 0.178
10UB29 2.44 234.96 235.2 249.2 254.8 268.8
Mean 1.640 1.719 1.485
3.05 185.12 166 176 184 198
T = 300 C SD 0.307 0.309 0.268
3.66 145.07 120.7 127.5 130.9 141.1
COV 0.187 0.180 0.180
6.1 56.96 49 50.4 51.8 52.5
Mean 1.559 1.639 1.377
T = 400 C SD 0.318 0.327 0.266
loading arrangements [15,16]. The beams are assumed to be COV 0.204 0.199 0.193
perfectly straight and remaining elastic. Five different UB Mean 0.983 1.035 0.879
sections are analysed, with slenderness ratios ranging from 100 T = 500 C SD 0.270 0.280 0.232
to 150. The elastic modulus is taken as 205 GPa. The theoretical COV 0.275 0.271 0.263
and finite element predictions are shown in Table 2(a) and (b) Mean 0.727 0.765 0.633
T = 600 C SD 0.198 0.203 0.158
for mid-span point loading and uniform moment loading cases,
COV 0.272 0.266 0.250
respectively. It can be seen that the error is almost negligible
(less than 1%).
After comparing the elastic solutions with theory, the increasingly until the member fails. Seven different section
finite element model is then validated with beams undergoing sizes of both UB and UC sections of grade S275 and S355
inelastic LTB based on experimental results from Kitipornchai steel, with various slenderness ratios ranging from 21 to 205,
and Trahair [22]. It can be seen in Table 3 that the buckling load producing a set of results representative of almost the whole
is affected by the assumed initial imperfections. Four assumed range of practical slenderness ratios, are considered in the
maximum initial imperfections in the form of a half sine numerical simulations. Two loading conditions, namely, equal
curve, ranging from a perfectly straight beam to the maximum and opposite moments and midspan point loads, are applied
imperfection of L/1000, are compared with actual test results. with five different temperatures: 400 C, 500 C, 600 C,
As can be seen, the value of L/1000 provides a good lower 700 C and 800 C.
bound for the test results. Hence, this value for imperfections is 5. Comparisons of various methods
subsequently used in the following numerical analyses.
Based on test results, a statistical comparison has been
4.2. Numerical simulations at elevated temperatures made for the Alternative, Rankine and EC3:1.2 [1] approaches.
As can be seen from Table 4, the Alternative approach gives
The validated model described in the previous section has the best mean value up to 400 C. For 500 C and 600 C,
been extended to elevated temperature conditions. The material the Alternative approach gives slightly higher predictions,
stressstrain curve at elevated temperatures comes from EC3 because of slight discrepancies of test results as discussed
formulations for steel at elevated temperatures. Besides, the before. The EC3 approach gives very conservative predictions
load is applied in two stages; firstly, the temperature is increased for temperatures up to 400 C, with as much as 70%
to the desired failure temperature; secondly, the load is applied underestimation of strength.
1074 R.B. Dharma, K.-H. Tan / Journal of Constructional Steel Research 63 (2007) 10661076

Table 5
Comparison of FE results from various methods

FEA/Rankine FEA/EC3 FEA/Alt.approach


Mean 1.166 1.255 1.073
T = 400 C SD 0.067 0.062 0.066
COV 0.057 0.050 0.062
Mean 1.185 1.271 1.098
T = 500 C SD 0.083 0.087 0.080
COV 0.070 0.068 0.073
Mean 1.155 1.241 1.052
T = 600 C SD 0.065 0.063 0.070
COV 0.056 0.051 0.067
Mean 1.182 1.271 1.063
T = 700 C SD 0.090 0.094 0.102
COV 0.076 0.074 0.096
Mean 1.154 1.236 1.076
T = 800 C SD 0.078 0.081 0.078
COV 0.068 0.065 0.072

Compared to the other approaches, the Alternative approach


gives the lowest standard deviation. The Rankine method
produces similar results as the EC3 approach, but gives
improved accuracy for temperatures less than 400 C. Taking
into consideration the simplest method of all, the Rankine
approach is useful for a quick estimation of the failure load.
More experimental results are needed for temperatures above
500 C, because of the limited range of results given by Vila
Reals et al. [4] test. Testing at temperatures above 600 C
is urgently needed, as steel temperatures can exceed 600 C
in a normal fire. Besides, a wider range of slenderness ratios
is required for a further study of LTB behaviour at elevated
temperatures.
The statistical comparisons of FE results from various
approaches are summarized in Table 5. The Alternative
approach provides very accurate results, with a maximum mean
error of 10%. The EC3 approach provides the most conservative
predictions. This is because the approach is calibrated
against a numerical analysis that uses a more conservative
imperfection model (considering the residual stresses in
addition to maximum lateral imperfection of L/1000). The
Rankine approach provides comparatively good results, and is
conservative with a maximum error of approximately 18%.
Several graphical comparisons of the results from various
approaches are shown in Fig. 8. These figures show clearly that
the Alternative approach matches quite well the FE predictions
for the range of slenderness ratios and temperatures considered,
while the predicted buckling moment based on the Rankine Fig. 8. Comparison of test [4] and FE results with various approaches at (a)
approach is always slightly lower than the one based on 400 C; (b) 500 C; (c) 600 C; (d) 700 C.
the Alternative approach. Out of these three approaches, the
EC3 approach gives the lowest prediction of the buckling Rankine approach is slightly lower. The lowest prediction of
moment. It is worth noting that there is practically no difference the buckling moment is given by the EC3:1.2 approach.
in
p the predicted buckling moment for slender section with
Mcx,fi /M E,fi 1.5. 6. Conclusions
A worked example in Appendix A, which is solved using the
Alternative, Rankine and EC3:1.2 approaches, shows that the Three weaknesses of the EC3:1.2 approach for the design
predicted buckling moment based on the Alternative approach of laterally unrestrained steel beams in fire are identified. In
is very close to the FE result, while that based on the order to overcome these weaknesses, a general approach, called
R.B. Dharma, K.-H. Tan / Journal of Constructional Steel Research 63 (2007) 10661076 1075

the Alternative approach, has been suggested. The Alternative Hence:


approach gives more accurate results and provides a logical link Mcr
between the ambient and the elevated temperature approaches. ME = q
1 E I y /E I x

In addition to the Alternative approach, the authors also
propose an analytical approach based on the Rankine principle. 5643
= q
This Rankine approach provides an easy and quick design
1 (3.1 108 /7.91 108 )

approach to estimate the lateral torsional buckling failure of
steel beams in fire, in which the complexity of the general = 7236.5 kN m.
approach has increased significantly at elevated temperatures. Step 2. Calculate Perry imperfection factor for LTB at elevated
Besides its simplicity, the use of the Rankine approach also temperature LT
T .
enables the failure temperature to be determined directly,

without an iterative procedure requiring guessing of the failure s s s !
temperature, as in EC3:1.2 approach. 2E Mcx 1.5 k E
LT = 0.007
T
p
0.4
Finite element models and published experimental results py ME ky
are used to validate these proposed design approaches. s
Comparisons with FE predictions show clearly that the 2 210 103
Alternative approach matches well the FE predictions, while the = 0.007 0.326
275
predicted buckling moment based on the Rankine approach is r r !
always slightly lower than that based on Alternative approach. 1290 1.5 0.13
0.4(0.326) = 0.12707.
The lowest prediction of the buckling moment is given by the 7236.5 0.23
EC3:1.2 approach.
Step 3. Calculate the value of p TE .
Acknowledgement
2k E E 2 0.13 210 103
The research study of the first author has been sponsored by p TE = =
2LT 2 210 103 /275 (1290/7236.5)

Singapore Millennium Scholarship.
= 200.5 MPa.
Appendix A Step 4. Calculate the value of LT
T .


A 6.5 m 356 406 UC 235 beam of S275 steel is simply
k y p y + LT
T + 1 pT

supported at both ends, such that lateral deflection and twist LT
T
= E
rotation at the supports are prevented, but the beam ends are free 2
to warp. A mid-span point loading is applied at the shear centre. 0.23 275 + (0.12707 + 1) 200.5
=
The beam is subjected to a maximum uniform temperature of 2
700 C during fire. The material and section properties are: = 144.61 MPa.
p y = 275 MPa; E = 210 GPa; G = 80.77 GPa; Sx =
Step 5. Determine the buckling stress at elevated temperature
4.69 106 mm3 ; I x = 7.91 108 mm4 ; I y = 3.1 108 mm4 ; pb,fi .
J = 8.12 106 mm4 ; Iw = 9.54 1012 mm6 .

The material reduction factors at 700 C [1]: k y = 0.23;
k E = 0.13; k p = 0.075. k y p y p TE
pb,fi = q
T + T 2 k p pT
LT LT y y E
A.1. Alternative approach
0.23 275 200.5
= p
Step 1. Calculate the value of Mcx and M E . 144.61 + 144.612 (0.23 275 200.5)
= 53.88 MPa.
Mcx = p y Sx = 275 4.69 106 = 1290 kN m.
Thus, the ultimate buckling moment capacity in fire Mb,fi
s s = pb,fi Sx = 252.7 kN m.
2 E Iy 2 E Iw

Mcr = C1 GJ + ;
L2 L2 A.2. Rankine approach
in which C1 is loading factor [1]
Step 1. Calculate the LTB slenderness ratio LT .
v !
u 2 210 103 3.1 108
u
= 1.365t
65002
q
LT = 2 E/ p y Mcx /M E
s p
2 210 103 9.54 1012
(80.77 103 8.12 106 ) +
65002
q
2 210 103 /275 1290/7236.5 = 36.65.
p
= 5643 kN m.
=
1076 R.B. Dharma, K.-H. Tan / Journal of Constructional Steel Research 63 (2007) 10661076

Step 2. Determine the normalised slenderness ratio LT . approach. EC3:1.2 [1] approach results in the lowest prediction
of buckling moment (Mb,fi = 195.2 kN m), which is only

77% of the FE result. Thus, the over-conservatism of EC3:1.2


2LT approach will result in uneconomical member design in fire.
LT =
2 E/ p y
36.652 References
= = 0.1782.
( 2 210 103 )/275 [1] European committee for standardization (CEN). Eurocode 3: Design of
steel structures, Part 1.2: General rules structural fire design, EN 1993-
Step 3. Determine the buckling stress at elevated temperature 1-2. Brussels (Belgium); 2005.
pb,fi . [2] Trahair NS, Bradford MA, Nethercot DA. The behaviour and design of
steel structures to BS5950. London: Spon Press; 2001.

[3] Bailey CG, Burgess IW, Plank RJ. The lateraltorsional buckling of
k y py 0.23 275 unrestrained steel beams in fire. Journal of Constructional Steel Research
pb,fi = ky
=  
k E LT
1+ 0.23 1996;36(2):10119.
1 + 0.13 0.1782
[4] Vila Real PMM, Piloto PAG, Franssen J-M. A new proposal of a
= 48.09 MPa. simple model for the lateraltorsional buckling of unrestrained steel I-
beams in case of fire: Experimental and numerical validation. Journal of
Thus, the ultimate buckling moment capacity in fire Mb,fi = Constructional Steel Research 2003;59:17999.
pb,fi Sx = 225.5 kN m. [5] Yin YZ, Wang YC. Numerical simulations of the effects of non-uniform
temperature distributions on lateral torsional buckling resistance of steel
I-beams. Journal of Constructional Steel Research 2003;59:100933.
A.3. EC3:1.2 approach
[6] (British standard institution) BSI. Structural use of steelwork in building,
Part 8: Code of practice for fire resistant design, BS5950. London (UK);
Step 1. Calculate the modified slenderness ratio for LTB in fire 2003.
LT,fi . [7] European committee for standardization (CEN). Eurocode 3: Design of
steel structures, Part 1.2: General rules structural fire design, ENV
1993-1-2. Brussels (Belgium); 1995.
[8] Toh WS, Tan KH, Fung TC. Compressive resistance of steel columns in
LT,fi = LT k y /k E = Mcx /Mcr k y /k E
p p p
fire: Rankine approach. Journal of Structural Engineering 2000;126(3):
p p
= 1290/5643 0.23/0.13 = 0.636. 398405.
[9] Toh WS, Tan KH, Fung TC. Strength and stability of steel frames in fire:
Step 2. Calculate the value of LT,fi
EC . Rankine approach. Journal of Structural Engineering 2001;127(4):4619.
[10] Tang CY, Tan KH, Ting SK. Basis and application of a simple interaction

= = 235/ p y = 0.65 235/275 = 0.6.
p
formula for steel columns under fire conditions. Journal of Structural
Engineering 2001;127(10):120613.
Hence:
  [11] Tang CY, Tan KH. Basis and application of simple interaction formula
2
LT,fi
EC
= 0.5 1 + LT,fi + LT,fi for steel frames under fire conditions. Journal of Structural Engineering
  2001;127(10):121420.
= 0.5 1 + (0.6 0.636) + 0.6362 = 0.893. [12] Huang Z-F, Tan KH. Rankine approach for fire resistance of axially-
and-flexurally restrained steel columns. Journal of Constructional Steel
Research 2003;59(12):155371.
Step 3. Determine the strength reduction factor for LTB in fire [13] (British standard institution) BSI. Fire Test on building materials and
LT,fi . structures, Part 20: Method for determination of the fire resistance of
elements of construction (general principles), BS476. London (UK);
1987.
1 [14] Wang YC. Steel and composite structures: Behaviour and design for fire
LT,fi = q
2 2 safety. London: Spon Press; 2002.
LT,fi
EC + LT,fi
EC
LT,fi [15] European committee for standardization (CEN). Eurocode 3: Design of
steel structures, Part 1.1: General rules and rules for buildings, EN 1993-
1
= p 1-1. Brussels (Belgium); 2005.
0.893 + 0.8932 0.6362 [16] (British standard institution) BSI. Structural use of steelwork in building,
Part 1: Code of practice for designRolled and welded sections, BS5950.
= 0.658.
London (UK); 2001.
[17] Kirby PA, Nethercot DA. Design for structural stability. Constrado
Step 4. Determine the ultimate buckling moment capacity in
Monographs, Suffolk (UK): Granada Publishing; 1979.
fire Mb,fi . [18] Rankine WJM. Useful rules and tables, London; 1866.
Mb,fi = LT,fi w Sx k y p y / M,fi = 0.658 1 4.69 106 [19] Merchant W. The failure loads of rigid jointed frameworks as influenced
by stability. The Structural Engineer 1954;32:18590.
0.23 275/1 = 195.2 kN m. [20] Horne MR, Merchant W. The stability of frames. Oxford: Pergamon Press
The FE prediction of buckling moment for this example is Ltd; 1979.
[21] Timoshenko SP, Gere JM. Theory of elastic stability. McGraw-Hill;
253.6 kN m which is very close to the Alternative approachs
1961.
prediction of 252.7 kN m. The Rankine approachs prediction is [22] Kitipornchai S, Trahair NS. Inelastic buckling of simply supported steel I-
225.5 kN m, which is slightly lower than that of the Alternative beams. Journal of the Structural Divison, ASCE 1975;101(ST7):133347.

You might also like