Thin-Walled Structures ()

Contents lists available at ScienceDirect

Thin-Walled Structures
journal homepage: www.elsevier.com/locate/tws

Fire exposed steel columns with a thermal gradient over

the cross-section
Oscar Delgado Ojeda a, Johan Maljaars b,c,n, Roland Abspoel a
a
Delft University of Technology, Delft, The Netherlands
b
Eindhoven University of Technology, Delft, The Netherlands
c
TNO, Delft, The Netherlands

art ic l e i nf o a b s t r a c t

Thermal gradients often occur in re exposed structures. This paper considers thermal gradients over the
Keywords: cross-section of steel columns. By means of nite element simulations, the paper demonstrates that these
Flexural buckling gradients reduce the exural buckling resistance of the columns. This is due to the eccentricity in the column
Thermal gradient created by the temperature gradient. Design equations in modern standards provide a gross approximation of
Non-uniform temperature the load bearing resistance of such columns in which the eccentricity is ignored and in order to compensate for
Fire this the yield stress and modulus of elasticity are to be determined at maximum temperature. Based on an in-
Steel depth analysis of the results of the nite element simulations, this paper provides an alternative design model
Column
which much better agrees with the actual behaviour of a re exposed column.
& 2015 Elsevier Ltd. All rights reserved.

1. Introduction
The temperature-dependent constitutive properties of steel cause
that exural buckling of columns at elevated temperature is different
from room temperature. Instead of the distinct yield plateau that
characterises (mild) steel at room temperature, a curved stressstrain
relationship occurs due to which the principles behind the elastic
critical (Euler) buckling load no longer holds. The 2% proof stress i.e.
the stress at a plastic strain of 2% is usually considered as an
alternative to the yield stress for steel structures exposed to re. In a
large temperature range, the modulus of elasticity reduces faster than
the 2% proof stress, implying that exural buckling at elevated
temperature may be more decisive as compared to room temperature.
On the other hand, residual stresses caused by rolling or welding may
relax as a consequence of the action of creep. A substantial number of
research activities starting in the years 80' of the previous century
have been devoted to exural buckling of steel columns at elevated
temperature. An extensive experimental and numerical research by
Talomana et al. [1] and Franssen et al. [2] forms the basis of the current
design rules in the European standard EN 1993-1-2 [3] and the design
rules in the American standard AISC 360-10 [4] provide practically
identical results.
The temperature over a member or structure is always non-
uniform in a real re situation. Columns with a temperature gradient
n
Correspondence to: TNO, Van Mourik Broekmanweg 6, Delft, The Netherlands.
Tel.: 31 88 8663464.
along the length of the column have been considered in more recent
research, amongst others [57]. These studies concluded that taking
the maximum temperature of the column provides a slightly con-
servative estimate of the failure load of individual columns. Flexural
buckling of columns with a temperature gradient over the cross-
section, which is the subject of the current study, has been considered
before in experimental studies [8,9], nite element studies [10,11], and
studies based on the principle of virtual work [12,13]. Three aspects in
such columns complicate the exural buckling behaviour:
The variation of temperature over the cross-section causes that
the modulus of elasticity and the proof stress vary over the
cross-section. The colder part of the section has a higher
modulus of elasticity and a higher proof stress as compared
to the hotter part. This makes the calculation of the (in)elastic
critical buckling load and the buckling resistance more difcult.
The higher stiffness of the colder part implies that the neutral
axis shifts towards the colder part. If the load is applied in the
geometric centroid of the original section, this causes a bending
moment that results in an increase of stress in the hot part and
a decrease in the cold part [14,15].
The difference in thermal expansion of the hot and the cold
part of the cross-section causes the column to bend towards
the re. In a pin-ended column this so-called thermal bowing
can be considered as a very large geometric imperfection or as
additional loading of the column by a bending moment causing
an increase of stress in the colder part and a decrease of stress
in the hotter part [15].

http://dx.doi.org/10.1016/j.tws.2015.02.009
0263-8231/& 2015 Elsevier Ltd. All rights reserved.

Please cite this article as: Delgado Ojeda O, et al. Fire exposed steel columns with a thermal gradient over the cross-section. Thin-
Walled Structures (2015), http://dx.doi.org/10.1016/j.tws.2015.02.009i
2 O. Delgado Ojeda et al. / Thin-Walled Structures ()
The latter two aspects are acting in opposite direction. Thermal
bowing increases with column length whereas the shift of the neutral
axis depends on the section type. Agarwal et al. [10] have demon-
strated that a slender column bends towards the hotter side whereas a
stockier column bends towards the cooler side before failure. The
results of the studies are ambiguous. Where references [8,10] con-
cluded that a thermal gradient reduces the resistance of a column in
comparison with the same column with uniform temperature, [12,13]
concluded that the critical temperature of a column with a thermal
gradient is higher than that without a thermal gradient. Thermal
bowing has been considered in all studies; however, the shift of the
centroid was considered in [8,10] but not in [12,13]. Although not
explicitly mentioned, it seems that the uniform temperature column
used for comparison was in these studies, the average temperature of
the non-uniform temperature case. Moura Correia et al. [11] have
demonstrated that a column with temperature gradient gives a higher
critical temperature than the same column with uniform temperature,
if the maximum temperature of the non-uniform case is considered
for the uniform case.
Most of the previous research into exural buckling of columns
with a temperature gradient across the cross-section was devoted
to columns with an abrupt jump in temperature. Agarwal et al. [10]
and Moura Correia et al. [11] considered edge columns that were
partially embedded in the (insulated) wall. Dwaikat et al. [8]
considered insulated columns for which a part of the insulation
was removed. These conditions result in very large temperature
gradients and temperature jumps. The temperature jumps result in
substantial internal stresses in the column in addition to the
residual stresses caused by rolling or welding. While relaxation of
residual stresses caused by creep is considered in the numerical
studies mentioned, relaxation of the internal stresses due to
temperature jumps is not. This may result in an overestimation of
the effects of the unequal temperature distribution on the buckling
resistance. It may also be the cause of the fact that the numerical
study in [10] shows local buckling deformations at the ultimate
resistance whereas the tests used to validate their model in [9] do
not show these deformations.
Numerical simulations to determine the member temperature are
applied more and more often for re exposed steel structures in order
to remove conservatism that is inherently present in the more simple
calculation methods. Temperature gradients occur in these structures
even if the column is an internal compartment column due to unequal
thickness of webs and anges and especially in case of local res in
which the column is exposed more severely from one side. These
temperature gradients are less severe as compared to gradients in
edge columns and abrupt temperature jumps are only marginal in
magnitude or not expected at all. It is interesting to determine if
existing simple design models for uniform temperature can be applied
for such cases. This paper provides a numerical study into I-shaped
columns with a linear temperature variation over the cross-section.
Table 1
Sections considered in the nite element parameter study.
Section ID Steel grade ha [mm] ba [mm] twa [mm]
S1 S460 492.0 199.4 18
S2 S460 152.4 88.7 4.5
S3 S460 1036.3 308.5 30
S4 S460 241.4 213.9 18.1
S5 S275 492.0 199.4 18
S6 S275 474.6 424.0 47.6
S7 S275 474.6 424.0 47.6
a
h section height, b section width, tw web thickness, tf ange thickness, Iz 2nd moment of area for bending about the weak axis and A section area.
Please cite this article as: Delgado Ojeda O, et al. Fire exposed steel columns with a thermal gradient over the cross-section. Thin-
Walled Structures (2015), http://dx.doi.org/10.1016/j.tws.2015.02.009i
2. Finite element simulations
2.1. Cases considered
Three sets of nite element simulation of rolled I-shaped columns
were carried out:
A set at room temperature consisting of 210 analysed columns.
The simulation results are compared with the design rules in
EN 1993-1-1 [3] in order to validate the nite element models.
A set at elevated temperatures with uniform temperature
consisting of 600 analysed columns. The simulation results
are compared with the results in Talamona et al. [1].
A set at elevated temperatures with a thermal gradient con-
sisting of 120 analysed columns. The simulation results are
compared with the uniform temperature results and with the
design models in standards.
The analysed 7 cross-sections were considered with various
dimensions according to Table 1. The sections are inspired on
standard American sections. Roots between web and anges were
ignored. For each cross-section different column lengths were
considered, covering a practical range of column slenderness. All
columns were restrained against buckling in the strong direction
and were pin-ended for buckling about the weak axis. Two steel
grades were considered. Grade S275 is a steel grade in between
the grades S235 and S355 that were considered in [1]. In addition
grade S460 has been considered in order to determine the effects
of a higher yield stress on the buckling resistance.
It is well known that geometrical imperfections and residual
stresses inuence the buckling resistance. The residual stresses
depend on the geometry of the cross-section. In order to consider
the difference in residual stresses, ve column design models are
considered at room temperature in EN 1993-1-1 depending on the
cross-section, indicted with so-called buckling curves a0, a, b, c
and d. The sections considered in this study are selected in such a
way that all ve buckling curves are covered.
The columns were modelled with eight-node shell elements with
reduced integration using the nite element programme DIANA [16].
Each ange outstand and the web were modelled with 9 nodes
(4 elements) along the cross-section and the elements were approxi-
mately square. Initial geometrical imperfections were applied with a
shape matching the rst Euler buckling mode and an amplitude
equal to L/1000, where L is the column length. This imperfection has
been identied as a sufciently accurate approximation for buckling
resistance calculations and has been used in many other studies
at room and at elevated temperatures, amongst others in [1].
The residual stresses and stressstrain curve considered are depend-
ing on temperature and are provided in the respective sections
hereafter.
tfa[mm] Iza[mm4] Aa[mm2] Buckling curve
32 4.25  107 2.10  104 a0
7.7 8.97  105 2.02  103 a0
54.1 2.67  108 6.28  104 a
30.1 4.92  107 1.67  104 a
32 4.25  107 2.10  104 b
77 9.82  108 8.42  104 c
110 1.40  109 1.11  105 d
 O. Delgado Ojeda et al. / Thin-Walled Structures () 3

Fig. 1. Residual stresses considered in the nite element models. Fig. 2. Results of the analyses at room temperature, provided as buckling factors, ,
as a function of relative slenderness, .

2.2. Room temperature analysis

A bi-linear stressstrain curve was applied in the room tempera-

ture simulations. The residual stress distribution used follows the
Lehigh pattern but with values according to the former Swedish and
Dutch construction standards and is provided in Fig. 1. The minimum
residual stress res,min  120 N/mm2 if h/br1.2 and res,min
70 N/mm2 if h/b41.2. The parameter follows from equilibrium.
The buckling resistances resulting from each analysis are evaluated
in agreement with the design model in EN 1993-1-1 where the
exural buckling reduction factor, , is provided as a function of the
relative slenderness,
Nu;buc
1
N pl
s
N pl
2
FE

where: Nu,buc ultimate buckling resistance, in this case originating

from the nite element simulation; Npl plastic squash load equal to
 y; FE Euler
Af  buckling load, for a pin-ended column equal to
2 EIz =L2 ; Emodulus of elasticity
The analyses results are displayed in Fig. 2. In agreement with
the design model in EN 1993-1-1, the relationship between
relative slenderness and buckling factor depends on the type of
cross-section, where the section classied with buckling curve a0
provides the highest buckling resistance and the resistance drops
for sections classied with buckling curves a, b, c and nally d.
Also from a quantitative point of view the results of the analyses
agree well with the design model in EN 1993-1-1: the coefcient
of determination is equal to R2 0.99. The match is perfect for
sections with buckling curves b, c and d and for slender sections
( 41) with buckling curves a0 and a. For stockier sections with
buckling curves a0 and a, the nite element results are slightly
lower as compared to the design model. This is attributed to the
omission of strain hardening in the nite element models. Overall
Fig. 3. Stressstrain relationships used in the simulations. (a) Stress versus strain.
the results give condence in the nite element models.
(b) Ratio stress/yield stress versus plastic strain.

2.3. Uniform elevated temperature analysis
The elevated temperature simulations were carried out with
material properties at 500 1C and at 700 1C. Round-housed stress
strain relationships in agreement with EN 1993-1-2 were consid-
ered for the columns at elevated temperature, Fig. 3. Because the
effects of creep are already incorporated in these stressstrain
relationships, [17], an explicit creep model was not considered in
the nite element simulations. This implies that the residual stress
pattern incorporated for the room temperature simulations needs
to be modied at elevated temperature because relaxation of
residual stresses, [18], is not explicitly modelled.
Different options exist for modelling residual stresses at elevated
temperature. Some studies considered initial and constant strains

Please cite this article as: Delgado Ojeda O, et al. Fire exposed steel columns with a thermal gradient over the cross-section. Thin-
Walled Structures (2015), http://dx.doi.org/10.1016/j.tws.2015.02.009i
4 O. Delgado Ojeda et al. / Thin-Walled Structures ()
with values equal to those at room temperature. The residual stresses
thus reduce with temperature as the modulus of elasticity. Others
have considered a reduction with temperature as the yield stress so
that the ratio res,min/fy remains unaltered at increasing temperature.
Both approaches do not consider relaxation of residual stresses and
thus overestimate the residual stresses at room temperature. There-
fore, lower residual stresses as the ones discussed before have been
considered in more recent studies. Takagi and Deierlein [19] used a
simplied approach by setting the minimum residual stress at room
temperature at a value of  69 MPa and reducing this stress with
temperature as the yield stress. Yang and Hsu [20] evaluated the
residual stresses in their test specimens as a percentage of the yield
strength. They found that the maximum residual stresses at 500 1C
had dropped to less than 10% of the yield stress. They concluded that
the effect of residual stresses on the column resistance can be
neglected at elevated temperatures.
In the current study, two residual stress cases were considered for
every column assessed at elevated temperature. The rst case repre-
sents an upper bound of the residual stresses to be expected at
elevated temperatures and consists of a minimum residual stress equal
to res,min  70 MPa with the pattern according to Fig. 1. The second
case, representing a lower bound, contains zero residual stresses.
The results of the nite element simulations carried out in the
current study are compared with those provided in [1], which used
beam elements to model column buckling of 180,000 centrally loaded
columns at uniform temperature. Fig. 4 provides the results at 500 1C
of the current study together with the maximum and minimum
values of all results in [1]. The relative slenderness and buckling factors
in the gures are determined with Eqs. (1) and (2), respectively, but
Fig. 4. Results of the analyses at 500 1C (uniform temperature). (a) With residual
stresses. (b) Without residual stresses.
Please cite this article as: Delgado Ojeda O, et al. Fire exposed steel columns with a thermal gradient over the cross-section. Thin-
Walled Structures (2015), http://dx.doi.org/10.1016/j.tws.2015.02.009i
using the elevated temperature properties E and fy, where the latter
is dened as the stress at 2% plastic strain (Fig. 3b). Fig. 3a and b
provides the results of simulations including residual stresses and
without residual stresses, respectively. Four observations are made:
At given relative slenderness, the buckling factor at elevated
temperature is considerably lower as that compared to room
temperature especially in the range 0.2 r r1.5.
The inuence of the section type or buckling curve on the
buckling factor for a given slenderness is negligible at elevated
temperature.
The steel grade does have an inuence on the buckling factor
the higher the steel grade, the higher is the buckling factor.
The results without residual stresses are within the maximum
and minimum limits provided in [1]. The results with residual
stresses are lower than the results in [1].
The rst three observations were already provided in [1]. By a
comparison with additional simulations with elasticplastic
stressstrain curves, we conclude that this is mainly a result of
the curvature of the stressstrain relationship. The difference in
residual stresses between room and elevated temperatures plays
an additional role.
Fig. 5 provides the results of the simulations at 700 1C. The
results of the simulations using S460 are on the lower bound of
the results for S355 in [1] (Fig. 5a). The results using S275 are
lower than this lower bound (Fig. 5a), but appear to be in between
the two bounds for S235 (Fig. 5b). In general, a reasonable
Fig. 5. Results of the analyses at 700 1C (uniform temperature) without residual
stresses. (a) Compared with S355 results in [1]. (b) Compared with S235 results in [1].
 O. Delgado Ojeda et al. / Thin-Walled Structures () 5

agreement is found between the results of the current model and I. Analysis with a temperature gradient along the strong axis.
the results in [1]. II. Analysis with a temperature gradient along the weak axis.

Two temperature distributions were considered for each type

of gradient analysis, Table 2.
2.4. Thermal gradient analysis
The type II gradient analysis is the realistic type of analysis
aimed for in this study. The type I gradient analysis is an articial
The temperature proles considered in the simulations are
case in which thermal bowing was not restrained during heating
taken from a previous CFD analysis in which the temperature of a
but the section was restrained against buckling in the strong
free standing column in a compartment with a localised pool re
direction after heating and before loading. The sections were thus
was determined, [21]. The temperature prole was almost linear
enforced to buckle in about the weak axis despite of thermal
over the cross-section and was further linearised in the nite
bowing and neutral axis shift in the strong direction. With this
element model in order to minimise internal stresses due to the
analysis type we are able to determine the effect of the varying
non-uniform temperature distribution, because these internal
material properties only, i.e. without the inuence of the eccen-
stresses may relax during heating. The coefcient of thermal
tricity caused by the thermal gradient.
expansion according to EN 1993-1-2 was used in the simulations
The results are rst evaluated using the average values of Eav and
and this resulted in thermal bowing of the columns. Two types of
f yav over the cross-section. In Fig. 6, the results of the type I gradient
gradient analysis were carried out:
analyses are compared with the uniform elevated temperature results
in [1]. The buckling factors for the simulations with a thermal gradient
Table 2 along the strong axis are lower than the results of [1] for S355 (Fig. 6a)
Temperature in the types of gradient analysis. and are in line with the results of [1] for S235 (Fig. 6b). Fig. 7 provides
the results of the type II gradient analyses together with the uniform
ID Analysis type Temperature [1C]
temperature results in [1]. The ratio between the buckling factor of the
Average Maximum max  min type I analysis and the buckling factor of the uniform temperature
analysis considered in this study is on average 0.94 and the standard
I-A I 452 490 76
deviation is 0.03. The lower resistance in case of a thermal gradient
I-B I 493 532 78
II-C II 500 635 270
conrms the conclusion in [15] that considering the average tempera-
II-D II 694 785 182 ture in such a case results in an overestimation of the cross-sectional
resistance. For the type II analysis the average value of this ratio is 0.82

Fig. 6. Results of the analyses with thermal gradient along the strong axis Fig. 7. Results of the analyses with thermal gradient along the weak axis compared
compared to the uniform temperature analyses in [1]. (a) Compared to the results to the uniform temperature analyses in [1]. (a) Compared to the results for S355 in
for S355 in [1]. (b) Compared to the results for S235 in [1]. [1]. (b) Compared to the results for S235 in [1].

Please cite this article as: Delgado Ojeda O, et al. Fire exposed steel columns with a thermal gradient over the cross-section. Thin-
Walled Structures (2015), http://dx.doi.org/10.1016/j.tws.2015.02.009i
6 O. Delgado Ojeda et al. / Thin-Walled Structures ()

and the standard deviation is 0.07. In line with [10] we conclude that it
is unconservative to evaluate sections with a thermal gradient over the
cross-section using the same design model as for uniform temperature
and the material properties at the average temperature.
In case of temperature gradients, EN 1993-1-2 [3] indicates that
the buckling resistance should be evaluated using the same design
equations as for uniform elevated temperature, but using material
properties at the maximum temperature throughout the section:
Emax and f ymax . The maximum temperature material properties
introduce a conservatism that is required to counterbalance the
fact that the bending moment caused by thermal bowing and a
shift of the neutral axis is not considered in these design
equations. The standards AISC 360-10 [4] and AS 4100 [22] do
not give provisions for the buckling resistance in case of tempera-
ture gradients; however, because we demonstrated that taking the
average temperature may be unconservative, the design equations
in these standards are also evaluated using the maximum tem-
perature. Fig. 8 provides the comparison of the ultimate resistance
between the type II gradient analysis using FEM, Nu,FEM, and using
the three standards, Nu,standard (vertical axis) as a function of the
ratio between Nu,FEM and the plastic capacity at maximum tem-
perature, N pl;max (horizontal axis). The partial or safety factors are

Fig. 8. Simulations with thermal gradient along the weak axis - ultimate resistance determined by FEM compared with design models in standards. (a) EN 1993-1-2 [3].
(b) AISC 2005 [4]. (c). AS4100 [22].

Please cite this article as: Delgado Ojeda O, et al. Fire exposed steel columns with a thermal gradient over the cross-section. Thin-
Walled Structures (2015), http://dx.doi.org/10.1016/j.tws.2015.02.009i
 O. Delgado Ojeda et al. / Thin-Walled Structures () 7

set to unity in the respective design equations. The graphs indicate
that the design equations in EN 1993-1-2 [3] (Fig. 8a) and AISC
360-10 [4] (Fig. 8b) are (very) conservative and that the design
equations in AS 4100 [22] is sometimes conservative and some-
times unconservative when evaluated on the basis of the maximum
temperature. The bad correlation with the AS 4100 [22] design
equation is mainly attributed to the fact that the reduction of the
modulus of elasticity at elevated temperature is not considered in this
standard. An additional reason is that the material properties in the
nite element simulations are taken from EN 1993-1-2 [3]. Whereas
the material properties in AISC 360-10 [4] are almost identical to those
in EN 1993-1-2 [3], the properties in AS 4100 [22] deviate signicantly.
The coefcients of determination are  0.13, 0.17, and  6.2
for the design equations in EN 1993-1-2 [3], AISC 360-10 [4] and
AS 4100 [22], respectively. In addition, the gure shows that many
simulated columns have a capacity larger than the upper bound
formed by the plastic capacity determined at maximum tempera-
ture, N pl;max . The results indicate that none of the design equations
matches the results of the nite element simulations.
s !
1 235 N=mm2 2
av 1 av av 5
2 fy
where: av coefcient in the buckling equation; fy yield
stress at room temperature; av relative slenderness according
to Eq. (2) with yield stress and modulus of elasticity at average
elevated temperature.
The eccentricity, e in Eq. (3), consists of the sum of the initial
imperfection, ei, thermal bowing eccentricity, eb, and the distance
between the shifted neutral axis and the load application point, ec. As
already indicated, eb and ec are acting in opposite direction and eb
increases with the column length. For short columns ec dominates the
total eccentricity whereas eb dominates in case of long columns. The
initial imperfection, ei, should be applied in that direction so that the
total eccentricity is increased, Fig. 9. An approximate equation is
considered for the eccentricity which is the maximum of eccentricities
e1 and e2 indicated in Fig. 9
e maxe1 ; e2 6

e1 ec 7

e2 eb  ec ei 8
3. Proposed design equation
The thermal bowing eccentricity, eb, is calculated using the coefcient
Alternative design equations have been considered in order to of thermal expansion, , or the length increase factors (L/L)
better match the load bearing resistance resulting from the nite provided in EN 1993-1-9
element simulations of columns with a thermal gradient. These
equations comprise:

a variation of the AISC 2005 equation for normal force and bending
moment interaction as proposed in [23] or modied in [10];
variations on the inelastic exural buckling resistance equation
in [24];
the PerryRobertson equation, [25,26], with a modied expres-
sion for the imperfection value;
a variation on the normal force and bending moment interac-
tion equation for beam-columns at elevated temperatures as
proposed in [1].

The best agreement with the nite element simulations was

found for the latter option. Subsequently a number of simplica-
tions were applied in the calculation of the required parameters in
order to ease the calculation for a practitioner. The simplications Fig. 9. Column eccentricities.
involve the calculation of the required material properties and
geometric parameters at average, maximum or minimum tem-
peratures in the section. The design equation is

Ns eN s
k r1 3
av N pl;av M u;av

where: Ns normal force due to the load; av buckling factor of a

centrally loaded column with material properties at the average
elevated temperature, av, from Eq. (4) eeccentricity from Eq.
(6); k interaction factor from Eq. (10); N pl;av normal force at
yielding with material properties at the average elevated tem-
perature, av.; M u;av bending moment capacity with material
properties at the average elevated temperature, av.
The buckling factor for a centrally loaded column in Eq. (3), av , is
a slightly modied version of the equation provided in EN 1993-1-2.
The modication is applied so that the equation closely matches the
results of the nite element simulations of pin-ended centrally loaded
columns with a uniform temperature as displayed in Figs. 4 and 5

1
av q 4
2 Fig. 10. Comparison between new design model and nite element simulations
av av 2  av with thermal gradient along the weak axis.

Please cite this article as: Delgado Ojeda O, et al. Fire exposed steel columns with a thermal gradient over the cross-section. Thin-
Walled Structures (2015), http://dx.doi.org/10.1016/j.tws.2015.02.009i
8 O. Delgado Ojeda et al. / Thin-Walled Structures ()
   
1   L2 L L
eb L2 max  min  9
8 8b L max L min
The shift of the neutral axis due to the thermal gradient can be
approximated by considering a linear decrease in the modulus of
elasticity from minimum to maximum temperature in the section.
The initial imperfection, ei, is taken as L/1000.
The interaction factor, k in Eq. (3), is a modied version of the
factor provided in Talamona, in order to better match the results of
the current nite element simulations. The equation proposed for
the interaction factor accounts for the fact that the eccentricity is
important for buckling of columns in the mid slenderness range
but not for very slender columns
!2
Ns Ns
k 7 7  0:55
Npl;av N pl;av
The new design Eqs. (3)(10) give a good agreement with the
nite element simulations, Fig. 10. The coefcient of determination
of the new design equations is R2 0.95.
4. Conclusions and recommendations
This paper presents a study into the effects of thermal gradients
over the cross-section on the exural buckling behaviour of re
exposed steel columns. The study shows that a thermal gradient
introduces an eccentricity into the column caused by a shift of the
neutral axis and due to thermal bowing. This eccentricity reduces the
exural buckling resistance of the columns. Design equations in
existing standards ignore this eccentricity and use the material
properties at maximum temperature in order to compensate for
this. The design equations appear to be very conservative in most
cases when compared to the results of nite element simulations. A
new design model is proposed consisting of simple equations which
takes the eccentricity into account. This new model gives a good
agreement with the results of nite element simulations. The study
considered pin-ended I-shaped columns with a linear temperature
gradient buckling around their weak axis. A further study is planned
to widen this scope.
Please cite this article as: Delgado Ojeda O, et al. Fire exposed steel columns with a thermal gradient over the cross-section. Thin-
Walled Structures (2015), http://dx.doi.org/10.1016/j.tws.2015.02.009i
!
References
[1] Talamona D, Franssen JM, Schleich JB, Kruppa J. Stability of steel columns in
case of re: numerical modeling. J Struct Eng ASCE 1997;123(6):71320.
[2] Franssen JM, Talamona D, Kruppa J, Cajot LG. Stability of steel columns in case
of re: experimental evaluation. J Struct Eng ASCE 1998;124:15863.
[3] EN 1993-1-2, Eurocode 3: design of steel structures - part 1-2: general rules
structural re design. Brussels: CEN; 2011.
[4] AISC 360-10. Specication for structural steel buildings. Chicago: American
Institute of Steel Construction; 2010.
[5] Becker R. Structural behavior of simple steel structures with non-uniform
longitudinal temperature distributions under re conditions. Fire Saf J
2002;37:495515.
[6] Talamona D. Numerical study of the re resistance of steel columns in axial
compression and uniform bending. In: Proceedings of the structures congress,
ASCE, 2009. p. 650659.
[7] Xu L, Zhuang Y. Storey stability of unbraced steel frames subjected to non-
uniform elevated temperature distribution. Eng Struct 2014;62:16473.
[8] Dwaikat MMS, Kodur VKR, Quiel SE, Garlock MEM. Experimental behavior of
steel-beam columns subjected to re induced thermal gradients. J Constr Steel
10
Res 2011;67:308.
[9] Choe L. Structural mechanics and behavior of steel members under re
loading. Ph.D. Thesis. Lafayette: Purdue University; 2011.
[10] Agarwal A, Choe L, Varma AH. Fire design of steel columns: effects of thermal
gradients. J Constr Steel Res 2014;93:10718.
[11] Moura Correia AJP, Rodrigues JPC, Vila Real P. Thermal bowing on steel
columns embedded on walls under re conditions. Fire Saf J 2014;67:5369.
[12] Cai J, Feng J. Thermal buckling of rotationally restrained steel columns. J Constr
Steel Res 2010;66:83541.
[13] Cai J, Feng J, Zhang J. Thermoelastic buckling of steel columns with load-
dependent supports. Int J Nonlinear Mech 2012;47:815.
[14] Garlock MEM, Quiel SE. Mechanics of wide-anged steel sections that
developthermal gradients due to re exposure. Int J Steel Struct KSSC
2007;7:15362.
[15] Garlock MEM, Quiel SE. Plastic axial load and moment interaction curves for
re exposed steel sections with thermal gradients. J Struct Eng ASCE
2008;134:87480.
[16] Manie J. DIANA User's manual release 9.4.4. Delft: TNO DIANA BV; 2012.
[17] Twilt L. Strength and deformation properties of steel at elevated temperatures:
some practical implications. Fire Saf J 1988;13:915.
[18] Yang K-C, Lee H-H, Chan O. Experimental study of re-resistant steel H-
columns at elevated temperature. J Constr Steel Res 2006;62:54453.
[19] Takagi J, Deierlein GG. Strength design criteria for steel members at elevated
temperatures. J Constr Steel Res 2007;63:103650.
[20] Yang K-C, Hsu R. Structural behavior of centrally loaded steel columns at
elevated temperature. J Constr Steel Res 2009;65:20628.
[21] Boon W. The Inuence of thermal gradients in steel columns due to pool res.
Thesis. Delft University of Technology, Delft; 2014.
[22] AS 4100. Australian Standards: Steel structures. Sydney: Standards Australia; 1998.
[23] Dwaikat MMS, Kodur VKR. Strength design criteria for steel beam-columns
with re induced thermal gradients. Eng J AISC 2011;48:12740.
[24] Maljaars J, Twilt L, Soetens F. Flexural buckling of re exposed aluminium. Fire
Saf J 2009;44:7117.
[25] Ayrton WE, Perry J. On struts. Engineer 1886;62:464-5513-5.
[26] Robertson A. The strength of struts. Inst Civ Eng 1925.