You are on page 1of 9

STRENGTH AND STABILITY OF STEEL FRAMES IN FIRE:

RANKINE APPROACH
By W. S. Toh,1 K. H. Tan,2 and T. C. Fung3

ABSTRACT: A simple analytical approach based on the Rankine principle has been developed to determine the
ultimate resistance of steel frames in fire. The proposed Rankine approach gives an approximation of the frames’
fire resistance through a simple interaction between two idealized structural behaviors—strength and stability.
Here, the strength and stability of the structures are evaluated using the rigid-plastic and the elastic buckling
analyses, both incorporating the thermal effects. The proposed approach is first verified using a finite-element
model. The verification studies include the effects of column and frame slenderness ratios, beam-column stiffness
ratio, steel grades, initial sway imperfections, and initial residual stresses. These studies indicate that frame
Downloaded from ascelibrary.org by New York University on 05/13/15. Copyright ASCE. For personal use only; all rights reserved.

slenderness ratio is an important parameter governing the behavior of simple frames in fire, and the performance
of the proposed approach is related to it. The Rankine approach is then applied to a series of 18 test frames
from the literature. It is shown that the proposed approach can provide predictions with sufficient accuracy for
simple frames in fire.

INTRODUCTION ical load factor ␭c. The evaluations of ␭p and ␭e are more
Among the various techniques for analyzing the behavior involved for frames under thermal effects, compared to frames
of steel frames subjected to elevated temperatures, numerical under normal conditions. Thus, attention is drawn to the ther-
modeling, in particular the finite-element method, has been mal effects in the plastic collapse and the elastic buckling anal-
widely used by researchers since the 1970s (Cheng and Mak yses. Basically, the thermal effects leading to the deterioration
1975; Furumura and Shinohara 1978; Rubert and Schaumann of material properties and additional axial compressive stresses
1986; Najjar and Burgess 1995). In general, numerical mod- in the system are the main concern. In the plastic analysis, the
eling requires significant computational effort. For practical steel yield strength, and therefore the plastic moment capacity
design purposes, a simple analytical tool to ascertain the ul- of a frame, reduce with increasing temperatures. In addition,
timate resistance and the failure modes of steel frames in fire with degradation of yield strength, the members’ axial force
is greatly favored. However, unlike the various design for- has a more pronounced effect on the sections’ plastic moment
mulas for isolated steel columns in fire (Commission 1995; of resistance. In the elastic buckling analysis, the reduction in
Talamona et al. 1997), to date, simple approaches are still not the steel elastic modulus with increasing temperatures, to-
available for steel frames in fire. On the other hand, simple gether with the additional compressive stresses arising from
approaches for frames at ambient temperature have been es- the thermal restraints (if any), lead to a reduction in the mem-
tablished since the 1950s; the most famous one is the Mer- bers’ stiffness, causing the system to become unstable.
chant-Rankine formula. Based on the empirical formula orig- This paper first outlines the determinations of ␭p and ␭e
inally proposed by Rankine in 1866 for perfectly straight incorporating the thermal effects, and the fundamental of the
columns, Merchant (1954) developed a more generalized for- FEM. This is followed by the verification study of the Rankine
mula, called the Merchant-Rankine formula, to determine the approach with the FEM, through a series of parametric studies.
failure load factor of rigid-jointed frames subjected to propor- These parameters comprise frame slenderness ratio, beam-col-
tional loading at ambient temperature. Later, Horne (1960, umn stiffness ratio, steel grade, initial sway imperfection, and
1963) further investigated the fundamental principles and ap- residual stresses. For the purpose of direct hand calculation in
plications of the Rankine approach. Although the Rankine ap- design, this study focuses on frames with only small thermal
proach is empirical, based on mathematical assumptions, restraint, in which the induced thermal stresses are insignifi-
Horne managed to give a formal proof to establish the rela- cant compared to the stresses due to the applied working loads.
tionship between the idealized elastic and rigid-plastic behav- In the frames studied, the magnitude of thermal stresses is
iors, and the actual nonlinear elastic-plastic behavior. around 5% of applied working stresses and can be ignored.
In this study, the Rankine approach is extended to frames Through the verification study, the basic structural response of
subjected to thermal effects. The proposed approach is an ex- simple steel frames due to the thermal effects can be better
tension of the Rankine formula for steel columns in fire, as understood. In addition, the reliability and the limits of the
addressed in a previous paper by Toh et al. (2000); it deals proposed Rankine approach can be established. It is found that
with rigid-jointed plane frames, subjected to uniform heating. the frame slenderness ratio, denoted by ⌳T (=␭p /␭e), is an im-
The proposed approach allows the interaction between the two portant parameter governing the behavior of steel frames under
idealized load factors for rigid-plastic ␭p and elastic buckling thermal effects. Moreover, ⌳T also relates to the performance
analyses ␭e, to determine an approximation to the actual crit- of the proposed Rankine approach. Lastly, the proposed ap-
proach is verified with 18 actual test frames. Through both
1
Res. Fellow, Nanyang Technol. Univ., School of Civ. Struct. Engrg., numerical and experimental verifications, it is shown that the
BLK N1, #1A-37, Nanyang Ave., Singapore 639798. proposed Rankine approach provides sufficiently accurate pre-
2
Assoc. Prof., Nanyang Technol. Univ., School of Civ. Struct. Engrg., dictions for steel frames in fire.
BLK N1, #1A-37, Nanyang Ave., Singapore 639798.
3
Assoc. Prof., Nanyang Technol. Univ., School of Civ. Struct. Engrg.,
BLK N1, #1A-37, Nanyang Ave., Singapore 639798.
PROPOSED RANKINE APPROACH
Note. Associate Editor: Peter Hoadley. Discussion open until Septem- Assumptions
ber 1, 2001. To extend the closing date one month, a written request must
be filed with the ASCE Manager of Journals. The manuscript for this The proposed approach makes the following assumptions:
paper was submitted for review and possible publication on February 22,
2000; revised December 19, 2000. This paper is part of the Journal of
Structural Engineering, Vol. 127, No. 4, April, 2001. 䉷ASCE, ISSN 1. Temperature distribution within a member is uniform.
0733-9445/01/0004-0461–0469/$8.00 ⫹ $.50 per page. Paper No. 22268. 2. Members are perfectly straight, isotropic, and prismatic.
JOURNAL OF STRUCTURAL ENGINEERING / APRIL 2001 / 461

J. Struct. Eng. 2001.127:461-469.


3. Members buckle in the plane of the frame only.
4. Local and lateral torsional buckling are not taken into
account.
5. Applied working loads are concentrated and propor-
tional.

Rankine Formula
At each particular temperature T, the proposed Rankine ap-
proach allows the strength and the stability aspects of a steel
frame to be separately determined. The rigid-plastic collapse
load factor ␭p is used to define the frame strength, while the
FIG. 2. Rankine Load Factor versus Frame Slenderness Ratio
elastic buckling load factor ␭e governs the frame stability. For
a frame subjected to proportional loading, the interaction of
these two idealized load factors gives rise to the critical load governed by stability. Second, plotting the ratio ␭c(T )/␭p(T )
versus ⌳T for (1) generates a unique Rankine curve as illus-
Downloaded from ascelibrary.org by New York University on 05/13/15. Copyright ASCE. For personal use only; all rights reserved.

factor ␭c at temperature T, as follows:


trated in Fig. 2, for any frames regardless of geometric prop-
1 1 1 erties, loading conditions, and critical temperatures attained.
= ⫹ (1) Based on the Rankine curve, the Rankine predictions can then
␭c(T ) ␭p(T ) ␭e(T )
be compared with FEM or test results.
Alternatively, if the critical temperature Tc is of interest
while the applied working loads remain constant [i.e., ␭c(Tc) Rigid-Plastic Collapse Load Factor
= 1.0], a trial-and-error procedure is necessary to determine Tc According to the plastic theorems derived for fire situations
such that the sum of 1/␭p(Tc) and 1/␭e(Tc) is equal to unity. (Tan et al. 1999), the collapse mechanism of a structure occurs
Fig. 1 illustrates the variation of ␭c versus T for a frame at when the three conditions of equilibrium, mechanism, and
elevated temperatures. Clearly, at any T < Tc, the determined yield are satisfied at a particular temperature. The correspond-
critical load factor is greater than unity, i.e., ␭c(T ) > 1.0. Thus, ing load factor is the rigid-plastic collapse load factor ␭p. In
the frame remains safe under its working loads. However, the rigid-plastic analysis, elastic deformations of the members
when T is increased to Tc, the frame turns into a state of in- are neglected, and plastic deformations are assumed to be con-
stability even under its working loads, i.e., ␭c(T ) = 1.0, and centrated at plastic hinge locations. There are several ways of
partial or overall collapse may occur. At any T > Tc, the critical obtaining ␭p for frames at elevated temperatures. For simple
load factor will be less than unity, i.e., ␭c(T ) < 1.0; the frame rigid frames consisting of horizontal beams and vertical col-
can be regarded as unsafe, as its stiffness and strength have umns only, all possible collapse mechanisms can be easily vi-
deteriorated to such an extent that the frame cannot even sus- sualized. The direct method of combined mechanisms is the
tain the working loads. most convenient manual way of deriving ␭p (Tan et al. 1999).
The simple interaction relationship among ␭c, ␭p, and ␭e in It is well known that the axial force effect on the plastic
the Rankine approach suggests that it is possible to derive a moment of resistance Mp of a member is usually negligible
general parameter to categorize the structural response of steel when the axial stress is small compared to the yield stress.
frames under thermal effects. Conveniently, the expression of However, at elevated temperatures, due to the degradation of
(1) can be rewritten as follows: yield strength, the effect of axial stress can be significant and
␭c(T ) 1 cannot be neglected in the determination of Mp. The axial force
= (2) of a member must be known a priori before determining its
␭p(T ) 1 ⫹ ⌳T
Mp. For fully heated frames with small thermal restraints, it
where the ratio ␭c(T )/␭p(T ) shows the extent to which the crit- can be assumed that the member axial forces at temperature T
ical load factor ␭c(T ) falls below the rigid-plastic collapse load are proportional to those values obtained from the ambient
factor ␭p(T ); and ⌳T = ␭p(T )/␭e(T ) is defined as the frame analysis, multiplied by the reduction factor for material yield
slenderness ratio, relating the plastic capacity to the stability strength. Based on this assumption, one need not perform te-
limit. dious computations to obtain the exact value of member axial
The introduction of the frame slenderness ratio ⌳T for frame forces corresponding to the actual state of deformation. How-
analysis has two advantages. First, the thermal effects on the ever, for more complex problems, for instance, partially heated
behavior of steel frames in fire can be fully incorporated by a frames or frames with significant thermal restraints, the entire
frame slenderness parameter. Here, a frame is classified as plastic analysis has to be performed to yield the exact ␭p(T )
stocky and its fire resistance is controlled by the strength cri- with respect to temperature T. In this case, the first-order elas-
terion when ⌳T is smaller than unity. Conversely, a slender tic-plastic hinge method proposed by Tan et al. (1999) can be
frame is one with ⌳T greater than unity and its behavior is used.
Elastic-Buckling Load Factor
According to the elastic buckling theory, the elastic insta-
bility of rigid-jointed frames subjected to both vertical and
lateral loads is associated with the effects of change of ge-
ometry and sway deflections on the equilibrium conditions.
Therefore, a realistic evaluation of the elastic buckling load
factor ␭e must incorporate these effects. For frames with small
thermal restraints, induced thermal stresses can be ignored. For
hand calculation purposes, the conventional stability functions
analysis is adopted here with modifications on the elastic mod-
ulus E; the member stiffness influence coefficients are ex-
FIG. 1. Variation of Critical Load Factor with Respect to Tem- pressed in terms of (1) the stiffness factor s; and (2) the car-
perature ryover factor c (Horne and Merchant 1965).
462 / JOURNAL OF STRUCTURAL ENGINEERING / APRIL 2001

J. Struct. Eng. 2001.127:461-469.


For a uniformly heated member of length L, s and c are
functions of ␳ = P/PE (T ), where P is the member axial force,
and PE (T ) = ␲2E(T )I/L 2 is the Euler load at temperature T,
giving the following expressions:
For compressive axial force (positive P)
(1 ⫺ 2␣ cot 2␣)␣
s= (3a)
tan ␣ ⫺ ␣
2␣ ⫺ sin 2␣ ␲
c= in which ␣ = 兹␳ (3b)
sin 2␣ ⫺ 2␣ cos 2␣ 2
FIG. 4. Reduction Factors for Steel fy and E at Elevated Tem-
For tensile axial force (negative P) peratures
(1 ⫺ 2␥ coth 2␥)␥
Downloaded from ascelibrary.org by New York University on 05/13/15. Copyright ASCE. For personal use only; all rights reserved.

s= (4a)
tanh ␥ ⫺ ␥ assumed to be linearly elastic which is only governed by E.
2␥ ⫺ sinh 2␥ ␲ In this study, the variations of E and fy with respect to elevated
c= in which ␥ = 兹⫺␳ (4b) temperatures comply with EC3-1.2 (Commission 1995). The
sinh 2␥ ⫺ 2␥ cosh 2␥ 2
code reduction factors for the slope of the linear elastic range
where E(T ) = kE (T )E = elastic modulus of the material at kE (T ), and the effective yield strength ky(T ) are applied to E
elevated temperatures; kE (T ) = reduction factor of the elastic and fy, respectively. The values of kE (T ) and ky(T ) with respect
modulus E; and I = section moment of inertia. to elevated temperatures are shown in Fig. 4.
According to the notations shown in Fig. 3, the equilibrium
equations of a member can then be expressed as follows:
FINITE-ELEMENT MODEL

A numerical model for simulating the nonlinear inelastic


behavior of plane-framed structures under thermal effects has
been developed to verify the proposed Rankine approach (Toh
2000). The proposed model is a displacement-based incremen-
tal FEM, comprising the assemblage of corotational beam el-
ements in a 2D plane, and taking into account both the ma-
terial and geometric nonlinearities. The corotational beam
element is derived from the Green strain and the second Piola-
Kirchhoff stress, based on the total Lagrangian formulation.
Geometric nonlinearities are modeled with a closed-form ge-
ometric stiffness matrix, while material nonlinearities are mod-
eled with the EC3-1.2 (Commission 1995) steel stress-strain
relationships in which the creep effect is implicitly incorpo-
where A = cross-sectional area; ␾4 = s; ␾3 = sc; ␾2 = s(1 ⫹ rated. The stress and strain states of an element are uniaxial
c); and ␾1 = 2␾2 ⫺ ␲2␳. and evaluated by means of a section discretization at the ele-
For frames with significant thermal restraints, induced ther- ment nodes in the form of a fiber model. The section discret-
mal stresses must be considered. This entails iterative trials ization technique also allows the consideration of temperature
and errors, and is not amendable to hand calculations. In this gradients across sections and initial residual stresses in the
case, the second-order elastic-plastic hinge method proposed analyses. However, along the member length, the temperature
by Toh (2000) can be modified to perform the elastic buckling distribution is constant. The Newton-Raphson iteration method
analysis. incorporating a simple temperature increment procedure is
adopted to deal with the nonlinearities.
Material Models The proposed FEM has been validated against a series of
published test and analytical results on various steel structures
The elastic modulus E and the yield strength fy are the two under thermal effects (Furumara and Shinohara 1978; Rubert
dominant parameters in the proposed Rankine approach. In the and Schaumann 1986; Franssen et al. 1994; Talamona 1995).
rigid-plastic analysis, the steel stress-strain relationship is as- These structures include isolated beams and columns, and sim-
sumed to be perfectly plastic which is only governed by fy. In ple portal frames. Generally, the FEM provided satisfactory
the elastic-buckling analysis, the stress-strain relationship is predictions of both ultimate fire resistance and deformations.

VERIFICATION STUDY

In this section, the accuracy of the proposed Rankine ap-


proach is examined against the FEM predictions. This is to
establish the range of application and the limitations of the
Rankine approach. Only simple portal frames with small ther-
mal restraints are considered, viz one braced frame and one
sway frame, as shown in Fig. 5. It is assumed that the whole
frames are subjected to a uniform heating regime and are made
of hot-rolled I-sections. Unless otherwise specified, the frame
members are assumed to be perfectly straight and initial resid-
FIG. 3. Local Elements: (a) Forces; (b) Displacements ual stresses are not considered.
JOURNAL OF STRUCTURAL ENGINEERING / APRIL 2001 / 463

J. Struct. Eng. 2001.127:461-469.


Geometric Properties
For frames subjected to static loading, the parameters influ-
encing the structural behavior can be categorized into the load-
ing conditions and the geometric properties. The loading con-
ditions consist of the intensity of lateral and gravity loads, the
ratios of lateral to gravity loads, and the loading patterns, etc.
The geometric properties consist of member length, section
properties, support conditions, column slenderness ratios Lc /r,
and beam-column stiffness ratios ␤, etc. The beam-column
stiffness ratio ␤ of a joint in a frame is defined as the relative
stiffness of the beams to the columns about the joint, as fol-

冘 冘
lows:
kb (Eb(T )Ib /Lb)
␤=
冘 冘
Downloaded from ascelibrary.org by New York University on 05/13/15. Copyright ASCE. For personal use only; all rights reserved.

= (6)
kc (Ec(T )Ic /Lc)
where I and L = section moment of inertia and the member
length, respectively; and subscripts b and c represent beam and
column, respectively. The ␤ value is relatively small for the
weak-beam strong-column design, and vice versa for the
strong-beam weak-column design.
In this paper, the effects of the geometric properties and the
loading conditions on the fire resistance of steel frames are
separately investigated. Furthermore, the effects of frame slen-
derness ratios ⌳T, steel grades, initial sway imperfections, and
residual stresses are also investigated.

Two-Member Braced Frame


To illustrate some aspects of the interaction between a beam
and a column in a frame under thermal effects, a simple two-
member braced frame in Fig. 5(a) is analyzed in the first series
of verification study. Two concentrated loads are applied at the
beam midspan and the column top, respectively, in which ␥
symbolizes the ratio of the beam load to the column load. The
steel fy and E at ambient temperature are taken as 235 N/mm2
and 210 kN/mm2, respectively.
For the frame type and the loading pattern considered, it is
the strength rather than the stability that governs the ultimate
FIG. 6. Comparison of FEM and Rankine Predictions for
resistance. By using the proposed Rankine approach, the ␭p is
Braced Frames: (a) Tc ⴝ 20ⴗC; (b) 100ⴗC ⱕ Tc ⱕ 400ⴗC; (c) Tc
determined by the beam mechanism, with one plastic hinge above 400ⴗC
forming at the beam midspan, and another at the joint. For the
latter, it may form on either the column or the beam section,
depending on the respective Mp values. In the elastic buckling from 0.25 and 0.50 to 0.75. Also, the frames bending about
analysis, the primary buckling mode is the column single-cur- the major and minor axes are separately considered in the anal-
vature bending mode. Due to the pinned-pinned support con- yses. As a result, a total of 36 test frames are considered in
ditions and the translational movement at the joint, the thermal this verification study. For each frame, the critical load factors
stresses induced in the frame are relatively small (<5%) com- at 20, 100, 200, and 800⬚C are determined using both the
pared to the stresses from the applied loads. Hence, the thermal proposed Rankine approach and the FEM. Plotting the ratio
stresses can be ignored in the determinations of the load fac- ␭c(T )/␭p(T ) versus ⌳T, the FEM predictions for all 36 braced
tors ␭p and ␭e. frames are presented in Fig. 6(a) for ambient temperature, and
The analyses are divided into two main groups according in Figs. 6(b and c) for elevated temperatures. It is noteworthy
to the beam-column stiffness ratios ␤ of 1.0 and 1/3, respec- that each point in the figures represents one frame result at a
tively. Within each main group, the frames are further divided particular temperature, with ␭c(T ) obtained from the FEM
into three subgroups according to the column slenderness ra- analysis, and ␭p(T ) and ␭e(T ) obtained from the plastic and
tios Lc /r of 30, 60, and 120, with the load ratios ␥ ranging elastic buckling analyses, respectively. The Rankine curve of
(2) is also shown in each figure for comparison. The analyses
show that the ⌳T vaues for all test frames are smaller than
unity, indicating that the strength criterion governs the ultimate
resistance.
Clearly, at ambient temperature, the Rankine curve forms a
lower bound to the FEM results [Fig. 6(a)]; actual ␭c from the
FEM are close to ␭p, regardless of the ⌳T values. This indicates
that the frames’ behavior is dominated by the strength and is
hardly affected by the second-order effects of instability. Un-
derestimation of ␭c for all frames suggests that the actual sec-
ond-order effects are much smaller than the Rankine estimates.
FIG. 5. Simple Portal Frames: (a) Braced; (b) Sway Nevertheless, the Rankine approach generates satisfactory pre-
464 / JOURNAL OF STRUCTURAL ENGINEERING / APRIL 2001

J. Struct. Eng. 2001.127:461-469.


diction when the ⌳T value is small enough. An accuracy of
below 10% error can be achieved for the minor and major
axes frames when ⌳T ⱕ 0.20 and ⌳T ⱕ 0.15, respectively.
For 100⬚C ⱕ Tc ⱕ 400⬚C, the Rankine predictions have
greater disparity with the FEM values. The gradual transition
from underestimation to overestimation of the predicted criti-
cal temperatures leads to scattering of data above the Rankine
curve in Fig. 6(b). It is noteworthy that in Figs. 6(b and c),
the FEM predictions of each frame for both axes of bending
are close to one another, and thereby grouped together. For Tc
above 400⬚C, the Rankine approach tends to overestimate the
ultimate resistance of the braced frames [Fig. 6(c)]. However,
the overpredictions are seldom >10% of the FEM values. For
␤ = 1.0, the accuracy of the Rankine predictions gradually
reduces with increasing ⌳T. For ␤ = 1/3, the predictions are
Downloaded from ascelibrary.org by New York University on 05/13/15. Copyright ASCE. For personal use only; all rights reserved.

very consistent and always fall between the zero and ⫹10%
error curve, for the whole range of ⌳T considered. This implies
that the Rankine approach yields more accurate predictions for
braced frames with weak-beam strong-column design when Tc
ⱖ 400⬚C.

One-Story Single-Bay Sway Frame


To demonstrate the applicability of the Rankine approach to
sway frames under thermal effects, consider the one-story sin-
gle-bay frame in Fig. 5(b). The lateral load of 0.01F is a small
disturbing force, which is equivalent to the initial sway im-
perfection required by EC3-1.1 (Commission 1992). The steel
fy and E are taken as 460 N/mm2 and 210 kN/mm2, respec-
tively.
In the Rankine approach analysis, the failure mode of the
frame is associated with sidesway instability. Fig. 7(a) illus-
trates the collapse mechanism corresponding to ␭p, in which
two plastic hinges form in the columns; the effect of axial
force on the columns Mp is the main concern. Similarly, Fig.
7(b) illustrates the sway mode of buckling corresponding to
the lowest ␭e, in which the joint rotations ␪1 and ␪2 are iden-
tical. It is noteworthy that the buckling resistance of this sway
frame is mainly contributed to by the combined rotational stiff-
ness of the beam-column joints. Hence, the beam-column stiff- FIG. 8. Comparison of FEM and Rankine Predictions for Sway
Frames: (a) Tc ⴝ 20ⴗC; (b) 100ⴗC ⱕ Tc ⱕ 400ⴗC; (c) Tc above 400ⴗC
ness ratio ␤ is the controlling parameter. Provided all the other
geometric properties remain constant, the larger the ␤ value,
the greater the sway resistance. Due to small thermal restraints, frame, than a minor-axis frame of the same geometric prop-
the thermal stresses can be ignored in the calculations of the erties. The conservative estimates are approximately larger
load factors ␭p and ␭e. than 20% for ⌳T < 2.0. That ⌳T having values exceeding unity
In this study, the frames of three ␤ values are separately implies that this study covers a wide range of sway frames,
considered, i.e., ␤ = 0.5, 1.0, and 2.0, respectively. For each including both the stocky frames governed by strength (0 <
␤ value considered, six different column slenderness ratios ⌳T < 1) and the slender frames governed by stability (1 < ⌳T
Lc /r of 10, 20, 30, 40, 50, and 60 are assigned. The frames < 5). Figs. 8(b and c) summarize the comparisons of the Ran-
bending about the major and minor axes are separately con- kine and the FEM results of the analyses with thermal effects.
sidered in the analyses as well. In total, there are 3 ⫻ 6 ⫻ 2 Focusing on Tc above 400⬚C, it shows that for most of the
= 36 frames being analyzed. For each frame, the critical load frames over the range of ⌳T considered, the Rankine predic-
factors corresponding to temperatures ranging from 20⬚C to tions are in good agreement with the FEM results, and many
800⬚C are determined using both the Rankine approach and comparisons are within ⫾10% error [Fig. 8(c)]. With the ex-
the FEM. ception of frames with ␤ = 2.0 and ⌳T ⱕ 1.0, the errors exceed
Fig. 8(a) summarizes the results of the frame analyses at 10% of FEM values. In short, the comparison study suggests
ambient temperature. Clearly, the Rankine approach produces that the Rankine approach is reliable in predicting the ultimate
a more conservative critical load factor ␭c for a major-axis resistance of sway frames at Tc above 400⬚C, with the excep-
tion of stocky sway frames with strong-beam weak-column
design. These conclusions are drawn for frames made of
I-sections with Grade S460 steel. The effects of different steel
grades are investigated in the following.

Effect of Steel Grade


The effects of steel grade on the Rankine approach perfor-
mance are now investigated through two braced and two sway
FIG. 7. Failure Modes: (a) Sway Collapse Mechanism; (b) frames. Assuming Grades S235 and S460 steel, respectively,
Sway Buckling Mode the FEM and the Rankine approach are employed to determine
JOURNAL OF STRUCTURAL ENGINEERING / APRIL 2001 / 465

J. Struct. Eng. 2001.127:461-469.


Consider the sway frame (diagram insert in Fig. 11) sub-
jected to two vertical loads F of 600 kN on the columns. The
sections and the dimensions are chosen so that ␤ = 1.0 and
Lc /r = 40. According to EC3-1.1, the recommended ␾ value
is therefore 1/200, and the equivalent lateral load is 6 kN. The
steel fy and E are taken as 460 N/mm2 and 210 kN/mm2, re-
spectively. Assuming the entire frame is subjected to uniform
heating, the FEM is used to analyze the structural behavior
prior to the critical temperature. For comparison purposes, the
␾ values of 1/100 and 1/50 are also considered. Fig. 11 shows
the temperature-displacement relationships for the horizontal
sway displacement at the top of the right column correspond-
ing to the ␾ values considered. The presence of initial sway
imperfections induces second-order P-⌬ effects to the col-
Downloaded from ascelibrary.org by New York University on 05/13/15. Copyright ASCE. For personal use only; all rights reserved.

umns, and thereby hastens the occurrence of sway instability.


Therefore, the sway deflection of the frames increases with
increasing ␾ value. It can be seen that the sway deflection
increases significantly when the ␾ value increases from 1/200
to 1/50. However, the increase of ␾ has little effect on the
ultimate fire resistance. The difference between the maximum
and the minimum Tc is only 37.1⬚C, which merely accounts
for a reduction of 6.0% in the ultimate fire resistance.
Next, consider two sway frames of the same ␤ value of 0.5,

FIG. 9. Effect of Steel Grade on Rankine Predictions: (a)


Braced Frames; (b) Sway Frames

the critical load factors for these frames corresponding to the


temperatures ranging from 20⬚C to 800⬚C.
First, consider two braced frames [Fig. 5(a)]—Frame 1: ␤
= 1/3, Lc /r = 30, ␥ = 0.75, and bending about the major axis;
and Frame 2: ␤ = 1.0, Lc /r = 60, ␥ = 0.25, and bending about
the minor axis. Fig. 9(a) shows that the Rankine approach
FIG. 10. Initial Sway Imperfections According to EC3-1.1: (a)
yields more conservative results for frames of higher steel Equivalent Geometric Imperfection; (b) Equivalent Horizontal
grade. In the Rankine approach, the steel grade only affects Forces
the rigid-plastic collapse load factor ␭p; the higher the steel
grade, the larger the value of ␭p. Therefore, Grade S460 frames
give larger ⌳T values than Grade S235 frames. This results in
the right shift of the FEM predictions for the higher steel-grade
frames as illustrated by the trend lines A and B in Fig. 9(a).
Hence, it can be concluded that the steel grade mainly affects
the frame slenderness and thereby the accuracy of the Rankine
predictions. These effects are similar for frames bending about
the major or minor axes.
Second, consider two sway frames [Fig. 5(b)] bending about
the major axis—stocky frame with ␤ = 0.5 and Lc /r = 20; and
slender frame with ␤ = 1.0 and Lc /r = 60. For both frames,
the effect of the steel grade on ⌳T is very obvious; replacing
Grade S460 with Grade S235 will reduce the ⌳T values by a
factor of 235/460. This results in the left shift of the FEM
predictions relative to the Rankine curve, as illustrated by the
FIG. 11. Effect of Initial Sway Imperfection on Sway Frame
trend lines A and B in Fig. 9(b). Compared to the slender
frames, the stocky frames are more sensitive to the steel grade
as the accuracy of the Rankine approach is significantly af-
fected by changing the steel grade. Normally, the Rankine pre-
dictions for higher steel grade are more conservative than the
predictions for lower steel grade, for both stocky and slender
sway frames.
Effect of Initial Sway Imperfection
Eurocode 3-Part 1.1 (Commission 1992) requires explicit
modeling of initial sway imperfections ␾ for sway frames
analysis under normal conditions, using either the equivalent
geometric imperfection or the equivalent forces, as shown in
Figs. 10(a and b), respectively. The equivalent force approach
conveniently allows the initial sway imperfection to be in- FIG. 12. Effect of Initial Sway Imperfection on Rankine Predic-
cluded in the Rankine approach. tions

466 / JOURNAL OF STRUCTURAL ENGINEERING / APRIL 2001

J. Struct. Eng. 2001.127:461-469.


bending about the major axis; one frame is stocky with Lc /r conditions, the frame slenderness ⌳T is equal to 2.10 at am-
= 20, while another is slender with Lc /r = 40. The steel fy and bient temperature using the Rankine approach, i.e., the frame
E are taken as 460 N/mm2 and 210 kN/mm2, respectively. is slender. Assuming the entire frame is subjected to uniform
Assuming the ␾ values of 1/50, 1/100, and 1/200, respectively, heating, the FEM analyses show that the frame behavior is
the FEM and the Rankine approach are then used to give the hardly affected by the presence of residual stresses. It should
critical load factors corresponding to the temperatures ranging be noted that the critical temperature of a frame with 0.5 fy
from 20⬚C to 800⬚C. The FEM analyses show that the slender residual stresses is only reduced by 1.7% or 10.4⬚C, compared
frames are more vulnerable to initial sway imperfection than to the one with no residual stresses. A similar response can be
the stocky frames. In the Rankine approach however, initial observed in the sway frame of the same geometric and material
sway imperfections, represented by the equivalent forces, will properties and loading pattern, but bending about the minor
only affect the ␭p of sway frames; hence, their effects are more axis. Fig. 13(b) shows the deflection curves corresponding to
significant for stocky frames. This difference leads to the rel- different residual stress magnitudes for the minor axis frame.
atively wide spread of data corresponding to various ␾ values Similarly, the residual stresses have little effect on the ultimate
for the slender frames, in comparison to the stocky frames resistance of the sway frame. Once again, the reduction in
critical temperature Tc corresponding to an increment of ␴residual
Downloaded from ascelibrary.org by New York University on 05/13/15. Copyright ASCE. For personal use only; all rights reserved.

predictions, as shown in Fig. 12. In both cases, the FEM pre-


dictions shift rightward relative to the Rankine curve with in- from zero to 0.5 fy, is only 2.9% or 17.7⬚C. According to the
creasing ␾. It means that the Rankine prediction for a frame plastic theorems, the strength of a structure is independent of
with a smaller ␾ value is more conservative relative to a larger residual stresses. However, this is not true for stability, as re-
␾ value. sidual stresses will reduce the structural stiffness. Therefore,
it is expected that residual stresses have a smaller effect on
Effect of Initial Residual Stresses stocky frames than on slender frames.
Due to uneven cooling during manufacturing, initial residual Placed in a broader context, the residual stresses in com-
stresses are present in hot-rolled I-sections. Generally, com- parison to initial sway imperfections, have insignificant effect
pressive residual stress exists at flange tips and at middepth of on the ultimate resistance of steel frames under fire conditions.
the web, while tensile residual stress exists in the flanges and Hence, although the Rankine approach does not consider re-
the web-flange junctions. In this study, bitriangular distribu- sidual stresses, this omission does not adversely affect its per-
tions of residual stresses (␴residual) with a maximum magnitude formance.
of 0.3 fy and 0.5 fy at ambient temperature are assumed, re- COMPARISON WITH TEST RESULTS
spectively. The residual strains corresponding to these residual
stress distributions are kept constant during the analyses at In this section, the proposed Rankine approach is verified
elevated temperatures. Consequently, the magnitude of resid- against the results of the 18 portal frames tested by Rubert
ual stresses reduces with increasing temperatures due to the and Schaumann (1986), as shown in Fig. 14. The dimensions
degradation of fy. and the applied loads in the tests are given in Table 1. Al-
Consider the sway frame insert in Fig. 13(a). The sections, though there are only 18 frame tests available in this compar-
the dimensions, and the bending axis are chosen so that ␤ = ison study, they represent a wide range of problems—from
0.5 and Lc /r = 40. The steel fy and E are taken as 460 N/mm2 braced to sway frames, from one- to two-bay frames, from
and 210 kN/mm2, respectively. Subjected to the given loading fully- to partially heated frames, and from major to minor axis
bending. It is noteworthy that the steel stress-strain relation-
ships at elevated temperatures provided by Rubert and Schau-
mann are adopted in the analyses.
Table 2 summarizes the Rankine results at ambient temper-
ature and at elevated temperatures of the frames. At ambient
temperature, only one frame (ZSR3) has ⌳T (=1.042) greater
than unity, indicating that all other frames are stocky. At the
predicted Tc, due to the uneven degradation of strength and
stability under thermal effects, four frames (EGR3, EGR4,
EGR5, and EGR6) now have ⌳T values greater than unity. The

FIG. 13. Effect of Residual Stresses on Sway Frames: (a) FIG. 14. Frame Types Tested by Rubert and Schaumann
Bending about Major Axis; (b) Bending about Minor Axis (1986): (a) EHR; (b) EGR; (c) ZSR

JOURNAL OF STRUCTURAL ENGINEERING / APRIL 2001 / 467

J. Struct. Eng. 2001.127:461-469.


TABLE 1. Test Parameters of Test Frames predictions of fire resistance of steel frames. This confirms the
l h fy F1 F2 accuracy and the reliability of the Rankine approach. It is note-
Frame (cm) (cm) (N/mm2) (kN) (kN) Remarks worthy that although residual stresses are not taken into ac-
(1) (2) (3) (4) (5) (6) (7) count in the Rankine approach for these 18 frames, the con-
sistently accurate predictions of Tc suggests that the effects of
EHR1 119 117 395 56 14 —a
EHR2 124 117 395 84 21 —a residual stresses may not be significant, and therefore, can be
EHR3 124 117 382 112 28 —a ignored in the analyses.
EHR4 125 150 389 20 5 —c
EHR5b 125 150 389 24 6 —c DISCUSSION
EHR6b 125 150 389 27 6.7 —c
EGR1b 122 117 382 65 2.5 —a At present, the verification study focuses on the behavior of
EGR1c 122 117 382 65 2.5 —a uniformly heated simple steel frames with small thermal re-
EGR2 122 117 385 40 1.6 —a straint. Both braced and sway frames at critical temperatures
EGR3 122 117 385 77 3.0 —a
EGR4 122 117 412 77 3.0 —a
ranging from 20⬚C to 800⬚C are investigated. The comparison
—a results show that the performance of the Rankine approach is
Downloaded from ascelibrary.org by New York University on 05/13/15. Copyright ASCE. For personal use only; all rights reserved.

EGR5 122 117 412 88 3.4


EGR6 122 117 412 88 3.4 —a different for the analyses at ambient temperature and at ele-
EGR7 122 117 320 68.5 2.6 —d vated temperatures. In summary, the critical load factors pre-
EGR8 122 117 385 77 3.0 —d dicted by the Rankine approach can be classified into three
ZSR1 120 118 355 74.0 2.85 —e temperature zones:
ZSR2 120 118 380 84.5 3.25 —e
ZSR3 120 118 432 68.5 2.64 —e
a
1. At ambient temperature, the Rankine predictions tend to
Frame fully heated and bending about major axis. be conservative. This is in good agreement with the find-
b
Frame partly in plastic range already at 20⬚C.
c
Frame fully heated and bending about minor axis. ings of Horne (1963).
d
Frame with cool beam and bending about major axis. 2. At Tc > 400⬚C, the Rankine predictions are mostly con-
e
Frame partly heated and bending about major axis. sistent and in good agreement with the FEM results. That
means that the simple interaction of the two idealized
load factors ␭p and ␭e takes account of the complexities
TABLE 2. Comparison of Results for Test Frames of actual structural behavior under thermal effects, aris-
Rankine Approach Test FEM Difference Between ing from both the material and the geometric nonline-
arities.
⌳T at ⌳T at Columns Columns
20ⴗC Tc (4) and (4) and
3. At 100⬚C ⱕ Tc ⱕ 400⬚C, the Rankine predictions be-
(nega- (nega- Tc Tc Tc (5) (6)
come less consistent.
Frame tive) tive) (ⴗC) (ⴗC) (ⴗC) (%) (%)
(1) (2) (3) (4) (5) (6) (7) (8) One striking observation is that these three temperature
zones coincide with the variation of the steel stress-strain re-
EHR1 0.104 0.105 625.4 600 620.5 4.24 0.80
EHR2 0.098 0.100 544.3 530 544.8 2.70 ⫺0.09 lationship at elevated temperatures. According to EC3-1.2
EHR3 0.094 0.119 451.4 475 458.2 ⫺4.97 ⫺1.48 (Commission 1995), the steel stress-strain relationship is
EHR4 0.357 0.463 521.7 562 530.7 ⫺7.16 ⫺1.68
EHR5 0.356 0.527 457.9 460 478.6 ⫺0.45 ⫺4.32 1. Elastic perfectly plastic with constant E and fy, at Tc <
EHR6 0.361 0.605 402.7 523 425.3 ⫺23.01 ⫺5.32 100⬚C
EGR1b 0.832 0.969 514.3 533 510.5 ⫺3.51 0.74
EGR1c 0.832 0.969 514.3 515 510.5 ⫺0.14 0.74
2. Nonlinear with varying E and constant fy, at 100⬚C ⱕ Tc
EGR2 0.826 0.595 643.9 612 596.7 5.21 7.90 ⱕ 400⬚C
EGR3 0.835 1.121 443.2 388 426.7 14.23 3.87 3. Nonlinear with varying E and fy, at Tc > 400⬚C
EGR4 0.893 1.169 458.2 424 451.3 8.06 1.53
EGR5 0.896 1.252 377.6 335 341.9 12.72 10.44 This suggests that the performance of the Rankine approach
EGR6 0.896 1.252 377.6 350 341.9 7.89 10.44 very much hinges on the steel material model. Further study
EGR7 0.624 0.801 476.7 454 456.3 4.99 4.46
EGR8 0.745 0.974 466.4 464 456.1 0.52 2.27 shall be conducted to investigate the relationship. Neverthe-
ZSR1 0.856 0.855 565.3 547 527.8 3.34 7.10 less, most of the frames fail at temperatures beyond 400⬚C,
ZSR2 0.874 0.940 519.9 479 495.4 8.54 4.94 and thus the Rankine approach can serve as a practical design
ZSR3 1.042 0.823 633.2 574 584.7 10.31 8.30 tool.
The verification study also shows that initial sway imper-
fections have detrimental effects on the frame fire resistance.
variation of ⌳T from 0.1 to 1.252 shows that the test data cover For simplicity, the initial sway imperfection of columns can
a wide range of frame slenderness ratios. The Rankine Tc are be represented by equivalent lateral force applied at the top of
then compared to test results and the FEM predictions. It the columns. This allows the Rankine approach to take sway
shows that only four out of the 18 Rankine predictions deviate imperfections into account. The FEM simulations show that
by more than ⫾10% from the test results (Table 2, column 7). the effects of sway imperfections are more significant in slen-
Three poor predictions are still within 15% accuracy, and the der frames than in stocky frames. The converse is true in the
maximum difference of 23% corresponds to frame EHR6. It Rankine approach, as only the frame strength will be affected
should be noted that frames EHR6 and EHR5 had the same by the magnitude of sway imperfections. It is also found that
geometric and material properties, but the former was sub- the effects of residual stresses are relatively insignificant and
jected to larger loads. However, the reported Tc of EHR6 was can be neglected in the analyses. Consequently, although the
higher than that of EHR5, thus suggesting that the test result Rankine approach does not take residual stresses into consid-
of EHR6 may not be representative. Compared to the FEM eration, this omission does not affect its performance.
results, the Rankine predictions seem to be more consistent
and accurate (Table 2, column 8); the maximum difference in CONCLUSIONS
the predictions merely exceeds 10% of the FEM values.
In short, in the comparison study with the test and the FEM In the proposed Rankine approach, the fire resistance of
results, the proposed Rankine approach provides satisfactory frames is determined through the simple interaction of two
468 / JOURNAL OF STRUCTURAL ENGINEERING / APRIL 2001

J. Struct. Eng. 2001.127:461-469.


idealized behaviors, viz the strength and the stability aspects. Horne, M. R. (1960). ‘‘Instability and the plastic theory of structures.’’
Herein, the rigid-plastic behavior represents the frame Trans. Engrg. Inst. Ca., 4(2).
Horne, M. R. (1963). ‘‘Elastic-plastic failure loads of plane frames.’’
strength, and the elastic-buckling behavior represents the Proc., Royal Soc., London, 274, 343–364.
frame stability. Basically, the deterioration of material prop- Horne, M. R., and Merchant, W. (1965). The stability of frames, Perga-
erties and the additional compressive stresses associated with mon, Tarrytown, N.Y.
thermal effects are the main concerns. It is found that the ratio Merchant, W. (1954). ‘‘The failure load of rigid jointed frameworks as
of the rigid-plastic collapse to the elastic-buckling load factors, influenced by stability.’’ The Struct. Engr., London, 32, 185–190.
defined as the frame slenderness ratio ⌳T, is the principal pa- Najjar, S. R., and Burgess, I. W. (1995). ‘‘A nonlinear analysis for three-
dimensional steel frames in fire conditions.’’ Engrg. Struct., 18(1),
rameter governing the structural behavior and the ultimate re- 77–89.
sistance of steel frames under thermal effects. When ⌳T < 1.0, Rubert, A., and Schaumann, P. (1986). ‘‘Structural steel and plane-frame
the frame is governed by strength and the primary failure assemblies under fire action.’’ Fire Safety J., 10, 173–184.
mode is a plastic mechanism. When ⌳T > 1.0, the frame is Talamona, D. (1995). ‘‘Buckling curves in case of fire—ECSC 7210 SA
governed by stability, and buckling becomes the primary fail- 316/515/931/618: Fire resistance of steel columns with eccentric load.
ure mode. Part 1.’’ Rep. No. INC-96/450-DT/VG Part 1, CTICM, Saint-Remy-les-
Chevreuse, Paris.
Downloaded from ascelibrary.org by New York University on 05/13/15. Copyright ASCE. For personal use only; all rights reserved.

This paper presents a preliminary study of the effects of Talamona, D., Franssen, J. M., Schleich, J. B., and Kruppa, J. (1997).
various parameters on the accuracy of the Rankine approach. ‘‘Stability of steel columns in case of fire: Numerical modeling.’’ J.
To facilitate hand calculations, the Rankine approach is only Struct. Engrg., ASCE, 123(6), 713–720.
applied to steel frames with small thermal restraints in this Tan, K. H., Fung, T. C., and Toh, W. S. (1999). ‘‘A first-order elastic-
study. When significant thermal restraints exist, iterations are plastic analysis of steel frames in fire.’’ Proc., 7th East Asia-Pacific
necessary and the procedure becomes too tedious for hand Conf. on Struct. Engrg. and Constr., H. Okamura and H. Shima, eds.,
Social System Institute, Tokyo, 423–428.
calculation. Through these studies, it is possible to establish Toh, W. S. (2000). ‘‘Strength and stability of steel structures under ther-
some limits on the applicability of the proposed method. A mal effects.’’ PhD thesis, Nanyang Technological University, Singa-
consistently good agreement with both numerical and test re- pore.
sults shows that the proposed Rankine approach can be utilized Toh, W. S., Tan, K. H., and Fung, T. C. (2000). ‘‘Compressive resistance
as a quick tool to assess the fire resistance of steel frames fully of steel columns in fire: Rankine approach.’’ J. Struct. Engrg., ASCE,
exposed to fire above 400⬚C. Furthermore, the same Rankine 126(3), 398–405.
approach can also be used to estimate fire resistance of steel APPENDIX II. NOTATION
columns (Toh et al. 2000). This consistency in approach for
both columns and frames is its greatest attribute. In addition, The following symbols are used in this paper:
more research work can be done to extend the Rankine ap- A = area (mm2);
proach to subframe analysis incorporating thermal effects. E = elastic modulus (N/mm2);
fy = yield strength (N/mm2);
APPENDIX I. REFERENCES h, L, l = member length (mm);
I = moment of inertia of section (mm4);
Cheng, W. C., and Mak, C. K. (1975). ‘‘Computer analysis of steel frame kE (T ) = reduction factor of elastic modulus (negative);
in fire.’’ J. Struct. Div., ASCE, 101(4), 854–856. ky (T ) = reduction factor of yield strength (negative);
Commission of European Communities. (1992). ‘‘Design of steel struc- Mp = plastic moment of resistance (N ⭈ mm);
tures: Part 1.1. General rules and rules for buildings (EC3-1.1).’’ Eu- P = member axial load (N);
rocode 3, Brussels. PE = Euler load (N);
Commission of European Communities. (1995). ‘‘Design of steel struc- s, c = stability functions (negative);
tures: Part 1.2. General rules. Structural fire design (EC3-1.2).’’ Euro- T = steel temperature (⬚C);
code 3, Brussels. Tc = critical temperature (⬚C);
Franssen, J. M., Schleich, J. B., Talamona, D., and Twilt, L. (1994). ‘‘A
␤ = beam-column stiffness ratio (negative);
comparison between five structural fire codes applied to steel ele-
ments.’’ Proc., 4th Int. Symp. Fire Safety Sci., T. Kashiwagi, ed., ␥ = load ratio (negative);
IAFSS, Gaithersburg, Md., 1125–1136. ⌳T = frame slenderness ratio (negative);
Furumura, F., and Shinohara, Y. (1978). ‘‘Inelastic behaviour of protected ␭c = critical load factor (negative);
steel beams and frames in fire.’’ Rep. No. 3, Res. Lab. of Engrg. Mat., ␭e = elastic buckling load factor (negative); and
Tokyo. ␭p = rigid-plastic collapse load factor (negative).

JOURNAL OF STRUCTURAL ENGINEERING / APRIL 2001 / 469

J. Struct. Eng. 2001.127:461-469.

You might also like