Fire Technology manuscript No.

(will be inserted by the editor)

⋆
Simulation Methodology for Coupled Fire-Structure
Analysis:
Modeling localized fire tests on a steel column

Chao Zhang · Julio G. Silva · Craig

Weinschenk · Daisuke Kamikawa · Yuji
Hasemi

Received: date / Accepted: date

Abstract Advanced simulation methods are needed to predict the complex

behavior of structures exposed to realistic fires. Fire Dynamics Simulator
(FDS) is a computational fluid dynamics (CFD) code, developed by NIST for
fire related simulations. In recent years, there has been an increase in use of
FDS for performance-based analysis in the area of structural fire research. This
paper discusses the FDS-FEM (finite element method) simulation methodol-
ogy for structural fire analysis. The general methodology is described and a
⋆ Accepted manuscript, Uncorrected proof. cite as Zhang, C. et al.. (2015). ”Simulation

Methodology for Coupled Fire-Structure Analysis: Modeling localized fire tests on a steel
column” Fire Technology. ,DOI: 10.1007/s10694-015-0495-9
C. Zhang
National Institute of Standards and Technology, Fire Research Division, 100 Bureau Drive,
Stop 8666, Gaithersburg, MD 20899-8666, USA
Tel.: +1 301 9756695
Fax: +1 301 9754052
E-mail: chao.zhang@nist.gov
J.G. Silva
National Institute of Standards and Technology, Fire Research Division, 100 Bureau Drive,
Stop 1070, Gaithersburg, MD 20899-1070, USA

C. Weinschenk
National Institute of Standards and Technology, Fire Research Division, 100 Bureau Drive,
Stop 1070, Gaithersburg, MD 20899-1070, USA

D. Kamikawa
Waseda University, Department of Architecture, Okubo 3-4-1, Shinjuku-ku, Tokyo, Japan
Present address: Forestry and Forest Products Research Institute, Matsunosato 1, Tsukuba,
Ibaraki, Japan

Y. Hasemi
Waseda University, Department of Architecture, Okubo 3-4-1, Shinjuku-ku, Tokyo, Japan.
2 Chao Zhang et al.

validation study is presented. A data element used to transfer data from FDS
to FEM codes, the adiabatic surface temperature, is discussed. A tool named
Fire-Thermomechanical Interface (FTMI) is applied to transfer data from FDS
to ANSYS. A high temperature stress-strain model for structural steel devel-
oped by NIST is included in the FEM analysis. Compared to experimental
results, the FDS-FEM method predicted both the thermal and structural re-
sponses of a steel column in a localized fire test. The column buckling time
was predicted with a maximum error of 7.8%. Based on these results, this
methodology has potential to be used in performance-based analysis.

Keywords CFD-FEM simulation method · Structural fire analysis · Fire

Dynamic Simulator (FDS) · Fire-Thermomechanical Interface (FTMI) ·
Adiabatic surface temperature · Finite element simulation · Localized fires ·
Steel column · Validation study

1 Introduction

In the event of fire, buildings should retain structural stability for a specified
period of time to allow for full evacuation. Traditionally, structural components
are required to fulfill the fire resistance ratings specified in prescriptive codes.
The fire resistance rating of a building component is determined by a standard
fire resistance test conducted on an isolated member subjected to a specified
time temperature curve. The standard fire resistance test, which was developed
more than a century ago, has a number of shortcomings [1]. For example, the
heating used in the test bears little resemblance to a real fire environment
and the behavior of isolated members cannot represent the behavior of the
components in an entire structure [2]. As a result, the design approach based on
prescriptive codes cannot assess the actual level of safety of a structure exposed
to a fire and usually yields a fire protection design that is too conservative
and lacks certainty of the value of the safety factors [3]. Considering green
construction is pushing design for more sensible usage and conservation of
materials, the use of fire protection materials which can be shown not to
be wasteful is important. It should be noted, however, that a fire protection
design based on a standard test or prescriptive code is not guaranteed to be
conservative [4–6].
Over the past 30 years, there have been significant advances in structural
fire research. New insights, data, and calculation methods have been reported,
which form the basis for modern performance-based (PB) codes for structural
fire safety [7]. The PB approach involves the assessment of the structural re-
sponse in real fires and, therefore, requires advanced computational approaches
for fire and structural modelings. Sophisticated computational fluid dynamics
(CFD) models are typically used to simulate realistic fires [8–10], while finite
element method (FEM) codes are mostly used for structural modeling [10,11].
An integrated CFD-FEM simulation approach is needed for advanced struc-
tural fire analysis [12].
⋆ Simulation Methodology for Coupled Fire-Structure Analysis: 3

Fire Dynamics Simulator (FDS) is an open source CFD code, developed

by NIST [13]. It has been widely used in fire engineering for modeling the
gas phase environment (temperature, heat flux, velocity, species concentra-
tions, etc.) in fires [14, 15]. Recently, there has been increased research in the
application of FDS for structural fire analysis [16–20]. Fire-structure inter-
face tools for transferring data from FDS to particular FEM codes (such as
ANSYS, ABAQUS, SAFIR) have been developed [19, 21, 22]. Although both
FDS and FEM codes have been separately validated, there needs to be a
direct validation for the integrated FDS-FEM simulation methodology. This
paper validates the integrated FDS-FEM simulation methodology against a
localized fire test on steel columns reported by Kamikawa et al. [23]. Se-
quential coupled fire-thermal-mechanical simulations are conducted. Two fire-
structure interfaces are used in this paper, the DEVICE approach and the
Fire-Thermomechanical Interface (FTMI) tool [24, 25]; both interfaces trans-
fer data between FDS and the commercial software ANSYS 1 . A new high
temperature stress-strain model for structural steel developed by Luecke et
al. [26] is used in the structural analysis.

Fires in the open or in large enclosures are characterized as localized fires,

e.g. vehicle fires in transportation infrastructure [27], small shop fires in trans-
port terminal halls [28], and workstation fires in open plan office buildings [29].
Compartment fires begin as fires with localized burning. The current structural
fire design approaches are developed for fully-developed compartment fires,
the gas temperatures of which can be approximated as uniformly distributed
in the compartment. In localized fires, the gas temperature distributions are
spatially non-uniform. Because of the thermal gradient, the failure model and
failure temperature of structural members in a localized fire might be different
than those of the members in uniform heating conditions (e.g. the standard
fire condition) [4–6]. Consideration of localized fires is important for safety
of structures in the modern buildings with large enclosures. From literature,
e.g. [1], realistic fire test data for model validation are quite limited, especially
for modeling structure response to non-uniform heating conditions. This lo-
calized fire test was selected for model validation because of the applicability
to real-world thermal conditions and because the test was well controlled (e.g.
the heat release rate of the fire was controlled by computer, and the axial load
was applied by oil jack controlled by electric hydraulic pump), and detailedly
measured (e.g. temperatures on all typical sides of the test specimen were
measured by series of thermocouples).

1 Certain commercial entities, equipment, or materials may be identified in this document

in order to describe an experimental procedure or concept adequately. Such identification is

not intended to imply recommendation or endorsement by the National Institute of Stan-
dards and Technology, nor is it intended to imply that the entities, materials, or equipment
are necessarily the best available for the purpose.
4 Chao Zhang et al.

2 Methodology

2.1 The CFD-FEM approach

Fig. 1 illustrates the CFD-FEM simulation approach for structural fire anal-
ysis. The fire-structure interaction is fundamentally two-way, while one-way
coupling may be advantageous under certain conditions [12]. In a one-way
coupling, the Navier-Stokes equations, radiation transport equations, etc., for
the fluid domain in a fire compartment are solved for the complete time du-
ration of interest by a CFD code to get gas temperatures, velocities, chemical
species, incident heat fluxes, film coefficients, etc. The heat equations for the
solid domain (building elements) use the thermal boundaries from the CFD
simulation to get the thermal response (temperature rise) within the build-
ing elements. Kinematics equations, constitutive equations, etc., for the solid
domain are solved by a FEM code to get the deformations, stresses, strains,
etc. Fire-structure and thermo-mechanical interfaces are used to transfer data
between different models. In a two-way coupling, the same set of equations
is solved except that at discrete time steps through the simulation the solid
phase FEM code provides feedback to update the CFD model.

2.2 The FDS code

Fire Dynamics Simulator is a large-eddy simulation (LES) based CFD code [13].
For the simulations performed in this study, FDS version 6.1.1 was used. LES
is a technique used to model the dissipative processes (viscosity, thermal con-
ductivity, material diffusivity) that occur at length scales smaller than those
that are explicitly resolved on the numerical grid. In FDS, the combustion is
based on the mixing-limited, infinitely fast reaction of lumped species, which
are reacting scalars that represent mixtures of species. Thermal radiation is
computed by solving the radiation transport equation for gray gas using the
Finite Volume Method (FVM) on the same grid as the flow solver. FVM is
based on a discretization of the integral forms of the conservation equations. It
divides the problem domain into a set of discrete control volumes (CVs) and
node points are used within these CVs for interpolating appropriate field vari-
ables. The governing equations are approximated on one or more rectilinear
grids. Obstructions with complex geometries are approximated with groups
of prescribed rectangles in FDS. One-dimensional (1D) heat conduction is as-
sumed for solid-phase calculations. Detailed descriptions of the mathematical
models used in FDS can be found in [30].

2.3 The fire-structure one-way coupling

Heat can be transferred from flames and hot gases to structures by radiation
and convection. The net heat flux q̇ ′′ can be defined by the sum of these two
⋆ Simulation Methodology for Coupled Fire-Structure Analysis: 5

terms:
q̇ ′′ = εs (q̇in
′′
− σTs4 ) + hc (Tg − Ts ) (1)
′′
where εs is emissivity of the exposed surface; q̇in is incident radiative flux;
Tg is temperature of the surrounding gas; Ts is temperature of the exposed
surface; hc is film coefficient; and σ is Stefan-Boltzmann constant.
Advanced fire simulation models (such as FDS) are capable of providing
the three-dimensional evolution of the fire, the incident radiative flux and the
temperature within the gas phase. Nevertheless, these software packages are
not typically capable of accurately evaluating temperature distributions in
solids. Therefore, the total heat flux to a solid may not be correctly calculated
at the end of the fire simulation and an additional approach is necessary.

2.3.1 The concept of adiabatic surface temperature

Consider an ideal adiabatic surface exposed to a heating condition; the net

heat flux to the surface is by definition zero, thus
′′
εAS (q̇in − σTAS
4
) + hc,AS (Tg − TAS ) = 0 (2)

where εAS is emissivity of the adiabatic surface; TAS is temperature of the

adiabatic surface or adiabatic surface temperature; and hc,AS is film coefficient
between the adiabatic surface and the surrounding gas.
From Eq. 2, the incident radiative flux to a surface can be calculated from
an adiabatic surface temperature,

′′ hc,AS (TAS − Tg ) 4
q̇in = + σTAS (3)
εAS
Consider a real surface exposed to the same heating condition, the net heat
flux to the surface can be calculated by
εs
q̇ ′′ = εs σ(TAS
4
− Ts4 ) + hc,AS (TAS − Tg ) + hc (Tg − Ts ) (4)
εAS
If the emissivity of the adiabatic surface is taken as the emissivity of the real
surface (εAS = εs ), and the film coefficient between the adiabatic surface and
the surrounding gas is equal to the film coefficient between the real surface
and the surrounding gas (hc,AS = hc ), we get

q̇ ′′ = εs σ(TAS
4
− Ts4 ) + hc (TAS − Ts ) (5)

Eq. 5 shows that the net heat flux to a surface can be approximately
calculated by using a single parameter TAS . In practice, the adiabatic surface
temperatures of interest can be approximately measured by a plate thermome-
ter [31]. Consider the case at high temperature (above about 400 ◦ C), where
convection is not the dominant mode of heat transfer in fire [32]; from Eq. 4
or 5 the adiabatic surface temperature measured by a plate thermometer can
be used to predict the net heat flux to a surface with a different emissivity.
6 Chao Zhang et al.

FDS [13] includes an output quantity of adiabatic surface temperature calcu-

lated by Eq. 5 according to the idea proposed by Wickstrom [33]. It should
be noted that the calculated adiabatic temperature of a surface is fundamen-
tally influenced by the convection or film coefficient (see Eq 4) so that the
value for film coefficient should be carefully selected when using the concept
for calculations where convection is important [34].

2.3.2 The DEVICE approach in FDS

In FDS, point measurements (devices) can be used to record the adiabatic

surface temperatures. The output adiabatic surface temperatures (with an
assumed constant film coefficient) are converted to inputs to the FEM thermal
model. The data points are interpreted as the effective black body temperature
and bulk (gas) temperature to calculate the radiative and convective heat
fluxes to the structural elements, respectively. Because point measurements
are used, it can be intractable to have measurement devices located in every
computational cell where the solid has an exposed surface. Therefore, linear
interpolation is used to determine the adiabatic surface temperature between
two recorded data. This is called the DEVICE approach for fire-structure
coupling.
One limitation of this approach is measurement density. It may not be
known a-priori where the point measurements need to be to capture the nec-
essary adiabatic surface temperature (AST) gradients. A lack of appropriate
AST resolution could have an impact on the FEM analysis for predicting
structural behavior. Additionally, for CFD models with complex geometry,
this issue could become more significant. A second limitation is the use of
a constant film coefficient. Following normative procedures [35] this variable
is usually considered a constant [17]. However, its value drives whether heat
transfer is dominated by radiation or convection. Also, the spatial distribution
and temporal evolution during the fire can be important for calculating the
heat flux from the fire simulation for thermo-mechanical analysis. As discussed
in Section 2.3.1, an improper selection of the film coefficient can influence the
heat flux and the resulting FEM analysis.

2.3.3 The Fire-Thermomechanical Interface (FTMI)

The goal of the FTMI model is to create the appropriate boundary condition
between the fire simulation and the thermo-mechanical analysis. Following
a CFD fire simulation, the heat flux is evaluated in the thermo-mechanical
model using Eq 5, depending on TAS and hc obtained by the fire simulation
and the surface temperature calculated at each time step during the thermo-
mechanical analysis.
To correctly calculate the heat flux, boundary condition information must
be passed from CFD to FEM models. In this task, the exposed surfaces in the
FEM model need to be mapped to the same surfaces from the CFD model [24].
⋆ Simulation Methodology for Coupled Fire-Structure Analysis: 7

This mapping is realized by a collection of I keypoints (of x coordinates) local-

ized at the center of each external face (with normal n), as illustrated in Fig. 2.
The position and the number of I keypoints is based on the mesh generated
for the thermo-mechanical analysis. In this way, the coupling procedure can
be achieved for different discretization levels. Also, small modifications and
dimensioning to the structural model does not require the CFD solution to
be recomputed, as the data can be remapped. This also allows for data to be
properly transfered from the CFD model to the FEM model without requiring
the two models to have identical computational meshes.
To compute this mapping, a Fortran code was written (fds2ftmi [24, 25]).
This code is based on the fds2ascii routine [13], which was written to parse FDS
binary output files. Basically, fds2ftmi traces the exposed surfaces in the FEM
model (ANSYS), defines and collects the I keypoints and the corresponding n
normals (related to the element surface). Based on each I keypoint position,
the code is able to search into the output data from FDS to iterate over time,
orientation, and mesh, and then pass the correct surface thermal exposure
results (TAS , hc) into a ANSYS APDL language script file.
In the ANSYS model, the goal of this interface procedure is to create an
iterative solution capable of using the surface temperature (obtained at each
time step during the thermo-mechanical analysis) to evaluate the heat flux
at each node of the exposed surface through Eq 5. Therefore, the surface
effect element SURF152 (ANSYS nomenclature) is employed. This element is
attached to the solid or shell elements to prescribe a boundary condition. In
this case, it applies a heat flux vector at each node of the exposed surface
based on Eq 5. The adiabatic surface temperature needs to be prescribed at
the element’s extra node and film coefficient is applied at the element’s surface.
If this procedure is used with shell elements, the heat flux vector is calculated
and prescribed at the top and bottom layers of the shell section.

2.4 Thermal-structure interface

ANSYS offers two distinct methods for thermal-structure analyses: the direct
method and the load transfer method. The direct method usually involves just
one analysis that contains all necessary degrees of freedom: displacements, ro-
tations, and temperatures. The thermal and structural calculations are cou-
pled by the interal solver in ANSYS. This method is advantageous when the
coupled interaction involves strongly-coupled physics or is highly nonlinear.
Unfortunately, the direct method works only for a specific set of bilinear solids
elements (hexahedrons and tetrahedrons). Use of these elements can become
computationally expensive.
In the load transfer method, the thermal and structural analyses are solved
independently. The results from the thermal analysis are applied as loads in
the structural analysis. For coupling situations which do not exhibit a high
degree of nonlinear interaction, the load transfer method is more efficient and
flexible [36]. Also, this method works with more element types, including shell
8 Chao Zhang et al.

and beams elements, which are extensively used to model global structures [22].
In this study, the load transfer method is used for coupled thermal-structure
analysis. The temperature data from heat transfer analysis are transferred to
the mechanical model for structural analysis. Thermal shell element SHELL131
and structural thermal element SHELL181 are used in the ANSYS analyses.
Previous work conducted at NIST [22] focused on transferring data between
the thermal analysis and the structural analysis, including cases with different
meshes and different element types. In this analysis, the element type and
mesh remained the same for both the thermal and structural computations.

3 Validation study

3.1 Description of the experiment from Kamikawa et al. [23]

Fig. 3 shows the experimental setup [23]. A 0.3 m square diffusion burner was
located just beside a square steel column (STKR400, 0.1 m × 0.1 m, 3.2 mm
thick and 1.6 m tall). Propane was used as the fuel, and the heat release rate
(HRR) was 52.5 kW. The height of the burner was 0.25 m (from the base of
the column). Four tests (cases) with various loading and restraint conditions
were conducted. A new column specimen was used for each test. In all cases,
the column was fixed at the base. In this paper, just case 1 and case 4 are
considered, the names are kept for easier comparison to the previous work.
Case 1 was intended to address the thermal expansion due to heating and
bending due to the thermal gradient. Therefore, the structure was unrestrained
except for the base. The fire was extinguished after 60 min, when the column
behavior (temperature and displacement) reached steady-state. In case 4, the
horizontal displacement toward the fire source was restrained; Fig. 4 shows
the locations of the vertical and horizontal displacement restraints. Also, a
vertical force was applied and increased gradually after the steel temperature
in the column became steady (about 52 min after ignition); the column buckled
about 90 min after ignition with a vertical force of approximately 374 kN. The
fire was extinguished only after the column buckled.
In all cases, the temperature distributions of the column side surfaces, the
horizontal displacements at various heights, and the vertical displacement at
column top were measured. Temperatures were measured by K-type (chromel-
alumel) thermocouples (0.65 mm bead diameter) fixed on column side surfaces
with spot-welded steel foil. Tension coupon tests were conducted to obtain the
mechanical properties of the steel. Fig. 5 shows the measured stress-strain
curves at various temperatures for the steel as well as those produced from
the material model developed by Luecke et al. [26]. The material model is
discussed in detail in Section 3.2. The elastic modulus and yield strength (offset
yield point at 0.2% strain) at ambient temperature are about 202,000 MPa and
400 MPa, respectively. The material model is discussed in detail in Section 3.2.
An uncertainty analysis was not conducted as part of the experimental
analysis. For discussion of the experiments, the nominal values for the fire size
⋆ Simulation Methodology for Coupled Fire-Structure Analysis: 9

and loads applied are presented. For experimental data presented, measure-
ment uncertainty presented is from representative values found in literature.
For thermocouple measurements of surface temperature on exposed steel, the
uncertainty, for elevated steel temperatures (T > 300 ◦ C), can be estimated
to be +1 ◦ C to -9 ◦ C [37]. For displacement measurements during fire tests,
the uncertainty can be estimated to ±3 % [38].

3.2 Material properties

To appropriately model the response of the exposed steel member, its material
properties need to be defined. Thermal properties for structural steel specified
in the Eurocode [39] were used. Figs. 6 and 7 plot the temperature-dependent
specific heat and thermal conductivity for structural steel, respectively. The
emissivity of steel was taken as 0.9, typical for steel in fire conditions. For
mechanical properties, the stress-strain model developed by Luecke et al. [26]
was used. While stress-strain measurements were performed as part of these
experiments [23], experimental data may not always be available. Therefore,
to be more representative of a practical engineering application, a model for
mechanical properties was used. The model is defined by the following set of
equations [26]
fyT
σ = εET (ε ≤ ) (6a)
ET
T k1 fyT n fyT
σ = fyT + (k3 − k4 fy20 ) exp[−( ) ](ε − ) (ε ≥ ) (6b)
k2 ET ET
with k1 = 7.820, k2 = 540 o C, k3 = 1006 MPa, k4 = 0.759, and n = 0.503.
The elastic modulus and yield strength at elevated temperature are give by
ET 1 T − 20 3.768 1 T − 20
= exp[− ( ) − ( )] (7)
E20 2 639 2 1650
and
fyT 1 T − 20 7.514 1 T − 20
= 0.09 + 0.91 exp[− ( ) − ( )] (8)
fy20 2 588 2 676
respectively. Where E20 , ET are elastic modulus of steel at room and elevated
temperatures, respectively; and fy20 , fyT are yield strength of steel at room
and elevated temperatures, respectively. Fig. 5 also shows the stress-strain
curves calculated by Eq. 6 using the measured ambient temperature elastic
modulus and yield strength. The constitutive model is a creep free model [26]
while creep strains have not been subtracted from the total measured strains.
Creep is usually regarded to be important for temperatures above 400 o C [40].
Therefore, there are divergences between the measured and calculated stress-
strain curves.
The equation for thermal expansion coefficient recommended by NIST TN
1681 [41] was used
αs = 1.17 × 10−5 + 1.34 × 10−8 T − 9.7 × 10−12 T 2 + 1.67 × 10−16 T 3 (9)
10 Chao Zhang et al.

3.3 Numerical models

Fig. 8 shows the FDS model geometry and computational mesh for Kamikawa
et al.’s experimental configuration. Dimensions of the computational domain
are 0.75 m × 0.45 m × 1.8 m. The grid size used is an important numerical
parameter in CFD because of its impact on numerical accuracy. The necessary
spatial resolution for a proper LES simulation of a free burning fire is custom-
arily defined in terms of the characteristic diameter of a plume, D∗ , which is
defined as [13]
Q̇
D∗ = ( √ )2/5 (10)
ρ∞ cp T∞ g
where Q̇ is the heat release rate; ρ∞ is ambient density; cp is the specific heat
of air at constant pressure; T∞ is ambient temperature; and g is acceleration
of gravity. The special resolution, R∗ , of a numerical grid is defined as,
dx
R∗ = (11)
D∗
where dx is characteristic length of a cell for a given grid. For the FDS model,
a uniform grid size of 0.025 m in XYZ directions (R∗ ≈ 1/12). The computa-
tional domain, therefore, consisted of 43,200 control volumes.
Fig. 9 shows the FEM models for thermal and structural analyses. The
FEM thermal model and the FEM structural model use the same mesh, as
shown in Fig. 9 (left). For both analyses the mesh was uniform with 10 equally
sized elements along the web and the flange and 120 elements along the height
of the column. The support conditions were applied to replicate the experi-
mental set (3.1) and are presented at Fig. 9 (center and right). In case 1, the
column is free to expand and bend, which means that except for the column
base, there are no vertical or horizontal restraints along the column. In case
4, a horizontal restraint at a height of 1400 mm is introduced. Additionally,
the load applied at the top of the column also acts as a restraint (3.1).
The boundary conditions from FDS will be translated to FEM by two
methods. In the first method, the thermal boundaries in the FEM thermal
model are represented using the adiabatic surface temperatures from FDS
simulation (recorded using Devices [13]), and the film coefficients on all side
surfaces of the column were taken as 9 W/m2 K according to [34]. In FTMI,
the thermal boundaries are evaluated automatically using the adiabatic sur-
face temperatures and the film coefficients (read from FDS output files [13])
calculated by FDS at each exposed FEM element.

4 Results

4.1 Fire environment

To visualize flame exposure of the column during the fire tests, Fig. 10 shows
the simulated flame geometry at three distinct times. Note that the flame
⋆ Simulation Methodology for Coupled Fire-Structure Analysis: 11

shape changes with time because of the turbulent combustion processes. Cor-
respondingly, Fig. 11 shows the adiabatic surface temperature spatial distri-
butions at the same three time instances as Fig. 10. To better quantify the
surface temperatures, Fig. 12 shows the time averaged adiabatic surface tem-
peratures along the mid-line of the column surfaces: front, back, and side.
The front surface directly faces the flame, including direct flame impingement,
and therefore has the highest adiabatic surface temperatures (AST). The AST
peaks at approximately 640 ◦ C at 0.4 m above the burner. Comparatively, the
back surface cannot “see” the flame (a view factor near zero) and therefore
remains significantly cooler than the front surface, < 50 ◦ C. The side surfaces
can see parts of the flame (the maximum AST is about 125 ◦ C). In addition
to the variation in view factor, the non-uniform gas temperature distributions
(see Fig. 13) contribute to the non-uniformly distributed ASTs.
The only difference between case 1 and case 4 for the fire environment and
the thermal response, is that in case 1 the fire was extinguished at 60 min
(after achieve a steady state behavior), while in case 4 the fire continued until
the column buckled (Section 3.1).

4.2 Thermal response

The predicted steel temperature distributions for both the DEVICE approach
and FTMI approach are shown in Fig. 14. In both cases, the distributions are
highly non-uniform across and along the column. The radial profile around
the maximum temperature observed with the DEVICE changed for FTMI
(Fig. 14). This is caused by the addition of film coefficient distribution in the
thermal boundaries in the FTMI compared to the constant value used with the
DEVICE. To assess the model predictions, the simulated temperatures were
compared to experimentally measured values at 4 locations over the duration
of the experiment: front and corner measurements 400 mm above the burner
and side and back measurements 600 mm above the burner. This compari-
son is presented in Fig. 15, which shows the agreement between the measured
and predicted maximum steel temperatures by both methods for case 1. The
temperature evolution of corner and side surfaces is highly influenced by the
heat conduction from the front surface (Fig. 15). Since their temperatures
are higher than the adiabatic surface temperature, Fig. 11, these surfaces are
in fact, emitting heat to the surrounding environment (Section 5). The heat
transfer at the back surface is dominated by convection while, as discussed be-
fore, the equation used in FDS to calculate the adiabatic surface temperature
is for radiation dominated situations. This might explain the under-prediction
of the steel temperature for the back surface.

4.3 Structural response

In case 1 the column has no support at the top. It is free to expand and bend
due to heating and thermal gradient respectively. The displacements at lateral
12 Chao Zhang et al.

(δx ) and vertical (δz ) directions were provided by the experimental measure-
ments and are used here to compare with the numerical results. Fig. 16 and
Fig. 17 compares the measured and predicted δx and δz , respectively. The max-
imum lateral displacement at a height of 1440 mm is 27 mm achieved during
the heating phase (≈ 7.5 min). After the fire achieves a steady state and the
temperatures were close to a constant value (from 10 to 15 min, Fig. 15), the
thermal gradient and the horizontal displacement decreased (Fig. 16), leading
to an increase in vertical displacement (Fig. 17). The vertical displacement ob-
tained with FTMI showed better agreement than the DEVICE for the heating
and the cooling of the column. However, the maximum displacement at 1 hr
of fire, obtained with the DEVICE is closer to the experimental data.

For case 4, the experimental work only presents the vertical displacement
at column top, and a discussion about the failure mode. The structural sup-
port conditions for case 4 are more complex than in case 1. The horizontal
displacement is restricted in the x direction at the height of 1392 mm (Fig. 4),
but there are no restraints (in x or y directions) at the top of the column. Nev-
ertheless, to apply the load in a proper manner, these displacements at column
top should not change the position of the experimental apparatus (Fig. 3 and
Fig. 4). Also, as the load increases, the generated forces will apply some re-
strictions to these horizontal displacements. In the numerical model, failing to
consider horizontal restrictions at column top leads to a global buckling be-
havior, bending in y direction. Despite this, the failure time was in agreement
with the experimental results. For both methods, Fig. 18 shows the agreement
between the measured and predicted column behavior for case 4. The pre-
dicted buckling time was about 88.5 min after ignition for using the DEVICE
method and 83 min for FTMI. The measured buckling time was about 90 min
after ignition, which means that the error was about 1.7% for AST and 7.8%
for FTMI. For the numerical models, buckling was determined when a lack of
convergence would lead to nonphysical displacements. In this situation, there
is a lack of confidence in the numerical response model due to nonlinearities
and instabilities. In numerical simulations, a lack of convergence does not nec-
essarily mean buckling has occured. The user should assess the results based
on knowledge and experience to find the cause for nonconvergence.

In practice, the applied force will produce some restraints to the horizon-
tal displacements at the column top. Modifying the FEM structural model
by adding horizontal restraints at the column top, the local buckling of plates
found in the test was observed in the numerical simulation, as shown in Fig. 19.
The predicted vertical displacement by the modified structural model using
FTMI is also presented in Fig. 18. The failure time was 86.5 min, which repre-
sents a 3.9% error. The comparison between the experiment and FTMI local
buckling is presented in Fig. 19.
⋆ Simulation Methodology for Coupled Fire-Structure Analysis: 13

5 Conclusions

This paper discusses the FDS-FEM simulation methodology for advanced

structural fire analysis. Two fire-structure interfaces were applied: the DE-
VICE approach and Fire-Thermomechanical Interface (FTMI) to transfer data
from FDS to ANSYS. By comparing the predicted and measured thermal and
structural responses of a steel column in a localized fire condition, the FDS-
FEM method was tested. For temperature measurements in case 1, predictions
were within estimated experimental uncertainty except for the more convec-
tively dominated rear side of the column. Also for case 1, the lateral and verti-
cal displacement were generally within the estimated experimental uncertainty
for both the DEVICE and FTMI methods. In case 4, however, the vertical dis-
placements underpredicted the vertical displacements. The final buckling time
was predicted within 10 % of the experimentally measured time for all of the
cases analyzed. The methods described in this study provide a feasible way to
study the complex behavior of structures in realistic fires.
A localized fire was used for validation based on the availability of the test
data and because localized fires are realistic fires which are easily controllable
and reproducible. While there are various types of realistic fires, the simula-
tion methodology described in this paper is regarded to be valid for all fire
scenarios because of the consideration that localized fires represent a univer-
sal fire condition in which both temperature rising and thermal gradient are
included.

Acknowledgements Valuable suggestions and review comments from Dr. Anthony Hamins,
Dr. Fahim Sadek, Dr. Matthew Bundy and Mr. Keith Stakes of NIST are acknowledged.

References

1. L. Bisby, J. Gales, and C. Maluk. A contemporary review of large-scale non-standard

structural fire testing. Fire Science Reviews, 2:1–27, 2013.
2. G.Q. Li and C. Zhang. Fire resistance of restrained steel components. In Proceedings
of the Fourth International Conference on Protection of Structures against Hazards,
pages 21–38, Beijing, China, 2009.
3. S. Lamont. The behaviour of multi-storey composite steel framed structures in response
to compartment fires. PhD thesis, University of Edinburgh, 2001.
4. C. Zhang, G.Q. Li, and A. Usmani. Simulating the behavior of restrained steel beams
to flame impinged localized fires. Journal of Constructional Steel Research, 83:156–65,
2013.
5. C. Zhang, J.L. Gross, and T. McAllister. Lateral torsional buckling of steel w-beams to
localized fires. Journal of Constructional Steel Research, 88:330–8, 2013.
6. C. Zhang, J.L. Gross, T.P. McAllister, and G.Q. Li. Behavior of unrestrained
and restrained bare steel columns subjected to localized fire. Journal of Structural
Engineering-ASCE, 2014.
7. G.Q. Li and C. Zhang. The chinese performance-based code for fire-resistance of steel
structures. International Journal of High-Rise Buildings, 2:123–30, 2013.
8. K.B. McGrattan. Fire modeling: where are we? where are we going? Fire Safety Science,
8:53–68, 2005.
14 Chao Zhang et al.

9. R.G. Gann, A. Hamins, K. McGrattan, H.E. Nelson, T.J. Ohlemiller, K.R. Prasad, and
W.M. Pitts. Reconstruction of the fires and thermal environment in World Trade Center
Buildings 1, 2, and 7. Fire Technology, 49:697–707, 2013.
10. N. Henneton, M. Roosefid, and B. Zhao. Application of structural fire safety engineering
to the canopy of the forum DES Halles, in Paris . In Proceedings of the 8th International
Conference on Structures in Fire, pages 1047–1054, 2014.
11. T.P. McAllister, F. Sadek, J.L. Gross, S. Kirkpatrick, R.A. MacNeil, R.T. Bocchieri,
M. Zarghamee, O.O. Erbay, and A.T. Sarawit. Structural analysis of impact damage
to World Trade Center Buildings 1, 2, and 7. Fire Technology, 49:615–642, 2013.
12. S. Welch, S.D. Miles, S. Kumar, T. Lemaire, and A. Chan. Firestruc - integrating
advanced three-dimensional modelling methodologies for predicting thermo-mechanical
behaviour of steel and composite structures subjected to natural fires. Fire Safety
Science, 9:1315–26, 2008.
13. K. McGrattan, S. Hostikka, R. McDermott, J. Floyd, C. Weinschenk, and K. Overholt.
Fire Dynamics Simulator, User’s Guide. National Institute of Standards and Technol-
ogy, Gaithersburg, Maryland, USA, and VTT Technical Research Centre of Finland,
Espoo, Finland, sixth edition, September 2013.
14. C. Zhang and G.Q. Li. Fire dynamic simulation on thermal actions in localized fires in
large enclosure. Advanced Steel Construction, 8:124–36, 2012.
15. W. Jahn, G. Rein, and J.L. Torero. A posteriori modelling of the growth phase of
dalmarnock fire test one. Building and Environment, 46:1065–73, 2011.
16. K. Prasad and H.R. Baum. Coupled fire dynamics and thermal response of complex
building structures. Proceedings of the Combustion Institute, 30:2255–62, 2005.
17. D. Duthinh, K.B. McGrattan, and A. Khaskia. Recent advances in fire-structure anal-
ysis. Fire Safety Journal, 43:161–7, 2008.
18. C. Zhang and G.Q. Li. Thermal response of steel columns exposed to localized fires -
numerical simulation and comparison with experimental results. Journal of Structural
Fire Engineering, 2:311–7, 2011.
19. N. Tondini, O. Vassart, and J.M. Franssen. Development of an interface between cfd
and fe software. In Proceedings of the Seventh International Conference on Stuctures
in Fire, pages 459–68, Zurich, Switzerland, 2012.
20. J. Alos-Moya, I. Paya-Zaforteza, M.E.M. Garlock, E. Loma-Ossorio, D. Schiffner, and
A. Hospitaler. Analysis of a bridge failure due to fire using computational fluid dynamics
and finite element models. Engineering Structures, 68:96–110, 2014.
21. L. Chen, C. Luo, and J. Lua. Fds and abaqus coupling toolkit for fire simulation and
thermal and mass flow prediction. Fire Safety Science, 10:1465–77, 2011.
22. D. Banerjee. A software indepdent tool for mapping thermal results to structure model.
Fire Safety Journal, 68:1–15, 2014.
23. D. Kamikawa, Y. Hasemi, K. Yamada, and M. Nakamura. Mechanical response of a
steel column exposed to a localized fire. In Proceedings of the Fourth International
Workshop on Stuctures in Fire, pages 225–34, Aveiro, Portugal, 2006.
24. J.G. Silva. Tridimensional Interface Model to Fire-Thermomechanical Analysis of
Structures under Fire Conditions (in Portuguese). D.Sc. Thesis, Civil Engineering
Program - COPPE/UFRJ, 2014.
25. J.G Silva, A. Landesmann, and F. Ribeiro. Interface model to fire-thermomechanical
performance-based analysis of structures under fire conditions. In Proceedings of the
Fire and Evacuation Modeling Technical Conference 2014, 2014.
26. W.E. Luecke, J.L. Gross, and J.D. Mccolskey. High-temperature, tensile, constitutive
models for structural steel in fire. AISC Engineering Journal, 2014. Accepted for
publication.
27. H. Mostafaei, M. Sultan, and A. Kashef. Resilience assessment of critical infrastructure
against extreme fires. In Proceedings of the 8th International Conference on Structures
in Fire, pages 1153–1160, 2014.
28. W.K. Chow. Assessment of fire hazard in small news agents in transport terminal halls.
Journal of Architectural Engineering, 11:35–38, 2005.
29. J. Stern-Gottfried and G. Rein. Travelling fires for structural design part 1: literature
review. Fire Safety Journal, 54:74–85, 2012.
⋆ Simulation Methodology for Coupled Fire-Structure Analysis: 15

30. K. McGrattan, S. Hostikka, R. McDermott, J. Floyd, C. Weinschenk, and K. Overholt.

Fire Dynamics Simulator, Technical Reference Guide. National Institute of Standards
and Technology, Gaithersburg, Maryland, USA, and VTT Technical Research Centre of
Finland, Espoo, Finland, sixth edition, September 2013. Vol. 1: Mathematical Model;
Vol. 2: Verification Guide; Vol. 3: Validation Guide; Vol. 4: Configuration Management
Plan.
31. U. Wickstrom. The plate thermometer - a simple instrument for reaching harmonized
fire resistance tests. Fire Technology, 30:195–208, 1994.
32. C. Zhang and A. Usmani. Thermal calculation of structures in fire: new insights from
heat transfer principles. Fire Safety Journal, 2014. Submitted.
33. U. Wickstrom, D. Duthinh, and K.B. McGrattan. Adiabatic surface temperature for
calculating heat transfer to fire exposed structures. In Proceedings of the 11the Inter-
national Interflam Conference, pages 943–53, London, England, 2007.
34. C. Zhang, G.Q. Li, and R.L. Wang. Using adiabatic surface temperature for thermal
calculation of steel members exposed to localized fires. International Journal of Steel
Structures, 13:547–58, 2013.
35. BSI. Eurocode 1: Actions on structures - Part 1-2: General rules - Actions on structures
exposed to fire. British Standard, 2002.
36. ANSYS. ANSYS User Mannual, Version 14.0. ANSYS Inc., 2012.
37. A. Hamins, A. Maranghides, K.B. McGrattan, E. Johnsson, T. Ohlemiller, M. Donnelly,
J. Yang, G. Mulholland, K. Prasad, S. Kukuck, R. Anleitner, and T. McAllister. Federal
Building and Fire Safety Investigation of the World Trade Center Disaster: Experiments
and Modeling of Structural Steel Elements Exposed to Fire. NIST NCSTAR 1-5B,
National Institute of Standards and Technology, Gaithersburg, Maryland, September
2005.
38. Y. Dong and K. Prasad. Experimental study on the behavior of full-scale composite steel
frames under furnace loading. Journal of Structural Engineering, 135(10):1278–1289,
2009.
39. BSI. Eurocode 3: Design of steel structures - Part 1-2: General rules - Structural fire
design. British Standard, 2005.
40. G.Q. Li and C. Zhang. Creep effect on buckling of axially restrained steel columns in
real fires. Journal of Constructional Steel Research, 71:182–88, 2012.
41. L.T. Phan, T.P. McAllister, J.L. Gross, and M.J. Hurley. Best practice guidelines
for structural fire resistance design of concrete and steel buildings. NIST technical note
1681, National Institute of Standards and Technology (NIST), Gaithersburg, Maryland,
2010.
16 Chao Zhang et al.

Fig. 1 Coupled CFD-FEM simulation approach for structural fire analysis

⋆ Simulation Methodology for Coupled Fire-Structure Analysis: 17

Fig. 2 Illustration of the exposed surfaces and the mapping procedure: a) thermomechanical
model b) fire simulation
18 Chao Zhang et al.

Fig. 3 Experimental setup in Kamikawa et al’s test [23]

⋆ Simulation Methodology for Coupled Fire-Structure Analysis: 19

Fig. 4 Loading and restraint conditions in Kamikawa et al’s test [23]

20 Chao Zhang et al.

Fig. 5 Stress-strain curves for the steel in Kamikawa et al’s test [23] (the stress-strain
curves calculated by the constitutive model given by Luecke et al. [26] are also presented.)
⋆ Simulation Methodology for Coupled Fire-Structure Analysis: 21

Fig. 6 Specific heat for structural steel specified in the Eurocode [39]
22 Chao Zhang et al.

Fig. 7 Thermal conductivity for structural steel specified in the Eurocode [39]
⋆ Simulation Methodology for Coupled Fire-Structure Analysis: 23

Fig. 8 FDS model for Kamikawa et al’s test [23]: Front view (left) side view (center) and
computational domain (right)
24 Chao Zhang et al.

Fig. 9 FEM models for Kamikawa et al’s test [23]: FE thermal/structural model (left) and
boundary conditions (center and right). The restraint conditions shown are for the test Case
4.
⋆ Simulation Methodology for Coupled Fire-Structure Analysis: 25

Fig. 10 FDS predicted flame behavior at 360 s (left), 900 s (center) and 3600 s (right)
26 Chao Zhang et al.

Temperature
◦C
720

620

520

420

320

220

120

20

Fig. 11 FDS predicted adiabatic surface temperature at 360 s (left), 900 s (center) and
3600 s (right)
⋆ Simulation Methodology for Coupled Fire-Structure Analysis: 27

Fig. 12 Time-averaged adiabatic surface temperature distributions along the mid-line of

three column surfaces
28 Chao Zhang et al.

Temperature
◦C
1,000
900
800
700
600
500
400
300
200
100
20

Fig. 13 FDS predicted gas temperature slices at 360 s (left), 900 s (center) and 3600 s
(right) along the centerline of the burner and column.
⋆ Simulation Methodology for Coupled Fire-Structure Analysis: 29

Fig. 14 FEM predicted steel temperature distributions using the AST method (top) and
the FTMI method (bottom) at three different times: 360 s (left), 900 s (center) and 3600 s
(right)
30 Chao Zhang et al.

Fig. 15 Comparison between measured and predicted steel temperatures incase 1 in

Kamikawa et al’s test [23]. Front and corner measurements are at z = 400 mm and the
side and back measurements are z = 600 mm.
⋆ Simulation Methodology for Coupled Fire-Structure Analysis: 31

Fig. 16 Comparison between measured and predicted lateral displacements at 4 different

heights along the column in case 1 in Kamikawa et al’s test [23]
32 Chao Zhang et al.

Fig. 17 Comparison between measured and predicted vertical displacements in case 1 in

Kamikawa et al’s test [23]
⋆ Simulation Methodology for Coupled Fire-Structure Analysis: 33

Fig. 18 Comparison between measured and predicted vertical displacements in case 4 in

Kamikawa et al’s test [23]
34 Chao Zhang et al.

Fig. 19 Illustration of x displacement distribution and local buckling achieved by the FTMI
coupling method: a) perspective view; b) expanded views of the buckling region; c) picture
from experiment [23]