Fire Safety Journal 83 (2016) 6678

Contents lists available at ScienceDirect

Fire Safety Journal

journal homepage: www.elsevier.com/locate/firesaf

Fire-thermomechanical interface model for performance-based

analysis of structures exposed to re
Julio Cesar G. Silva a, Alexandre Landesmann b,n, Fernando Luiz B. Ribeiro b
a
b

Fire Research Division, National Institute of Standards and Technology (NIST/USA), United States of America
Civil Engineering Program, Federal University of Rio de Janeiro (COPPE/UFRJ), POB 68506, 21945-970, Brazil

art ic l e i nf o

a b s t r a c t

Article history:
Received 31 March 2015
Received in revised form
4 April 2016
Accepted 23 April 2016
Available online 27 May 2016

This paper presents an interface model to perform a one-way coupling between a re simulation (CFD
model) and a structural analysis (FEM model) aimed at performance-based analysis of structures exposed
to re. The Fire-Thermomechanical Interface (FTMI) model is capable of processing the results from a re
simulation to properly account for the heat transfer by convection and radiation, between the re and
the exposed surfaces, based on Adiabatic Surface Temperature concept. The methodology is presented
and veried against simple cases, and the improvements required to achieve complex geometries are
introduced. An application is also presented evaluating the re-thermomechanical behavior of an
H-prole column under a localized re. At the end of the analysis, it is possible to obtain the structural
behavior under specic re scenarios. An automated procedure is created to surpass the isolated member
analysis, allowing the simulation of the behavior of global structures discretized with shell and/or solid
elements under re conditions. In these examples, both solid and shell elements are used to demonstrate
that the procedure can be applied to evaluate the global behavior of structures. The results also suggest
that the methodology can provide reliable performance-based analyses.
& 2016 Elsevier Ltd. All rights reserved.

Keywords:
Coupled analysis
Fire-structural interaction
Thermomechanical behavior

1. Introduction
Traditionally, the performance of structures under re conditions has been achieved by prescriptive procedures, available from
international codes and standards [13]. Those procedures are
generally based on furnace test results, and focused on checking if
isolated structural members meet a required re resistance time
by means of analytical formulations. These do not account for the
system behavior including connections, second order effects (catenary or membrane action) or large displacements. Nowadays,
advanced numerical models based on Finite Element Method
(FEM) are capable of predicting the global behavior of structures
including large displacements and material nonlinearities. However, the application of these models to re conditions is generally
based on simplied temperature-time curves [2]. These temperature-time curves may not accurately represent the re development, and generally do not account for the three-dimensional fuel
distribution or the re compartment geometry.
On the other hand, numerical models based on the Computational Fluid Dynamics (CFD) are capable of providing a reliable
n

Corresponding author.
E-mail addresses: [email protected] (J.C.G. Silva),
[email protected] (A. Landesmann), [email protected] (F.L.B. Ribeiro).
http://dx.doi.org/10.1016/j.resaf.2016.04.007
0379-7112/& 2016 Elsevier Ltd. All rights reserved.

description of re evolution and have the capability to simulate

the actual re dynamics for different scenarios. Fire Dynamics Simulator (FDS, [4]) is an open source CFD code developed by National Institute of Standards and Technology (NIST) for re simulations. Over the last decade, this code has been extensively used
for re engineering and has been validated for a wide range of
applications [5,6].
Despite CFD and FEM being mature research elds, coupled rethermomechanical analysis (CFD-FEM) is a relatively new area of research [712]. After the collapse of the World Trade Center towers
Prasad and Baum [7] proposed an interface between the FDS [4] and
the ANSYS1 package [13] to investigate the behavior of structural
elements during this event. Their method was called Fire Structural
Interface (FSI) and assumed that the heat transfer between re and
exposed surfaces was given only by radiation. Following Prasad and
Baum, a European research project called FIRESTRUC [8] analyzed a
number of ways to perform an interaction between CFD and FEM
codes focusing on the behavior of structures under re. Among the
1
Certain commercial entities, equipment, or materials may be identied in this
document in order to describe an experimental procedure or concept adequately.
Such identication is not intended to imply recommendation or endorsement by
the National Institute of Standards and Technology, nor is it intended to imply that
the entities, materials, or equipment are necessarily the best available for the
purpose.

J.C.G. Silva et al. / Fire Safety Journal 83 (2016) 6678

CFD codes considered, FDS has the advantage of including a combustion model to address the re growth.
One of the big concerns stated by FIRESTRUC was related to determining which variables should be transferred from re simulations
to FEM models. The Adiabatic Surface Temperature (AST) was proposed by Wickstrm [14] as a variable capable of describing complex
convective and radiative conditions into one single scalar quantity. In
sequence, an interface between FDS and ANSYS was presented by
Wickstrm et al. [15] and Duthinh et al. [16] using the AST concept. At
this point, the coupling procedure was only applied to isolated
structural members such as trussed beams or columns. The main issues related to this coupling include the handling of different geometries, the time scale, and the amount and format, of data to be

The procedure decomposes this

domain into two parts:

67

transferred between models.

The main purpose of this paper is to provide an interface model
to performance-based analysis of structures under re conditions.
The Fire-Thermomechanical Interface (FTMI) improves the reach
of re engineering by providing an automated code to extract the
data from re simulations, transform the information, and prescribe the correct boundary conditions to the thermomechanical
analysis. This automated procedure surpasses the isolated member
analysis allowing the simulation of the behavior of global structures discretized with shell and/or solid elements under re conditions. In the next sections the FTMI methodology is presented. In
this paper, FDS [4] is used for the re simulation and the commercial package ANSYS [13] is used for the thermomechanical
analysis. The interface method, however, can be easily applied to
other computer models. Verication examples and an application
case are provided to show FTMI applicability for solids and shell
elements.

2. Fire-Thermomechanical Interface: FTMI

a)
Fire simulation
Thermal exposure of structures
domain boundaries

b)
exposed surface

Thermomechanical model
Global behavior of the structure

The description of structural behavior under re conditions by

a re-thermomechanical model is related to a domain that includes the structure and its components. The boundary conditions
include the thermal loads (re model) and mechanical loads
(structural model). The thermomechanical problem needs to address the differences in the physical phenomena involved between
the two types of analysis.
The procedure described in this paper decomposes re-thermomechanical model domain, illustrated in Fig. 1, into two parts: the rst
part is devoted to re simulation and the second is focused on the
thermomechanical behavior. In the re simulation, the structure
geometry is simplied and the overall domain size extends beyond
the structure to properly capture the re propagation and the smoke
and hot gas ow (cf. Fig. 1b). For the thermomechanical analysis, only
the structure is modeled and the re simulation is represented by heat
uxes, applied as boundary conditions at the exposed surfaces, as
shown in Fig. 1c. In order to exchange data, both models have the
same coordinate system and a consistent geometry (Fig. 1).
This approach is commonly referred to as one-way coupling. In a
two-way coupling strategy, the thermomechanical results are transposed back to the re simulation (i.e. displacements, collapses, etc).
The two-way approach can lead to a more complex simulation, increasing the amount of data to be transferred between the models. Its
advantages are related to cases where displacements and/or collapses
can change the ventilation and thereby the re source or the uid
ow pattern, creating a different re scenario. With one-way coupling
it is possible to develop each model separately, by different model
users/developers, each one with their own expertise. In addition, the
one-way procedure can be achieved for different discretization levels
(FEM mesh) and small modications or dimensioning does not imply
the entire calculation to be restarted, as the structures geometry is
simplied for the re simulation.
2.1. Heat transfer from res
Heat can be transferred from ames and hot gases to a structure's surfaces by radiation and convection, as illustrated in Fig. 2.
The total heat ux (qtot
) is dened by the sum of these two parcels:

qtot
= qrad
+ qconv

c)

Beam, shell and solid elements

Fig. 1. Illustration of the coupled eld domain decomposition: (a) problem description; (b) re simulation domain; (c) thermomechanical analysis discretization.

(1)

) is obtained through the balance

The radiative heat ux ( qrad
between the radiative energy absorbed ( er, abs ) and emitted ( er, emi )
by the surface and can be represented by the following equation:

= er, abs er, emi

qrad

(2)

68

J.C.G. Silva et al. / Fire Safety Journal 83 (2016) 6678

2.2. Surface thermal exposure

flames and hot gases

In order to establish an accurate interface between the re simulation and the thermomechanical analysis, the Adiabatic Surface Temperature concept is used [14]. In this approach the real
surface is replaced by a perfectly insulated surface exposed to the
same heating conditions3; the total heat ux to this ideal surface is
by denition zero [15]:

exposed surface

4
er, inc ( TAST ) + h ( Tg TAST ) = 0

Fig. 2. Illustration of the heat ux from the re source (ame and hot gases) to the
exposed surfaces of the structure.

The radiative energy absorbed by the exposed surface can be

expressed as part of the radiative incident energy ( er,inc ), in other
words, the absorbed part can be considered by multiplication of
the incident energy by a coefcient, absorptivity (), in the following equation:

er, abs = er, inc

(3)

Note, the reected energy ( er, ref ) is discarded and does not take
part in this balance. According to the StefanBoltzmann law, a
surface emits radiative thermal energy following:

er, emi = (Ts )4

(4)

where is the emissivity, s is the StefanBoltzmann constant and

Ts is the temperature of the exposed surface in Kelvin2.
Kirchhoff's law of thermal radiation states that emissivity and
absorptivity are equal. Thus Eqs. (3) and (4) can be combined and
the radiative heat ux can be written as:

= er, inc (Ts )4

qrad

(5)

) is dened by Newton's law of

The convective heat ux ( qconv
cooling, and depends on the difference between the gas temperature adjacent to the exposed surface (Tg) and the surface
temperature; it can be expressed as:

= h ( Tg Ts )
qconv

(6)

where h is the convective heat transfer coefcient, which is proportional to the velocity eld around the surface.
Based on the presented denitions of radiative and convective
heat uxes, Eq. (1) can be reorganized as:

qtot
= er, inc (Ts )4 + h ( Tg Ts )

(7)

Advanced re simulation models, such as the code used in this

paper, provide results that characterize the three dimensional evolution of the re, including the radiative thermal energy incident on the
exposed surfaces and the gas temperatures. However, the models are
not capable of accurately characterizing the temperature distribution
on solids, and thus the surface temperature. Consequently, the total
heat ux in Eq. (7) cannot be precisely calculated at the end of the re
simulation. Therefore, an alternative approach is required.
2

The methodology is presented in Kelvin, but the results are scaled to Celsius
degrees to facilitate the reading.

(8)

and the temperature of this idealized surface, referred to here as

the adiabatic surface temperature (TAST), can be obtained as an
output from the re simulation.
Since the real and idealized surfaces are exposed to the same
heating conditions, the total heat ux that will be applied to the
thermomechanical model can be obtained by the solution of the
system formed by (Eqs. (7) and 8):

q = e (T )4 + h T T
r, inc
( g s)
s
tot

er, inc ( TAST ) + h ( Tg TAST ) = 0

(1)

4
4
= ( TAST ) ( Ts ) + h ( TAST Ts )
qtot

(9)

This single variable (TAST) is considered capable of reducing the

complexity of the re simulation into one simple scalar by several
previous researches [1518].
To achieve a correct denition of the total heat ux, accounting for
both the convective and radiative heat uxes, it is also necessary to
add the convective heat transfer coefcient (h). Following normative
procedures (EC1p1.2 [2]) h is usually considered as a constant [16], but
depending on the chosen value, h may diminish or amplify the relative importance of the convective heat ux for a specic re scenario
(relative to the radiative heat ux). By including its spatial distribution
and evolution throughout the duration of a re, h can help to precisely
transcribe the heat ux from a re simulation to thermomechanical
analysis.
The two variables (TAST, h) are denominated here as surface
thermal exposure. After the re simulation, the heat ux is evaluated
at the thermomechanical model using Eq. (9), depending on the surface thermal exposure (TAST, h) obtained by the re simulation, and the
surface temperature (Ts), calculated at each time step during the
thermomechanical analysis.
2.3. Models in which geometries do not match perfectly
One of the intrinsic differences between CFD and FEM models
is related to their discretization approaches (i.e., the size and shape
of the cells or elements used in each eld). For an FDS re simulation, the domain is discretized into orthogonal cells (or hexahedrons) in a rectilinear grid. For the thermomechanical analysis,
the structure is usually discretized with solids (tetrahedrons or
hexahedrons) or shells (3D plane) elements. Also, the element size
required to achieve a good resolution in the thermomechanical
model would lead to an unfeasible re simulation. Therefore, even
for the most modest structures, with the simplest geometries, the
models generated for each side of this interface do not match
perfectly.
In order to establish the connection between these two models,
the available re simulation results need to be mapped to the external
faces of the elements (FEM). The re simulation results, TAST and h,
3
The adiabatic surface temperature concept does not imply that all the surfaces are modelled as adiabatic surfaces; this replacement is part of the methodology formulation.

J.C.G. Silva et al. / Fire Safety Journal 83 (2016) 6678

depend on the orientation of the cell face where they are calculated
and this orientation needs to be accounted for before the results can
be prescribed as boundary conditions. This mapping is realized by a
collection of I keypoints (of x coordinates) localized at the center of
each external face (with normal n), as illustrated in Fig. 3. The position
and the number of I keypoints is based on the mesh generated for the
thermomechanical analysis (Fig. 3a). In this way, the coupling procedure can be achieved for different discretization levels or even slightly
different structural models (by dimensioning process or small
changes).
With the intention to map the different models and get these results automatically, a code called fds2ftmi was created. This code is
based on fds2ascii routine, available as part the FDS package [4]. Basically, fds2ftmi traces the exposed surfaces from ANSYS and then
collects the I keypoints and their corresponding normal (n) directions
(related to the element surface). Based on each I keypoint position, the
code is able to search into the boundary results le (.bf) from FDS,
iterate over time, orientation, and meshes to transcribe the correct
surface thermal exposure results (TAST, h) into ASCII text les. To
prescribe these variables correctly as boundary conditions in the
thermomechanical model, the code also outputs a le in ANSYS APDL
language script format.
The described procedure can be easily understood for solid elements with aligned surfaces as illustrated in Fig. 3. Nevertheless, the
extension of this procedure to more types of elements and geometries
is a more challenging task. The temperature distribution through a
beam element cross section is generally handled by a separate thermal analysis. In this procedure, beam elements can be used in the
thermomechanical analysis, but only in areas not directly affected by
the re; making use of its longitudinal thermal conduction. For geometries discretized with shell elements, fds2ftmi places I keypoints on

nodes

I keypoints
x, n

a)

69

both sides of the planar elements to capture the correct orientation by

the normal direction and the thermal gradient through the shell layers. Further discussion about the application of this procedure for
complex geometries, curved or oblique to the Cartesian axis, will be
handled together with the verication cases.
2.4. Heat ux into FEM models
At the FEM model, the main process related to this interface procedure is an iterative solution capable of using the surface temperature, obtained at each time step during the thermomechanical analysis, to evaluate the heat ux at each node of the exposed surface
through Eq. (9). Therefore, the surface effect element SURF152 (ANSYS
nomenclature) is used; this element can apply a heat ux vector at
each node of the exposed surface based on Eq. (9), as illustrated in
Fig. 4. In this task, the TAST needs to be prescribed at the element's
extra node and h is applied at the element's surface (Fig. 4). If this
procedure is used with shell elements, the heat ux vector is calculated and prescribed at the top and bottom layers of the shell section.
2.5. Verication cases
2.5.1. Panel A
In order to verify the FTMI, a steel panel is exposed to a localized
re. The temperature distribution obtained is compared with FDS
results. FDS has a 1D thermal conduction model for solids, so to reduce the inuence of this simplication, a simple plane geometry is
considered. The structure is aligned with Cartesian coordinates (parallel to yz plane) with a 1.5 m width, 1.0 m height and 1 cm thickness
(cf. Fig. 5). The simulation domain is a 1.5 m  1.5 m  2 m open box
with 36,000 cubic cells (5 cm edges). The re scenario is the leakage of
0.1 l/min of methane at 2 m/s, corresponding to a re of approximately 62.5 kW. In this example, the thermal properties are considered constant, with the values taken from EC3p1.2 [3] for ambient
temperature. The re source is on the ground and the distance between the plane and the re source is 30 cm. The goal of this simulation is to evaluate whether FTMI is capable of creating an interface
that reproduces the heat transfer (item 2.1) between ames and hot
gases (from CFD) to structures surfaces (FEM).
The surface temperature obtained by both methods (FDS and
FTMI) is compared. Because of its 1D heat transfer model, FDS is not
able to capture heat conduction across the panel (in yz plane Fig. 5).
This is important if the material has a high thermal conductivity (e.g.
steel). To match both models and to keep the comparison at the level
of the interface between the gas and solid phases, the ANSYS thermal
model was adapted to perform just 1D conduction through the panel
(x axis).

FEM model

TAST

qtot

I keypoints

TAST

Ts

h TAST Ts

TAST

qtot
h

Available results
TAST, h
b)

q node

qtot Aelem
n
nnodes

CFD model

Fig. 3. Illustration of the exposed surfaces and the mapping procedure: (a) thermomechanical model; (b) re simulation.

Fig. 4. Illustration of the thermal exposure (TAST, h) transcription into boundary

conditions by SURF152 element.

70

J.C.G. Silva et al. / Fire Safety Journal 83 (2016) 6678

TAST

(C)

450

400

350
300

250

200

E
C

150

F
100

50

1.2m

a)

a)

0
0

h7
A

10
t (min)

(W/m.K)

0.50m

0.25m

B
E

b)
20C

0.25m

0.25m

0.25m

A B C

180C

300C

420C

0.5m
D E F

0.25m
0.25m 0.25m
0.25m

Fig. 5. Illustration of the re scenario for panel A with TAST distribution at 5 min:
(a) 3D model; (b) front view.

b)
The adiabatic surface temperature (TAST) distribution at 5 min of
re elapsed time is presented in Fig. 5a. The central region of the panel
has the highest temperature values and the distribution is approximately radial. The ame shape changes the circular pattern to a more
elliptical silhouette. The evolution of the thermal exposure is presented in Fig. 6. As expected, the TAST is almost steady, around 400 C
at A and D and 150 C at C and F (Fig. 6b). The convective heat transfer
coefcient is dependent of the velocity eld close to the surface,
generating oscillations at the higher elevations of the plane (A, B and
C - Fig. 6b).
After the re simulation, the thermal exposure results are prescribed as boundary conditions at the thermal model based on FTMI
(by fds2ftmi code). The surface temperature (Ts) distribution for this
1D thermal model is illustrated in Fig. 7. The Ts distribution has the
same prole as TAST (Fig. 5), but with higher values in the central region. The addition of a h distribution in FTMI (which also has higher
values at central region Fig. 6b) corrects the heat ux distribution
(Eq. 9), increasing the temperature gradient in this region. At 10 min
into the simulation, Ts at A is 164.4 C and just 68.7 C at C, which
correspond to 42% of Ts at A; at the same instant, the TAST at C is 60% of
the TAST at A.
The comparison between the surface temperature evolution obtained by FTMI and FDS is presented in Fig. 8. The maximum difference between the results is about 0.5% (at A). These results, show that

t (min)

Fig. 6. Evolution of the thermal exposure: (a) adiabatic surface temperature (TAST);
(b) convective heat transfer coefcient (h).

the methodology is capable of replicating the temperature boundary

conditions in FEM.
After comparison between FTMI and FDS temperature results, it
is important to keep in mind that the temperature distribution
calculated by FDS (and presented in Fig. 8) does not address the 3D
heat conduction in solids, even for simple geometries such as this
panel. In fact, that is the motivation to create a different approach
to calculate the heat ux on exposed surfaces (Section 2.2). This
study (panel A) was conducted to compare just the interface
methodology, and therefore the FEM thermal model was adapted
to consider just the unidirectional heat conduction across the
panel.
Implementing the 3D thermal model with temperature dependent material properties (as suggested in EC3p1.2 [3]), FTMI is
now applied to the same case to compare the distribution and
evolution of surface temperature, both presented in Figs. 9 and 10
respectively, with the 1D case presented previously. With the
addition of the 3D thermal model, it was expected that the thermal energy would be able to dissipate through the yz plane
creating a decrease in the central area's temperature and an

J.C.G. Silva et al. / Fire Safety Journal 83 (2016) 6678

Ts

71

(C)

A B C
A

160

D E F

0.5m
D

0.25m
0.25m 0.25m
0.25m

140
120

FDS
FTMI

100
80

a)

60

40

F
20
0
0

t (min)

Fig. 8. Evolution of surface temperature, comparison between FTMI (1D) and FDS
results.

b)

in general complex geometries and even simple panels have stiffeners or adjacent elements that can change the temperature
distribution. The next example discusses the applicability of this
methodology to complex geometries and the results are compared
with the panel A 3D thermal results (Figs. 9 and 10).

c)

d)
20C

84C

132

180

Fig. 7. Distribution of the surface temperature: (a) 2 min; (b) 5 min; (c) 7.5 min; (d)
10 min (1D thermal model).

increase in the lateral area's temperatures, (see Figs. 7 and 9 for

comparison). The temperature at A achieved 134.3 C after 10 min
whereas the maximum temperature at the panel was 154 C, about
82% and 85% of the temperature calculated by the 1D thermal
model, 163.4 C and 181.5 C, respectively (Fig. 10).
In this simplied case, the structure was represented by a plane
surface, aligned with the Cartesian axes. However, structures have

2.5.2. Panel B
Panel B has the same geometry as panel A, but it is not aligned
with the Cartesian axes. It is generated by the rotation of the panel
A about z axis, as shown in Fig. 11.
The re scenario and the distance from the plane to the
re source were kept the same as in panel A to allow direct
comparison between results. Fig. 12 shows the re scenario
and the adiabatic surface temperature distribution at 5 min
of elapsed time. This panel is oblique to the xy axis. FDS currently
requires the geometry to be discretized by regular hexahedrons,
so this inclined panel is therefore built by rectangular blocks
in a technique often referred to as stair-stepping. The goal of this
case is to analyze how FTMI can handle complex geometries and
how the results are affected by the stair stepping technique
(Fig. 12b).
The adiabatic surface temperature distribution in Fig. 12 shows
that the maximum temperature values are not in the center as
they were in panel A. The highest values are associated with the
shape factor between the ame position and the cell face alignment (numerical discretization). According to Eq. (8), the TAST is
dependent on radiative incident energy and the convective heat
transfer coefcient. Since those quantities are both orientation
dependent, TAST becomes orientation dependent too.
To include this form factor between the ame and the cell face
orientation and the actual surface alignment, another approach is
introduced for complex geometries. Fig. 13 presents an illustration
of this procedure for a generic surface. If there are contributions in
x, y and z for variables TAST and h, those contributions are considered dependent on the orientation for which they were calculated by the multiplication of the variable by the unit vector in
each direction. The vectors vAST and vh can be dened by the sum
of these contributions:

vAST = TAST , xi + TAST , yj + TAST , zk

(10)

72

J.C.G. Silva et al. / Fire Safety Journal 83 (2016) 6678

Ts (C)

A B C

160

D E F

A (1D)

0.5m

0.25m
0.25m 0.25m
0.25m

140
120

100
80

a)

B
E

60

40

F
20
0
0

t (min)

Fig. 10. Evolution of surface temperature, comparison between FTMI (3D) and FDS
(1D) results.

b)

c)
a)

d)
20C

72C

111C

150C

Fig. 9. Distribution of the surface temperature (3D thermal model): (a) 2 min; (b)
5 min; (c) 7.5 min; (d) 10 min.

b)

Fig. 11. Sketch of the panel rotation around z axis: (a) panel A; (b) panel B.

vh = hxi + hyj + hz k

(11)

where i is the unit vector in x direction, j is the unit vector in y

direction and k is the unit vector in z direction.

After this sum of contributions, the orientation of the real

surface is included by projecting the vectors vAST and vh onto the
normal direction and considering that the modulus of these

J.C.G. Silva et al. / Fire Safety Journal 83 (2016) 6678

73

a)
Fig. 13. Illustration of the thermal exposure characterization for complex
geometries.

b)
140C

20C

230C

320C

Fig. 12. Illustration of the re scenario for panel B with TAST distribution at
5 min: (a) 3D model; (b) frontal view.

projections can represent the thermal exposure at n direction:

vAST , n = TAST , n =

vh, n = hn =

vAST n
vAST
vAST n

vhn
vh
vh n

(12)

(13)

The total heat ux can be calculated by Eq. (9), substituting the

thermal exposure variables by the values calculated by Eqs. (12)
and (13). This denition of the heat ux by the vector composition
of the thermal exposure can lead to some inaccuracy since the
radiative heat ux is dependent on the forth power of TAST.
However, this is inversely proportional to the level of discretization of the complex geometries.
This additional treatment for complex geometries corrected the
prole presented in Fig. 12 and created a distribution prole where
the highest temperatures are concentrated in central region. A
comparison between the thermal exposure variables obtained for
panels A 3D and B is presented in Fig. 14. At A (central region), the
results for panel B showed a good agreement with the panel A 3D

results for both variables (TAST and h). At the lateral regions, at C,
the vector treatment created higher temperature results; this
difference is amplied for convective heat transfer coefcient as
can be seen on Fig. 14b. Since h is dependent on the velocity eld
around the surface, the stair stepping technique can create errors
in the velocity eld around the blocks.
The spatial distribution of surface temperature for panel B is
presented in Fig. 15. Comparing this distribution with the obtained
results for panel A 3D, it is possible to observe that the vector
treatment created a hotter central region. The maximum temperature achieved at the panel surface was 165.8 C at 10 min of
re elapsed time, 7.6% higher than the maximum temperature
calculated for panel A 3D. The evolution of surface temperature is
presented in Fig. 16. The maximum temperature at A for panel B is
152.8 C, about 13.6% higher than panel A 3D. The same behavior
can be observed at C, D and F. It is important to point that even
with this difference, the results achieved for panel B are closer
than the results obtained by panel A 1D. The maximum temperature for panel A 1D (FDS) is 181.5 C, 17.6% higher than panel A
3D (at A this difference escalates to 21.8%). Also, the unidirectional
heat conduction model required geometry aligned with the Cartesian axes to generate even an approximation of the temperature
distribution in the solid phase. If FTMI and fds2ftmi are applied to
complex structures, such as curved geometries, sloped ceilings,
etc., the obtained results will be dependent on the FDS solution
around the structure (geometry), e.g. radiation, convective heat
transfer coefcient and the ow motion, which need to be handled
carefully.

3. Application
3.1. H-prole column
In this case a simply supported H-prole column is exposed to
a localized re. The steel column is 3 m tall. The cross section is a
0.3 m (ange)  0.4 m (web) with a 12.5 mm thick web and 16 mm
thick anges. The re scenario is a 200 kW pool re

74

J.C.G. Silva et al. / Fire Safety Journal 83 (2016) 6678

T500
AST

(C)

450

400
350
300

250
200

a)

C
150
100
50
0
0

a)

t (min)

8h (W/m.K)

b)

Panel B
Panel A 3D

6
5
4

c)

1
0

b)

t (min)

Fig. 14. Evolution of the thermal exposure, comparison between panel B and panel
A 3D: (a) adiabatic surface temperature (TAST); (b) convective heat transfer coefcient (h).

(20 cm  20 cm) located 40 cm (from pool center) away from the

web, as shown in Fig. 17. At ambient temperature (20 C) the yield
strength of the steel is 250 MPa and its Young's Modulus is
205 GPa. Thermal and mechanical properties are considered
temperature dependent as suggested in EC3p1.2 [3].
The column is discretized with shell elements (FEM) to demonstrate the applicability of FTMI to shell structures. Each side of
these plane elements will have a different thermal exposure, incident radiation and gas temperature around the surface. The
proposed methodology is designed to address this capability and
prescribe the correspondent heat ux at the shell elements top
and bottom layers.
The TAST distribution at 15 min of elapsed re time is presented
on Fig. 17. The pool re is close to the column, so the portion of the

d)
20C

72C

111C

150C

Fig. 15. Distribution of the surface temperature for panel B: (a) 2 min; (b) 5 min; (c)
7.5 min; (d) 10 min.

web facing the re will be hotter than the other parts of the cross
section (Fig. 17a). The FEM model is presented in Fig. 17b where
some points are highlighted to link with the results.
The adiabatic surface temperature results for each side of the
ange elements can be compared to illustrate difference of the

J.C.G. Silva et al. / Fire Safety Journal 83 (2016) 6678

Ts (C)

TAST

160

A
D

140
120

75

(C)
D1

500

A (Panel A 3D)

E1

400

100
300

80

F1
60
200

C (Panel A 3D)

40

20
0
0

t (min)

Fig. 16. Evolution of surface temperature, comparison between panel A 3D and

panel B results.

not exposed surfaces

C
F

web and flanges

facing fire

0.75m
B
E

0.75m
A
D

0.75m

a)

b)
20C

260C

440C

F2

E2

D2

100

620C

Fig. 17. Illustration of the H-prole column localized re scenario with TAST distribution at 15 min: (a) 3D model; (b) FEM model.

thermal exposure at the top and bottom layers of this plane element. The TAST evolution for points: D, E and F, located at a ange
close to the re source (Fig. 17b), are presented in Fig. 18 (the
subscript 1 is related to the layer facing the re source and subscript 2 to the layer facing outwards). At D1, the TAST is about
550 C during the re growth phase, and for D2 the average result
is 35 C. Since D2 is not facing the re, this result comes from the
hot gases that ow around that point. Heat ux needs to be accounted for at the points not facing the re to include the cooling

10

20

30

40

50

t (min)

Fig. 18. Evolution of the adiabatic surface temperature (TAST) as a function of re

elapsed time.

provide by heat exchange at those points. This can increase the

temperature gradient inside the ange. At E and F this effect is also
present, at E1 and F1 the TAST are around 325 C and 175 C, for E2
and F2 they are 32 C and around 20 C, respectively.
According to the prescriptive values presented at EC1p1.2 [2] a
convective heat transfer coefcient of 25 W/(m2 C) should be
used for ISO834 curve, this value was also used by Duthinh et al.
[16]. In this case, the maximum h for the points at the ange is
achieved at F1, with an average of 11.4 W/(m2 C). At D1 and E1, h is
around 9 W/(m2 C). At F2 h is around 5 W/(m2 C), which is closer
to the 4 W/(m2 C) presented in EC1p1.2 [2] for unexposed surfaces. For the remaining unexposed points, D2 and E2, h is between
7 W/(m2 C) and 6 W/(m2 C). These differences illustrate that the
convective heat transfer coefcient distribution can impact the
temperature distribution.
The distribution of the surface temperature is illustrated in
Fig. 19 and the evolution of this variable is presented in Fig. 20. The
maximum temperatures are achieved at the points close to the re
source, at A and D. At A1, the temperature is about 430 C at 1 h of
re exposure, and 425 C at A2 at the same time. At this temperature, the steel already presents yield strength reduction. For
D1 and D2, the temperature is about 369 C and 363 C respectively. The temperature decreases with the distance from re
source (B and E Fig. 20) and those points are more affected by the
thermal conduction (Fig. 19).
This simply supported column is subjected to a vertical load of
325 kN, which correspond to 1/50 of the Euler's buckling critical
load. Under this loading condition, the column will be close to a
uniform stress state and remain straight at the beginning of the
simulation; the horizontal displacements generated will be due to
the thermal gradient applied by the localized re. The von Mises
stress (S) distribution is illustrated in Fig. 21, the evolution of von
Mises stress and displacements () at keypoints in the column
(Fig. 17) follows in Fig. 22. At the beginning of the re, the stresses
are higher at the areas close to the re source (which also have the

76

J.C.G. Silva et al. / Fire Safety Journal 83 (2016) 6678

Ts

(C)

A1

400

A2
D1

350

D2

300
250

B1,B2

200

E1,E2
150

100
50
0
0

a)

b)

c)

10

20

30

40

50

t (min)

Fig. 20. Evolution of the surface temperature in function of re elapsed time.

As this column is simply supported, the vertical displacement is

not constrained, and the thermal expansion will make this column
to increase its size. The vertical displacement of about 7 mm after
1 hour of re simulation at the top of the column (G Fig. 17b) is
shown in Fig. 22.
This exercise demonstrated the applicability of FTMI to evaluate the
behavior of structures under re conditions. The routines and codes
created in this methodology were included in the FDS repository [4].
The predicted results compare to expected behavior, and future works
will be devoted to comparison against experimental data.

4. Conclusions

d)
20C

e)
110C

300C

430C

Fig. 19. Distribution of the surface temperature: (a) 5 min; (b) 10 min; (c) 15 min;
(d) 30 min; (e) 60 min.

highest temperatures Fig. 19). The expansion of the heated ange

areas create a bending moment due to the temperature gradient
(Fig. 21c), which will generate horizontal displacements (A, D - Fig.
21d, e and Fig. 22b). The Young's modulus and the proportional
limit of stress start to decrease after the temperature achieves
100 C (EC3p1.2 [3]). Those changes in the material properties will
reduce the stress state and increase the deformations at some
points of the structure (Fig. 20, Fig. 21d, e, f and Fig. 22a). As an
example, at point A, a peak in the von Mises stress is observed at
12 min (Fig. 22a) when the temperature is around 220 C. This
abrupt change is motivated by reduction in proportional limit and
in Young's modulus, which affects the structure even before the
yield strength reduction (which starts at 400 C, or 34 min for
A Fig. 20).

This paper presented the results of a proposed interface model

that links re simulation and thermomechanical analysis to deliver
a performance-based analysis of structures under re conditions.
The proposed one-way coupling procedure was able to extract the
needed variables to translate the surfaces' thermal exposure and
transform this information to overcome the intrinsic differences
between these two types of numerical models (CFD-FEM). This
interface approach has several advantages, including allowing the
models to be run independently, which means that small changes,
and/or redimensioning of the structure does not require a restart
of the re simulation. The interface was veried against simple
cases, showing that FTMI is capable of reproducing the thermal
exposure obtained by the re simulation. This methodology was
designed to allow the use of other FEM codes by changing the way
it writes the results. The usage of FTMI for complex geometries,
such as inclined or curved surfaces, was also discussed. Application cases showed that FTMI can be used for performance-based
analysis of structures under re conditions by means of solid and/
or shell elements.4 Beam elements can also be used in areas not
directly subjected to the re. This approach can increase the reach
of re engineering, allowing the simulation of the behavior of
entire structures under re conditions.
4
The authors suggest the use of shell elements in areas close to the re source,
in association with bar elements in regions not inuenced directly by the re. Also,
in the author's experience, the processing power used to run CFD simulations of
complex scenarios and large structures has been proved to be large enough to run
thermomechanical simulations of the structure behavior.

J.C.G. Silva et al. / Fire Safety Journal 83 (2016) 6678

a)

b)

77

c)

Fig. 22. Evolution of thermomechanical behavior: (a) von Mises stress; (b)
displacements.

d)
2MPa

e)
60MPa

f)
160MPa 250MPa
[3]

Fig. 21. Distribution of von Mises stress (displacements amplied 15  ): (a) start;
(b) 5 min; (c) 10 min; (d) 15 min; (e) 30 min; (f) 60 min.

[4]

[5]

Acknowledgement
The authors thankfully acknowledge the National Council for
Scientic and Technological Development, (Grant no. 229645/
2013-5) Brazil (in Portuguese: Conselho Nacional de Desenvolvimento Cientco e Tecnolgico, CNPq) for the nancial support
during this work.

[6]

References

[9]

[1] AISC/LFRD, Manual of Steel Construction Load and resistance factor design
specication for structural steel buildings, Chicago, 2005.
[2] EN 1991-1-2, Eurocode 1-Actions on Structures Part 12: General Actions

[7]

[8]

[10]

Actions on Structures Exposed to Fire, Comit Europen de Normalisation,

Brussels, 2002.
EN 1993-1 2, Eurocode 3-Design of Steel Structures Part 1.2: Structural Fire
Design, Comit Europen de Normalisation, Brussel, 2005.
K. McGrattan, S. Hostikka, R. McDermott, J. Floyd, C. Weinschenk, K. Overholt,
Fire Dynamics Simulator, User's Guide, National Institute of Standards and
Technology, Gaithersburg, Maryland, USA and VTT Technical Research Centre
of Finland, NIST Special Publication 1019, Espoo, Finland, 6th edition, 2013.
K. McGrattan, S. Hostikka, R. McDermott, J. Floyd, C. Weinschenk, K. Overholt,
Fire Dynamics Simulator, Technical Reference Guide, Volume 3: Validation,
National Institute of Standards and Technology, Gaithersburg, Maryland, USA
and VTT Technical Research Centre of Finland, NIST Special Publication 1018-3,
Espoo, Finland, 6th edition, 2013.
U.S. NRC, Verication and Validation of Selected Fire Models for Nuclear Power
Plant Applications, Volume 7: Fire Dynamics Simulator (FDS), Rockville, MD,
2007.
K. Prasad, H.R. Baum, Coupled re dynamics and thermal response of complex
building structures, in: Proceedings of the Combustion Institute, vol. 30, 2005,
pp. 22552262.
S. Kumar, S. Miles, S. Welch, O. Vassart, B. Zhao, A. Lemaire, L. Noordijk, J.
Fellinger, J. Franssen, FIRESTRUC integrating advanced three-dimensional
modelling methodologies for predicting thermo-mechanical behavior of steel
and composite structures subjected to natural res, 2006.
N. Tondini, O. Vassart and J.-M. Franssen, Development of an interface Between CFD and FE software, in: Proceedings of the 7th International Conference on Structures in Fire, Zurich, Switzerland, 2012.
Y. Xiaojun, A. Jeffers, A comparison of subcycling algorithms for bridging
disparities in temporal scale between the re and solid domains, Fire Saf. J. 59
(2013) 5561.

78

J.C.G. Silva et al. / Fire Safety Journal 83 (2016) 6678

[11] J. Alos-Moya, I. Paya-Zaforteza, M. Garlock, E. Loma-Ossorio, D. Schiffner,

A. Hospitaler, Analysis of a bridge failure due to re using computational uid
dynamics and nite element models, Eng. Struct. 68 (2014) 96110.
[12] J. Silva, A. Landesmann, F. Ribeiro, Performance-based analysis of cylindrical
steel containment vessels exposed to re, Fire Saf. J. 69 (2014) 126135.
[13] S.A.S. Ansys, Reference manual version 12, Swanson Anal. Syst. Inc (2009).
[14] U. Wickstrm, Heat transfer by radiation and convection in re testing, Fire
Mater. 28 (2004) 411415.
[15] U. Wickstrm, D. Duthinh, K. McGrattan, Adiabatic surface temperature for
calculating heat transfer to re exposed structures, in: Proceedings of

Interam, 2007.
[16] D. Duthinh, K. McGrattan, A. Khaskia, Recent advances in restructure analysis, Fire Saf. J. vol. 43 (2008) 161167.
[17] J. Sandstrm, U. Wickstrm, M. Velkovic, Adiabatic surface temperature: a
sufcient input data for a thermal model, in: Proceedings of Application of
Structural Fire Engineering, Prague, Czech Republic, 2009.
[18] U. Wickstrm, A. Robbins, G. Baker, The use of adiabatic surface temperature
to design structures for re, in: Proceedings of Structures in Fire (SIF), Michigan, USA, 2010.