Download as pdf or txt
Download as pdf or txt
You are on page 1of 25

Journal of Structural Fire Engineering

®
Modeling structures in fire with SAFIR : theoretical background and capabilities
Jean-Marc Franssen, Thomas Gernay,
Article information:
To cite this document:
Jean-Marc Franssen, Thomas Gernay, (2017) "Modeling structures in fire with SAFIR®: theoretical
background and capabilities", Journal of Structural Fire Engineering, https://doi.org/10.1108/
JSFE-07-2016-0010
Permanent link to this document:
https://doi.org/10.1108/JSFE-07-2016-0010
Downloaded by Cornell University Library At 00:41 05 July 2017 (PT)

Downloaded on: 05 July 2017, At: 00:41 (PT)


References: this document contains references to 25 other documents.
To copy this document: permissions@emeraldinsight.com
The fulltext of this document has been downloaded 6 times since 2017*
Access to this document was granted through an Emerald subscription provided by emerald-
srm:333301 []
For Authors
If you would like to write for this, or any other Emerald publication, then please use our Emerald
for Authors service information about how to choose which publication to write for and submission
guidelines are available for all. Please visit www.emeraldinsight.com/authors for more information.
About Emerald www.emeraldinsight.com
Emerald is a global publisher linking research and practice to the benefit of society. The company
manages a portfolio of more than 290 journals and over 2,350 books and book series volumes, as
well as providing an extensive range of online products and additional customer resources and
services.
Emerald is both COUNTER 4 and TRANSFER compliant. The organization is a partner of the
Committee on Publication Ethics (COPE) and also works with Portico and the LOCKSS initiative for
digital archive preservation.

*Related content and download information correct at time of download.


The current issue and full text archive of this journal is available on Emerald Insight at:
www.emeraldinsight.com/2040-2317.htm

Modeling structures in fire with Modeling


structures
SAFIR®: theoretical background
and capabilities
Jean-Marc Franssen
Structural Engineering Division, University of Liege, Liege, Belgium, and
Received 13 July 2016
Thomas Gernay Revised 18 August 2016
Accepted 21 October 2016
The National Fund for Scientific Research F.R.S.-FNRS,
Structural Engineering Division, University of Liege, Liege, Belgium
Downloaded by Cornell University Library At 00:41 05 July 2017 (PT)

Abstract
Purpose – This paper aims to describe the theoretical background and main hypotheses at the basis of
SAFIR®, a nonlinear finite element software for modeling structures in fire. The paper also explains how to
use the software at its full extent. The discussed numerical modeling principles can be applied with other
similar software.
Design/methodology/approach – Following a general overview of the organization of the software, the
thermal analysis part is explained, with the basic equations and the different possibilities to apply thermal
boundary conditions (compartment fire, localized fire, etc.). Next, the mechanical analysis part is detailed,
including the time integration procedures and the different types of finite elements: beam, truss, shell, spring
and solid. Finally, the material laws are described. The software capabilities and limitations are discussed
throughout the paper.
Findings – By accommodating multiple types of finite elements and materials, by allowing the user to
consider virtually any section type and to input the fire attack in multiple forms, the software SAFIR® is a
comprehensive tool for investigating the behavior of structures in the fire situation. Meanwhile, being
developed exclusively for its well-defined field of application, it remains relatively easy to use.
Originality value – The paper will improve the knowledge of readers (researchers, designers and
authorities) about numerical modeling used in structural fire engineering in general and the capabilities of a
particular software largely used in the fire engineering community.
Keywords Finite element analysis, Thermal analysis, Fire, Structures, Numerical modeling, SAFIR
Paper type Research paper

1. Introduction
SAFIR® is a software that is largely used in the fire engineering community for research
purposes, as well as for real applications. Based on the finite element method, the software
allows modelling the behavior of structures in fire, taking into account material and
geometrical non linearities, the thermal elongation, as well as the reduction of strength and
stiffness of the materials at elevated temperature. As of June 2016, 119 licenses have been
acquired by universities or research centers and 83 by design offices around the world. In
addition to that, an even higher number of demonstration versions with limited capabilities
have been distributed over the years. Google Scholar counted 317 scientific papers in which
SAFIR is quoted.
It is essential that reviewers and readers of scientific papers published on the basis of Journal of Structural Fire
Engineering
SAFIR simulations, as well as authorities who have to approve design based on this © Emerald Publishing Limited
2040-2317
software, are aware and understand all capabilities and limits of this software. The objective DOI 10.1108/JSFE-07-2016-0010
JSFE of this paper is thus to give information about the software, to detail its fundamental
assumptions and equations and to state its capabilities and limits. One previous paper by
Franssen (2005) gave some information about the basic theories and hypotheses, but it did
not include other important aspects such as information about the material laws, which are
given here. Furthermore, there have been a lot of developments within the past 10 years
leading to new functionalities that will be explained here, such as the possibility to consider
the effect of localized fires or to conduct 3D thermo-mechanical analyses with solid finite
elements.
Another objective of the paper is to highlight some capabilities of the software that are
not straightforward but may be very useful. It is shown how, with the tools, elements and
materials embedded in the code, the user can represent the behavior of a joint, model some
phenomena such as the shadow effect in concave sections or impose specific boundary
condition such as a limited thermal expansion at the extremity of a structural member
(buttress). This paper will thus provide a reference for users of SAFIR who want to use the
Downloaded by Cornell University Library At 00:41 05 July 2017 (PT)

software at its full extent. Meanwhile, as most of the discussed principles are valid
regardless of the software that is being used, it will also be a valuable source of information
for users of other software modelling structures in fire.

2. General organization of the software


From the beginning, the objective behind the development of SAFIR was to model building
structures subject to fire. To serve as a tool for the scientific community as a whole, it was
desired for SAFIR to cover a range of applications as wide as possible. In that sense, the
objective was different from that of specific software developed for particular member types.
To achieve this objective, the software was conceived to accommodate multiple types of
finite elements and materials. The modular character of SAFIR allows continuous
enrichment in the library of elements and materials. Nevertheless, the selection of
capabilities introduced in the code is dictated by the service they can provide in the fire
situation to have a software that is easy to use for applications in its well-defined field of
application. This distinguishes it from general purpose finite element codes that may treat a
wider range of applications at the cost of higher level of complexity.
The analysis of a structure in the fire situation requires determination of the
temperatures in the structure (thermal analysis), as well as evaluation of stresses, strains
and displacements (mechanical analysis). SAFIR allows performing these two types of
analyses with an easy transfer of information between the two. The strategy that has been
adopted is a weak coupling from the thermal analysis to the mechanical analysis. The
temperature field in the different parts of the structure is evaluated in a first step up to the
end of the fire, then the mechanical behavior is determined in a subsequent analysis.
The temperatures influence the mechanical behavior, but the results of the mechanical
analysis have no influence on the temperature distribution. This strategy is valid in most
cases because temperatures affect strongly the mechanical response of structures, whereas
the opposite influence is negligible except in very specific situations. Examples of effects
which cannot be accounted for with this strategy include plasticity generated heat;
anisotropy of conductivity in cracked materials; contact resistance between two adjacent
materials that have separated; stress-induced change of geometry (e.g. spalling, local
buckling in a concrete filled hollow steel section, etc.); large displacements of the structure
influencing its position relative to the fire source; effects of the loss of stability, insulation
performance or integrity of a separating member on the fire scenario. Regarding the latter, it
would be a real step forward if the discipline could provide a tool that would allow
performing a fully coupled analysis in which the fire would attack the partitions between
different fire compartments, and the failure of these partitions would influence the spread of Modeling
fire in the building. Yet this objective seems to be far from us if only because of the structures
stochastic nature of fire propagation through partitions.
Before conducting the thermal analysis, the fire attack must be defined by the user and
input in SAFIR as an entry. Several methods can be used to characterize the fire attack and
transfer the information to the program. It is essential to know the capabilities that can be
accepted by the code and the underlying hypotheses and equations used to calculate the heat
flux that will be applied at the boundary of the structure; these will be presented in Section 3.2.
For beam elements, each section is drawn by arranging a set of triangular and
quadrangular elements side by side, each one made of its own material, which allows the
user to draw virtually any section types (Figure 1). This offers more flexibility than relying
on a library of predefined section types. This discretization of the section is used for the
thermal analysis (calculation of the temperature at each node) as well as for the mechanical
analysis (determination of forces and stiffness in the section based on the temperatures in
Downloaded by Cornell University Library At 00:41 05 July 2017 (PT)

each element used in the thermal analysis which will form a fiber in the beam element).
For 3D solids, the same discretization is used for the thermal and mechanical analyses so
that the temperatures are directly mapped on the mechanical model (Figure 2). Six-noded
and eight-noded elements can be used. For shell elements, a uniaxial temperature
distribution is calculated across the thickness of the slab. The temperature at the through
thickness points of integration for the mechanical analysis is linearly interpolated between
the nodal temperatures.
In most simulations, the structure is mechanically loaded at room temperature to be
heated thereafter with the applied loads being kept constant. This type of analysis is
referred to as transient state and yields results in the time domain. During this process, large
displacements of the structure and temperature dependent nonlinear material behavior are
considered.
SAFIR does not have any failure criterion embedded as such. The simulation will run
until the time specified by the user is reached or until it cannot converge to a state of
equilibrium for the structure or it encounters numerical problems at the material level. It is
then the responsibility of the user to judge whether the last converged step corresponds to
the fire resistance time or to a premature numerical failure. Conversely, excessively ductile
behavior must also be detected in which cases user-defined deflection criteria have to be
applied. A typical case is a beam that exhibits such horizontal displacement at its extremity
that it would lose its support in a real application.
The problem to be solved is defined in input ASCII files. These can be constructed using
a text editor or using the general purpose graphic pre-processor GiD (Coll et al., 2016) for
which specific interface routines with SAFIR have been developed. SAFIR is written in
FORTRAN language, versions 77 and 90, and is provided as an executable file to run in a
Windows© environment. The results are provided in two ASCII files: one user-readable file
that can be directly examined in a text editor and one file organized in a XML format that
can be visualized by the accompanying graphic post-processor DIAMOND. This latter
format allows visualizing the results also by any other graphic post-processor.
To accommodate large models in reasonable computational times, the system of
equations inherent to the finite element method is solved using a sparse matrix solver. The
direct solver PARDISO (Luisier et al., 2013) is adopted here using a symmetrical matrix of
the system.
The next sections give more detailed information about the thermal analysis (Section 3),
the mechanical analysis (Section 4) and the constitutive material models available in SAFIR
(Section 5).
JSFE
Downloaded by Cornell University Library At 00:41 05 July 2017 (PT)

Figure 1.
Various sections of
beam elements, with
representation of the
mesh and materials
(a-b) and temperature
distribution (c)
3. Thermal analysis Modeling
3.1 Thermal equilibrium in the structure structures
Linear isoparametric finite elements with classical shape functions are used to represent the
geometry, as a function of the coordinates of the nodes, and the temperature field in the
element, based on the temperature at the nodes (Franssen, 2005). Heat transfer by
conduction according to the Fourier law is assumed in the material. C0 continuity is ensured
at the boundaries between adjacent elements, meaning that there is no discontinuity in the
temperature field between adjacent elements. The temperature varies linearly on all the
borders of the elements and the temperature at a common border is the same in both
elements. Yet the first derivative of the temperature perpendicularly to the border is not
continuous which means that the heat flux is not continuous from one element to the other.
The weighted residual method is used in the finite element formulation of the problem
and, summing the contributions from all the elements, equation (1) is obtained, which
describes in matrix form the equilibrium of heat fluxes in the structure at any given instant
Downloaded by Cornell University Library At 00:41 05 July 2017 (PT)

in time:

½ K fT g þ ½C  fT g ¼ f gg (1)

In equation (1), K and C are the matrices of conductivity and capacity, respectively, T is the
vector of the temperatures at the nodes, and g is the vector accounting for the heat
exchanges at the boundaries. Numerical integrations by the method of Gauss are performed
on the volume of the element to evaluate the conductivity and the capacity matrix. The
number of points of integration can be selected by the user (e.g. typically 2  2 points for a
four-noded 2D element). The fact that the thermal properties are temperature-dependent is
taken into account, including the fact that these thermal properties vary in the element, i.e.
from one point of integration to another. It must be stressed that the coefficient of heat
transfer by convection and the emissivity are considered as material properties in SAFIR.
Equation (1), expressing the thermal equilibrium at a given time, has to be integrated in
time to yield the evolution of the temperatures during the course of the fire. This is done by
the implicit single step scheme of the generalized central point (Franssen, 2005). The user
can select the temporal integration parameter u comprised between 0 and 1 to choose
between a completely explicit scheme to a completely implicit one, or any intermediate
solution. A value of 0.9 is typically used by the authors.
Convergence is checked considering the residue in equation (1), i.e. the out of balance
thermal flux. If the equilibrium is not satisfied, a correction of the temperatures has to be
applied. This correction is evaluated by setting the linearized form of the residue (using
Newton method) to zero. Iterations are then performed until the residue and/or the
corrections made on the temperature reach an acceptable low level.

Figure 2.
3D Thermal model of
a part of a composite
floor system
JSFE The size of the time steps is given by the user, with different sizes being possible during
different phases of the fire.
Heat exchanges in internal cavities can also be taken into account according to the same
principles as those described by Wickström (1979). The main hypotheses are:
H1. Heat transfer in the cavity by conduction in the gas is negligible.

H2. Specific heat of the gas in the cavity is negligible.

H3. The gas in the cavity is transparent (it does not absorb radiation energy).

H4. The boundaries of the cavity are gray surfaces.


The case of complex cavity shapes, for example, with surfaces of the cavity not seeing each
other due to concavity, or objects not touching the border of the cavity, can be considered in
Downloaded by Cornell University Library At 00:41 05 July 2017 (PT)

SAFIR (Figure 3). This allows analyzing, for example, a steel-rolled column encased in a
circular stainless-steel tube used for decorative purposes or a false ceiling under a concrete
slab.
The shadow effect can be taken into account for concave sections. This effect comes from
the fact that, for concave sections, the amount of radiative energy that crosses the so called
“box section”, i.e. the smallest convex section that can encompass the real section, cannot
have increased when it meets the surface of the steel element. At least, this is the case if the
section receives radiation energy, as in a standard test furnace, by the far field more than by
the gas surrounding the section. For convection, on the other hand, mass transfer provided
by air circulating in the concave re-entrant parts of the section can bring in energy (except
for very closed cavities) and the convective heat transfer in these cavities is of the same
order of magnitude as the one that occurs on the convex parts of the section. A modification
should thus happen essentially on the radiative part of the net het flux. Different techniques
exist to consider the concave nature of a section in a thermal analysis. The most
straightforward relies on a modification of the boundary properties of the material
(coefficient of heat transfer by convection and/or emissivity). The modification can be made

Figure 3.
Effect of radiation in
internal cavity on the
temperature
distribution for a
section heated on its
lower face (2D
thermal analysis)
with average values for the whole section, but it is also possible to differentiate between Modeling
different boundaries of the section, for example, convex part of the section on one hand (with structures
no modification on this part) and concave part of the section on the other hand. Also, the
concave part can be differentiated, for example, between the web and the interior sides of the
flanges for an H section. Another technique consists in modeling the “box section” using
fictitious radiative elements that close the cavity in the concave parts of the section but are
not in contact with the section (no heat transfer by conduction) (Figure 4). These radiative
elements have an emissivity equal to one. They absorb a certain amount of radiative energy
from the fire and then re-emit this same amount in the cavity toward the section. In that
case, mutual radiation between the web and the inner part of the flanges is considered,
whereas it is neglected with the other technique that relies on a modification of the
emissivity of the material.

3.2 Boundary conditions


Downloaded by Cornell University Library At 00:41 05 July 2017 (PT)

The thermal attack from the fire is given as an input data in SAFIR. To model the effect of
the fire on the structure, the user prescribes boundary conditions for the thermal analysis
that may take various forms. Note that if no boundary condition is given at a border, an
adiabatic boundary condition is assumed. This is used to model an axis of symmetry. In this
case, the isotherms are perpendicular to this boundary. The following boundary conditions
can be introduced:
 The user may directly prescribe the evolution of the temperature as a function
of time at different nodes. This capability is particularly useful for benchmark
purposes, or when the temperatures have been recorded during a furnace test
and the user wants to follow these temperature in the mechanical model as
closely as possible because the focus of the analysis is on the mechanical
response.
 The user may prescribe the evolution of the temperature of hot gases that
surround the section, in the form of a time–temperature relationship. Standard
fire curves are proposed in SAFIR, but the user also has the possibility to define
his own time–temperature relationship. The heat flux exchanged between a

Figure 4.
2 Fictitious radiating
surfaces for an
unprotected H-shape
steel profile
JSFE boundary and the hot gas is computed using the Eurocode formula, with a
linear convective term and a radiation term (equation 2). In this equation, h is
the coefficient of heat transfer by convection in W/m2K, Tg and Ts are the
temperature of the gas and of the boundary in K, s is the Stefan–Boltzman
coefficient and « * is the relative emissivity of the material:
  
qn ¼ h Tg  Ts þ s «   Tg 4  Ts 4 (2)

 The user may prescribe the heat flux at the boundary as a function of time. A
constant incoming flux must be combined with a gas temperature as
described under Point 2 to allow energy reemission to the far field. If not, a
continuous incoming flux will increase the temperature continuously to
infinite values.
Downloaded by Cornell University Library At 00:41 05 July 2017 (PT)

 The heat flux from one or several localized fires can be considered. Different
models are available to represent localized fires and directly transfer the
generated heat flux in SAFIR. One can obtain the thermal flux from a local fire
to a beam or a ceiling slab (Hasemi model). It is also possible to obtain the
thermal flux from a local fire to a column; this capability has been developed
recently in the framework of the LOCAFI project funded by the research
fund for coal and steel (Vassart et al., 2016). Finally, the thermal flux can be
imported from a computational fluid dynamics (CFD) calculation owing to a
dedicated interface (Tondini et al., 2012).
For the Hasemi and the LOCAFI models, each fire source is described by the 3D position of
the source in the structure (where the flame originates), the vertical elevation of the ceiling
(used to check whether the flame touches the ceiling or not) and, as a function of time, the
diameter of the circular fire source and the rate of heat release of the fire. In case of multiple
fires, the input fluxes from each fire are simply added.
The Hasemi model is in accordance with the model from Annex C of EN 1991-1-2 (2002),
assuming that the flame touches the ceiling. With this model, the heat flux received by the
section is isotropic, which means that the direction from the fire to the section is not
considered.
In the LOCAFI fire model, each fire source can be assumed as cylindrical or conical. At
each time step, the length of the flame is calculated according to equation C.1 of EN1991-1-2.
The trunk of the flame extends from the elevation of the fire source to the tip of the flame or
to the elevation of the ceiling, whichever is the shortest. This distance is divided into a
number of horizontal slices, and, for each slice, the temperature of the flame is calculated at
mid-level of the slice according to equation C.2 of EN1991-1-2. This temperature, calculated
along the centerline of the flame, is assumed to be uniform in the slice. If the length of the
flame exceeds the source to ceiling distance, a disk will form underneath the ceiling, and the
Hasemi model is used to calculate the radius of this disk and the heat flux generated by
the hot gases in this zone. Each slice and, if relevant, each concentric ring of the disk is
divided along the circumference into a large number (typically 180) of sectors. This double
division creates a series of rectangular facets that form the boundary of the fire source.
For each section, the heat flux at any time is calculated separately for each surface of the
boundary. This means that the thermal attack from the LOCAFI fire to the section is
anisotropic: the boundaries that are facing the fire receive the highest flux, whereas the
boundaries on the opposite side receive no flux at all. The radiative flux from the fire to the
section is calculated considering the radiative flux from each facet of the fire to each Modeling
boundary of the section with due consideration of the view factors between them. Heat structures
losses from the boundary to the far field are automatically considered. Consideration of a
LOCAFI fire will thus result in a transient and largely non-uniform temperature distribution
in the structure, which can then be used by SAFIR in a subsequent structural analysis, with
the elements having any orientation in space.
The CFD-SAFIR interface has been developed to allow easy transfer of the results from a
CFD simulation to a SAFIR thermal analysis. Results are transferred for every time step and
multiple positions in space in the form of the temperature of the gas (used for convective
exchanges), the coefficient of heat transfer by convection (not used in SAFIR for which a
constant material dependent coefficient is considered) and the radiant intensities for a given
number of directions. It was chosen to use the radiant intensities from different directions
(rather than impinging radiant fluxes on predefined surfaces or adiabatic surface
temperatures) because this allows computing the fluxes at each surface of the structural
Downloaded by Cornell University Library At 00:41 05 July 2017 (PT)

elements by integration of the radiant intensities taking into account possible shadow
effects in concave sections. A weak coupling strategy is used, i.e. the CFD calculation does
not depend on the mechanical response of the structure. Structural elements may be present
in the CFD model if they influence the flow of mass or of energy (typically, when the
structural elements also form the boundary of the compartment or for deep concrete beams)
or they can be omitted if their influence is negligible (typically for linear steel elements).
In practice, the CFD analysis and the SAFIR analyses can be performed by two different
persons. The CFD specialist makes his model in which the mesh can be refined where
required by the fire. The results are written in a transfer file according to a well-defined
format. Then, the SAFIR specialist makes his structural model without considering the fire
in which the mesh can be refined where required by the structure (it is independent from the
mesh used in the CFD analysis). SAFIR reads the results from the transfer file and uses them
to calculate the temperatures in the structural elements and subsequently the structural
response. As the results transferred from CFD are given at certain time steps, at certain
locations in the space and for certain directions, interpolations are performed by SAFIR, in
time and space (including in spherical coordinates for the appropriate directions of the
radiant intensities) (Tondini et al., 2012).
When a local fire model is being used, the thermal model of SAFIR calculates the
temperatures by a series of 2D thermal analyses performed at each longitudinal point of
integration of each beam finite element of the structure, which means that there is no
longitudinal heat flux in the elements.

3.3 Modification of the geometry


Most thermal calculations are performed on a fixed geometry. It is nevertheless possible to
modify the geometry by removing some finite elements after certain duration of fire.
Typically, elements in peripheral zones of the section are removed to represent falling off or
spalling of concrete. The thermal analysis resumes with this modified geometry including
the position of the thermal boundary conditions which have been moved inward. It is up to
the user to decide which element layers are removed and at what time it has to be done. This
capability thus allows representing the effect of spalling but is by no means a way to predict
the occurrence of the phenomenon.
If a separation occurs between two adjacent materials during the course of a fire and a
contact resistance to conduction develops, this can be modelled by a thin layer of a fictitious
material. The thickness of this layer will be constant, but it is possible to vary the contact
resistance by making its thermal conductivity temperature dependent.
JSFE 4. Mechanical analysis
4.1 Basic equations
4.1.1 Quasi-static solution. The mechanical problem can be solved in a quasi-static manner
using equation (3):

DF ¼ K Du (3)

with K the stiffness matrix of the structure, DF the vector of incremental applied nodal
forces and Du the vector of incremental nodal displacements.
The vector of incremental displacements Du is divided in two parts, one for which the
displacements have to be determined from equation (1) and one that contains the
displacements imposed by the user (possibly time-dependent). If the user imposes a constant
value equal to 0, this creates a usual fixed support, but the user can also impose any
variation as a function of time, which allows working in a displacement control mode.
Downloaded by Cornell University Library At 00:41 05 July 2017 (PT)

Different nodes can be forced to have the same value of the corresponding degree of freedom
(master slave relationships). This capability is used to introduce internal hinges in frame
type structures. Two nodes are located at the same position with a master–slave
relationship on the displacements but not on the rotations. This procedure to create internal
hinges is very straightforward, but, because the degrees of freedom are defined with respect
to the global axes of coordinates, in 3D models, strictly speaking, it is only valid in small
displacements; in fact, the hinge should follow the rotation of the extremities of the
members.
The capability to impose a certain time-dependent variation of the displacement at a
node can be used to model a contact between the structure and its environment, for example,
if a beam has a limited space for thermal elongation. A first simulation is performed when
the displacement under consideration is not restrained and the evolution of the displacement
is observed. A second simulation is performed in which the displacement of the node is
specified to follow the evolution computed in the first simulation but limited to the gap
between the structure and the environment; when this value has been reached, the
displacement is kept constant for the rest of the simulation.
The vector of incremental forces DF is also divided in two parts, one where the variation
as a function of time is imposed by the user and one where the reaction is calculated (at the
nodes where the displacement is imposed). A variety of possibilities is given to introduce the
loads as distributed loads on the shells or beam elements, such as uniformly distributed
loads, trapezoidal loads, in the local or global system of coordinates and pressure loads from
water table that follow the structure. Increasing the loads proportionally as a function of
time at constant temperature allows determining the load bearing capacity of the structure
at room temperature. This exercise is highly recommended before modelling the fire
behavior. First, this allows verifying the model (type of supports, of connections) in a
situation that is more familiar to many users than the fire situation, and, also, this gives an
idea of the load level during the fire situation and, hence, of the fire resistance time that can
be expected. During a fire situation, the applied loads are normally kept constant, but SAFIR
allows varying the forces as a function of time if desired. In fact, the applied forces are
separated into different groups, each group being multiplied by a function of time. It is thus
possible to have a group of loads which are permanently applied to the structure (design
values of the dead load and life load), another group of forces that disappear progressively
during the course of the fire (modelling, for example, the burning of some combustible mass
or melting and evacuation of snow), a group that appears at a certain time (a truck appears
on the deck of an underground car park), whereas another group is being pulsed by a
sinusoidal function of time (wind load). In the vast majority of situations, only permanent Modeling
loads are considered. structures
In practice, the procedure followed by the authors is:
 determining the load bearing capacity at room temperature; and
 performing another simulation in which the load is applied within 20 s while the
heating of the structure is marginal and then letting the heating being applied
under constant load. This leads to the fact that the time in the simulation
corresponds to the fire duration. Loading the structure before the fire starts
would either require starting the simulation at a negative value of time (which
has not been foreseen in the code) or delaying the beginning of the fire after a
certain period utilized for loading the structure.
The stiffness matrix is usually taken as the tangent matrix, recalculated at every iteration of
every time step (pure Newton–Raphson procedure). When shell finite elements are being
Downloaded by Cornell University Library At 00:41 05 July 2017 (PT)

used, experience has shown that convergence is facilitated if an approached Newton–


Raphson procedure is being used where the stiffness matrix of the first iteration of every
step is taken as the matrix of the last iteration of the previous step.
4.1.2 Time integration procedure. The iterative procedure to integrate in time from one
converged time step n (point A on Figure 5) to the next step n þ 1 is illustrated on Figure 5.
At converged time step n, the stress-related strain is noted e s (Tn) and the plastic strain e pl.
The procedure is as follows:
 The increment of thermal expansion at every point of integration is calculated
on the base of temperature increments from n to n þ 1.
 The properties of the materials (strength, stiffness, etc). are updated
corresponding to the new temperatures. As a result, the virgin stress-strain law
(s – e s ) is different at temperature Tnþ1 compared with Tn (Figure 5).
 Plastic strains e pl at the points of integration, as well as nodal displacements,
are kept constant at the beginning of the step. Because the displacements are
“frozen”, the total strains are also constant. However, the thermal strain
component in the total strain has changed by De th as a result of the temperature
variation. Therefore, a new value of the stress related strain e s (Tnþ1) is
calculated [equation (5)]. In case of temperature increase, the thermal strain
increases so the stress related strain decreases (considering elongation as
positive); compared to the converged step, the structure is in an artificial state of
compression.

A
σ
Tn+1
Tn

B
Figure 5.
–Δεth First step of the time
integration procedure
in SAFIR
εpl εσ(Tn+1) εσ(Tn) εσ
JSFE  A new stress and a new tangent modulus are calculated as a function of the new
stress related strain e s (Tnþ1), taking into account the new material properties
(stress-strain law at Tnþ1) and the fact that the plastic strain is constant (see
point B on Figure 5).
 These stresses are integrated on the volume of the elements to compute the
internal nodal forces which are not anymore in equilibrium with the applied
nodal forces. In case of temperature increase because the displacements are
constrained, the structure undergoes an internal state of out of balance
compression.
 The stiffness at the integration points is integrated to compute the stiffness
matrix of the structure.
 The out-of-balance forces are applied to the structure, leading to incremental
displacements, new strains (stress related component), new stresses and new
Downloaded by Cornell University Library At 00:41 05 July 2017 (PT)

nodal forces.
 The procedure described under point 5 to 7 is repeated several times at constant
temperature until the convergence criteria is satisfied.
 Plastic strains are updated after convergence.
The convergence criteria are based on the energy norm rate (ENR) calculated as the vector
product of the incremental forces and the incremental displacements at each iteration. The
ENR of the current iteration is compared to the cumulated ENR from the beginning of the
calculation and convergence is assumed when the ratio of both is lower than a limit defined
by the user. The fact that the displacements are fixed during the first iteration implies that
at least two iterations must be performed at each time step to reach equilibrium. On the
other hand, in case of a completely restrained structure, convergence will always be
assumed after two iterations.
It must be stressed that at every iteration to reach equilibrium in the structure, internal
iterations may be required in each point of integration to find the stress that corresponds to
the strain and temperature (Step 4 in the procedure above), in accordance with the material
constitutive law. In some cases, such as in concrete structures where some parts are
subjected to tension, the internal iterations at the material level may be significantly more
time consuming than the equilibrium iterations at the structural level.
4.1.3 Dynamic formulation. The quasi-static procedure that has been described here is
valuable as long as the structure is relatively stable. Yet this procedure fails when the
structure or a part of it is not statically stable. The displacements increase, whereas
the internal nodal forces decrease, and it is impossible to find a position for which the
equilibrium is satisfied. This may prevent the software to provide a good insight into the
failure mode because the computation stops at a time when the displacements are limited
and their evolution in the post-critical phase are not accessible. More than that it may occur
that simulation stops at an early stage because of the occurrence of a local instability
(buckling of an individual bar in a statically indeterminate structure, for example). To
circumvent this problem, a dynamic equation has been introduced, see equation (4):


F ¼ K u þ Mu (4)

with M the matrix of masses of the structure (rendered diagonal in SAFIR) and u€ the vector
of nodal accelerations.
The damping matrix has not been introduced in equation (4) because it is usually Modeling
computed as a linear combination of the mass matrix and the stiffness matrix, and this may structures
create problems when the stiffness matrix is not anymore positively defined. Instead,
numerical damping is being used (Newmark method with a = 0.45 and d = 0.80). More
information can be found in (Franssen and Gens, 2004).
If the user selects a dynamic analysis, equation (4) is being used consistently during the
simulation. Because the mass matrix is constant the procedure simply implies computation
of the nodal velocities and accelerations and utilization of a modified matrix for the system
of equations. When the evolution of displacements is slow, velocity and acceleration are
negligible, the time steps can be as long as in a quasi-static analysis and the additional
computational cost is negligible. Yet when velocity increases, much shorter time steps must
be used. An automatic procedure has been established in the software that continuously
monitors and controls size of the time steps based on the number of iterations required to
converge and on an automatic return to the previously converged step when convergence is
Downloaded by Cornell University Library At 00:41 05 July 2017 (PT)

not possible within a limited number of iterations. The time step is automatically reduced or
increased depending on the difficult or easy convergence. The user must introduce the value
of the initial time step, the maximum value of the time step to be accepted, the minimum
value at which the simulation stops if convergence cannot be achieved and the final time
where the simulation stops even if stability is still ensured.
4.1.4 Time step selection. Whereas time is fictitious in a quasi-static procedure (only the
temperature increments are relevant), this is not the case anymore in a dynamic calculation.
This is why a duration of 20 s is typically allowed for application of the loads. Applying the
loads within, say, 1 s would be equivalent to a shock being applied on the structure with
some undesirable oscillations that could lead to failure.
In usual structures, it is common to start the calculation with small values of the time
step to allow a progressive application of the load, for example, in four steps of 5 s each. This
is particularly advised if the applied load is close to the load bearing capacity of the
structure, and this induces a nonlinear response of the structure. The situation is different in
case of a prestressed structure. Prestressing is in fact introduced via initial strains, and these
are balanced during the first step, whatever the size of this step. It is thus required to apply
the full value of at least the dead load also at the first step because, in most prestressed
structures, the dead load counterbalances the effect of prestressing. A simply supported
prestressed hollow core slab, for example, is typically prestressed at the lower level in the
section and needs the sagging moment from the dead load to limit the tensile stresses at
the higher level. Yet if this first step that accommodates the effect of prestressing and of the
loads is short, it will also be seen as a shock by the structure. It is thus preferable to start
with a first step of sufficient duration, for example, 20 s.

4.2 Beam element


The beam finite element is a prismatic straight Bernoulli type element. To the traditional
rotations and translations at the two end nodes has been added a warping DoF to allow non-
uniform torsion to be considered. A third node that bears the non-linear part of the axial
displacement has also been added. The non-linear part of the strain component is averaged
on the length of the element to eliminate shear locking (de Ville de Goyet, 1986).
Longitudinal integration is performed numerically using two or three points of Gauss,
whereas integration in the section is based on a fiber model (Section 2).
The warping function and the torsion stiffness can be calculated by SAFIR on the base of
the discretization of the section used for the thermal analysis. It is essential to fix the value
of the warping function to 0 for at least one node (e.g. on the center of rotation or on an axis
JSFE of symmetry of the section) because although the torsion stiffness is computed from the first
derivative of the warping function and is thus not sensitive to a uniform offset of the
function, the warping function itself is also used in the treatment of the torsion DoF’s and
incorrect behavior develops if the value of the function is increased by a same value at all
nodes (which can occur if the value is not fixed in at least one node).
The torsion stiffness is pre-calculated on the base of the elastic properties of the materials
in the section at room temperature. It is then considered as constant during the whole
simulation. This beam element is thus not appropriate for members in which torsion is the
dominant load transfer mode, which is rarely the case in civil engineering structures.
Nevertheless, for steel members that fail by lateral torsional buckling, it is advised to reduce
the stiffness of the material during the calculation of the torsional stiffness depending on the
critical temperature that is expected. One iterative calculation may be required if a high level
of precision is requested.
This element is used to represent linear members such as beams and columns. It is also
Downloaded by Cornell University Library At 00:41 05 July 2017 (PT)

used for representing bars in truss girders, with a minimum of two finite elements per bar to
capture buckling of individual bars. In 3D truss girders, the three rotations at the two end
nodes cannot be all released because this would generate an energy-free longitudinal
rotation.
This element has also been used to model steel studs in composite steel-concrete
members with partial connection. Vertical beam elements are rigidly fixed to the upper
flange of the steel beam and hinged at mid-level of the concrete slab. They are given
appropriate section, material properties and temperature to represent the desired strength
and stiffness. Each stud can be modeled individually or one beam element is used to
represent a group of studs.
The modeling of the section into fibers can be used to represent semi-rigid connections.
Each fiber is given appropriate area, material type, mechanical properties and temperature
to represent the behavior of a particular component, for example, a bolt or bolt row. The
integration on the section of the behavior of all fibers will result in the global behavior of
the joint submitted to the combination of bending and axial forces. The effect of shear, on
the other hand, cannot be accounted for by this method.

4.3 Truss element


A truss finite element is also provided, made of one single point of integration, i.e. one
material, one temperature and one stress level. This element has been used essentially for
modelling external prestressing tendons (Franssen, 1993a). It can also be used in
conjunction with solid 3D elements to represent individual re-bars in a solid matrix. It has
to be understood yet that, in that case, the bars would have a uniaxial behavior. If a truss
element is used in isolation (e.g. for representing a bracing bar in a building or a
suspension bar in a bow string bridge), it has to be verified that the bar remains in tension
because this element cannot represent buckling. This is why this element is not used to
model internal bars of a truss girder; beam finite elements should be preferred in that case
as mentioned in Section 4.2.
This element can also be used to create a linear relationship between two nodes.

4.4 Shell element


The shell element is a quadrangle based on four nodes, each bearing three translations and
three rotations. Integration on the section is numerically done by four points of Gauss,
whereas the user can choose the number of Gauss points that are used for the numerical
integration on the thickness, from two if the membrane behavior is dominant to ten if
bending is dominant. A spurious mode with zero energy may develop if the model is totally Modeling
planar made of rectangular elements that have all the same dimensions and no other element structures
and no support provides stiffness or restrain with respect to the rotations in the plane of the
model (drilling rotations). More information, as well as validation examples, on the shell
element can be found in Talamona (2000).
This element can be used for representing plates of steel members, especially for
modelling local buckling in slender members. It can also be used to model reinforced
concrete walls or slabs by combining a concrete matrix with different layers of steel rebars.
In that case, the steel reinforcements are smeared laterally and their contribution is taken
into account by a uniaxial model; only an elongation in the direction of their axis will
produce a stress, whereas an elongation perpendicular to the axis or a shear strain in the
element do not produce any stress in the bars.
Two options are possible for definition of the steel rebars orientation. This orientation
can be defined with respect to the local system of coordinates of each element by specifying
Downloaded by Cornell University Library At 00:41 05 July 2017 (PT)

the angle between the local x axis and the layer of rebars (Figure 6). This method is
convenient, for instance, for meshing of a circular ring with a structured mesh based on
radial lines and concentric lines. In this case, local angles of 0 and 90 degrees are used for
specifying the orientation of the rebars.
Alternatively, the orientation of steel rebars can be defined with respect to the global
system of coordinates of the structure (Figure 7). The user specifies the orientation of a
vector in the global system of coordinates. The bars have, in each element, the orientation of
the perpendicular to this vector that is contained in the shell element. When two layers are
present, the norm of the vector can be set to 0 for the second layer; then, the orientation of
this second bar layer is perpendicular, in each element, to the previous one. This method is

Figure 6.
Bars orientation
defined in local axes

Figure 7.
Bars orientation
defined in global axes
JSFE convenient, for instance, for meshing particular shapes such as hyperbolic paraboloids,
domes or for slabs or walls in which the mesh is unstructured due to openings or
particularities in the geometric shape. For instance, Figure 8 shows an example of
calculation for which, given the irregular shape of the slab, the rebar orientation is more
conveniently defined in global axes.

4.5 Spring element


A spring element has been introduced to link the structure to the external world via a non-
linear relationship. The element has a single node, one that belongs to the structure, and it is
characterized by a direction. This is the direction of the translation u that influences the
element and the direction in which the force F of the element is applied (Figure 9).
On the contrary to other elements that can accommodate any material, this one does not
have a material defined as such; it has its own behavior directly described by a
displacement-force relationship. This relationship is characterized by a linear branch with
Downloaded by Cornell University Library At 00:41 05 July 2017 (PT)

stiffness E and two plastic plateaus at levels Fsup and Finf (Figure 10). These can be used, for
example, to represent the active pressure and the passive pressure that can develop in the
ground at the interface with a retaining wall as a function of the relative displacements
between the wall and the soil. The initial condition ui and Fi that prevail in the element at the
initial stage (when the displacement of the structure is 0) can also be given, allowing to
represent the erection and loading history of the wall before the fire. These properties are not
temperature-dependent. Behavior during displacement reversal is plastic, with unloading
parallel to the loading curve.
This element has been used to represented soil pressure on the walls and soil reaction
under the foundations of cut and cover concrete tunnels. For instance, Figure 11 shows
the deformed shape and soil pressures just before failure (plastic hinges have already

Figure 8.
A slab of japan
tobacco building
designed using
SAFIR (courtesy
INGENI)

Figure 9.
Node and direction
defining the spring
element
developed) and after failure for such a tunnel (note that the pressures under the Modeling
foundations are not uniform, but they have been truncated in the Figure). The element structures
can also be used for vertical walls in underground car parks. It offers a better
representation of the evolution of action from the soil than a predefined constant
trapezoidal distribution.

4.6 Solid element


The 3D solid element is based on six or eight nodes, with each node bearing three
translations. The user can select from one to three Gauss integration points in each
direction. This element allows analyzing structures under triaxial stress states and

F
Downloaded by Cornell University Library At 00:41 05 July 2017 (PT)

Fsup

Fi
Figure 10.
E Five parameters
displacement-force
Finf
relationship for the
ui u spring element

Figure 11.
2D Model of a tunnel
using beam elements
for the structure and
springs for the soil
JSFE elevated temperature. However, in the mechanical analysis, only the quasi-static
procedure is available and large displacements are not taken into account.
Applications with 3D solid elements in SAFIR are various and may include, for instance,
the analysis of prestressed hollow core slabs, concrete masses (e.g. in nuclear applications)
or the study of shear punching in concrete flat slabs. Figure 12, for example, shows the
response of a flat slab supported on a central column (1/4 of the slab modeled). The steel
rebars in the model have also been represented by lines of solid elements. Other examples
with steel structures include the analysis of the flame straightening process of heavy steel
sections or the detailed analysis of a joint.
It is intended by the authors to implement a four-noded (tetrahedron) solid element in
SAFIR to facilitate the meshing of unstructured volumes.

5. Materials
Table I shows an overview of the materials implemented in SAFIR. Most commonly used
Downloaded by Cornell University Library At 00:41 05 July 2017 (PT)

materials in structural engineering are available for thermal and mechanical analyses,
namely, concrete, steel and wood. Several types of stainless steel and aluminum can also be
used. In addition, thermal properties of gypsum plaster boards have been implemented, and
it is possible to introduce user defined materials for thermal analyses (either constant or
temperature-dependent properties), for example, for insulating products.

Figure 12.
3D Solid elements
used for studying
shear punching in a
flat slab

Thermal Structural analysis


Type of FE 2D-3D Solid Spring Beam Truss Shell 3D Solid

Type of law Uniaxial Uniaxial Plane stress Triaxial


Material
Steel X X X X
Concrete X X X X
Wood X X
HSC X X
Stainless steel X X
Aluminum X X
Elastic X
Table I. Bi-plastic X
Material models Gypsum X
implemented in Insulation X
SAFIR User X
Material models embedded in the code are based on the Eurocodes recommendations, at Modeling
least during heating. Some hypotheses had to be made for the behavior during cooling, structures
which is hardly mentioned in the Eurocodes (these hypotheses are mentioned in the
subsequent sections). This is necessary for investigating the behavior of structures under
real fire scenarios that include a cooling phase (Gernay and Dimia, 2013; Gernay and
Franssen, 2015a).

5.1 Thermal
The temperature-dependent thermal properties have been programmed in the code for
concrete, carbon steel and stainless steel, aluminum, wood and gypsum plaster boards. For
concrete, steels, aluminum and wood, the thermal models are based on the corresponding
Eurocodes.
The thermal algorithm is based on the computation of the enthalpy. Use of the
Downloaded by Cornell University Library At 00:41 05 July 2017 (PT)

enthalpy formulation, instead of the specific heat, makes the software much more
stable in cases where the specific heat curve shows sudden and severe variations as
is the case, for example, in gypsum, in carbon steel or with the evaporation of
moisture.
Parameters to be introduced for all materials are the coefficient of heat transfer by
convection on heated surface and the one on unheated surface (W/m2K) plus the emissivity (-).
Additional inputs may be required for some materials. More detailed information is given
hereafter.
For concrete materials, the additional parameters are the type of aggregate (siliceous
or calcareous), the specific mass of dry concrete (kg/m3), the free water content (kg/m3)
and a last parameter that allows tuning the thermal conductivity between the lower limit
and the upper limit (see clause 3.3.3 of EN 1992-1-2). The evaporation of moisture is
considered in the enthalpy formulation, but the subsequent movements and participation
in the heat balance of vapor are neglected. During cooling, there is no re-condensation of
the water, and the thermal conductivity is considered as not reversible and taken at the
value of the maximum reached temperature.
Carbon steel used for structural steel and for reinforcing bars have the same thermal
properties. The same holds for all types of stainless steel that have been embedded. For
aluminum, on the other hand, materials of series 5000 and 6000 that have been programmed
differ by their thermal conductivity. The thermal properties of all metallic materials are
considered as reversible during cooling in SAFIR.
The behavior of wood is considered as purely conductive using modified thermal
properties that represent complex phenomena, according to Annex B of EN 1995-1-2.
Parameters to be introduced are the specific mass including water content (kg/m3), the water
content in per cent of the dry mass (-), the ratio of conductivity along the grain to
conductivity perpendicular to the grain (-) and the vector giving the direction of the grain.
The direction of the grain is especially important in 3D analyses, as the conductivity is
usually larger along the grain. On the other hand, the material is isotropic in most 2D
analyses performed on the section of a beam or a column, where the grain is perpendicular to
the section. The thermal properties (conductivity, specific heat, density) are not reversible
during cooling; they keep the values corresponding to the maximum reached temperature.
For gypsum, the properties have been adopted from Cooper (1997). Two materials are
programmed in the code, which only differ by their density at 20°C (732 kg/m3 or 648 kg/m3).
The specific heat, equal to 1,500 J/kgK at 20°C, and the thermal conductivity, equal to 0.25 W/
mK at room temperature, vary dramatically with temperature. As a consequence, a slow
JSFE convergence is observed during the time integration process and this may lead to a time step as
small as 1 s being used.
One material with constant thermal properties, named “insulation”, can be used. Parameters
to be introduced are the thermal conductivity (W/mK), the specific heat (J/kgK), the specific
mass of the dry material (kg/m3) and the water content (kg/m3). In addition, up to five user-
defined materials with either constant or temperature-dependent thermal conductivity, specific
heat and specific mass can be used. The properties are specified at a certain number of
temperatures with linear interpolation in between. The properties can be defined by the user as
reversible, meaning that their values only depend on the current temperature, be it during
heating or cooling, or as irreversible, in which case during cooling from a maximum
temperature Tmax, the properties will keep the value that was valid for Tmax. The water content
has to be given at the initial temperature. The energy dissipated by evaporation of the water is
considered in the enthalpy formulation using the same method as for concrete materials.
Downloaded by Cornell University Library At 00:41 05 July 2017 (PT)

5.2 Structural
The constitutive relationships are based on the strain decomposition model of equation (5):

e tot ¼ e th þ e s þ e tr þ e i (5)

with « tot the total strain, obtained from spatial derivatives of the displacement field; « th the
thermal elongation, dependent only on the temperature; « s the stress related strain that
contains the elastic and plastic part of the strain (Figure 5); « tr the transient creep, a
particular term that appears in concrete during first heating under load; and « i an initial
strain that can be used either for initial prestressing or for the strain that exists in in situ
concrete when it hardens at a moment when loads already exist in the structure.
5.2.1 Uniaxial laws. Predefined uniaxial material models are embedded in the code for
the temperature-dependent nonlinear mechanical behavior of concrete, (carbon and
stainless) steel, wood and aluminum. The uniaxial models are to be used with truss and
beam finite elements, as well as for reinforcing bars in shell finite elements.
5.2.1.1 Concrete. The concrete models are based on the laws of Eurocode (EN 1992-1-2,
2005). Parameters to be introduced are the aggregate type (siliceous or calcareous), the
compressive strength and the tensile strength. In addition, the user can select if the transient
creep (equation 5) is treated implicitly or explicitly in the model. The implicit formulation
corresponds strictly to the current Eurocode model. The explicit formulation is a refinement
of the model which is calibrated to yield the same response as the current Eurocode model
during heating under constant stress (which is the situation considered in the Eurocode
model) but which, in addition, is able to take into account the non-reversibility of transient
creep strain when the stress and/or the temperature is decreasing (Gernay and Franssen,
2012) (Gernay, 2012). It must be highlighted that even with continuously increasing
temperatures and constant external applied loads, the stress will decrease in some parts of a
concrete section due to differential thermal expansion. Therefore, the explicit formulation is
recommended in any case.
The user can also select between normal strength concrete and high strength concrete
(HSC). The only difference lies in the factors used for reduction of compressive strength with
temperature. For HSC, these factors are as defined in Section 6 of EN 1992-1-2 for the three
strength classes.
During cooling, the mechanical properties of strength and strain at peak stress are not
reversible. An additional loss of 10 per cent of the concrete compressive strength with
respect to the value at maximum reached temperature is considered during cooling, as
prescribed in EN 1994-1-2 (2005). A residual thermal expansion or shrinkage is considered Modeling
when the concrete is back to ambient temperature, the value of which is taken as a function structures
of the maximum temperature according to experimental tests (Franssen, 1993b).
5.2.1.2 Steel. The steel models are based on the corresponding Eurocodes. Parameters to
be introduced are the Young modulus and the yield strength. Two additional parameters are
defined by the user to specify the behavior during cooling: the maximum temperature
beyond which the behavior is not reversible during cooling (threshold) and the rate of
decrease of the residual yield strength when the maximum temperature has exceeded the
threshold (in Pa/K).
For structural carbon steel, the model (based on EN 1993-1-2, 2005) is elastoplastic with a
limiting strain at the effective yield strength of 0.15 and an ultimate strain of 0.20 (linear
descending branch in between). For reinforcing steel, the models for ductility class A, B and
C (Figure 3.3 of EN 1992-1-2) with class N values for hot rolled and for cold worked steel
(Table 3.2a of EN 1992-1-2) are available. Prestressing steel of cold worked class B type
Downloaded by Cornell University Library At 00:41 05 July 2017 (PT)

(Table 3.3 of EN 1992-1-2) is also embedded in the code.


A specific material for structural steel has been introduced for modeling slender sections
with beam finite elements. This material has been developed to take into account local
buckling based on the concept of effective stress. The stress–strain relationship from
Eurocode 3 is adjusted in compression to take into account the effects of instabilities for each
combination of temperature-slenderness-support conditions (Franssen et al., 2014). Two
additional parameters have to be introduced: the slenderness and the number of supports of
the plate where the material is present (three supports for outstand plates such as flanges,
four supports for internal plates such as webs). This material is very useful when modeling
large structures made of slender steel elements for which a shell FE model would be too
computationally expensive.
A user-defined steel material can also be used. It has the same general equations of
stress–strain relationship as the structural steel from Eurocode, but the user can choose
the evolution of properties with temperature. Temperature dependent reduction factors
for the Young modulus, yield strength and proportional strength, together with the
thermal strain, are specified at specified temperatures with linear interpolation in
between.
For stainless steel, the mechanical properties have been programmed for the
following types according to Annex C of EN 1993-1-2: 1.4301, 1.4401, 1.4404, 1.4571,
1.4003, 1.4462 and 1.4311.
5.2.1.3 Aluminum. The mechanical properties of aluminum follow the equations of EN
1999-1-2 (2007). The different alloys and tempers that have been programmed are: 6061-T6,
6063-T6, 5083-H12 and 5083-O.
5.2.1.4 Wood. For wood, the model of Annex B of EN 1995-1-2 (2004) is adopted.
Parameters to be introduced are the modulus of elasticity, the compressive strength and the
tensile strength. The strength and stiffness start to decrease as soon as the temperature
exceeds 20°C and they reduce to zero at 300°C. In the range 20°C-300°C, different reduction
factors apply to tension and compression for the strength and modulus of elasticity. The
behavior is not reversible during cooling. The thermal strain is null.
5.2.2 Biaxial (plane stress) laws. Predefined plane stress material models are embedded
in the code for the temperature-dependent nonlinear mechanical behavior of concrete and
steel. These models are to be used with the shell finite element.
For concrete, the model is based on a plastic-damage formulation (Gernay et al., 2013).
Plasticity is based on a Drucker–Prager yield function in compression and a Rankine cut off
in tension. Damage is formulated using a fourth-order tensor to capture the different damage
JSFE processes in tension and compression including the effect of stress reversal on the concrete
stiffness (crack closure). Transient creep is computed explicitly and not recovered during
cooling or unloading. The variation of compressive and tensile strengths with temperature
is according to EN 1992-1-2. The user selects the type of aggregate (calcareous or siliceous)
and eight parameters: Poisson’s ratio, compressive strength, tensile strength, strain at peak
stress, dilatancy parameter, compressive ductility parameter, compressive damage at peak
stress and tensile ductility parameter. These parameters allow calibrating the model on a
specific type of concrete. For generic applications (when the concrete type is not known or
has not been tested), predefined values of the parameters are suggested (Gernay and
Franssen, 2015b).
For steel, the model is elastoplastic with a von Mises yield function and isotropic
nonlinear hardening. The variation of Young modulus, effective yield strength and
proportional limit with temperature follow the EN 1993-1-2 and the hardening function is
chosen to match as closely as possible the Eurocode uniaxial stress strain relationship.
Downloaded by Cornell University Library At 00:41 05 July 2017 (PT)

Poison’s ratio is assumed to remain constant.


5.2.3 Triaxial laws. Predefined triaxial stress material models are embedded in the code
for concrete and steel to be used with the solid finite element. These models are an extension
of the plane stress models described in the previous section, with the only difference lying in
the number of stress and strain components.

6. Conclusion
SAFIR is a computer software for modeling the response of structures subjected to fire.
It incorporates the results of several decades of research works into a tool that offers the
latest scientific developments in the field while, at the same time, because it has been
tailored for a specific field of application, it does not require long and extensive learning
process to be mastered. Its structure is sufficiently flexible to incorporate further
developments, either in terms of finite element types, of constitutive material models or
of solution algorithms.
This duality of, on one hand, offering a good level of technical capabilities and scientific
content and, on the other hand, being affordable and accessible to the vast majority of actors
in the field is the key feature of this program. It helped structural fire engineering to find its
place in fire safety engineering for researchers in many universities and research centers, for
authorities in many countries and for small- and mid-size design offices throughout the
world (although big companies with a branch that specialized in fire safety engineering are
also using it more and more).
This code is not distributed as open source although some material models specifically
developed at the University of Liege may be offered in the future. It is nevertheless the desire of
the authors to give the opportunity to develop the code to well-identified and reliable partners
for the benefit of the community of all users.
The main challenge that the authors are facing lies in continuing the development of new
theories and concepts and continuing the analysis of new structural systems (which are the
prime tasks expected from researchers, those that can lead to the production of scientific
publications), while at the same time developing and maintaining the software, i.e.
implementing new theories and models that continuously appear and offering as much help as
possible to the users. Nowadays, developing and maintaining a software is not recognized as a
scientific production in the academic world but the authors committed themselves to this task
because they believe that the outcome favors the development of structural fire engineering for
the benefit of the community.
References Modeling
Coll, A., Ribo, R., Pasenau, M., Escolano, E., Perez, J., Melendo, A., Monros, A. and Garate, J. (2016), “GiD structures
v.13 User Manual”, CIMNE (Eds).
Cooper, L.Y. (1997), “The Thermal response of Gypsum-Panel/Steel-Stud Wall Systems exposed to Fire
Environments – a simulation for use in Zone-Type Fire Models, NISTIR 6027”, National
Institute of Standards and Technology, Gaithersburg, MD.
De Ville de Goyet, V. (1986), “Nonlinear elastic frame analysis by finite element”, Journal of Structural
Engineering (ASCE), Vol. 112 No. 10, pp. 2362-2365.
EN 1991-1-2 (2002), Eurocode 1: Actions on Structures – Part 1–2: General Actions – Actions on
Structures Exposed to Fire, CEN, Brussels.
EN 1992-1-2 (2005), Eurocode 2: Design of Concrete Structures – Part 1–2: General Rules – Structural
Fire Design, CEN, Brussels.
EN 1993-1-2 (2005), Eurocode 3: Design of Steel Structures – Part 1–2: General Rules – Structural Fire
Downloaded by Cornell University Library At 00:41 05 July 2017 (PT)

Design, CEN, Brussels.


EN 1994-1-2 (2005), Eurocode 4: Design of Composite Steel and Concrete Structures. Part 1–2: General
Rules – Structural Fire Design, CEN, Brussels.
EN 1995-1-2 (2004), Eurocode 5: Design of Timber Structures – Part 1–2: General – Structural Fire
Design, CEN, Brussels.
EN 1999-1-2 (2007), Eurocode 9: Design of Aluminum Structures – Part 1–2: Structural Fire Design,
CEN, Brussels.
Franssen, J.M. (1993a), “SAFIR applied to prestressed beams at ambient temperature”, External
Prestressing in Structures. Non-Linear Calculation Tests of Prestressed Beams, Association
Française pour la Construction, pp. 67-73.
Franssen, J.M. (1993b), “Thermal elongation of concrete during heating up to 700°C and cooling”,
University of Liege, available at: http://hdl.handle.net/2268/531
Franssen, J.M. (2005), “SAFIR: a thermal-structural program for modelling structures under fire”,
Engineering Journal, Vol. 42 No. 3, pp. 143-158.
Franssen, J.M. and Gens, F. (2004), “Dynamic analysis used to cope with partial and temporary
failures”, in Franssen, J.M., Benichou, N., Kodur, V. and Sultan, M. (Eds), Proceedings of SiF'04:
Third International Structures in Fire Workshop, IRC, Ottawa, pp. 297-310.
Franssen, J.M., Cowez, B. and Gernay, T. (2014), “Effective stress method to be used in beam
finite elements to take local instabilities into account”, Fire Safety Science, Vol. 11,
pp. 544-557.
Gernay, T. (2012), “Effect of transient creep strain model on the behavior of concrete columns subjected
to heating and cooling”, Fire Technology, Vol. 48 No. 2, pp. 313-329.
Gernay, T. and Dimia, M.S. (2013), “Structural behaviour of concrete columns under natural fires”,
Engineering Computations, Vol. 30 No. 6, pp. 854-872.
Gernay, T. and Franssen, J.M. (2012), “A formulation of the eurocode 2 concrete model at elevated
temperature that includes an explicit term for transient creep”, Fire Safety Journal, Vol. 51,
pp. 1-9.
Gernay, T. and Franssen, J.M. (2015a), “A performance indicator for structures under natural fire”,
Engineering Structures, Vol. 100, pp. 94-103.
Gernay, T. and Franssen, J.M. (2015b), “A plastic-damage model for concrete in fire: applications in
structural fire engineering”, Fire Safety Journal, Vol. 71, pp. 268-278.
Gernay, T., Millard, A. and Franssen, J.M. (2013), “A multiaxial constitutive model for concrete in
the fire situation: theoretical formulation”, International Journal of Solids and Structures,
Vol. 50 Nos 22/23, pp. 3659-3673.
JSFE Luisier, M., Schenk, O., et al. (2013), “Fast methods for computing selected elements of the green’s function
in massively parallel nanoelectronic device simulations”, in Wolf, F., Mohr, B. and an Ney, D. (Eds),
Euro-Par 2013, LNCS 8097, Springer-Verlag Berlin, Heidelberg, pp. 533-544.
Talamona, D. and Franssen, J.M. (2000), “New quadrangular shell element in SAFIR”, Proceedings of
the 1st International Structures in Fire workshop, in Franssen, J.M. (Ed.), Copenhagen, Liege,
pp. 195-210.
Tondini, N., Vassart, O. and Franssen, J.M. (2012), “Development of an interface between CFD and FE
software”, Proceedings of the 7th International Conference on Structures in Fire, ETH Zurich,
Switzerland, pp. 459-468.
Vassart, O., Hanus, F., Brasseur, M., Obiala, R., Franssen, J.-M., Scifo, A., Zhao, B., Thauvoye, C., Nadjai,
A. and Sanghoon, H. (2016), “Temperature assessment of a vertical steel member subjected to
localised fire, acronym LOCAFI”, Report RFSR-CT-2012-00023, European Commission
Research Programme of the Research Fund for Coal and Steel, Esch/Alzette.
Wickström, U. (1979), “A numerical procedure for calculating temperature in hollow structures exposed
Downloaded by Cornell University Library At 00:41 05 July 2017 (PT)

to fire”, Report N°79-2, Dpt of Struct. Mech., Lund.

Corresponding author
Thomas Gernay can be contacted at: thomas.gernay@ulg.ac.be

For instructions on how to order reprints of this article, please visit our website:
www.emeraldgrouppublishing.com/licensing/reprints.htm
Or contact us for further details: permissions@emeraldinsight.com

You might also like