Sunteți pe pagina 1din 17

NUMERICAL MODELLING OF THE BEHAVIOUR OF CONCRETE

STRUCTURES IN FIRE
Terro, Mohamad J.

Finite element thermal and structural computer programs were developed to predict the behavior of
three dimensional reinforced concrete structures in fire. The thermal program uses an implicit
recurrence scheme to improve convergence. The structural program was based on recent
experimental data obtained at Imperial College on the transient thermal strain behavior of concrete
heated for the first time under load. The thermal program was validated against data from an
experimental test on a steel ceramic element. The structural model was also validated against data
from experimental tests on concrete slabs and a concrete column. Good correlation between
predicted and experimental results were obtained for both temperature distribution in the thermal
module and deformations in the structural model. The results highlight the important role that
transient thermal creep plays, particularly in the column example.

Keywords: Finite element analysis, reinforced concrete, thermal analysis, structural analysis, fire,
temperature, transient thermal creep, free thermal strain, load induced thermal strain, stress-strain
relation, bi-axial loading.

1. INTRODUCTION

The analysis of structures in fire consists of two calculation stages. First, a nonlinear thermal
analysis is carried out to determine the temperature distribution history within the elements
forming the structure under study. Second, the variations in the stiffness matrix due to
changes in the material properties of the structure are calculated accordingly, and a static
analysis is performed in a number of time-steps until failure.

Both the thermal and structural analyses can be achieved using either the finite element or
finite difference methods. However, most available computer programs in this field use the
finite element method, although a few of them use the finite difference method for the
thermal analysis.

The programs described in this paper comprise two modules: TEMP which determines
incremental temperature gradients and STRUCT for structural analysis calculations [1].

2. SIGNIFICANCE OF RESEARCH

Numerical methods have been applied to structural analysis since the 1960s, but computer programs
dedicated to the analysis of buildings exposed to fire did not appear until a decade later. Most of the
currently available programs have grown out of a number of research projects within universities or
research establishments. Consequently, the programs tend to be inadequately structured, because the
aim of the researcher is to further the science rather than to produce commercial user-friendly
software. This problem has been avoided in STRUCT being developed within the shell of a well-
established commercial finite element program called LUSAS (London University Stress Analysis
System) [3]. To the knowledge of the author, STRUCT is the only available fire-dedicated program
capable of simulating the structural response of fully three-dimensional building structures under fire
attack. Also, the constitutive material model employed in STRUCT is based on up-to-date
experimental data on the behaviour of concrete at transient high temperatures.
3. TEMPERATURE ANALYSIS

An accurate prediction of the temperature variation in the structural element with time is essential for
determining the corresponding material properties (thermal and mechanical) at the relevant
temperatures and for carrying out the structural analysis.

3.1 Description of Computer Program TEMP

TEMP is a non-linear finite element heat transfer analysis computer program designed for the
prediction of the temperature distribution history in two-dimensional continua and based on the
Fourier transient heat transfer equation. The finite element library consists of four-noded general
quadrilateral isoparametric elements and three-noded triangular elements.

The temperature dependent thermal properties of a number of different building materials are used as
input data into program. The following thermal properties are allowed to vary as functions of the
average temperature of a finite element:

- Thermal Conductivity k
- Volumetric Heat Capacity .c
- Convection Factor 
- Convection Power 
- Resultant Emissivity 

Within each time step, the residual of the heat balance is checked for convergence against a user-
defined value. The limiting number of iterations that can be carried out until convergence is achieved
is also defined by the user in order to optimize computer time.

TEMP employs a backward difference time integration algorithm (implicit). This integration scheme
gives a stable solution and therefore has no limitation on the size of the time step for stability and
convergence unlike the more commonly used explicit solutions.

The thermal boundary conditions, which are the temperature heating or cooling inputs at the surfaces
of the element, can be set from either specified temperatures at the nodes at the surfaces exposed to
the environment or from a heat flow defined as a temperature-time relation (in tabular form, or in
accordance with an equation relating temperature to time as defined by the fire test standards in BS
476) using convection and radiation properties of the surface/gas interface derived from experimental
data.

3.2 Mathematical Formulation

In a real fire, or a fire test, heat will flow to the surface of the structure exposed to fire by means of
radiation and convection. Then heat will be transferred internally away from the surface by means of
conduction. Because of the time dependency of the gas temperature, this heat transfer is classified as a
“transient” temperature analysis problem.

The Fourier transient heat conduction second order partial differential equation for flow in a
homogenous media is represented for the two-dimensional situation as follows:

  2T  2T  T
K 2    Q  C (1)
 x y 
2
t

2
Where:

K = Thermal conductivity of the material (in W/m oK).


T = Field variable, or temperature (in oK).
Q = The internally generated heat (in W/m3 ).
C = The specific heat of the material (in J/kg oK).
 = The density of the material (in kg/m3 ).
t = The time variable (in seconds).

In problems of structures exposed to fire, the internally generated heat component Q is not active. The
transient heat flux at the boundary conditions can be a constant heat flux, or variable convection and
radiation heat fluxes. This is represented by the following equation:

 T T 
q  qc  qr  K  lx  l (2)
 x y y
Where:

lx , ly = The direction cosines of the normal to the boundary surface.


q = Imposed heat flux other than fire.
qc = The convective heat flux from the fire.
qr = The radiative heat flux from the fire.

The convection heat fluxes can be expressed as follows:

qc = ( Te - Ts ) (3)

Where:

 = Convection factor, (in W/m C).


 = Convection power.
Te = Environmental or gas temperature, C.
Ts = Average surface temperature, C.

The radiation heat flux can be expressed by Stefan's equation:

qr = ( as.f.Te4 - s.Ts4 ) (4)

Where:

as = Absorptivity of the exposed surface.


 = Stefan-Boltzman constant, (5.67x10-8 W/m .oK).
f = Emissivity of the flame.
s = Emissivity of the surface (= as).

3
In most heat transfer calculation methods, equation (4) is simplified by the use of a resultant
emissivity as follows:

qr =  (Te4 - Ts4 ) (5)

Where  is the resultant emissivity of the surface and fire which ideally should be temperature
dependent.

In the case of prescribed temperatures, a fixed value of temperature is given at points in the material
(analogous to prescribed displacements in structural analysis).

Using a finite element formulation, the temperature matrix T, which represents the temperature values
at nodes, is expressed by:

T = N T (6)

Where:

Ni = The shape or nodal interpolating function at node i.


Ti = Value of temperature at node i.
n = Total number of nodes.

Using the Galerkin weighted residual or the variational method, equation (1) can be rewritten as
follows in matrix notation:


H.T + P. T + F = 0 (7)

Where:

H = Conductivity matrix.
P = Capacitance matrix.
F = Heat load vector.
T = Temperature vector.

T = Temperature rate vector.

To evaluate the temperature rate term, a recurrence or time-stepping scheme has to be adopted. This
can be achieved using a finite element or a finite difference method where the time domain is
discretised into time elements (time-steps), and an interpolating scheme is employed at nodes or
stationary times. Therefore, functions H and F will be expressed by:

H =  . Ht+t + (1-) . Ht and F =  . Ft+t + (1-) . Ft (8)

Where  [0,1] is the characteristic value of the time discretisation scheme.

And the temperature rate term is expressed by:

4
 T  -T
T = (9)
t

Therefore, equation (7) is rewritten as follows in matrix notations:

 P   P 
 T   H  
 t   t  T
T  (1   ) H Tt   Ft   t  (1   ) Ft  0
   
(10)

An iteration procedure within each time-step t is performed and the temperature dependent material
properties are continuously updated until convergence is achieved within a given tolerance.

It can be noted that the choice of the parameter  affects equation (10) and thus it also affects the
amount of computation involved within a time-step and the stability and convergence requirements of
the numerical model.

The stability of the numerical scheme is achieved if the following condition is satisfied:

wi . t(2-1) > -2 (11)

Where:
 ki 
wi = Eigenvalue of the system corresponding to mode i.  
 c i .i .v i 

Being inversely proportional to the volume (vi), the largest eigenvalue of the problem corresponds to
the smallest element, which is normally at the fire exposed boundary.

If  = 0, the stiffness matrix used to calculate T in equation (10) is simplified to a lumped diagonal
matrix and the stability requirements would dictate the use of a time-step value smaller than 2/wi .
This scheme is called an explicit scheme or the `Forward Difference (Euler)' scheme. The use of such
schemes would require experienced users since convergence is hard to achieve if an inadequate time
stepping routine or finite element mesh are adopted.

TEMP employs the `Backward Difference (Euler)' method, which is an implicit scheme, where =1.
Since a unit value for  is employed, equation (9) becomes unconditionally satisfied and therefore
would give stable results without restrictions on the time-step. Most of the available thermal analysis
programs, in the field of fire research, use the forward difference (Euler), or explicit, recurrence
scheme, i.e. with a value of  = 0.

3.3 Validation of Temp

TEMP was used to predict the temperature history at the exposed and unexposed steel surfaces of a
steel/ceramic measuring element exposed to a standard furnace test [2] as shown in figure 1. This
element was originally used in the validation of TASEF and TEMPCALC at the Fire Research Station

5
due to its high sensitivity to temperature gradients. A parametric study was conducted with resultant
emissivity values of 0.55, 0.6, 0.65, 0.7 and 0.75 at the exposed surface. Results from TEMP were
compared with those of other established programs and experimental data (see table 1). It can be
observed from table 1 that the experimental and predicted values are in close agreement at the
exposed surface. In the case of TEMP it was not necessary to use such a fine mesh and small time step
as was required in the other programs. Figure 2 shows a comparison between experimental and
predicted temperature results using TEMP with a resultant emissivity value of 0.55 at the exposed
surface. The value of the resultant emissivity is a function of the properties of the material exposed to
fire, the geometry of its surface, and the media (air) through which temperature, or fire, is being
transferred. Therefore, such a parametric study is always necessary to establish the closest value of
the resultant emissivity to the real case under consideration.

4. STRUCTURAL ANALYSIS.

The problem of structures exposed to fire is both geometrically and materially non-linear. The
geometrical non-linearity is mostly caused by the large displacements of the elements of the structure
near failure. The temperature dependency of the material properties adds another dimension to the
non-linear nature of the problem.

4.1. Description of Computer Program STRUCT

The non-linear finite element program STRUCT is devoted to the analysis of general three
dimensional building structures under fire attack. STRUCT was developed within the shell of a
general purpose finite element program called LUSAS [3].

STRUCT can model reinforced concrete, steel, and composite concrete-steel structures exposed to
high thermal loading. This includes columns, beams, one-way and two-way spanning slabs, slabs
connected with beams and columns etc..

The thermal analysis is performed separately using TEMP. The integration points for the structural
analysis are pre-defined by the user and input in the thermal analysis program. An automatic search
and two-dimensional interpolation is then carried out in the thermal analysis program and a
corresponding temperature file is created and interfaced automatically with STRUCT.

The iterative procedures employed in LUSAS include the Standard and Modified Newton-Raphson or
any combination of the two methods, in addition to a Line Searches technique which is normally
coupled with the Newton-Raphson based iterations.

The finite element library includes three-dimensional three-noded beam elements and eight-noded
shell elements.

Each element is associated with a temperature file depending on the geometry of the cross-section and
the heat exposure type at the boundaries. The temperature is assumed to be uniform along the length
of the beam element and across the horizontal plane of the shell element. Based on these assumptions,
the required thermal analysis is two-dimensional for the beam element and one-dimensional for the
shell element.

The through-thickness integration for the beam element is performed by assigning characteristic areas
to Gauss points across the section. The shell element is subdivided into a number of layers having
different material properties. The layers corresponding to steel bars in the shell element are assigned
an angle in the global reference system to take into account the two-way spanning effect.

6
The support conditions can be specified by the user as free, fully restrained, or assigned any value of
restraint or prescribed displacement at the degree of freedom of individual nodes. The loading
conditions can be uniformly distributed, concentrated, trapezoidal, or any given function of the span
applied in the three-dimensional global reference system.

4.2. Constitutive Model of Concrete.

The constitutive model for concrete is described in the following relations:

Based on measurements on a phenomenological level, the strain of a concrete specimen heated for the
first time under load can be described by the following strain components:

tot = th + tot.cr +  (12)

Where:

tot = Total strain.


th = Free thermal strain, including shrinkage.
tot.cr = Total creep strain, including basic creep, transitional thermal creep (ttc),
and drying creep.
 = Instantaneous stress-related strain.

Equation (12) can be simplified to:

tot = FTS + LITS (13)

Where:

FTS = Free Thermal Strain = th


LITS = Load Induced Thermal Strain = tot.cr + 

Both FTS and LITS can be determined experimentally. However, for numerical calculation we require
the stress-related strain (). This is derived numerically as follows:

 (i)
  LITS(i) -  (i-1)
tot.cr
(14)

Where the superscript (i-1) refers to the previous iteration step. To start this algorithm, the first value
of the stress-related strain is estimated using the following linear relation:

  /E
(i) (i) (15)

Where E is the modulus of elasticity.


The stress value in equation (15) is estimated at the first iteration in the first step from the value of
LITS using the relations developed between LITS and stress.

In the following sections, models for each component are developed as functions of temperature (T),
percent aggregate content by volume (Va) and stress level during heating (). Therefore:

FTS = f(T, Va)


LITS = f(T, , Va)
 = f(T,, Va)

7
Free Thermal Strain

The Free Thermal Strain (FTS), is the strain occurring during heating an unstressed specimen. It
includes the thermal 'expansion' and drying shrinkage components. FTS of concrete is dominated by
that of the aggregate [4] and, therefore, both aggregate type and content are important in this respect.
In this model, FTS is considered to be a function of temperature, aggregate type, and aggregate
content. Strictly, however, the FTS model should also be a function of moisture content and rate of
heating, which at relatively low rates of heatings are second order effects. The inclusion of moisture
content and pore pressures will be the subject of future development at Imperial College.

With regard to modelling aggregate content, a tentative approach is to assume for practical purposes a
linear FTS-Aggregate content relation in the range 60-75% aggregate content. The relation, based on
test results from Schneider (70% & 73%), Anderberg (60%), and Khoury (65-65%) is assumed to be
in the general form [1]:

FTS(T, Va ) = FTS(T,65%) x (A Va + B) (16)

where A = 4, B = -1.6, and Va is the fraction volume of aggregate (coarse and fine). In this equation,
the effect of aggregate content on FTS is assumed to be independent of temperature.

Load Induced Thermal Strain

Concrete heated under load experiences considerable contraction compared with the free (unloaded)
thermal strains (FTS). The Load Induced Thermal Strain (LITS) is determined by subtracting the two
strain values [4]. LITS comprises Transitional Thermal Creep (TTC), Drying Creep, Basic Creep,
and changes in elastic strain with temperature contribute to LITS. However, by far the largest
component of LITS for mature concrete is TTC which is irrecoverable and occurs only on first
heating. Research at Imperial College [4] has indicated that the LITS of concrete in general can be
represented by a "Master" LITS curve independent of the type of aggregate (for a given aggregate
content by volume) used for temperature up to about 450C. LITS can also be considered as a linear
function of stress, and is also dependent on the heating rate and age at heating for ages less than about
3 months [5].

Based on experimental observations, the components of LITS were assumed to be linearly related to
stress level up to nearly 0.6-0.7 stress/cold strength ratio (/uo). This was found for transient creep
[4,6] and basic creep of concrete at high temperatures [7,8]. Therefore, an idealized linear relation
between LITS and stress/cold strength ratio, was assumed. In the following equation, LITS is
normalised for values obtained at (/uo) = 0.3, and is based on the master LITS curve for
temperatures up to 450C [8]:

LITS (T,  / uo ) = LITS (T,0.3).(0.032+ 3.226.  / uo ) (17)

The LITS(T,0.3) can be expressed as a fourth order polynomial function of temperature (in C) based
on master LITS data for temperatures up to 590C [1]:

LITS (T,0.3) = A0 + A .T+A1 .T + A2 .T3 + A4 .T4 (18)

where:

A0 = -43.87
A1 = 2.73
A2 = 6.3510-2

8
A3 = -2.1910-4
A4 = 2.7710-7

However, the LITS of Thames gravel concrete departs from the master curve at temperatures above
400C. A different relation for this concrete was, therefore, developed:

LITS (T,0.3) = 1.48(B0 + B1 .T + B2 .T2 + B3 .T3 + B4 .T4 + B5 .T5 ) (19)

where:

B0 = -1098.50
B1 = 39.21
B2 = -0.43
B3 = 2.4410-3
B4 = -6.27 10-6
B5 = 5.95 10-9

The data given above applies for concrete containing 65% aggregate by volume. The effect of
aggregate content on LITS is assumed to be linear and similar in magnitude to the effect on basic
creep [10]. The relation is therefore [1]:

LITS = LITS(T,65%)  (3.05 . Va / 100) (20)

It should be noted that all temperatures in equations of LITS are in C.

Creep In Tension

There do not exist any data on LITS under tensile state of stress. This is probably due to the fact that,
firstly concrete is usually designed to be used in compression, and secondly the tensile properties of
concrete are difficult to measure. In fact, creep in tension has never been considered in previous
concrete models for fire-dedicated computer programs.

At ambient temperatures, the creep behavior of concrete in tension has been studied by few research
workers but with conflicting results. This is due to the difficulty in maintaining long-term tensile load
on concrete, while simultaneously measuring very small strains developed due to the low values of
stress [11]. The majority of the experimental results suggest that the ratio (creep in tension)/(creep in
compression) is 1.2 to 1.3 for the same stress level [11,12]. Assuming that tot.cr in tension will
also follow the same trend, then it should not be ignored in the strain model. However, it should be
emphasized that this is only an assumption taken due to the absence of experimental results in tension
at high temperatures.

Stress-Strain Relation

The general form of the stress-strain curve of concrete in compression at ambient temperatures, as
described by part 2 of the British standard for reinforced concrete design BS8110, is maintained at
elevated temperatures.

For concrete heated without load, the stress-strain curve experiences a reduction in the ultimate stress
(u), modulus of elasticity and slope of the descending portion, but an increase in the strain at ultimate
stress (u) and the crushing strain. However, these effects are reduced for concrete heated under stress
such that at 30% stress level the stress-strain curve at elevated temperatures may be little different
from that at ambient temperatures. This is confirmed by results from several researchers that show
after heating under load, the strength [12], elastic strains [4], and strain at maximum stress [14,15]

9
experience relatively small changes with temperature. These are important results and are quite
frequently ignored in previous models that take the stress-strain relation of concrete at high
temperatures from data obtained on specimens heated without load.

The compressive stress-strain relation, in its ascending and descending branches, is assumed in this
model to be in the general form:

E. 
 = (21)
1 +  /  u 
2

where E = 2u/u (= initial tangent E assumed to be = 2 x secant E).

The stress-strain relation in tension is on the other hand represented by two linear branches and can be
expressed by:

 = E for 0 <  = <  ut (21a)

 =  ut (crack - )/(crack -  ut ) for  ut <  < crack (21b)

=0 for  > = crack (21c)

Figure 3 shows the uniaxial stress-strain model for concrete in compression and tension. It also
depicts the loading and unloading paths. This model is defined by the following variables:

In the compression zone:

u = Ultimate compressive stress


u = Strain at ultimate compressive stress
crush = Crushing strain.

In the tension zone:

 ut = Ultimate tensile strain


 ut = Strain at ultimate tensile stress
crack = Cracking strain.

u is taken as a function of temperature and stress level during heating. However, above a certain
load level during heating the compressive strength of concrete does not decline with increase in
temperature to the same degree as when it is heated without load for temperatures up to about 600
C. A number of researchers have reported that improved strength was recorded for stress/cold
strength ratios above about 0.2 [4,13,15].

Therefore, the generalised relation between u, T, and the stress level /uc is an interpolation from
known values of u at the two pre-stress levels during heating, 0% and 20%, which are derived from
existing experimental data at different temperatures, where uc is the ultimate stress at ambient
temperatures (c for cold) with 0 load history.

For 0 < /uc < 0.2 a linear interpolation between the two curves is used to determine the value of
u(T,/uc), which for higher stress levels is taken to be equal to u(T,0.2). The latter assumption can
be conservative as further improvements could be obtained at higher load levels. However, the

01
improvements are expected to be reversed at even higher load levels during heating. Data are lacking
in this area and further testing is required at load levels in excess of 0.2.

u, like the compressive strength, is a function of the stress temperature levels during heating.
Schneider [15] has measured u for normal concretes loaded at three stress levels (0.1, 0.2, 0.3) during
heating. His results showed that u is nearly independent of temperature for stress level of 0.3 with
temperatures up to approximately 650C. However, this was shown to be also true for basalt and
limestone concretes under 0.2 stress level up to 600C [15]. The following u model was developed
using Schneider's results, but adopting the 0.2 stress level (rather than 0.3) as the limiting value above
which it is considered independent of temperature.

At 0 stress level:

 u0(T) = C0 + C1.T + C2.T2 +C3.T3 (22)

where:
C0 = 2.0510-3 C1 = 3.0810-6
C2 = 6.17 10-9 C3 = 6.58 10-12

At 0.1 stress level:

2 3
u1(T) = D0 + D1.T + D2.T +D3.T (23)

where:
D0 = 2.0310-3D1 = 1.2710-6
D2 = 2.17 10-9D3 = 1.64 10-12
And at 0.2 stress level

u2(T) is assumed constant independent of temperature. This constant is taken as 0.002 as given in
BS8110 for ambient temperature.

For 0  /uc  0.2 a second order interpolation scheme in /uc has been used, therefore, the
general formula for u(T,/uc) will take the following form:

 u T, /  uc  = N 0 ( /  uc ). u0 (T) + N1 ( /  uc ). u1 (T) + N2 ( /  uc ). u2 (T) (24)

where:
N0 (  / uc ) = 2.Li(Li-1/2) N1 (  / uc ) = 4.Li.Lj

N2 (  / uc ) = 2.Lj(Lj-1/2)

and Li =1-5(/uc) Lj = 5(/uc).

The criterion for failure of concrete in compression is the crushing strain crush. It is still not clear in
the literature if crush varies with temperatures and whether it is dependent on the stress history during
heating. Anderberg [6,16] in his concrete model uses crush as a function of temperature only and
assumes a linear increase from the value of 0.005 at room temperature to 0.015 at 1000C for
quartzite aggregate concretes. At ambient temperatures crush is related to the rate of loading and
concrete strength; the higher the strength, the lower the crushing strain [12].

00
In the absence of any contrary evidence at high temperatures crush is related to u (expressed in
MPa) based on information obtained at ambient temperature. It is also related to temperature:

 crush = 4.942  10-3  6.995  105 u  3.993  107  2u (25)

Where u is also a function of temperature. It should be noted here that the failure criterion is
approximate and therefore crush has been taken to be the same for all types of concretes.

The available data on the direct tensile strength of concrete is limited. Whatever information is
available was obtained from split cylinder or flexural tests. The latter test yields higher tensile
strength values than the former because of the higher maximum fibre stresses in flexural specimens at
the moment of incipient failure due to the tension stiffening effect in which the surrounding material
blocks the propagation of a crack [12]. The effect of the stress pre-history during heating on the
tensile strength of concrete has not been referred to in the literature and does not appear to have been
investigated. Therefore, as a first approximation  ut has been taken as equal to 0.1u(T), and hence
related to temperature.

The strain-softening effect in concrete models have been investigated in many research works.
Values of crack of between 0.002 and 0.004, independent of temperature, have been assumed in the
literature [6,16,17].

Therefore, due to the lack of comprehensive information about crack and how it is related to
temperature or stress history, a constant value of 0.004 as used by Anderberg for crack has been
assumed in the present model.

Poisson's Ratio

Elastic

At ambient temperatures the values of the elastic Poisson's ratio varies in general between 0.15 and
0.20 [12]. Values of between 0.1 and 0.3 have been reported for the Poisson ratio at elevated
temperatures up to 600C [15].

There exists in fact very few experimental results on the effect of elevated temperatures on the elastic
Poisson's ratio [15]. However, these results vary to an extent that makes it difficult to discern a clear
trend of the elastic Poisson's ratio as a function of temperature. Furthermore, other factors such as the
effect of load level during heating, likely to effect the variation of the elastic Poisson's ratio with
temperature, have not been investigated. Therefore, an average value of 0.15 is assumed here
independent of temperature.

Creep

The creep Poisson's ratio has also not been fully investigated at ambient temperatures, and there is
disagreement on its behavior. In general, the creep Poisson's ratio, under a uniaxial state of stress,
does not appear to substantially differ from the elastic Poisson's ratio [11]. Creep Poisson's ratio has
not been studied at all at elevated temperatures. Therefore, in the model vcreep and velastic are
assumed to be of equal magnitude.

The Bi-axial Model

The strength of concrete under bi-axial compressive stress is about 16% higher than in the uniaxial
state [12,18].

02
Similarly, published works show that the mechanical properties of concrete improve when exposed to
a bi-axial compression state of stress (plane stress) at elevated temperatures [19,20,21]. However,
concrete becomes more liable to explosive spalling if it is exposed to a bi-axial compression state of
stress at elevated temperatures [20].

Therefore, in order to achieve a reliable bi-axial model for concrete at elevated temperature, a
nonlinear bi-axial stress interaction curve is used. The Von-Mises yield criterion is employed in the
compression-compression zone of the bi-axial stress envelope. In the tension-tension zone, no
interaction is assumed between the stresses in the two directions, and in the compression-tension and
tension-compression zones a linear interaction is assumed. This model can be best explained by
figure 4.

4.3 Constitutive Model of Reinforcing Steel

The total strain for steel in STRUCT is given by the sum of the thermal and creep strains together
with the stress-related strain:

tot = th + cr +  (26)

where:

 tot = Total strain, compatible with the virtual displacement


 th = Free thermal strain
cr = Creep strain
 = Instantaneous stress-related strain.

Thermal Strain

The free thermal strain, or "thermal expansion", of steel is expressed by Anderberg [6.16]:

 th = a T2 + b T + c (27)

Where:

a = 5.128 x 10-9C-2
b = 1.179x10-5C-1
c = 2.379 x 10-4

Creep

The creep model adopted in the constitutive steel model of STRUCT is the Dorn-Plem theory [15]:

 cr =  cr,0  2. Z. /  cr,0  if 0 <    0 (28a)

 cr =  cr,0  1.Z. /  cr,0  if  >  0 (28b)

Where:

cr,0 = Projection of the secondary creep line on the axis of creep


Z = Rate of secondary creep

03
 = Temperature compensated time.
0 = The transitional value of  between primary and secondary creep = cr,0/Z.

The temperature compensated time is expressed by the following equation:

H
 = 0 e R  T dt
t
(29)

where:

H = Activation energy of creep, in J/mol


R = Gas constant, in J/mol K
t = Time, in hours
T = Temperature, in K.

Stress-Strain Relation

The stress-strain for reinforcing steel is determined using the following equations:

 =  Es if   y (30a)

 =  y + E*s     y  if     y (30b)

where:

Es = Modulus of elasticity
y = Yield stress
y = Yield strain
E*s = Slope of strain-hardening lines.

Es , E*s , y and y are assumed to be functions of temperature only. This stress-strain model is
bounded by two parallel lines defined by:

Upper Bound:

 u =  y + E*s .     y  (31)

Lower Bound

 1 = -  y + E*s .     y  (32)

Failure is assumed when  reaches a maximum value of 0.02.

4.4 Validation of STRUCT

Example 1: Normal Weight Concrete Slab

04
STRUCT was validated against experimental data using two beam elements with nine through
thickness integration points to analyse the loaded and unloaded slab units tested at Warrington Fire
Research Station [22]. The variation of the deflection of the mid-point of the structural elements with
and without including the creep response model (tot.cr) at high temperatures was examined. The
central vertical deflection and axial displacement results were compared with experimental data in
figures 5 and 6 for the unloaded and loaded slab units respectively. The predicted results showed close
agreement with experimental data.

It can be observed from the results that the effect of creep at high temperatures was not significant in
this example. The reason for this is because creep, being proportional to the stress level, would only
be of significance in regions of both high stress and temperatures. However, highly stressed regions
lie mainly in the compressed top portion of the cross section of the slab which remains relatively cool
during the testing period.

Example 2: Light Weight Concrete Slab

A 150 mm thick light weight aggregate reinforced concrete slab unit was fire tested at Warrington
Fire Research Centre [23]. The slab unit was simply supported over the furnace walls on the two short
edges and had the same dimensions and reinforcement as the normal weight slab described in the
previous section. Two beam finite elements were used to describe the lightweight concrete slab. The
results of this analysis were compared with experimental data. This comparison is shown in figure 7.
The predicted results were in very good agreement with experimental data. The significant
improvement in the overall behaviour of the slab in fire by using lightweight concrete can be noted by
comparing figures 5 and 6.

Example 3: Normal Weight Concrete Column

Validation of STRUCT using beam elements was made against a loaded normal weight reinforced
concrete column [24]. The results of a first study of the problem, carried out with and without creep,
showed the significance of the effect of transient thermal creep on column deformations during first
heating [1]. A parametric study of the effect of eccentricity of loading was conducted using STRUCT
to predict the behaviour of the column. An improved correlation with experimental results was
obtained by introducing eccentricity values of 2.5% and 5% of the depth of the square column as
shown in figure 8. This inherent estimated eccentricity could be explained by the initial imperfection
of the column at ambient temperatures, which is usually assumed in design. Moreover, the observed
temperatures of the longitudinal reinforcing steel bars throughout the fire test were far from
symmetric [24]. The non-symmetry of those temperatures is caused by the ingress of flames into the
concrete section through the experimentally observed cracks distributed at random along the surface
of the tested column. This indicates a thermally induced out-of -straightness that would further
increase the eccentricity value which is not originally accounted for and should, therefore, be
estimated.

5. CONCLUSIONS

 The use of the implicit recurrence scheme in the thermal analysis program allows accurate
predictions to be made using less computer time. This is because the mesh size can be coarser,
and the time increments can be larger and not restricted in size to achieve convergence.

 The concrete constitutive model developed is based on up-to-date experimental data on the
strain behaviour of concrete in compression at transient high temperatures. Simplified
assumptions had to be made with regard to the high temperature strain behaviour of concrete in
tension due to the lack of experimental data.

05
 The structural analysis program has been developed to simulate the structural response of fully
three-dimensional shell and building structures under fire. Validation tests on slabs and a
column show good agreement with experimental results. The incorporation of creep is
important, particularly in the column example. Most other structural programs do not
incorporate transient thermal creep

6. ACKNOWLEDGEMENTS

The author wishes to thank the LUSAS team of FEA Ltd., in particular Dr. T. Crook and Dr. P.
Lyons, for their assistance in the interfacing of the programs with LUSAS.

7. REFERENCES

[1] Terro M.J.: `Numerical Modelling Of Thermal & Structural Response Of Reinforced Concrete
Structures In Fire'. PhD thesis, Department of Civil Engineering, Imperial College, January 1991.

[2] Sterner E., Wickstrom U.: `Calibration Of Fire Resistance Test Furnaces'. Nordset Project No.
785-88. Report 1989:36, Fire Technology, Bor2s 1989.

[3] LUSAS: `Advanced Procedures Training Course'. FEA.

[4] Khoury G.A., Grainger B.N., Sullivan P.J.E.: "Transient Thermal Strain of Concrete During
First Heating Cycle to 600C'. Three papers reprinted from the Magazine of Concrete Research,
CIB W14/86/45 (UK), 1986.

[5] Khoury, G.A., Sullivan, P.E.E., Gungoosingh, R., and Malektojar, S.: "Effect of Age and
Specimen Size on the Transient Thermal Strain of Concrete". Transactions of the 9th
International Conference on Structural Mechanics in Reactor Technology, Lausanne, 17-21
August, 1987.

[6] Anderberg, Y., Thelandersson, S. "Stress and Deformation Characteristics of Concrete at


High Temperatures". Division of Structural Mechanics and Concrete Construction, Lund,
Institute of Technology, Sweden, 1976.

[7] Nasser, K.W., and Neville, A.M.; "Creep of Concrete at Elevated Temperatures". Proceedings,
American Concrete Institute, Vol. 62, 1965.

[8] Khoury, G.A., Dias, W.P.S., Sullivan, P.J.E.S.; "Deformation of concrete and cement paste
loaded at constant temperatures from 140 to 724C". Materials of Structures, Vol. 19, No. 110,
1986, pp. 97-104.

[9] Khoury, G.A.: "Transient Thermal Creep of Nuclear Reactor Pressure Vessel Type Concretes'.
Ph.D. thesis, Department of Civil Engineering, Imperial College, 1983.

[10] Working Party of the Material Technology Divisional Committee: "The Creep of Structural
Concrete". Concrete Society Technical Paper 101, January, 1973.

[11] Neville, A.M., Digler, W.H., and Brooks, J.J.: "Creep of Plain and Structural Concrete".
Construction Press, 1983.

[12] Neville, A.M.: "Properties of Concrete". Longman Scientific and Technical, Third Edition,
1987.

[13] Abrams, M.S.: "Compressive Strength of Concrete at Temperatures to 1600F". Temperature


and Concrete, ACI Special Publication SP-25, Detroit, Michigan, 1971.

06
[14] Schneider, U. et al.: "Strength Losses of Concrete at High Temperatures, Cause and Effects".
Mitteilungsblat fur die Amtliche Material Prufung Niedersachsen, Jahrg. 18/19, Hochschule
Goslar, S. 50-57, 1979.

[15] Schneider, U.: "Properties of Materials at High Temperatures, Concrete". Rilem, June 1986.
(Results up to 1981).

[16] Anderberg, Y.: "Free-Exposed Hyperstatic Concrete Structures - An Experimental and


Theoretical Study". Bulletin 55, Lund, Institute of Technology, 1976.

[17] Gedling, J., Mistry, N., and Welch, A.: "Evaluation of Material Models for Reinforced
Concrete Structures". Computers and Structures, Vol. 24:2, pp. 225-232, 1986.

[18] Kupfer, H., Hilsdorf, H.K., and Rush, H.: "Behaviour of Concrete Under Biaxial Stress". J.
Amer. Concr. Inst., 66, pp. 656-666, August 1969.

[19] Haksever, A.: Ehm, C.: "Application of Biaxial Concrete Data for Bearing Members Under
Fire Attack". Fire Safety Journal, 12; pp 109-119, 1987.

[20] Kordina, K.; Ehm, C.; and Schneider, U." "Effects of Biaxial Loading on the High
Temperature Behaviour of Concrete". Fire Safety Science-Proceedings of the First International
Symposium, pp. 281-290, 1984-1985.

[21] Thienel, K.-Ch.; Rostasy, F.S.; Becker, G.: "Strength and Deformation of Sealed HTR-
Concrete Under Biaxial Stress and Elevated Temperatures", SMIRT 11 Transactions Vol. H
(August 1991), Tokyo, 1991.

[22] Warrington Fire Research Centre (WFRC): 'Ad-Hoc Fire Test Using the Heating Conditions
of BS 476: Part 8: 1972 on Two 150 mm Thick Normal Weight Concrete Slabs". A special
investigation sponsored by the Building Research Establishment, Fire Research Station,
WARRES No. 40728, 20th May 1987.

[23] Warrington Fire Research Centre (WFRC): 'Ad-Hoc Fire Test Using the Heating Conditions
of BS476: Part 8: 1972 on a 150 mm and 250 mm Thick Light Weight Reinforced Concrete
Slab'. A special investigations sponsored by the Building Research Establishment, Fire Research
Station. WARRES No. 40812, 20th May 1987.

[24] Joint Fire Research Organization (FROSI): "A Standard Fire Resistance Test on a Gravel
Aggregate Concrete Column (Reference B1)". FROSI No. 5510, FRO/118/010. Tested-August
1972.

[25] Becker, J. amd Bresler, B.: "FIRES-RC, A Computer Program for the Fire Response of
Structures-Reinforced Concrete Frames". University of California, Berkeley, Fire Research
Group, Report No. UCB FRG 74-3, July 1974.

07

S-ar putea să vă placă și