You are on page 1of 25

INTERNATIONAL JOURNAL FOR NUMERICAL METHODS IN ENGINEERING

Int. J. Numer. Meth. Engng 2000; 48:761–785

Fractional step methods for thermo-mechanical damage


analyses at transient elevated temperatures

J. Stabler1 and G. Baker2; ∗; †


1 Department of Civil Engineering; The University of Queensland; 4072; Queensland; Australia
2 School of Engineering; The University of Warwick; Coventry CV4 7AL; U.K.

SUMMARY

In the interest of computational eciency this paper describes the implementation of a coupled thermo-
damage constitutive model into a coupled time-stepping analysis using fractional step methods. To begin
it is demonstrated that a thermo-damage model can be presented in a thermodynamic framework with the
evolution equations satisfying the ÿrst and second laws of thermodynamics. The equations of evolution are
partitioned in two ways, thus deÿning two fractional step methods: an isothermal method and an isentropic
method. When implemented into a time-stepping algorithm the isentropic method maintains a precise energy
balance for the entire analysis where as the isothermal method can only provide an energy balance at the end
of each thermal time step. In addition, a stability analysis shows that the isentropic analysis is unconditionally
stable while a isothermal analysis is at best conditionally stable. Simulations of thermal fracture in a restrained
specimen under heat show stable growth of damage to failure. Copyright ? 2000 John Wiley & Sons, Ltd.

KEY WORDS: thermo-mechanics; coupled problems; time integration; fractional step methods

1. INTRODUCTION

The behaviour of concrete exposed to high temperatures is a complex phenomenon and is important
in many applications in civil infrastructure and the process industry. Mechanical properties are
known to be highly temperature dependent [1]: thermal damage reduces elastic modulus, and the
overall constitutive response changes from relatively brittle at room temperature to very ductile
at temperatures like 600◦ C. Thermal expansion is in turn dependent on the stress state and stress
history in the material, with additional coupling being provided through transient thermal creep [2].
Research into the appropriate constitutive models, and coupling of heat conduction with vapour
di usion, is an ongoing line of enquiry.
However, in this paper we focus on algorithmic aspects for the solution of thermo-mechanical
damage implemented within a full transient analysis for heat conduction. In particular, we in-
vestigate fractional-step methods for the time integration of this strongly coupled problem. For
the detail in this work, we adopt a non-local damage model based on the scalar damage model

∗ Correspondence to: G. Baker, School of Engineering; The University of Warwick; Coventry CV4 7AL; U.K.
† E-mail: gb@eng.warwick.ac.uk

Received 18 November 1998


Copyright ? 2000 John Wiley & Sons, Ltd. Revised 6 September 1999
762 J. STABLER AND G. BAKER

of Mazars and Pijaudier-Cabot [3], extended to include temperature reliance. The thermal damage
component is linked to known data for specimens heated before loading, and provides an increased
compliance. Importantly, the model complies with a formal thermodynamic framework and satis es
the complete dissipation inequalities under thermal and mechanical damage. Complete details of
the constitutive model, including validation with data, will be given elsewhere. The important
ingredients for this contribution are the strong coupling, and transient elevated temperatures.
Integration of the coupled equations for thermo-mechanical problems can be carried out simul-
taneously. However, the thermal and mechanical elds involve quite di erent time scales, with
the time steps required for the non-linear mechanical behaviour being much smaller than for the
heat conduction analysis. If the thermal analysis is forced to use the same small steps as the me-
chanical analysis, and the degrees of freedom are coupled, then the solution will be unnecessarily
expensive.
Armero and Simo [4; 5] showed that the integration can be performed by splitting the governing
operators (i.e. load–temperature history increments) into two phases, and that the solution of these
is staggered. Further, they investigated two generic types of operator split:
(i) an isothermal phase, i.e. constant temperature with increasing load, followed by heat con-
duction at constant con guration, i.e. a constant load phase at varying temperature; and
(ii) an isentropic phase in which entropy is held constant but temperature may vary along with
load increments, followed again by a thermal phase at constant con guration.
The notion of a staggered (or split) algorithm dates back to Argyris and Doltsinis [6] in which the
isothermal split seems the obvious choice; Park and Felippa [7] give further references to earlier
studies of coupled problems. However, stability has always been recognized as a major challenge to
the isothermal staggered algorithms. Recently, Schre er et al. [8] studied these standard staggered
schemes for coupled thermal, mechanical and vapour di usion problems, and concluded there was
no obvious scheme that was superior for all conditions.
For both thermo-elasticity and thermo-plasticity, Armero and Simo [4; 5] showed that the isen-
tropic split is unconditionally stable, whereas the isothermal one is only conditionally stable. An
analysis of the dissipation during the isothermal phase showed that the nite element displacement
norms lose their contractivity property in the discrete form, so that only conditional stability can
be assured. On the other hand, with entropy constant, the isentropic phase reverts to a mechanical
theory with adiabatic internal energy, and the thermal phase is just heat conduction at constant con-
guration. Hence, unconditional stability of the non-linear thermo-elastic problem is preserved [5].
In the Armero and Simo work [4; 5], there was no interaction between the phases, in the sense
that heat conduction is carried out at xed displacements, which means that the method is ostensibly
only suited to small temperature increments. At the end of the constant load phase, no increment of
strain due to thermal expansion is introduced into the mechanical phase. Hence, there must be an
explicit jump in strain at the beginning of the next loading phase, which has not been considered
for plastic redistribution. For small temperature increments, as might be expected from plastic
dissipation, this strain jump apparently causes no numerical instability. However, for signi cant
temperature changes as expected in our thermo-mechanical analysis at elevated temperatures, this
technique generally leads to complete divergence of the solution.
The approach developed here is to employ the operator splits, but at the end of the separate
phases, a nal simultaneous solution is performed in order to provide consistency between the
thermal and mechanical displacement increments. In that sense, we prefer the term fractional-step
methods, since the integration schemes do not involve operator splits entirely.

Copyright ? 2000 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2000; 48:761–785
FRACTIONAL STEP METHODS FOR THERMO-MECHANICAL DAMAGE 763

The paper is organized as follows. We begin with the thermodynamic framework for our thermo-
mechanical analysis under transient temperature and stress, including the explicit form of the
internal energy function which satis es the complete dissipation inequality. Speci cally, we outline
details of the non-local thermo-damage model for concrete. Next, the discretized forms of the local
energy and momentum balance equations for solution on a nite element mesh are given. Then
we present full algorithmic details of the integration schemes that have been developed, including
explicit forms of discrete equations for the isothermal and isentropic splits; in this, we also include
speci c details of approximations to the logarithmic form of the thermal power term in the internal
energy function which does not requires local iterations. This is followed by a consideration of
stability of the fractional step schemes, showing that the isentropic form is in fact the natural
choice and is completely stable. We illustrate stability with energy balance calculations comparing
internal and applied energy in model problems exposed to heat and stress. Finally, we present an
analysis of thermal fracture in a four point bend specimen heated under restraint to failure. The
growth of damage under restraint stress exhibits no numerical instability even under quite large
time steps.

2. THERMO-DAMAGE FORMULATION

2.1. Free energy


We will assume that the governing energy function is the logarithmic form of the Helmholtz free
energy function. For small displacements, this is given by
  
1 
= U : E( ) : U − ( − 0 ) m( ) : U + c( )  − 0 −  ln (1)
2 0
where E represents the fourth-order elasticity tensor, m the second-order symmetric coupling tensor,
c the volumetric speci c heat and 0 the initial temperature of the system. The independent state
variables are the total strain U, the absolute temperature , and a vector , which is used to represent
any damage variables that may be required in the constitutive model. In this paper it will also be
convenient to de ne the relative temperature as # =  − 0 . The thermodynamic properties of the
system are the stress b, the entropy , and the thermodynamic force A, which is equivalent to an
energy release rate. These properties can be obtained from the energy function as follows:

@
b= = E : U − #m (2)
@U  
@ 
=− = m : U + c ln (3)
@ 0
@
A=− (4)
@
Note that the coupling tensor is given by
m = 3K 1 (5)
where K represents the bulk material modulus and the coecient of thermal expansion; 1 repre-
sents the second-order identity tensor. The total strain eld can be decomposed into a stress-based

Copyright ? 2000 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2000; 48:761–785
764 J. STABLER AND G. BAKER

component U stress (we will call this the stress-related strain) and a free thermal strain component
which is written U free = #1 for linear thermal expansion. As an alternative to Equation (2) the
stress tensor can be obtained from the stress related strain using
b = E : U stress with U stress = U − U free (6)
The two equilibrium equations governing this problem are the force equilibrium equation and the
energy balance equation. The local forms of these equations are
 uˆ t = div[b t ] + b (7)
ˆˆ˙ t = −div[ q̂ t ] + r t + mech (8)
where û t and ˆ t represent the displacement and the entropy at any point within the region
at
any time t. The body forces and heat sources in the material are denoted by b and r, respectively.
The heat ux is denoted by q̂ and the density by . The nal term mech represents the mechanical
dissipation of the system. Because we are concerned with a full transient analysis the momentum
terms are maintained in (7). The energy balance will be developed more fully in the next section.

2.2. First and second laws of thermodynamics


Consider an in nitesimal element within a di erential continuum. The rst law of thermodynamics
can be represented by the following energy balance [9]:
ė + ˙ = p + qc + r (9)
In this equation e represents the internal energy,  is the kinetic energy, p is the stress power, qc
is the rate of heat ow into the element by heat conduction, and r is the rate of heat ow into
the element by a heat source. Using the de nition of stress power, the energy balance equation
can be written in the following form:
ė = b : U̇ − div q + r (10)
The relationship between Helmholtz free energy and the internal energy is e = + . Taking the
time derivative yields
ė = ˙ +  ˙ +  ˙ (11)
Now we write the time derivatives of free energy in terms of the independent variables by applying
the chain rule with constitutive laws (2)–(4) to give
˙ = b : U̇ −  ˙ − A · ˙ (12)
By considering the second law of thermodynamics the condition governing the evolution of the
system can be derived. For solids it is usual to start with the Clausius–Duhem inequality, which
is given as
 
 = −( ˙ + ) ˙ + b : U̇ + q·∇ 1 ¿0 (13)

This equation represents a dissipation condition with the total dissipation consisting of two compo-
nents. The dissipation due to heat conduction is represented by the nal term while the remaining
terms when grouped together represent the mechanical dissipation, i.e.  = mech + cond .

Copyright ? 2000 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2000; 48:761–785
FRACTIONAL STEP METHODS FOR THERMO-MECHANICAL DAMAGE 765

After substituting Equation (12) into Equation (13) we identify the mechanical dissipation as
mech = A · ˙ (14)
Finally, substituting Equations (11) and (12) into Equation (10) and using the result in Equation
(14) the format of the energy balance Equation in (8) is obtained.

2.3. Non-local damage model


At this stage we should record the details of the non-local thermo-mechanical scalar damage model
used as the basis for the computational developments discussed in this paper. This is an extension
of the Mazars’ scalar damage model [3]. The assumptions are that the material remains isotropic
and that damage a ects the Young’s modulus and the bulk modulus equally i.e. Poisson’s ratio
is not a ected by damage. Two internal variables = (d; g) are used to describe the damage in
the material: a mechanically induced damage component, d, and a thermal damage component, g.
Equations (15) and (16) give the reduction of the Young’s modulus and the bulk modulus which
can be used to generate the sti ness and coupling tensors used in the energy functions and the
constitutive relationships,
E = (1 − g)(1 − d)E0 (15)
K = (1 − g)(1 − d)K0 (16)
Expression (16) governs the degradation of the coupling tensor through (5). The thermo-damage
model also assumes that the speci c heat is in uenced by the thermal damage parameter as given
by
 
x2
c = x1 + c0 (17)
(x3 − g)
where x1 ; x2 and x3 are parameters obtained by considering the experimental data.
For the internal variables = (d; g) used to describe damage, the thermodynamic forces can be
identi ed as

−@ 1
Ad = = U : (1 − g)E0 : U − ( − 0 )(1 − g)m0 : U (18)
@d 2
−@ 1
Ag = = U : (1 − d)E0 : U − ( − 0 )(1 − d)m0 : U
@g 2
  
x2 
− c0  − 0 −  ln (19)
(x3 − g) 2 0

Using the above expressions in conjunction with (14) results in the following expression for the
mechanical dissipation:
mech = Ad ḋ + Ag ġ (20)
The mechanical damage parameter represents damage in the material due to mechanical work and
is assumed to be a function of the positive stress-related principal strains. To achieve this, a scalar

Copyright ? 2000 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2000; 48:761–785
766 J. STABLER AND G. BAKER

quantity called the local equivalent strain is introduced and is given by


s  2
P3 |istress | + istress
˜ = (21)
i=1 2

where istress refers to the principal values of the stress-related strains. The value of this quantity
is rounded to the nearest microstrain [10]. To avoid ill-posedness due to strain softening the
mechanical model must include an internal length scale lc [11]. This leads to the de nition of
a non-local equivalent strain variable, calculated as the weighted average of the local equivalent
strain in a representative region about the point considered at position s:

Z
1
 = (x − s) ˜ d
(22)
Vr

Z
Vr = (x − s) d
(23)

is the volume of the structure and the weighting function (x − s) is de ned by a suitable bell
function:
 
4kx − sk2
(x − s) = exp − (24)
lc2
The representative region, over which  is calculated, can be taken as a circle of radius 1:2lc
about point s. Outside of this region the value of the weighting function is taken to be zero. The
non-local equivalent strain is used to control the growth of damage by the de nition of a damage
surface f =  − k(d; ): The conditions for damage growth are


if f¡0
ḋ = 0 (25)
or if f = 0 and ḟ¡0

ḋ ¿ 0 if f = 0 and ḟ = 0

The softening parameter, k(d; ), takes the maximum of the largest value of  reached during the
previous loading history at a point, and the threshold value k0 () which is given by
(Ac − 1)
k0 () = (26)
Ac F()Bc
where Ac ; Bc are material parameters and F() is a function yet to be determined. To distinguish
between tensile and compressive damage the principal stresses are decomposed to b = b+ + b−
where b+ contains only the positive principal stresses and b− contains only the negative principal
stresses. The corresponding strain tensors are then given by
U stress
t = E−1 : b+ and Ucstress = E−1 : b− (27)
It should be noted that while b+ contains only positive values, U stress
t will contain both positive
and negative values, and similarly for the strains associated with b− .

Copyright ? 2000 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2000; 48:761–785
FRACTIONAL STEP METHODS FOR THERMO-MECHANICAL DAMAGE 767

The damage parameter, d, is taken as the weighted average of a uniaxial tension damage
parameter, d t , and a uniaxial compression damage parameter, dc
d = t d t + c dc (28)
where

(1 − Ac )k0 () Ac
dc = 1 − − [Bc F()(−k
 0 ())]
(29)
 e
(1 − A t )k0 () At
dt = 1 − − [B t (−k (30)
 e  0 ())]
 
P3 cistress (tistress + cistress )
c = Hi (31)
i=1 ˜2
 
P3 tistress (tistress + cistress )
t = Hi (32)
i=1 ˜2
Z3
F() = (33)
( + Z4 )

where Ac ; A t ; Bc ; B t ; Z3 and Z4 are material damage parameters. In these expressions, F()


controls the decay in the peak stress with temperature, as shown in the next section. The term Hi
used in Equations (31) and (32) is given by Hi = 1 if tistress + cistress ¿0, and Hi = 0 otherwise.
The thermal damage parameter, g, represents damage due to thermal expansion and is assumed
to be a function of the free thermal strain and is thus given by
g = 1 − (Z1 − Z2 )F() (34)
where Z1 and Z2 are additional material parameters. Evolution of thermal damage is controlled by
the following condition:

ġ=0 ˙
if 60
(35)
ġ¿0 ˙
if ¿0

2.4. Uniaxial stress–strain response


The thermo-damage model presented here is an isotropic model and consequently the appropriate
validation test is the uniaxial stress strain test. There are eight material damage parameters that
need to be speci ed for this model. Four of the parameters are present in Mazars’ damage model
and Saouridis [12] determined the variation of these parameters for standard concrete:
16Ac 61:5; 5006Bc 62000; 0:76At 61:2; 10 0006Bt 650 000
It should be noted that these values where obtained by considering the local version of Mazars’
model. There are many instances in the literature [13; 14] where researchers, using a non-local
version of Mazars model, have used values of Bt ranging as low as 3500. Values of this order

Copyright ? 2000 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2000; 48:761–785
768 J. STABLER AND G. BAKER

Figure 1. Constitutive law: compression. Figure 2. Constitutive law: tension.

of magnitude signi cantly alter the shape of the stress–strain curve. The values for the parameters
Z1 to Z4 were obtained by considering the experimental data available in the literature for concrete
subjected to elevated temperatures. Special attention was given to uniaxial compression data [1; 2].
The values used in this paper are: Z1 = 1:296; Z2 = 992×10−6 ; Z3 = 312:5 and Z4 = 14:5. The initial
values for Young’s modulus and Poisson’s ratio were E0 = 31 GPa and  = 0:2.
Figure 1 shows the predicted stress–strain relationships for concrete specimens that have been
tested in uniaxial compression at elevated temperatures. These results capture the main trends
shown in the experimental data, and are suciently accurate for the current computational exercise.
However, it should be noted that the reduction of the peak compressive stress at 350 and 450◦ C
is larger than that published in the experimental data. This is due to the linear form of Equation
(34). The use of a tri-linear function would probably improve the results obtained from the thermo-
damage model.
The uniaxial tensile stress–strain behaviour predicted by the thermo-damage model is presented
in Figure 2. There are no post-peak experimental results for concrete tested in direct tension to
validate these results. There are two main temperature dependent features in the tensile response
predicted by the model. These are the reduction in the peak tensile stress and the reduction in the
area under the stress strain curve as temperature increases. This is consistent with our hypothesis
for the tensile behaviour of concrete at elevated temperature. The damage parameters used to
obtain these curves were: Ac = 1:4; At = 0:8; Bc = 1850 and Bt = 2:1 × 104 .

3. INTEGRATION SCHEMES AND TIME-STEPPING TECHNIQUES

The thermo-damage model described previously has been incorporated into a transient analysis.
The equation representing mechanical behaviour (7), is a second-order di erential equation and is
solved in the time domain using a Newmark time-stepping approach. The time integrating factors
used are = 12 and = 14 which corresponds to enforcing the di erential equation at time tn+1=2 .
The thermal evolution equation (8), is a rst-order di erential equation and is solved in the time
domain using a Crank–Nicolson scheme with = 12 . This ensures that the solution of the thermal

Copyright ? 2000 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2000; 48:761–785
FRACTIONAL STEP METHODS FOR THERMO-MECHANICAL DAMAGE 769

phase is consistent with the solution of the mechanical phase. Separately both of these schemes
are unconditionally stable. However, when these schemes are coupled and material non-linearities
introduced, stability of the resulting time-stepping scheme should be examined.
Three di erent methods of coupling the mechanical and thermal equations will be discussed
in this section, but rstly the governing equations for the mechanical and thermal equilibrium
equations will be presented in a secant sti ness form that is suitable for nite element discretization.

3.1. Governing equations


The equation governing mechanical equilibrium at any point within the region
at any instant
in time, t, was given in (7). First substitute (2) into (7) and employ a Newmark time-stepping
scheme between time tj and tj+1 , then pre-multiply the equation by a continuous and smooth vector
function, âT , and integrate over the volume to obtain

Z Z
2 1
âT ûj+1 d
+ ∇( â) : [Ej+1 : U[ûj+1 ]]T d

(t)2
2

I I
1 1
= âT bj+1 n ds + âT bj n ds
2 @
2 @

Z Z
2 1
+ â ûj d

T
∇( â) : [ Ej : U[ûj ]]T d

(t)2
2

Z Z
1 1
+ ∇( â) : [mj+1 #̂j+1 ]T d
+ ∇( â) : [mj #̂j ]T d

2
2

Z Z
2
+ âT  v̂j d
+ âT b d
(36)
t

Similarly, the Crank–Nicolson scheme is applied to the thermal equilibrium equation for the time
interval tn to tn+1 . Using the chain rule on (3) and substituting into equation (8), then pre-
multiplying the equation by a smooth continuous scalar eld variable, â, integrating over the
volume and apply Fourier’s law for heat conduction gives

Z   Z
1 x2 1
a x1 + c0 n+1 d
+ ∇(a) ·∇(k n+1 ) d

t
(x3 − gn+1=2 ) 2

Z
1
+ a[(1 − dn+1=2 )(1 − gn+1=2 )m0 : U̇]n+1 d

Z
1
− a[(1 − dn+1=2 )m0 : Un+1=2 ] ġn+1 d

Z
1
− a[(1 − gn+1=2 )m0 : Un+1=2 ] ḋn+1 d

Z    
1 x2 (n+1 + n )
+ a c0 ln ġ n+1 d

2
(x3 − gn+1=2 )2 20

Copyright ? 2000 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2000; 48:761–785
770 J. STABLER AND G. BAKER

I I Z
1 1 1
=− a q n+1 · n ds − a q n · n ds − ∇(a) · ∇(k n ) d

2 s 2 s 2

Z  
1 x2
+ a x1 + c0 n d

t
(x3 − gn+1=2 )
Z
1
− a[(1 − dn+1=2 )(1 − gn+1=2 )m0 : U̇] n d

Z
1
+ a[(1 − dn+1=2 )m0 : Un+1=2 ]ġ n d

Z
1
+ a[(1 − gn+1=2 )m0 : Un+1=2 ] ḋn d

Z    
1 x2 (n+1 + n )
− a c 0 ln ġ n d

2
(x3 − gn+1=2 )2 20
Z Z
+ a rn+1=2 d
+ a mech d
(37)

In Equations (36) and (37), n, represents an outward unit vector that is normal to the surface and,
k represents Fourier’s coecient of heat conduction. mech is the weighted average mechanical
dissipation calculated over the thermal time interval and is obtained from (20). These equations
represent the linearized weak form of the equilibrium equations and can be solved after a nite
element discretization into nodal variables.

3.2. Integration schemes


When investigating the transient behaviour of a body subject to mechanical and thermal loading,
a suitable time scale must be selected. From physical observations we are aware that the time
scale for thermal problems is several orders of magnitude larger than the time scale for mechan-
ical problems. It would be advantageous to make use of these di erent time scales in numerical
computations.
Unfortunately, due to the coupling between the thermal and mechanical equilibrium equations,
it is not simple to take advantage of these di erent time scales. In this paper we investigate three
schemes which were inspired by the work of Armero and Simo [4]. These are solutions based on
(i) an uncoupled isothermal phase, (ii) an uncoupled isentropic phase and (iii) a fully coupled
(simultaneous) solution. The general concept of these methods is that the operator governing
velocity and thermal rates is split and that the mechanical and thermal loading are (in some sense)
analysed in separate phases.
Both the isothermal and isentropic methods have the same structure and are illustrated in
Figures 3 and 4. The thermal evolution is calculated based on a thermal time step tn . In solv-
ing the mechanical evolution the thermal time step is divided into J sub-time steps tj and, by
applying either an isothermal or isentropic condition the rst J − 1 of these sub-time steps can be
solved independently of the thermal equilibrium equation. The last sub-time step in the mechanical
phase is then solved iteratively with the thermal time step to give the displacements u n+1 and the
temperatures n+1 .

Copyright ? 2000 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2000; 48:761–785
FRACTIONAL STEP METHODS FOR THERMO-MECHANICAL DAMAGE 771

Figure 3. Isothermal analysis. Figure 4. Isentropic analysis.

During the evolution of the system, mechanical damage is allowed to grow over each sub-time
steps while thermal damage is only allowed to evolve over the thermal time step.

3.3. Isothermal methods


As a rst consideration when considering Equation (36) the isothermal method may appear to be
the more obvious choice. It is certainly the easiest to implement. In solving the isothermal phase
(sub-time step 0 to J − 1) it is only necessary to modify the equation by setting #j+1 = #j . In
the nal mechanical time step this condition is removed and as a result the equation is coupled
with the thermal equilibrium equation. The structure of this method is shown diagrammatically in
Figure 3.

3.4. Isentropic method


On the other hand, if the concepts of thermodynamics are considered more carefully then the
isentropic method becomes a more obvious choice. In this case the rst J − 1 mechanical time
steps are uncoupled from the thermal equilibrium equation by applying an isentropic condition
j = j+1 . For the nal mechanical time step this condition is removed and the coupling is restored.
Figure 3 shows the structure of this method.
It is now necessary to develop the mechanical equilibrium equation that governs the evolution
of the system during the isentropic phase. Using Equation (3) the isentropic condition can be
written as
   
j+1 1 1 cj j
ln = [mj : Uj ] − [mj+1 : Uj+1 ] + ln (38)
0 cj+1 cj+1 cj+1 0
Because g evolves during the thermal time step only the speci c heat will remain constant during
the isentropic phase, i.e. cj+1 = cj = cn . Now taking the exponential of both sides in Equation (38)

Copyright ? 2000 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2000; 48:761–785
772 J. STABLER AND G. BAKER

it is possible to write
 
1
#j+1 = j exp [mj : Uj − mj+1 : Uj+1 ] − 0 (39)
cn
Unfortunately, if we substitute this equation into Equation (36) we cannot transfer Uj+1 to the
left-hand side of the equation. Therefore, the isentropic phase would require an iterative solution
and as a result be too expensive.
It avoid this problem it is necessary to realize that during the adiabatic phase the temperature
change in the structure is quite small and so a linear approximation can be established. Firstly,
we nd some constant n such that
 
n #n
ln = n (40)
0 0
Consequently, at sub-time step j + 1 the following approximation can be made:
 
(#n + #j+1 ) n + #j+1
n ≈ ln for small # (41)
0 0
As a result the adiabatic condition can be approximated by
#j+1 #j
mj+1 : Uj+1 + cn n = mj : Uj + cn n (42)
0 0
and hence the relative temperature can be written as
0 0
#j+1 = mj : Uj − mj+1 : Uj+1 + #j (43)
n cn n cn
Substituting this into Equation (36) gives
Z Z
2 1
âT ûj+1 d
+ ∇( â) : [Ej+1 : U[ûj+1 ]]T d

(t)2
2

Z   T
1 0
+ ∇( â) : mj+1 ⊗ mj+1 : U[ûj+1 ] d

2
n cn
I I
1 1
= â bj+1 n ds +
T
âT bj n ds
2 @
2 @

Z Z
2 1
+ â ûj d

T
∇( â) : [Ej : U[ûj ]]T d

(t)2
2

Z   T
1 0
+ ∇( â) : mj+1 ⊗ mj : U[ûj ] d

2
n cn
Z Z
1 1
+ ∇( â) : [mj+1 #̂j ]T d
+ ∇( â) : [mj #̂j ]T d

2
2

Z Z
2
+ âT  v̂j d
+ âT b d
(44)
t

Copyright ? 2000 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2000; 48:761–785
FRACTIONAL STEP METHODS FOR THERMO-MECHANICAL DAMAGE 773

It is this equation that controls the mechanical evolution during the isentropic phase of mechanical
analysis. Our experiences have shown negligible di erences in computational cost between the
isothermal and isentropic methods.

3.5. Simultaneous analysis


This method of analysis does not take advantage of the di erent time steps for the mechanical and
thermal phases. The time step is, therefore, governed by the mechanical analysis and at each time
step there is an iterative solution between the thermal and the mechanical equilibrium equations. It
would be expected that a problem solved using the simultaneous analysis would take signi cantly
longer to complete than a problem solved using an isentropic or isothermal methods. For the
numerical problem analysed in Section 5 the authors found the solution time for the simultaneous
analysis was 6 times longer that the solution time for the isentropic analysis.

3.6. Structure of a numerical algorithm


As mentioned earlier the numerical algorithm used to solve the thermo-mechanical damage problem
is essentially the same for both the isothermal and isentropic methods. The overall numerical
structure is shown in Algorithm Box 1.
Each time that the mechanical equilibrium equation is solved it is possible that mechanical
damage may occur. If damage does occur then it is necessary to iterate to obtain a converged
solution for the displacement eld. Algorithm Box 2 gives the structure of the algorithm required
to nd the damage solution. For the damage model presented here, we use a secant sti ness
method in solving for the displacement eld. For this method convergence of the mechanical
solution is based on the norm of a residual force vector. There are two reasons why the secant
sti ness method is used in this paper. Firstly a tangent operator method requires the derivative of
the equivalent strain which is possible in a gradient damage approach [15] but not for the type of
non-local model used here. Secondly, the damage function used in this paper is calculated from
weighting functions which, when di erentiated, introduce current stress increments into the tangent
operator.

Copyright ? 2000 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2000; 48:761–785
774 J. STABLER AND G. BAKER

Both the isothermal and isentropic methods require the thermal evolution equation and the nal
sub step solution of the mechanical phase to be solved simultaneously and the coupling between
these two equations requires that this is done iteratively. Algorithm Box 3 gives the structure for
this algorithm. Again thermal damage may occur during in the thermal solution and mechanical
damage may occur in the mechanical solution.

4. DISCUSSION ON THE TIME-STEPPING SCHEMES

Because of the di erent time scales associated with the mechanical and thermal problems the
isentropic and isothermal approaches are computationally more attractive than the simultaneous
analysis. The isentropic method has the further advantage that it is unconditionally stable for a
linear thermo-elastic analysis. For this case the size of the time step is governed by accuracy
considerations alone. Unfortunately, it is not intuitively obvious that this unconditional stability is
retained when a thermo-damage model is incorporated into the analysis. It is therefore desirable
to include a stability check in the analysis. A common check used in structural dynamics is an
energy balance calculation, which seems appropriate in a thermo-mechanical setting.
In this section we will formulate the appropriate global energy balance equation and then use this
equation to discuss the implications of the isothermal and isentropic methods for the solution of
the coupled thermo-damage problem. During this discussion it will become clear that the isentropic
method is a more natural formulation than the isothermal method, although a constant temperature
phase may have seemed a more obvious choice. We conclude this section with a brief stability
estimate for the isothermal and isentropic methods.

Copyright ? 2000 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2000; 48:761–785
FRACTIONAL STEP METHODS FOR THERMO-MECHANICAL DAMAGE 775

4.1. Global energy balance equation


The time derivative of the energy balance equation per unit volume was given in Equation (9).
This expression can be written in a slightly di erent form as

ė + ˙ = ’˙ + !˙ (45)

where ’ represents the heat received by an in nitesimal element and ! represents the work done.
Substituting in Equation (12), integrating over the volume and making use of (14), results in the
following global energy balance:
Z
(b : U̇ +  )
˙ d
− mech + K̇ = Ẇ + Q̇ (46)

In this equation K represents the total kinetic energy of the structure; W the total work done by
the external forces on the structure; Q the total heat received by the structure and mech represents
the energy dissipated over the entire volume of the structure. Equation (46) allows an energy
balance check to be completed at the end of each sub-time step and in our scheme is computed
explicitly from the nodal variables.

Copyright ? 2000 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2000; 48:761–785
776 J. STABLER AND G. BAKER

4.2. Implications for the mechanical solution methods


We will now consider the implications of using the isentropic and isothermal methods for the
solution of the coupled thermo-mechanical problem. To begin we return to some of the fundamen-
tal principles of thermodynamics. Firstly consider the physical geometry of the structure as the
thermodynamic system. The internal energy is the potential from which all of the thermodynamic
properties of the system can be obtained. Note that other energy potentials, such as the Helmholtz
free energy, can be associated with the internal energy by de ning the appropriate transformation.
For example, the Helmholtz free energy is de ned by the transformation = e − .
There are only two mechanisms for energy to be transferred between the system and the sur-
roundings; these are by either heat or work transfer. The second law of thermodynamics recognizes
that heat and work are not equivalent quantities. Now assume that the internal energy is a function
of a set of independent state variables. It is possible to specify one state variable, the entropy,
to represent the total elementary heat input into the system. All of the other state variables must
therefore represent input of elementary work.
Now if we consider an insulated body subject to mechanical work only (i.e. no heat transfer
between the body and the environment), and if we assume further that no damage occurs in the
material (i.e. no mechanical energy is dissipated to heat), then the entropy of the body should
remain constant throughout the loading cycle. Recall that it may be necessary for the temperature
to change to maintain this constant entropy condition; refer to Equation (43).
Consider now the isentropic method. During the mechanical phase the intrinsic assumption is
that the entropy remains constant. Consequently, this method exactly emulates the physical situation
and therefore appears to be a natural choice. The isothermal method does the opposite. By keeping
the temperature constant during the mechanical phase it forces the entropy to change. It is only
in the nal coupled mechanical step that the situation is reversed and the adiabatic condition is
enforced. Thus, this method is a less natural choice in solving the coupled thermo-elastic problem.
The argument can now be extended to the thermo-damage case. When damage occurs there is
a conversion of energy from work to heat. In both the isentropic and isothermal methods, the
amount of heat generated during the mechanical phase is stored and applied as a heat input for the
solution of the thermal equilibrium equation. Therefore, sub-time steps j = 1 to J −1 are essentially
conducted under an adiabatic conditions. It is only at the nal mechanical sub-time step that there
is a heat input into the system and the entropy is allowed to evolve. Once again the isentropic
method is more representative of the physical situation.
The argument also applies to the general case where there may be heat transfer across the
boundary of the system. In e ect, the heat input is applied in pulses during the nal mechanical
time step, but is allowed to conduct through the system over the full thermal time step.
An estimate of the thermal time step, based on the heat conduction problem, can be obtained
from

h2 c
t = (47)
4k
where h is a representitive element size, c the speci c heat per unit volume and k is Fourier’s
coecient of heat conduction.
We will now consider a numerical experiment where a uniaxial compressive load is applied
to a panel and then removed. The load is applied under displacement control and the maximum
strain reached during the loading process is 3000 . The load is then removed from the panel.

Copyright ? 2000 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2000; 48:761–785
FRACTIONAL STEP METHODS FOR THERMO-MECHANICAL DAMAGE 777

Figure 5. Uniaxial compression: evolution Figure 6. Uniaxial compression: energy balance


in temperature. checks based on isentropic and isothermal analyses.

The analysis is conducted over a period of 200 s and a time increment of 10 s is used for the
the thermal time step. It is assumed that there is no transfer of heat between the system and the
surroundings.
The relevant results are shown in Figures 5 and 6. Figure 5 shows the evolution of the temper-
ature in the system and Figure 6 illustrates the di erences between the isentropic and isothermal
methods by plotting the corresponding energy evolution with time. The curve labelled ‘applied
energy’ represents the r.h.s. of Equation (46) while the curves labelled ‘energy balance isentropic
method’ and ‘energy balance isothermal method’ represent the l.h.s. of Equation (46). It can be
seen that the energy for isentropic method corresponds precisely to the applied energy curve for
the entire analysis where as the energy for isothermal method only matches the applied energy
curve at the end of each thermal time step; moreover, the variation is a signi cant proportion
of the work done. The gure demonstrates that only with the isentropic method is it possible to
conduct a continuous energy balance check, and enforces the point that this method is a more
natural solution of the coupled problem.
For a second numerical experiment we will consider a specimen where energy is applied in
the form of heat and work. Figure 7 shows the set up of the problem. The left-hand face of the
specimen is heated with a constant heat ux of 1 kW=m2 . All other faces are insulated. There are
vertical supports on the bottom face and horizontal supports on the right-hand face. At time 7500 s
a vertical stress is applied increasing from zero to 30 MPa over a period of 50 s. The load is
then maintained at 30 MPa for the remainder of the analysis. It is evident from Figure 7 that the
amount of energy received by the specimen due to work is much smaller than the energy received
from heat since otherwise there would be a noticeable change in slope when the compressive load
is applied. It should be noted further that in this example the heat ux in quite low, causing a
temperature increase of approximately 0:24◦ C=min. This heat ux was chosen to help visualize
the di erences between the isentropic and isothermal analyses.
An energy balance was conducted for both the isentropic and isothermal analyses. A portion
of these results are shown in Figure 8. Again the energy for the isentropic method corresponds
precisely with the applied energy for the entire analysis. For the isothermal method there is a

Copyright ? 2000 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2000; 48:761–785
778 J. STABLER AND G. BAKER

Figure 7. Heating and loading of a panel: total applied energy.

Figure 8. Heating and loading of a panel: energy balance during mechanical loading.

balance of the energy equation for all sub-time steps before and after the load is applied. The
reason for this is that before and after the load is applied there is no energy input into the system
during the isothermal phase of the mechanical analysis and hence the energy balance is maintained.
However, for the time during which the load is increasing the energy balance is only maintained
at the end of each thermal time step. This is because there is an energy input into the system
during the isothermal phase.
As argued above the formation of a fractional step method inherently keeps the heat (and hence
the entropy) constant during the purely mechanical phase. The isentropic method is consistent with
this formulation by keeping the entropy constant while the isothermal method keeps the temperature
constant while the entropy changes. Thus, the isothermal method results with an imbalance in the
energy function during the isothermal phase. This imbalance is corrected in the nal mechanical
sub-time step.

4.3. In uence of relative time steps


Both Armero and Simo [5] and Simo and Miehe [16] use the operator splits similar to those
described here, but in a staggered algorithm. That is, they essentially apply a mechanical step
while freezing the temperature (or entropy), and then apply the thermal step while freezing the
deformation. Moreover, they use reasonably large strain increments, typically greater than 10 000 
occurring over a time interval of 10 s. This is possible since convergence of thermo-plasticity
problems with hardening can be achieved with larger time steps than those used in thermo-damage
problems where softening occurs due to mechanical damage. In this paper, we allow for sub-
increments in the mechanical phase relative to one thermal time step to allow for convergence of
the mechanical softening problem. The method is a generalization of staggered algorithms.

Copyright ? 2000 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2000; 48:761–785
FRACTIONAL STEP METHODS FOR THERMO-MECHANICAL DAMAGE 779

This raises the important issue of the relative size of the time steps used in the mechanical and
thermal phases of the fractional step methods described in this paper. Of particular importance is
the rate at which heat is generated during the evolution of damage. In the physical system, the
dissipated energy associated with the mechanical damage causes a continual increase in the entropy
of the system. The split methods do not input this dissipated energy until the nal mechanical time
step and therefore does not allow the entropy to evolve until the nal mechanical time step. To
maintain the accuracy of the solution process, the size of the thermal time step will be governed
by two criteria. The rst is the time scale associated with the heat conduction problem. The second
is the time taken for the heat, generated by mechanical dissipation during the purely mechanical
phase, to accumulate so that it has a signi cant in uence on both the local thermal strains and the
local degree of thermal softening of the yield or damage surface, and therefore on the stress state
of the system. For materials such as concrete, it is the experience of the authors that the time
scale associated with the heat conduction problem is the governing time scale. This is illustrated
by the results of the uniaxial compression example presented in the previous section.
In that example, a compressive load is applied under displacement control until a strain of
3000  is reached. During this loading process the mechanical damage reaches a value of d = 0:68
and the total energy dissipated is 35 kJ. Using a speci c heat of 1:05 J=(g◦ C) and a density of
2450 kg=m3 , the change in temperature due to the total amount of dissipated energy can be
calculated to be 0:0136◦ C, which has a very small in uence on the stress state of the system.
The temperature pro le during the loading state of this system is shown in Figure 5. In this
gure the smooth change in temperature is due to the elastic heating e ect and the discrete jump
at the end of each thermal time step is due to the dissipated energy being injected into the
system.
The strains in concrete rarely exceed 10 000  without total damage occurring to the material
and therefore it can be seen from the above example that the heat conduction problem will govern
the size of the thermal time step. Even at larger strains where the material is signi cantly damaged,
the change in damage, and hence dissipation, corresponding to a strain increment reduces as
the strains continue to increase, so that again we do not see an in uence on the di erent time
scales.
Although nite strains are not studied in this paper, it is worth commenting that both Armero and
Simo [5], and Simo and Miehe [16] used relatively large time steps and observed no diculties;
for example even for the expansion of a thick-walled cylinder they use strain increments up to
105 000  over a time interval of 10:5 s but only see a temperature rise of 2:5◦ C per step. The
important factor is again the size of strain increment per step which in uences the dissipation and
hence the heat generated. In fact, Armero and Simo [5] studied the problem of adiabatic shear
banding, where they show that under very high strain rates the heat conduction is too slow for
plasticity and softening to distribute, and so localisation results in a short time scale. They simply
reduced the size of time step according to strain rate to give a consistent strain increment per step
for the various loading rates studied. This reduction is required for both a simultaneous solution
and a split solution in order to capture the localization.
In these extreme cases, we can also reduce the size of the thermal time step so that it is closer
to the size of the mechanical sub-increments, in order to capture localization by the rapid thermal
softening in the shear band. In other words, the method here can recover a staggered algorithm
like Armero and Simo [5] if required under extremes loading rates. Indeed, one might argue that
the time scale for the conduction problem is a function of the quasi-static case modi ed by the
strain rate.

Copyright ? 2000 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2000; 48:761–785
780 J. STABLER AND G. BAKER

4.4. Stability estimates


The objective of this section is to comment on the stability of the numerical algorithms for initial
boundary value problems. For more details on the stability of the isentropic and isothermal phases
the authors refer to the work of Armero and Simo [4; 5] and references therein.
The procedure here is to de ne a suitable function that evolves as the initial boundary value
problem evolves through time. Stability of the numerical algorithm is then maintained if this
function is non-increasing along the ow generated by the initial boundary value problem.
The appropriate function [5] is given by
Z   Z I
1
L= e − 0  + 0 v · v d
+ Vext d
− T · u ds (48)
2 s

where Vext is the energy potential of the body force (i.e. b = −@Vext =@u) and T is the traction
vector (T = bn with n as the outward unit normal). The other symbols have their usual meanings.
Taking the time derivative of this function by using Equations (11) and (12), the identity
b :U̇ = b : ∇(v), and Green’s theorem for integration by parts, gives
Z Z
dL
= v · [v̇ − div[b] − b] d
+ [−mech + ( − 0 )]
˙ d
(49)
dt
The mechanical equilibrium equation causes the rst integral to vanishes leaving
Z    
dL 0
= −mech + 1 − ˙ d
(50)
dt 

In obtaining this equation use has been made of the mechanical equilibrium equation and the
de nition of the Helmholtz free energy. For the simultaneous method and the coupled phase of
the isothermal and isentropic method we can use the thermal equilibrium equation, given by (8),
and Green’s theorem to obtain

Z I   Z   Z  
dL 0 0 0 1 0
= −mech d
− 1− n · q ds + − q · ∇ d
+ 1− r d

dt  s    
Z I   Z  
0 0 0
= −  d
− 1− n · q ds + 1− r d
(51)
 s  

where the quantity q · ∇(1=) is the dissipation due to heat conduction. The total dissipation is
given by  = mech + cond . Now if there are no heat sources and the boundary conditions are
either q = 0 or  = 0 ∀t then the above equation will reduce to
Z
dL 0
=−  d
(52)
dt 
Because we always satisfy the second law of thermodynamics, which requires ¿0, the require-
ment for the energy function L to be non-increasing along the ow is satis ed. From this we
can conclude that the simultaneous method and the coupled phase of the isothermal and isentropic
methods are stable. For the isothermal and isentropic methods we still need to investigate the

Copyright ? 2000 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2000; 48:761–785
FRACTIONAL STEP METHODS FOR THERMO-MECHANICAL DAMAGE 781

stability of the purely mechanical phase of the solution procedure (mechanical sub-time steps 1 to
J − 1). During this phase the evolution equations are uncoupled and can be written as
Isothermal phase

v = u̇
v̇ = div[t ] + b (53)
c˙ = 0

Isentropic phase

v = u̇
v̇ = div[t ] + b (54)
˙ = 0

4.4.1. Isothermal method. We will now consider the evolution of L subject to the constraints of
Equations (53). Rewriting Equation (50) as
Z
dL
= [−mech + #] ˙ d
(55)
dt
For the Helmholtz free energy function used in this paper the entropy is given by
 

 = m : U + c ln (56)
0
and the time derivative can be written as
 
 c
˙ = ṁ : U + m : U̇ + ċ ln + ˙ (57)
0 

Enforcing the isothermal condition ˙ = 0 results in


Z    
dL 
= −mech + #ṁ : U + ċ ln + (#m) : ∇(v) d
(58)
dt 0
Using Greens theorem we write
Z    I Z
dL 
=− mech + #ṁ : U + ċ ln d
+ [#m] v · n ds − v · div[#m] d

T
(59)
dt 0 S

In the thermo-elastic case with homogeneous boundary conditions this reduces to


Z
dL
= − v · [m∇(#)] d
(60)
dt
As discussed by Armero and Simo [4] this does not ensure that L is a non-increasing function
and therefore the isothermal stage is at best conditionally stable. It should be noted that while
the nal simultaneous step is essential for accurate solutions involving transient temperatures, it
cannot overcome the conditional stability of this isothermal phase.

Copyright ? 2000 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2000; 48:761–785
782 J. STABLER AND G. BAKER

Figure 9. Finite element mesh for the restrained heating problem.

4.4.2. Isentropic method. To investigate the evolution of L during the isentropic phase, Equation
(50) is reproduced below
Z
dL
= [−mech + #]˙ d
where # =  − 0 (61)
dt
The constant entropy condition ˙ = 0 reduces the time derivative of L to
Z
dL
= − mech d
(62)
dt
and because thermodynamics requires that mech ¿0, L will be non-increasing along the ow
generated by the initial boundary value problem. This reinforces that point that the isentropic
method is a more appropriate choice than the isothermal method for the purely mechanical phase.
Finally, it should be noted that if the thermal boundary conditions di er from those used to
simplify Equation (51) then it is possible that the numerical solution for the fully coupled problem
will not be unconditionally stable. However, when analysing the examples in this paper no stability
problems were evident.

5. THERMAL FRACTURE PROBLEM

In this section an analysis of a four-point mixed-mode bending specimen is presented. The aim is
to heat the underside of the specimen, forming a thermal gradient which will drive a crack from
the notch until failure. The specimen is 440 mm long and 100 mm deep. The notch is located in
the centre of the upper face and is 20 mm deep and 2 mm wide. Figure 9 shows the reference
con guration of the discretised specimen. The nite element mesh contains 1378 linear plane stress
elements and 759 nodes. The restrained nodes are indicated by the arrows in the gure. For the
central supports the nodes on both sides of the arrow are restrained. The specimen is subject to
heating on the bottom face that causes a temperature increase of 0:7◦ C per minute; the other three
faces are assumed to be insulated. An internal length of 15 mm was used.
The analysis was run rstly with a thermal time step of 20 s and then run with a thermal time
step of 4 s to con rm the results. In both cases there were 200 sub-time steps in the mechanical
phase and in each case an isentropic and isothermal analysis was conducted. The results are
consistent for all of the analyses and are summarized in Figures 10–13. In both cases the analysis
failed after a modest temperature increase of 60◦ C.
For this problem both the temperature pro le and restraining forces cause tensile stresses in the
vicinity of the notch. The development of the temperature eld induces horizontal self equilibrating

Copyright ? 2000 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2000; 48:761–785
FRACTIONAL STEP METHODS FOR THERMO-MECHANICAL DAMAGE 783

Figure 10. Maximum restraining forces: time 3560 (s).

Figure 11. Restraining forces and damage at time 3600 (s).

Figure 12. Restraining forces and damage near failure: time 5120 (s).

Figure 13. Restraining forces at lower right support.

Copyright ? 2000 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2000; 48:761–785
784 J. STABLER AND G. BAKER

stresses to develop. These stresses would be tensile near the top of the beam where the material
is forced to expand more than its free thermal strains would indicate, and compressive near the
bottom where the expansion would be less than the free thermal strains. Superimposed on this
is the e ect of the restraints, which cause a band of compressive stress to develop between the
internal two supports and some tensile horizontal stresses under the notch due to the curvature of
the deformed shape. The combination of these stresses explains the development of the damage
zone in the specimen.
During the initial stages the increase in temperature causes both the stresses and the restrain-
ing forces to increase. Immediately under the notch the stresses (and the corresponding strains)
will contain tensile components and will grow until the strains reach the threshold value for me-
chanical damage to occur. Damage gradually grows under the tip of the notch until the tensile
stresses around this zone become large enough to cause a sudden propagation of the damage zone.
Figure 10 shows the situation in the beam when the restraining forces are at a maximum. Figure
11 shows the situation a short time afterwards when the damage zone has propagated through
much of the cross-section. As the temperature increases further the damage zone will continue
to propagate slowly towards the lower right support which results in a redistribution of stresses
and the nal situation is show in Figure 12. A history of the restraining force at the lower right
support is given in Figure 13. The gure gives some indication of the speed at which the damage
zone propagates through the section and also suggest the possibility of a snap back phenomenon.

6. CONCLUSION

In this paper, we have focused on algorithmic details of the solution of non-linear strongly coupled
thermo-mechanical problems, our initial motivation being the transient analysis of concrete under
combined stress and elevated temperature. Speci cally, we have presented details of fractional step
methods developed for the time integration of this strongly coupled system.
With reference to operator split techniques, we nd that a thermal phase at constant con guration
often failed to converge under the increments of temperature relevant to our application. Hence,
we employ operator splits, but add one nal simultaneous step (after the fractional step) to bring
thermal and mechanical displacement increments to a consistent stage. We note for the non-
linear thermo-damage model, with the fractional step methods employed, that the isentropic split
is still unconditionally stable, whereas the isothermal split is only conditionally stable; the nal
simultaneous phase cannot improve that status. Nonetheless, for the examples considered here, we
found no real numerical instability with the method based on the isothermal split.
However, we have shown that the isentropic operator split is the natural choice, because of
the precise energy balance maintained throughout the iterative process. In numerical experiments
on model problems, and in a case study of thermal fracture of a restrained specimen, we found
complete stability and accuracy even with quite large time steps.

REFERENCES
1. Schneider U. Concrete at high temperatures—a general review. Fire Safety Journal 1988; 13:55 – 68.
2. Anderberg Y, Thelandersson S. Stress and Deformation of Concrete at High Temperatures. Part 2: Experimental
investigation and Material Behaviour, Bullentin 54. Lind Institute of Technology: Lund, Sweden, 1976.
3. Mazars J, Pijaudier-Cabot G. Continuum damage theory—application to concrete. Journal of Engineering Mechanics
1989; 115:345–365.

Copyright ? 2000 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2000; 48:761–785
FRACTIONAL STEP METHODS FOR THERMO-MECHANICAL DAMAGE 785

4. Amero F, Simo JC. A new unconditionally stable fractional step method for non-linear themomechanical problems.
International Journal for Numerical Methods in Engineering 1992; 35:737–766.
5. Amero F, Simo JC. A priori stability estimates and unconditionally stable product formula algorithms for nonlinear
coupled thermoplasticity. International Journal of Plasticity 1993; 9:749–782.
6. Argyris JH, Doltsinis J. On the natural formulation and analysis of large deformation coupled thermomechanical
problems. Computer Methods in Applied Mechanics and Engineering 1981; 25:195–253.
7. Park K, Felippa C. Partitioned analysis of coupled problems. Computational Methods in Transient Analysis. In:
Belytschko T, Hughes TJR. (eds). North-Holland: Amsterdam, 1983.
8. Schre er BA, Simoni L, Turska E. Standard staggered and staggered newton schemes in thermo-hydro-mechanical
problems. Computer Methods in Applied Mechanics and Engineering 1997; 114:93–109.
9. Lia WM, Rubin D, Krempl E. Introduction to Continuum Mechanics (3rd edn). Pergamon Press: Oxford, 1993.
10. Stabler J, Baker G. Spurious bifurcations in transient analysis with isotropic damage. Journal of Engineering Mechanics
1999; 125:1090–1093.
11. Pjaudier-Cabot G, Bazant Z. Nonlocal damage theory. Journal of Engineering Mechanics 1987; 113:1512–1533.
12. Saouridis C. Identi cation et Numerisation Objectives des Comportements Adoucissants: Une Approche Multiechelle
de L’Endommagement du Beton. Ph.D. Thesis, l’ Universite Pierre et Marie Curie—Paris 6, Laboratoire de Mecanique
et Technologie E.N.S. de Cachan—Universite Pierre et Marie Curie—C.N.R.S., 61, Avenue du President Wilson 94230
Cachan, France, 1988.
13. Pijaudier-Cabot G, Gerard B, Molez L. Damage mechanics of concrete structures subjected to mechanical and
environmental actions. In: di Borst R, Bicanic N, Mang H, Meschke G. (eds). Computational Modeling of Concrete
Structures. vol. 1 A.A. Balkema: Rotterdam, 1998; 567–576.
14. Mazars J, Pijaudier-Cabot G. From damage to fracture mechanics and conversly: a combined approach. International
Journal of Solids and Structures 1996; 33:3327–3342.
15. de Boarst R, Geers MGD, Kuhl E, Peerlings RHJ. Enhanced damage models for concrete fracture. In: Computational
Modelling of Concrete Structures, vol 1, de Boarst R, Bicanic N, Mang H, Meschke G. (eds). A.A. Balkema:
Rotterdam, 1998; 231–248.
16. Simo JC, Miehe C. Associative coupled thermoplasticity at nite strains: Formulation, numerical analysis and
implementation. Computer Methods in Applied Mechanics and Engineering 1992; 98:41–104

Copyright ? 2000 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2000; 48:761–785

You might also like