Sei sulla pagina 1di 15

doi: 10.1111/j.1460-2695.2007.01154.

Strain-life approach in thermo-mechanical fatigue evaluation


of complex structures
URO S ROSA, MARKO NAGODE and MATIJA FAJDIGA
University of Ljubljana, Faculty of Mechanical Engineering, Askerceva 6, SI-1000 Ljubljana, Slovenia
Received in final form 23 May 2007

A B S T R A C T This paper is a contribution to strain-life approach evaluation of thermo-mechanically

loaded structures. It takes into consideration the uncoupling of stress and damage evaluation and has the option of importing non-linear or linear stress results from finite element
analysis (FEA). The multiaxiality is considered with the signed von Mises method. In
the developed Damage Calculation Program (DCP) local temperature-stress-strain behaviour is modelled with an operator of the Prandtl type and damage is estimated by use
of the strain-life approach and Skeltons energy criterion. Material data were obtained
from standard isothermal strain-controlled low cycle fatigue (LCF) tests, with linear parameter interpolation or piecewise cubic Hermite interpolation being used to estimate
values at unmeasured temperature points. The model is shown with examples of constant
temperature loading and random force-temperature history. Additional research was done
regarding the temperature dependency of the K p used in the Neuber approximate formula for stress-strain estimation from linear FEA results. The proposed model enables
computationally fast thermo-mechanical fatigue (TMF) damage estimations for random
load and temperature histories.
Keywords damage accumulation; elastoplasticity; finite element analysis; thermomechanical fatigue.
N O M E N C L A T U R E Amin = minimal cross-section

b = the fatigue strength exponent


c = the ductility exponent
ck = load influence factor
D = damage
D = elasticity matrix
E = Youngs modulus

e = nominal strain
g = cyclic stress-strain curve
i = data point
j = spring-slider segment index
k = temperature index
K  = cyclic hardening coefficient
K p = limit load ratio
L = load
LP = full plastification force
LF = initial plastification force

Correspondence: U. Rosa, E-mail: uros.rosa@fs.uni-lj.si

808


c 2007 The Authors. Journal compilation 
c 2007 Blackwell Publishing Ltd. Fatigue Fract Engng Mater Struct 30, 808822

S T R A I N - L I F E A P P R OAC H I N T H E R M O - M E C H A N I C A L FAT I G U E E VA L UAT I O N

809

n = index of the top data point


N f = number of cycles to crack initiation
n = cyclic hardening exponent
P SWT = Smith-Watson-Topper parameter
R = load ratio
r j = fictive yield stress
Rm = tensile strength

= cyclically stable yield stress
Rp,0.2
S = nominal stress
t = time
T = temperature
j = the Prandtl density
r = fictive yield stress class width
T = temperature increment
W p = dissipated energy per cycle
 = cycle strain range
 = cycle stress range
= strain tensor
= total strain
a = strain amplitude
f = ductility coefficient
j = spring-slider strain
= stress tensor
= total stress
a = stress amplitude
e = pseudo elastic stress
f = fatigue strength coefficient
m = mean stress
SVM = signed von Mises stress
x , y , z = normal stress in X, Y and Z direction
j = spring stress
xy , xz , yz = shear stress in XY, XZ and YZ plane

INTRODUCTION

For machines and components under variable multiaxial


loading, fatigue evaluation is one of the most important
steps in the design process. Appropriate material testing
and simulation is the key to efficient life prediction.
There are several requirements for making accurate
life predictions: a proper material model with material
parameters derived from testing, modelling of stress
strain response during cycle loading, a multiaxial fatigue
criterion, a proper damage accumulation model and component testing to evaluate the correlation between life
prediction models and experiments.
In circumstances where service loads create complex
stressstrain fields, the first challenge is to properly model
the multiaxial fatigue. At present, there are several different approaches that can be divided in four major groups:
1 The approaches that calculate equivalent stress or strain13
in each node of the finite element model.

2 The use of stress or strain invariants.3


3 The critical plane approach (at first suggested by Brown
and Miller and later on modified by several researchers
(e.g. Socie, FatemiSocie and Papadopoulous,1,3,4 ) that
considers the critical plane in each selected node for life
estimation.
4 The energy criterion5,6 that takes into account the dissipated energy during cycle loading.

Besides the four categories above, approaches can be


divided according to their field of usage or analyses requirements. Historically, the first division was between
7

high-cycle (HCF) and low-cycle fatigue (LCF). Wohler


pioneered the field of HCF and the development of stresslife approach with S/N curves, and later Manson8 and Coffin9 made substantial developments in the field of LCF
with the establishment of the strain-life approach. This
classic differentiation is still present nowadays mainly due
to the tendencies for elastic deformation in HCF and inelastic deformation in LCF, as can be seen in the critical


c 2007 The Authors. Journal compilation 
c 2007 Blackwell Publishing Ltd. Fatigue Fract Engng Mater Struct 30, 808822

810

U. R O S A et al.

plane approach in Ref. [3] or energy criteria in Ref. [5]. An


important field of research concerns the development of a
unified approach to LCF and HCF, such as the energy criteria summarized in Ref. [5] or the unified approach based
on shakedown theories and dissipated energy proposed by
Constantinescu et al.10
Materials exhibit different failure mechanisms that are
commonly divided into tensile-type failure and shear-type
failure. This division can be easily made in the critical
plane approach, where there are tensile critical plane models, such as the tensile critical plane model of Socie,1,5
and shear critical plane models such as those of BrownMiller,1,3 Fatemi-Socie3 and Socie.3
Regarding the analyses requirements, both proportional and non-proportional loading situations have to
be considered, as inappropriate models can lead to nonconservative life estimations.1,11 In commercial software12,13 the signed von Mises stress method is frequently
used for modelling the effects of proportional loading as it
generally gives conservative results by additionally taking
into account the stresses in the directions that do not contribute to damage. For simulations associated with nonproportional loading the critical plane approach is widely
used,12 especially the normal or shear strain model, the
Socie tensile model and the WangBrown model. The
Socie tensile model, based on Smith-Watson-Topper,14
can give non-conservative results under non-proportional
loadings as shown in Ref. [1], whereas better agreement
with measurements has been obtained with the Fatemi
Socie approach. The WangBrown model for multiaxial cycle counting and damage accumulation according to
Ref. [3] and Ref. [15] gives good correlation with experiments for multiaxial non-proportional loading. However,
a general problem with critical plane methods is that the
computational process can be very time-consuming, requiring damage calculation in all planes of each selected
node.
Improving computation times is an important issue in
the development of these methods, as can be seen in Li
et al.2 and Amiable et al.16 Currently, calculations for
simple models can last as long as 1012 h or even few
days.2,16 This is why fast computation methods like those
proposed by Nagode and others1719 are under development. Choosing a method that gives the best compromise between computational speed and good correlation
with experiments is still an open question for the design
engineer.
The purpose of this paper is to extend the models already
developed by Nagode an others1719 to multiaxial states
for use in computationally fast thermo-mechanical fatigue
(TMF) life-predictions of complex 3D structures. The
method uses the strain or stress-controlled rheological
spring-slider model developed by Nagode and others that
enables elastoplastic material behaviour modelling with

an operator of the Prandtl type under constant or variable


temperature. It is based on isothermal strain-controlled
LCF tests with the assumption that hysteresis loops are
stabilized. The major drawback of the proposed model is
that it is basically uniaxial and that at the moment it is
capable of elastoplastic modelling with stabilized hysteresis loops without the consideration of cyclic creep, cyclic
hardening and cyclic softening.
Multiaxiality is considered with the calculation of signed
von Mises stress in every node and damage is calculated
with the Smith-Watson-Topper parameter14 and Miner
linear damage accumulation rule.20 Skeltons energy criterion6,21,22 is also applied as a damage indicator; it is
temperature independent and based on the stabilized cycle. This approach has been successfully used in complex
multiaxial TMF analyses as shown by Amiable et al.16 ,
Constantinescu et al.,23 Charklauk et al.24 and Thomas
et al.25

APPROACH TO TMF PROBLEMS

Uncoupled analyses
In TMF evaluations, the thermal and mechanical loadings
acting on the specified structure must be considered. In
order to obtain the temperature fields that are later applied together with the mechanical loads in the structural
analyses, an assumption must be made regarding the uncoupling of the thermal and structural analyses. Transient
thermal analyses are performed in order to obtain the desired temperature fields for all load cases. The computed
temperature fields are then applied in combination with
the mechanical loading in the stressstrain finite element
analysis (FEA) analyses.
The second uncoupling regards the separation of the
stressstrain response from the damage calculation. The
stressstrain response together with the temperature fields
are exported for the final damage evaluation into the developed Damage Calculation Program (DCP).
This non-unified approach has been widely and successfully used in TMF evaluations with the Skelton energy
criterion in the automotive industry.2325

Evaluation procedure
The proposed evaluation process is explained in detail below and includes the following important steps:
1 Calculation of the temperature fields for the load history
with transient thermal FEA.
2 Calculation of the stressstrain response for the turning
points in the given load history with FEA. The analysis
can be either linear or nonlinear.


c 2007 The Authors. Journal compilation 
c 2007 Blackwell Publishing Ltd. Fatigue Fract Engng Mater Struct 30, 808822

S T R A I N - L I F E A P P R OAC H I N T H E R M O - M E C H A N I C A L FAT I G U E E VA L UAT I O N

811

Supposing that cyclically stable cyclic stressstrain curves


are available, elastoplastic strain can be modelled with an
operator of the Prandtl type that connects several elementary hysteresis operators and it is used for modelling real
hysteresis phenomena.29
The proposed stressstrain modelling has been presented thoroughly by Nagode and others 18 , which is
why only the final set of equations is listed below.
The total elastoplastic strain can be expressed in a form
known as the operator of the Prandtl type18,29
(ti ) =

nr


j (Ti )j (ti )

(2)

j =0

Fig. 1 Rheological spring-slider model.

3 From the FEA the stress and temperature histories for all
nodes are exported to the specifically developed DCP.
4 If the stresses have been computed with the linear FEA
the DCP estimates the elastoplastic stresses in the turning
points with the Neuber approximate formula.
5 In the DCP a stressstrain modelling of complete hysteresis loops is performed using the stress-controlled springslider model.18
6 After the stressstrain modelling the rainflow cycle counting is performed, by utilizing the cycle closure method
used in Ref. [18] and the calculation of the equivalent cycle temperature.17,26
7 Damage estimation using the Smith-Watson-Topper parameter14 and the estimation of the dissipated energy is
performed.

Stress-controlled modelling of elastoplasticity


The stressstrain modelling in DCP is based on the stresscontrolled serially connected spring-slider model18,27
shown in Fig. 1. It is capable of modelling elastoplastic
hardening solids and nonlinear kinematic hardening for
isothermal or non-isothermal cases.
If isotropic hardening can be neglected, cyclically stable
(or half-life) cyclic stressstrain curves (g) with RambergOsgood18,28 relation

1/n (T)

(1)
+
= g(, T) =
E(T)
K  (T)
are not necessary but are commonly used in fatigue analyses. T, E(T), K  (T) and n (T) are temperature, the Young
modulus, the cyclic-hardening coefficient and the cyclichardening exponent, respectively. There is no limitation
however against using any other stressstrain relation that
exhibits elastoplastic behaviour with nonlinear kinematic
hardening.

for 0 t1 t2 ti , where T i = T(ti ) and


j (ti ) is the play operator with general initial value

j (ti ) = max (ti ) rj ,


j (Ti1 )
min (ti ) + rj ,
j (ti1 ) .
j (T1 )

(3)

Presumably, there is no residual stress initially, so j (0) =


0 and (0) = 0. The Prandtl densities j (T k ) in range j =
0, . . . , nr and k = 0, . . . , nT
j (Tk ) =

1
(j+1 (Tk ) 2j (Tk ) + j1 (Tk ))
r

(4)

are gained from the available isothermal cyclically stable cyclic stressstrain curves, where 1 (Tk ) = 0 (Tk ) = 0.
Fictive yield stresses r j are usually dispersed equidistantly with constant fictive yield stress class width r
between the zero stress and the maximal expected stress
(Fig. 2).
To speed up the computation, input histories of (t), T(t),
material parameters and the Prandtl densities are tabulated by setting the stress increment to r and choosing
a temperature increment of T. The tabulated material
parameters and the Prandtl densities are calculated only
once and stored before the , modelling process beginning at Eq. 2.
Stressstrain modelling
From the FEA results the signed von Mises stress12,13,30 is
calculated using Eq. 5 and exported to the DCP where the
stressstrain modelling with the stress-controlled model
is performed. The symbol SGN in Eq. 5 equals the sign
of the principal stress with the largest magnitude.
1 
SVM (t) = SGN (x y )2 + (x z )2
2
 2

2
2 1/2
+ (y z )2 + 6 xy
+ xz
+ yz
.

(5)

Stresses can be computed with nonlinear or linear FEA.


In the case of nonlinearly computed stresses, the same


c 2007 The Authors. Journal compilation 
c 2007 Blackwell Publishing Ltd. Fatigue Fract Engng Mater Struct 30, 808822

U. R O S A et al.

812

where Amin is the minimal cross-section of the specimen



the yield stress
where the plastification begins and Rp,0.2
for the cyclically stable cyclic stressstrain curve.
According to LMS Falancs Theory Manual Version 2.912
the useable range of K p is between: K p = 1 and K p = 30.
The first value does not take into account the Neuber approximate formula, which means that the input stresses are
directly used for the stressstrain modelling. This value is
suitable when the input stresses are calculated with nonlinear FEA or when extremely conservative predictions are
required with the linearly calculated stresses. The second
value is used only in combination with linearly calculated
stresses when there is no influence from the plastic limit
load.12
Due to temperature dependant mechanical characteristics of the material, some temperature dependency of LP
and LF values has also been expected. In an attempt to
obtain better agreement between the results from linear
and nonlinear FEA, a temperature dependant K p has been
introduced
K p (T) =

Fig. 2 Prandtl density assessment, stress control.

material model must be applied in both the FEA software


and in the DCP program in order to obtain consistent results. At the moment, due to the limitations of the springslider model elastoplastic kinemetic hardening material
models are allowed.
In order to achieve a computationally fast evaluation
method, great care has been given to the linearly computed stresses in combination with approximate formulae
to estimate the elastoplastic material behaviour. There are
several approximate methods from which the frequently
used12,18,31 Neuber approximate formula is applied,


E(T)

2

e
S /E(T)

(6)

where the nominal stress and strains are defined as S =


e /K p and e = g(S , T). K p is the limit load ratio and g(S ,
T) is the cyclic stressstrain curve as defined in Eq. 1. The
limit load ratio gives the ratio between the limiting load
and the load when yielding starts. Its value is calculated
by,12
Kp =

LP
,
LF

(7)

where LP is the force that causes the full plastification of


the analysed cross-section and LF is the force that starts
the plastification in the most stressed node. In our case LP
is computed as,

LP = Amin Rp,0.2
,

(8)

LP (T)
.
LF (T)

(9)

This has been calculated for the maximal surface temperature for the constant temperature field cases shown in
the Examples chapter.
The value K p = 1 has been used for the evaluation of
the nonlinear analyses results and K p = 30, K p = 1 and
the calculated K p (T) for the evaluation of linear analyses
results.
At the moment, the proposed temperature dependency
of K p (T) can be applied with constant temperatures only.
Damage estimation
An appropriate damage estimation method provides the
key to an efficient life-prediction. In the proposed method,
the focus is on employing computationally fast and widely
accepted approaches. This is the reason why the wellknown Smith-Watson-Topper damage parameter1,11,14,32
has been applied as the first damage estimation method.
It includes the influence of the stress and strain amplitude
and the mean stress on the damage and is in accordance
with physical tests in a variety of cases.11,12 Since the temperature during an individual cycle can vary, the equivalent
cycle temperature17,26 T e , as first proposed by Taira,26 and
the cycle closure method, as described by Nagode and others 17,18 , have been used. The cycle closure problem can
be observed in non-isothermal cases as can be seen for two
simulated conditions in Fig. 3 with temperature increase
and temperature decrease during one cycle.
That is why a cycle closure method was introduced by
Nagode and others.18 Given that it is time consuming
to estimate the true cycle closure point, a simple and


c 2007 The Authors. Journal compilation 
c 2007 Blackwell Publishing Ltd. Fatigue Fract Engng Mater Struct 30, 808822

S T R A I N - L I F E A P P R OAC H I N T H E R M O - M E C H A N I C A L FAT I G U E E VA L UAT I O N

813

part operates for the majority of its lifetime, and that the
cumulated dissipated energy at stabilization can be considered as a material constant used as a crack initiation
criterion.6,2325
In accordance with this, it has been considered that the
load history is applied in the stabilized state and the dissipated energy has been used for the identification of the
most critical area in the evaluated structure. At the moment in the DCP, the simplified equation16
Wp  p

(13)

is used for calculating the dissipated energy per counted


cycle. Through the calculation of the load history in the
stabilized state it is linearly accumulated

Wp,i
(14)
Wp =
cycles

and the value W p used for the critical area identification.


Both implemented damage assessment methods are
suitable for any combination of load and temperature
histories.
Fig. 3 Left: cycle closure and temperature increase. Right: cycle
closure and temperature decrease.

EXAMPLES DEFINITIONS

conservative cycle definition has been applied as18


|from to |
|from to |
from + to
a =
, m =
, a =
,
2
2
2
(10)
where from denotes the starting point of the cycle and
to its reversal point.
For each closed hysteresis loop, the Smith-WatsonTopper damage parameter is calculated1,11,14,32
2
PSWT
= (a + m ) a E(Te )

(11)

and the number of cycles to crack initiation is obtained


from1,11,14,32
2
PSWT
=  2f (Te )(2Nf )2b(Te )

+ f (Te )E(Te )f (Te )(2Nf )b(Te )+c(Te ) ,

(12)

where f is the fatigue strength coefficient, b the fatigue


strength exponent, f the ductility coefficient and c the
ductility exponent. The material parameters depend on
the equivalent cycle temperature and can be linearly or
piecewise cubic Hermite33 interpolated. The latter is very
efficient when there is information on the function values
and the first derivatives values at a set of data points. The
number of cycles to crack initiation is used for damage
estimation with the Miner linear damage accumulation
rule.
The Skelton energy criterion6,21,22 is gaining considerable importance in TMF evaluations.16,2325 It is based on
the assumption that after the initial hardening or softening the material reaches the stabilized state, in which the

The aim of the following section is to show that the abovedescribed procedure can be successfully used in TMF
analyses of complex 3D structures. To enable DCP validation, linear and nonlinear FEA with the renowned ANSYS software have been performed and compared to the
results gained by DCP. The stressstrain fields obtained
from the nonlinear FEA have been used as reference to
evaluate the agreement between the material model implemented in the DCP and the material model used in
ANSYS.
In order to facilitate the reduction of computational
times, great care has been put in to the linear FEA in
combination with the Neuber approximate formula. The
influence of K p upon stressstrain trajectory, damage and
dissipated energy has also been studied.
Specimen and material
Figure 4 shows the modified 5 mm thick ASTM International34 specimen with an additional hole of diameter
3 mm; boundary conditions and applied loads are also
shown. The specimen was shortened for the FEA evaluation in the fixation-gripping area to reduce the size of
the model.
The specimen is a commonly used flat specimen for standard LCF tests and is also suitable for tests at elevated
temperatures. A hole was introduced in the centre of the
sample in order to explore the stresses and strains around
such a weak point in the structure. With such a specimen, a
uniaxial test can be used to simulate complex stressstrain


c 2007 The Authors. Journal compilation 
c 2007 Blackwell Publishing Ltd. Fatigue Fract Engng Mater Struct 30, 808822

814

U. R O S A et al.

stress-strain curve. It represents the material behaviour


with a combination of various portions (or subvolumes),
all subjected to the same total strain, but each having a
different yield stress. Each subvolume has a simple stressstrain response but when these are combined, the model
can represent kinematic hardening.13,36 The input points
for the material definition in nonlinear FEA are shown
in Fig. 5. For the damage estimation, temperature dependent P SWT curves have been calculated from the material
parameters in Table 1.

Loading
In fatigue evaluations with linear FEA and multiple load
channels, the load superposition is used.11,12 For each applied load direction, a unity load Lk is applied separately on
the structure and the tensor for the load influence factor
ck is calculated. This is later used for the stress calculation
concerning a specific time and location11,12

Fig. 4 Modified ASTM34 specimen with boundary conditions and


applied loads.

states and observe the stress-concentration phenomena


and easily predict the highly damaged areas. All tests were
carried out using the same finite element model, consisting of 7700 linear cubic and prismatic elements. The mesh
was refined around the hole. All analyses were performed
with a 2.8 GHz Pentium IV, 4 GB RAM, Windows XP
based workstation.
For validation of the proposed approach, the following
material was chosen: Steel 10 CrMo 9 10 (heat treatment
930 C/1.5 h air, 710 C/1.5 h air, 680 C furnace) with
RambergOsgood and Manson-Coffin-Morrow parameters (Table 1) obtained by Boller and others35 for 23, 300,
400, 500 and 600 C.
For the nonlinear analyses in the ANSYS program,
the linearization of the RamberOsgood curve has been
made with the multi-linear kinematic hardening Besseling
model,13,36 which cannot exceed 20 input points per one

(t) =

n


ck Lk (t).

(15)

k=1

Considering the basic equation for stress calculation in


linear FEA37
= D( 0 ) + 0 ; D = f (T),

(16)

where the elasticity matrix D is temperature dependant,


and supposing that no initial stresses 0 are present, Eq. 15
and 16 can be rewritten for force with magnitude |L1 | =
L1 = 1 in Eq. 17.
1 = D(1 0 ) = c1 L1 c1 ,

(17)

where the quantities related to the force are the stress


tensor 1 , the mechanical strain tensor 1 and the initial
strain tensor 0 , which includes the influence of the temperature change. Taking, for example, a second force with
the same direction as L1 and of different magnitude
L1 = k L1

(18)

with the same temperature change as in case of Eq. 17.


Rewriting Eqs. 15 and 16 presents a correct formulation

Table 1 Temperature dependant material parameters35


T ( C)

E(T) (MPa)

K  (T) (MPa)

n (T) ()

f (T) (MPa)

f (T) ()

b(T) ()

c(T) ()

Rp,0.2 (T) (MPa)

23
300
400
500
600

210 000
204 100
187 800
184 800
162 000

842
691
681
497
327

0.118
0.102
0.102
0.077
0.057

736
675
638
473
316

0.266
0.782
0.424
0.320
0.576

0.065
0.064
0.066
0.051
0.038

0.0527
0.6280
0.6170
0.6010
0.6810

405
366
362
308
230


c 2007 The Authors. Journal compilation 
c 2007 Blackwell Publishing Ltd. Fatigue Fract Engng Mater Struct 30, 808822

S T R A I N - L I F E A P P R OAC H I N T H E R M O - M E C H A N I C A L FAT I G U E E VA L UAT I O N

Fig. 5 Isothermal cyclically stable cyclic stressstrain curves; with


points used for the multi-linear curve in ANSYS program. Material
10 CrMo 9 10, heat treatment 930 C/1.5 h air, 710 C/1.5 h air,
680 C furnace.35

on the left side and the stresses formulated with Eq. 15 on


the right side.
1 = D(1 0 ) = D(1 0 ) kL1 = c1 kL1 .

(19)

From Eq. 19 it follows that, in general, load superposition cannot be used in thermo-mechanical situations due
to the simultaneous interaction of mechanical and thermal loads. However, load superposition can be used only
with the linear FEA if there are no initial stresses, and if
mechanical and thermal strain change from one time step
to another for the same factor. It is also not possible to
superimpose stresses and strains obtained from separate
force applications with stresses and strains obtained from
separate thermal loads. This is due to different elasticity
matrices, which in the first case is not temperature influenced D = f (T), but is indeed temperature influenced in
the second case D = f (T). The superposition is therefore only possible for linearly computed pure mechanical
stresses and strains that are not influenced with temperature changes.
The later explanation leads to the conclusion that FEA
analyses have to be carried out for each change in temperature and force throughout the load history.
In the paper, two types of validation tests are shown.
At first, this is done with constant temperature fields for
one cycle at the following maximal surface temperatures
on the heated region 20, 300, 400, 500 and 600 C (see
Fig. 4). In these cases, analyses have been performed for

815

Fig. 6 Force and temperature histories with hold times; random


loading.

one reversed cycle with R = 1, firstly with force amplitude L1 = 12 kN and secondly with force amplitude
L2 = 25 kN. For all temperatures at the lowest force L1 , the

. Yieldgross stresses never exceed cyclic yield stress Rp,0.2
ing is present only in the net section with stress peaks near
the hole. The second load L2 leads to larger yield zones. It
should be emphasized that the heating process produces
different temperature distributions in the cross-sections
causing regions with higher temperature to yield first.
Secondly, the random force and temperature history
(Fig. 6) has been applied with linear force and temperature changes between turning points. This is a synthetic
load history with a combination of highlow heating rates
and hold times. Time dependent phenomena like heating
rates and strain rates do not influence elastoplastic material models at all. The temperature history is related to the
highest surface temperature on the specimen. The exact
applied force-temperature combination has high heating
rates and its exact path is rarely found in any industrial
application, but is employed due to its complexity and
ability to demonstrate the broad applicability of the proposed model. The turning points define the number of
structural analyses required, which amounts to a total of
22 analyses for the 30 s history. The analysed force and
temperature loading combinations are shown with circle
points in Fig. 6.
R E S U LT S

Damage and dissipated energy have been used to compare


fatigue evaluations of nonlinear FEA and those of linear


c 2007 The Authors. Journal compilation 
c 2007 Blackwell Publishing Ltd. Fatigue Fract Engng Mater Struct 30, 808822

816

U. R O S A et al.

Fig. 8 Stressstrain trajectories at node 7266; 20 C, L1 = 12 kN.


Fig. 7 Meshed specimen showing nodes 2288 and 7266.

FEA for different K p -s. Due to different interactions between mechanical and thermal strains, the location of the
most damaged node can move if the temperature field
changes even if the mechanical force remains unchanged.
For validation purposes, two nodes were selected (see
Fig. 7). The most damaged node in the linear analysis at
20 C (node number 2288) and the node with the average
damage value in the highly damaged area (node number
7266). Node 2288 was also used for stressstrain trajectory
validation for the most complex random loading. More
detailed analyses were made at 20 C in order to eliminate
errors caused by interpolation of temperature-dependant
material parameters.
A complete set of damage and dissipated energy values for L1 = 12 kN and random loading are given in
Appendix A. The values for the second load level L2 =
25 kN and constant temperatures are tabulated in
Appendix B.
Stressstrain trajectories
Stressstrain trajectories were obtained directly from the
exported signed von Mises stresses and the signed total
mechanical von Mises strains as provided by the nonlinear solutions in the ANSYS software. Stressstrain paths
obtained by DCP from nonlinear FEA stress results, as
well as those from linear FEA stresses for two distinct
K p -s, are also presented. In the figures below, the following notation is used
1 Stresses and strains that were gained from nonlinear FEA.

2 Stresses in turning points were taken from nonlinear FEA,


and these values were then processed by DCP to produce
complete stressstrain trajectories for K p = 1.
3 Stresses in turning points were taken from linear FEA,
and these values were then processed by DCP to produce
complete stressstrain trajectories for K p = 30.
4 Stresses in turning points were taken from linear FEA,
and these values were then processed by DCP to produce
complete stressstrain trajectories for K p (T) (see Eq. 9).

Stressstrain trajectories for L1 and L2 are given in


Figs 8 and 9, respectively.
The stressstrain trajectories and the temperature history at node 2288 for random loading are shown in
Fig. 10.
Nonlinear FEA
By comparing the Besseling material model13 built in ANSYS with the DCP material model (see Figs 810) it can be
seen that good agreement has been reached at the turning
points. However, differences have arisen for the following reasons: Firstly, the trajectories drawn from the stress
and strain values exported from ANSYS are influenced
by the linearization of the RambergOsgood curves. Secondly, in the FEA software, the equivalent stresses and
strains are computed from the complete stress and strain
tensors, whereas in DCP the strain is calculated with the
proposed uniaxial model from the signed equivalent von
Mises stresses imported from ANSYS.


c 2007 The Authors. Journal compilation 
c 2007 Blackwell Publishing Ltd. Fatigue Fract Engng Mater Struct 30, 808822

S T R A I N - L I F E A P P R OAC H I N T H E R M O - M E C H A N I C A L FAT I G U E E VA L UAT I O N

817

curves. More pronounced differences were observed for


L2 . In this case, (Fig. 9) the calculated K p (T) resulted in
better agreement with the nonlinear FEA as K p = 30. It
can be concluded that at larger yielding, the influence of
K p becomes significant and may lead to non-conservative
results in subsequent damage estimations for high K p .
For random loading, stresses were overestimated at certain turning points, as shown in Fig. 10. Strain history C
obtained from linear FEA for K p = 30 is quite similar to A
from nonlinear FEA, except for stresses which exceeded
the yield stress. This is in agreement with previous observations for constant temperature and L2 .
Damage and dissipated energy

Fig. 9 Stressstrain trajectories at node 7266; 20 C, L2 = 25 kN.

The previous section provides the basis for damage and


dissipated energy estimation. It is assumed that the damage and dissipated energy estimated with DCP based on
nonlinear FEA are the best and considered as reference
values in relation to those obtained from the linear FEA.
Visual comparison of the damage and dissipated energy
distributions for random loading and constant temperature loading at T max = 500 C and L1 = 12 kN is given in
Figs 11 and 12. Both linear and nonlinear FEA resulted
in the same highly damaged areas, with more pronounced
damage around the hole being observed for the nonlinear
FEA.
The damage distributions for the random loading
(Figs 13 and 14) show two critical areas with a wider critical region near the specimens constraint for the linear

Fig. 10 Temperature, stress and strain history at node 2288;


random loading.

Linear FEA
By comparing stressstrain trajectories first computed
with the linear FEA and then transformed with the
Neuber approximate formula, the influence of K p upon
nonlinear material behaviour can be studied. For L1
(Fig. 8), where negligible yielding has occurred, K p = 30
and K p (T) gained from Eq. 9 result in nearly the same

Fig. 11 Damage distribution for constant temperature; T max =


500 C, L1 = 12 kN, linear FEA, K p = 30.


c 2007 The Authors. Journal compilation 
c 2007 Blackwell Publishing Ltd. Fatigue Fract Engng Mater Struct 30, 808822

818

U. R O S A et al.

Fig. 12 Damage distribution for constant temperature; T max =


500 C, L1 = 12 kN, nonlinear FEA, K p = 1.

Fig. 14 Damage distribution for random loading; nonlinear FEA,


K p = 1.

Fig. 13 Damage distribution for random loading; linear FEA,


K p = 30.

Fig. 15 Dissipated energy distribution for random loading; linear


FEA, K p = 30.

FEA. The dissipated energy distribution is also used as an


indicator of the most critical areas. As shown in Figs 15
and 16, this approach gives the same critical regions as
those based on damage estimation.
By comparing the values given in Appendices A and B,
it can be seen that for both load levels, an increase in the
damage and in the dissipated energy occurred at higher

temperatures. This happens both globally and also at node


2288, which is in accordance with the temperature dependent P SWT and RambergOsgood parameters given in
Table 1. For the random loading, the maximal global damage and dissipated energy gained from linear FEA and
K p = 30 are 4.45 times and 3.25 times higher as compared to the nonlinear FEA, respectively.


c 2007 The Authors. Journal compilation 
c 2007 Blackwell Publishing Ltd. Fatigue Fract Engng Mater Struct 30, 808822

S T R A I N - L I F E A P P R OAC H I N T H E R M O - M E C H A N I C A L FAT I G U E E VA L UAT I O N

Fig. 16 Dissipated energy distribution for random loading;


nonlinear FEA, K p = 1.

819

Fig. 18 Comparison between estimated dissipated energies based


on linear and nonlinear FEA.

observed, whereas conservative damages were estimated


for K p (T).
The influence of K p at node 2288 is less pronounced
in the case of dissipated energy estimation, as shown in
Fig. 18. Except for L2 and T max = 600 C, all other values
are conservative and within five times the best fit line.
The findings shown in Figs 17 and 18 are in accordance
with those described in the section Stressstrain trajectories. The damages are more scattered than the dissipated
energies. This is due to the fact that small errors in strain
estimation lead to larger errors in damage estimations especially for small values of P SWT .
Computational time

Fig. 17 Comparison between estimated damages based on linear


and nonlinear FEA.

The comparison of the calculated damages for the node


number 2288 is given in Fig. 17. It can be seen that the
majority of points are within the zone of five times the
damage value given in dashed lines. For L1 , both K p -s
give similar damage scores that are all on the conservative
side. For L2 and K p = 30 non-conservative results were

In fatigue evaluation, the computational time is also of key


importance in assessing the utility of the method. Computation time was measured for one reversed cycle with
R = 1, L2 = 25 kN and T max = 600 C, and also for
the random loading shown in Fig. 6. For the reversed cycle, the linear FEA took 46 s, whereas the nonlinear FEA
lasted 140 s. The damage estimation took approximately
7 s for linear and nonlinear FEA, as the input data for the
DCP is the same for both cases. For the random loading, the linear FEA took 200 s to run, whilst the nonlinear
took 2260 s. The corresponding damage estimation lasted
approximately 20 s. It should be noted that in the first
case, the loading history had three turning points and in
the second case there were 22, but the damage estimation
time is higher only by a factor of 2.8. This is due to data


c 2007 The Authors. Journal compilation 
c 2007 Blackwell Publishing Ltd. Fatigue Fract Engng Mater Struct 30, 808822

820

U. R O S A et al.

preprocessing in the DCP that does not depend on the


number of turning points.
CONCLUSIONS

The current paper is an extension of three previous papers by Nagode and others1719 and deals with the stresscontrolled strain-life approach for the damage and dissipated energy estimation of arbitrary thermo-mechanically
loaded 3D structures. It is a non-unified approach incorporating the uncoupling of transient thermal calculations,
stressstrain calculations and damage or dissipated energy
estimations. The method uses material data obtained from
standard strain-controlled LCF tests carried out at distinct but constant temperatures. The multiaxiality is considered with the signed von Mises equivalent stress that is
exported to the DCP, allowing computationally fast material behaviour modelling and damage or dissipated energy
estimation. The main focus has been put on the application of the proposed method in combination with stresses
computed using linear FEA. To improve the agreement
between the nonlinear FEA and those from linear FEA
in combination with the Neuber approximate formula,
a temperature dependant K p (T) has been introduced. It
has been established that K p has significant impact upon
the estimated damage and the dissipated energy especially
at larger yielding. This may lead to non-conservative estimates for high K p . However, if K p approaches unity,
conservative estimates can be expected. Linear FEA in
combination with DCP is appropriate if yielding is not
pronounced. This will be further investigated on the cold
part of the exhaust system. Linear FEA and DCP can only
be used if K p is properly selected.

REFERENCES
1 Chen, X., Xu, S. and Huang, D. (1999) A critical plane-strain
energy density criterion for multiaxial low-cycle fatigue under
non-proportional loading. Fatigue Fract. Engng Mater. Struct.
22, 679686.
2 Li, B., Reis, L. and de Freitas, M. (2006) Simulation of cyclic
stress/strain evolutions for multiaxial fatigue life prediction. Int.
J. Fatigue 28, 451458.
3 You, B. R. and Lee, S. B. (1996) A critical review on multiaxial
fatigue assessments of metals. Int. J. Fatigue 18, 235244.
4 Papadopoulos, I. V., Davoli, P., Gorla, C., Filippini, M. and
Bernasconi, A. (1997) A comparative study of multiaxial
high-cycle fatigue for metals. Int. J. Fatigue 19, 219235.
5 Macha, E. and Sonsino, C. M. (1999) Energy criteria of
multiaxial fatigue failure. Fatigue Fract. Engng Mater. Struct. 22,
10531070.
6 Skelton, R. P. (1991) Energy criterion for high temperature low
cycle fatigue failure. Mater. Sci. Tech. 7, 42739.

7 Wohler,
A. (1870) Uber
die Festigkeits-Versuche mit Eisen und
Stahl. Zeitschrift fur
Bauwesen XX, 73106.

8 Manson, S. S. (1965) Fatigue: A complex subject some simple


approximations. Experimental Mechanics 5, 193226.
9 Coffin, L. F. (1954) A study of the effects of cyclic thermal
stresses on a ductile metal. Transactions of ASME 76, 931
950.
10 Constantinescu, A., Dang Van, K. and Maitournam, M. H.
(2003) A unified approach for high and low cycle fatigue based
on shakedown concepts. Fatigue Fract. Engng Mater. Struct. 26,
561568.
11 Kocabicak, U. and Firat, M. (2004) A simple approach for
multiaxial fatigue damage prediction based on FEM
post-processing. Materials and Design 25, 7382.
12 LMS Falancs Theory Manual Version 2.9 (2000), LMS Durability
Technologies GmbH.
13 ANSYS Inc. (2005) ANSYS Relase 10.0 Documentation. ANSYS
Inc.
14 Smith, K. N., Watson, P. and Topper, T. H. (1970) A
stressstrain function for the fatigue of metals. J. Mater. 5,
76778.
15 Heyes, P., Lin, X., Buczynski, A. and Brown, M. W. (1999)
Application of biaxial plasticity and damage modeling to the life
prediction and testing of automotive components, in: Multiaxial
fatigue and fracture, International Conference Biaxial/Multiaxial
Fatigue and Fracture (Edited by E. Macha, W. Bedkowski and T.
Lagoda). Elsevier, Amsterdam, 179194.
16 Amiable, S., Chapuliot, S., Constantinescu, A. and Fissolo, A.
(2006) A comparison of lifetime prediction methods for a
thermal fatigue experiment. Int. J. Fatigue 28, 692706.
17 Nagode, M. and Hack, M. (2004) An online algorithm for
temperature influenced fatigue life estimation: stress-life
approach. Int. J. Fatigue 26, 163171.
18 Nagode, M. and Zingsheim, F. (2004) An online algorithm for
temperature influenced fatigue life estimation: strain-life
approach. Int. J. Fatigue 26, 155161.
19 Nagode, M. and Fajdiga, M. (2006) Temperaturestressstrain
trajectory modeling during thermo-mechanical fatigue. Fatigue
Fract. Engng Mater. Struct. 29, 175182.
20 Miner, M. A. (1945) Cumulative damage in fatigue. J. Appl.
Mech. 12, A159A164.
21 Skelton, R. P., Vilhelmsen, T. and Webster, G. A. (1998) Energy
criteria and cumulative damage during fatigue crack growth.
Int. J. Fatigue 20, 641649.
22 Skelton, R. P. (2004) Hysteresis, yield and energy dissipation
during thermo-mechanical fatigue of a ferritic steel. Int. J.
Fatigue 26, 253264.
23 Constantinescu, A., Charkaluk, E., Lederer, G. and Verger, L.
(2004) A computational approach to thermomecahnical fatigue.
Int. J. Fatigue 26, 805818.
24 Charklauk, E., Bignonnet, A., Constantinescu, A. and Dang
Van, K. (2002) Fatigue design of structures under
thermomechanical loadings. Fatigue Fract. Engng Mater. Struc.
25, 11991206.
25 Thomas, J. J., Verger, L., Bignonnet, A. and Charkaluk, E.
(2003) Thermomechanical design in the automotive industry.
Fatigue Fract. Engng Mater. Struct. 27, 887895.
26 Taira, S. (1973) Relationship between thermal fatigue and
low-cycle fatigue at elevated temperatures, in: Fatigue at elevated
temperatures (Edited by A. E. Carden, A. J. McEvily and C. H.
Wells). ASTM STP, 80101.
27 Conle, A., Oxland, T. R. and Topper, T. H. (1988)
Computer-based prediction of cyclic deformation and fatigue


c 2007 The Authors. Journal compilation 
c 2007 Blackwell Publishing Ltd. Fatigue Fract Engng Mater Struct 30, 808822

S T R A I N - L I F E A P P R OAC H I N T H E R M O - M E C H A N I C A L FAT I G U E E VA L UAT I O N

28

29

30

31

32

behaviour, in Low cycle fatigue (Edited by H. D. Soloman, G. R.


Halford, L. R. Kaisand and B. N. Leis). ASTM STP 942,
12181236.
Ramberg, W. and Osgood, W. R. (1943) Description of
stress-strain curves by three parameters. Technical note no. 902,
NACA.
Brokate, M. and Sprekels, J. (1996) Hysteresis and phase
transitions. Applied Mathematical Sciences, 121. Springer
Verlag, New York, USA.
Chen, H. F. and Ponter, A. R. S. (2004) Integrity assessment for
a tubeplate using the linear matching method. Int. J. Press Vessels
Piping 81, 327336.
Neuber, H. (1961) Theory of stress concentration for shear
strained prismatic bodies with arbitrary stress-strain law. J.
Appl. Mechan. 28, 544550.
Riedler, M., Leitner, H., Prillhofer, B., Winter, G. and
Eichlseder, W. (2007). Lifetime simulation of
thermo-mechanically loaded components. Meccanica 42, 4759

821

33 Moler, C. B. (2004) Numerical computing with MATLAB.


Electronic edition: The MathWorks, Inc., Natick, MA,
2004.
34 ASTM E 606-92 (Reapproved 1998) (1992) Standard Practice for
Strain-Controlled Fatigue Testing. ASTM International.
35 Boller, C. and Seeger, T. (1987) Materials data for cyclic loading
Part B: low-alloy steels. Elsevier Science Publishers B.V.,
Amsterdam, 238294.
36 Besseling, J. F. (1958) A theory of elastic, plastic, and creep
deformations of an initially isotropic material showing
aisotropic strain-hardening creep recovery and secondary creep.
J. Appl. Mechan. 25, 529536
37 Zienkiewicz, O. C. and Taylor, R. L. (2000) The Finite Element
Method, Fifth edition, Volume 1: The Basis. ButterworthHeinemann, Oxford, UK.
38 Lee, K. O., Hong, S. G., Yoon, S. and Lee, S. B. (2005) A new
high temperature life correlation model for austenitic and
ferritic stainless steels. Int. J. Fatigue 27, 15591563.

A P P E N D I X A : R E S U LT S F O R L 1 A N D R A N D O M L O A D I N G

Table A1 Test results for L1 = 12 kN and random loading

Constant temperature field,


T max = 20 C L1 = 12 kN

Constant temperature field,


T max = 300 C L1 = 12 kN

Constant temperature field,


T max = 400 C L1 = 12 kN

Constant temperature field,


T max = 500 C L1 = 12 kN

Constant temperature field,


T max = 600 C L1 = 12 kN

Random loading

FEA

K p ()

DN2288 ()

W p(N2288) (105 J/mm3 )

Dmax ()

W p,max (105 J/mm3 )

Linear

30

9.75105

2.17

9.75105

2.17

Linear
Linear
Nonlinear
Linear

1
2.16
1
30

8.79103
1.04104
2.68105
9.66105

3.16101
2.25
0.99
2.51

8.79103
1.04104
2.68105
1.07104

3.16101
2.25
0.99
2.60

Linear
Linear
Nonlinear
Linear

1
1.97
1
30

1.51102
1.05104
3.17105
1.31104

8.48101
2.74
1.19
2.66

1.99102
1.22104
4.29105
1.31104

8.98101
2.86
1.41
2.66

Linear
Linear
nonlinear
Linear

1
1.95
1
30

2.51102
1.40104
3.84105
2.47104

1.17102
2.94
1.19
2.70

2.74102
1.40104
4.86105
2.47104

1.17102
2.94
1.44
2.91

Linear
Linear
Nonlinear
Linear

1
2.06
1
30

4.09102
2.49104
1.07104
5.05104

1.10102
2.86
1.59
3.42

6.79102
2.70104
1.07104
5.05104

1.44102
3.12
1.59
3.42

Linear
Linear
Nonlinear
Linear
Linear
Nonlinear

1
30
1
30
1
1

1.52
5.05104
8.53105
9.33104
2.14
1.07103

8.88102
3.42
1.11
1.40101
1.81104
1.61101

1.52
5.05104
4.26104
4.78103
12
1.07103

8.88102
3.42
2.65
5.33101
7.30105
1.64101


c 2007 The Authors. Journal compilation 
c 2007 Blackwell Publishing Ltd. Fatigue Fract Engng Mater Struct 30, 808822

822

U. R O S A et al.

A P P E N D I X B : R E S U LT S F O R L2

Table B1 Test results for L2 = 25 kN

Constant temperature field,


T max = 20 C L2 = 25 kN

Constant temperature field,


T max = 300 C L2 = 25 kN

Constant temperature field,


T max = 400 C L2 = 25 kN

Constant temperature field,


T max = 500 C L2 = 25 kN

Constant temperature field,


T max = 600 C L2 = 25 kN

FEA

K p ()

DN2288 ()

W p(N2288) (105 J/mm3 )

Dmax ()

W p,max (105 J/mm3 )

Linear

30

3.29103

1.77101

3.29103

1.77101

Linear
Linear
Nonlinear
Linear

1
2.16
1
30

1
9.15102
3.16102
3.32103

3.49104
1.26102
6.74101
1.77101

2.66
9.15102
3.16102
3.32103

3.49104
1.26102
6.74101
1.77101

Linear
Linear
Nonlinear
Linear

1
1.97
1
30

1
2.75101
3.07102
2.58103

3.55104
2.42102
1.13102
1.93101

3.12
2.75101
4.14102
2.74103

3.55104
2.42102
1.36102
1.93101

Linear
Linear
Nonlinear
Linear

1
1.95
1
30

1
6.54101
5.62102
5.09103

3.38105
9.61102
1.57102
1.97101

1.06
6.54101
7.27102
5.40103

3.38105
9.61102
2.06102
1.99101

Linear
Linear
Nonlinear
Linear

1
2.06
1
30

1
5.58101
1.25101
8.97103

3.04105
5.43102
1.71102
2.16101

2.53
8.41101
1.25101
9.73103

7.55105
6.11102
1.71102
2.16101

Linear
Linear
Nonlinear

1
30
1

1
8.97103
3.64101

8.43106
2.16101
2.45102

5.11
9.73103
4.84101

1.96108
2.16101
2.57102


c 2007 The Authors. Journal compilation 
c 2007 Blackwell Publishing Ltd. Fatigue Fract Engng Mater Struct 30, 808822

Potrebbero piacerti anche