Numerical lifetime assessment of engine parts submitted

to thermomechanical fatigue
Guy Lederer, Eric Charkaluk, Laetitia Verger, Andrei Constantinescu

To cite this version:

Guy Lederer, Eric Charkaluk, Laetitia Verger, Andrei Constantinescu. Numerical lifetime assessment
of engine parts submitted to thermomechanical fatigue: Application to exhaust manifolds design. SAE
2000 World Congress, Mar 2000, Detroit, United States. �10.4271/2000-01-0789�. �hal-00116123�

HAL Id: hal-00116123

https://hal.science/hal-00116123
Submitted on 9 Nov 2022

HAL is a multi-disciplinary open access L’archive ouverte pluridisciplinaire HAL, est

archive for the deposit and dissemination of sci- destinée au dépôt et à la diffusion de documents
entific research documents, whether they are pub- scientifiques de niveau recherche, publiés ou non,
lished or not. The documents may come from émanant des établissements d’enseignement et de
teaching and research institutions in France or recherche français ou étrangers, des laboratoires
abroad, or from public or private research centers. publics ou privés.

Distributed under a Creative Commons Attribution - NonCommercial| 4.0 International

License
 Numerical Lifetime Assessment of Engine Parts Submitted to
Thermomechanical Fatigue, Application
to Exhaust Manifold Design

Guy Lederer, Eric Charkaluk and Laetitia Verger

PSA Peugeot-Citröen

Andrei Constantinescu
Ecole Polytechnique

ABSTRACT the complex physical phenomena in a series of simple

models.
In this work a numerical method for the design of
This paper focuses on the applications of computational
components submitted to severe cyclic
mechanics on fatigue life assessment and presents the
thermomechanical loading is developed. This tool is
case study of an exhaust manifold. Therefore a series of
based on a Finite Element (FE) analysis. In a first part
intermediary results, such as material behavior or
the temperature distribution is obtained and used in the
temperature distribution, is included as a part of the
second part for the mechanical computation. The
numerical method section.
analyses use the description of the geometry of the part,
a precise information of the thermal properties, an
CASE STUDY PRESENTATION
appropriate behavior of the material at low and high
temperature and a good assessment of the boundary
The presented method can be considered as a general
conditions (heat transfer coefficients, contact, ...). This
approach for numerical lifetime prediction of parts under
method is applied to asses the low cycle fatigue design of
thermomechanical loading. In the sequel the example of
a diesel turbo-charged exhaust manifold in cast iron.
an exhaust manifold has been chosen to illustrate this
These calculations, failure location and lifetime
method.
estimation, obtained on two versions of this component,
are compared to experimental data. The results show a
ENGINE AND MANIFOLD DATA – Two prototype
good agreement in terms of critical zones location and of
versions of a four cylinder 2.1 l turbocharged diesel
lifetime comparison in both versions and permit thus to
engine have been studied both experimentally and
classify the versions.
numerically.
INTRODUCTION

Computational fatigue design of engine parts submitted

to thermomechanical loading includes a wide range of
difficulties conducting to a complex problem. These
difficulties stem from the precise identification of the
thermal loading, the mechanical constitutive behavior
which has to perform at both low and high temperatures
and the fatigue criterion itself. Even if the previous points
have extensively been studied in the last decades, one
still lacks a global robust approach for lifetime
assessment in an industrial design context. The
constrains of the industrial are the matching of
experimental results and numerical lifetime computations
as well a project duration compatible with the ever
shortening motor development schedules. Therefore one Figure 1. Exhaust manifold version 1
needs to concentrate in each step of the design method

1
The mesh of version 1 of this manifold is presented figure Steady-state temperature field – If the thermal
1. The second version (called version 2) differs only conductivity is given as a function of temperature, the
slightly from version 1 by typical modifications such as steady state diffusion problem can easily be solved
ribs or local thickness modifications (figure 11). provided that the boundary conditions are precisely
evaluated. Three “families” of boundary conditions can be
Both manifolds are made of a spheroidal graphite silicon/
marked out: internal surface, external surface and
molybdenum cast iron (4 wt% Si, 0.8 % Mo). The surface
contact conductance on contact surfaces.
of the part has not been treated with the exception of the
contact surfaces (cylinder head, screws, turbo) which
1. Internal surface – The major heat flux contribution to
been machined.
the heating of the manifold comes from the forced
convection with the exhaust gas. This is a tricky problem
EXPERIMENTAL CONDITIONS – These two manifolds
as the estimation of local convection coefficients needs a
have been submitted to standard prototype durability
fair knowledge of both the fluid flow near the manifold
cycles on a classical motor testing device. The test rig
wall and the gas temperature. The resolution of this fully
includes the whole motor, with its intake and exhaust
turbulent anisothermal fluid problem is performed using
ports, coupled with a controlled brake. The device is
the commercial FIRE code [1] and its description is out of
located in a ventilated room in order to keep a room
the scope of this paper. However, let us simply
temperature of 30°C, but with no special air blow system
summarize the assumptions at the basis of these
on the outer surface of the motor or the manifold. The
computations :
cycle alternates continuously full load at 4000 rpm and
light load at the same regime. The transient period from • The mesh of the fluid part is generated automatically
full to light load and light to full load takes 30s and the and the cell thickness in the wall neighborhood is
total cycle takes approximately 700 s. The external adapted during the computations in order to get a
surface of the manifold is monitored by regular visual correct evaluation of the boundary layer.
inspection and to detect crack apparition. • The boundary conditions for the fluid computation
are directly derived from engine and turbo-charger
NUMERICAL METHOD data, i.e. imposed gas temperature and fluid flow at
the intake of the manifold and imposed pressure at
The numerical method will be presented here just in its the exhaust given by the turbocharger. The
global aspects without computational details. presented temperature distributions (see figures 2
The crucial points from the numerical point of view are: and 3) are obtained under full load conditions.
• The anisothermal k-ε analysis does not show
• the sensitiveness of the result to the temperature
important differences in gas temperature between
distribution as it acts through the severe coupling
intake and exhaust for this kind of manifolds.
with the mechanical material parameters determining
the mechanical fields • The local heat transfer coefficient between the gas
and the wall at full load has been obtained with the
• the choice of a mechanical constitutive law which has
Diwakar law [2] based on a logarithmic description of
to represent the behavior at low and high
the wall model. The resulting heat transfer
temperatures and has to be identified from simple
coefficients at full load are then projected on the
experiments
internal surface of the FEM mesh presented figure 1.
• the fatigue criterion which has to take into account The values of these coefficients can vary from
the 3D nature of the mechanical fields and has to approximately 200 W/m²°K to more than 1500 W/
integrate these through an anisothermal cycle. m²°K near the turbocharger and depend on the local
flow rate and the temperature of the gas.
THERMAL COMPUTATION – The aim of the thermal
computations is the prediction of the temperature field 2. External surface – The heat exchange on the external
distribution on the manifold from engine data (i.e. exhaust surface has been modeled using both natural convection
gas temperature and mass flow rate) which can already with air at room temperature and radiation.
be estimated in the early phases of the project.
The natural convection coefficient h is derived from the
The experimental cycle proposed here is a clearly classical empirical correlation [3]:
transient one. However, in order to simplify the analysis
we first determine a steady state temperature field Ra0 .25 λ f
distribution based on full load data and only afterwards h = 0.53
compute the transient response under a series of φ ext
(1)
assumptions concerning the evolution of the boundary
conditions described are described in the Transient where λ f is the air conductivity at an estimated "film"
temperature field paragraph.
T p − Text
temperature T f = (Tw and Text denoting
2
2
respectively the wall temperature of the manifold and the Transient temperature field – Supposing that all boundary
surrounding air temperature); φ ext is a characteristic conditions are known at full load, the transient results are
obtained under the following assumptions:
external diameter and Ra is the Rayleigh number
evaluated at Tf . Equation (1) is valid for a laminar flow • At our time scale, the exchange temperature on the
internal surface is directly dependent of the load
(103 < Ra < 109) and leads in our case to values of h
evolution.
typically around 20 W/m²°K.
• The variation of convection coefficient on the internal
The radiative part φr of the heat flow is simply computed surface between small and full load is neglected. This
by the classical radiation law: is due to the regime which remains slightly constant
during the test and the assumption that, in first
φ r = ϖ( T ) β ( T 4 − Text
4
) approximation, the convection coefficient follows this
(2) regime.
where ϖ( T) is the emissivity of the cast iron as a • The conduction coefficient on the external surface
and contact conductance remain constant
function of temperature and β is Stefan’s constant. The
radiant contribution to the heat flux is quite significant in The computations are performed using the transient
analysis in ABAQUS/standard. The CPU time on a HP-V
hot zones of the manifold as ϖ( T) reaches 0.9 for Tw class computer is about 800s.
around 700°C.
A comparison between the computed and measured
3. Contact surfaces – The thermal conduction between temperature evolution is presented figure 3. The spatial
the manifold and the cylinder head or the turbo-charger is location of the points quoted 1 and 2 is sketched on the
taken into account by means of a thermal contact manifold in figure 1. We can observe a good agreement
conductance. For example the temperature field of the between measured and computed temperature, even if
cylinder head in the vicinity of the manifold, Tcyl, being the temperature rate reaches 15 °C/s.
known from earlier analyses, the boundary condition,
φcont , on the contact surface between cylinder head and
manifold stands :

1
φ cont = (Tw − Tcyl )
RTc
(3)
where Tw is as defined above and RTc denotes the
contact resistance. This contact resistance integrates
globally the thermal characteristics of the joint and of the
interface. This parameter is obtained from measurements
performed under representative clamping conditions.

FE computation – The conduction problem is solved Figure 3. Steady-state temperature field at full load
using the ABAQUS Version 5.7 commercial FE code [4]. (°C). On this particular manifold, only the
The convergence of this nonlinear problem takes some heating has been experimentally measured.
few iterations. The resulting temperature field is plotted
figure 2. One can remark that the temperature exceeds MATERIAL CHARACTERISTICS – The constitutive law
700°C near the turbo-charger and presents a spatial has to characterize the material from 20 °C to 700 °C. At
gradient of more than 400°C between the input and the 20 °C it presents an elasto-plastic behavior and at 700 °C
output of the manifold (less than 0.2 m). an almost purely viscous one. The fact that one has to
deal with a simple unifying law over the whole
temperature domain is dictated both by the FE transient
analysis and the limited time devoted to the lifetime
prediction during the development of a new manifold.
This conducts to the choice of an elasto-visco-plastic
material model. A similar conclusion has also been drawn
by [5].

Elasto-Visco-Plastic (EVP) behavior – A reasonably

Figure 2. Steady-state temperature field at full load (°C) representative model of the high temperature EVP
behavior of the material is achieved with a viscoelastic-
elastoplastic constitutive law with six temperature-
depending parameters.
3
The stress state is described by the sum of two Numerical implementation – The numerical
contributions of the elastoplastic part and the viscoelastic implementation of this EVP law is performed in ABAQUS
part of the model : by means of a User Material (UMAT) [4]. This procedure
allowed the optimization of both CPU time and
( ) (
σ = σ p + σ v = C p: ε − ε p + C p: ε − ε p ) (4)
robustness by use of a radial return algorithm [6-7] for the
integration of the material behavior. This totally implicit
implementation overcomes the difficulties of the standard
where Cp and Cv are the forth order tensors of the elastic
« VISCO » analysis of ABAQUS which always tries an
moduli. The corresponding young moduli are denoted by
explicit algorithm before switching in implicit mode. Our
Kp and Kv in the one-dimensional rheological
algorithm is proved unconditionally stable whereas
representation of figure 4. εvp and εvp represent
ABAQUS standard often presents slow convergence
respectively the plastic and viscous strain tensor.
(leading to very little step sizes and high CPU costs) or
even failure of convergence in severe transient loading
conditions.
The numerical behavior of the material law has been
tested on isothermal and anisothermal conditions before
the complete manifold computation. The anisothermal
computations correspond to a simple Thermo-
Mechanical Fatigue specimen undergoing thermal cycles
from 20 °C to 700 °C (the test is described later in the
fatigue life estimation section). Computed temperatures
Figure 4. Rheological model for the cast iron and stresses have been compared with measured values
and correlated satisfactorily.
Both inelastic strains follow a flow rule described by a
A
strain rate distance γ ≥ 0 and a direction as
A
A
ε&in = γ ⋅ .
A
The plastic law is given by a von Mises equivalent stress
yield criterion J2 and a linear kinematic hardening rule:


( ) 2
 A p = dev σ p − 3 H . ε p

γ p =

3
( )
J A −σy
2 2 p
Figure 5. Evolution of the identified yield limit σy of the
plastic part of the model
(5)
The viscous law is a three-dimensional Norton-Hoff rule : STRUCTURAL ANALYSIS – Finite Element Model and
Boundary Conditions – The FE mesh of the manifolds
A v = dev (σ v ) contained about 15000 linear 3D brick elements,
 corresponding to about 66000 Degrees Of Freedom
3 J 2 (A v )
m
 (DOF). The cylinder head has been modeled by a thick
γ v = plate.
 2 η
(6) The coupling between the cylinder head and the manifold
has been represented by a number of screws and a
This model and its parameters have been identified from
standard small sliding friction contact model [4].
isothermal cyclic traction-relaxation tests at several
temperatures. The parameters are then linearly
Analysis – The analysis has been performed in two
interpolated between these identification temperatures.
steps: first the manifold has been screwed on the cylinder
Figure 5 presents, for instance, the dependence of σy on
head and second a series of thermal cycles have been
temperature. One can remark that the yield limit lowers
computed. The imposed temperature distribution was a
tremendously with increasing temperature and this will
result of the previous presented transient analysis. The
demand special care in the numerical integration
number of cycles has been chosen in order to obtain a
algorithm.
stable limit cycle for the mechanical fields.

4
FATIGUE LIFE ESTIMATION – The major challenge was constant temperatures between 200°C and 700°C under
to determine a low cycle fatigue (LCF) criterion in a strain control. The total strain rate was of 10-3 s-1 and the
multiaxial context. The lifetime prediction of the parts strain ranges were ± 0,25 %, ± 0,5 % and ± 1 % (R=-1).
should be included within the standard deviation of the TMF tests were conducted on clamped specimens
experimental results. The experimental lifetime has been heated by the Joule effect. The tests are then in an out-
considered as the number of cycles to failure of a of-phase mode. Maximum temperatures have been
representative elementary volume. This approach also varied between 40°C and 700°C with an heating rate of
permits a separation of the structural aspects (multi-axial 20°C/s. The maximum temperature has been obtained in
stress state, anisothermal loading, …) and the fatigue a region of approximately 10 mm in the center of the
properties of the given material which can be established specimen. The maximum temperature gradient was 30-
from simple classical laboratory LCF tests. 40°C/mm. The parameters of the test are the clamp value
(183000 N/mm and 227000 N/mm) and the dwell time at
A close inspection of classical LCF laws shows that the
700°C (60 seconds and 900 seconds).
Manson-Coffin [8-9], the Strain-Range Partitionning
(SRP) [10] or the Smith-Watson-Topper (SWT) function For all tests the dissipated energy has been evaluated on
[11] established in an isothermal and uniaxial context are the stabilized cycle by a numerical integration procedure
not appropriate for this anisothermal problem. Two types from the stress-strain experimental curves for the LCF
of difficulties can be pointed out. The first one comes test and by FEM computations for the TMF tests. The
from the multiaxial aspect of our problem which makes experimental versus predicted lifetime is presented in
difficult a generalization of these damage functions. The log-log plot in Figure 6.
second one comes from the high range of attained
temperatures: classical Manson-Coffin interpretation of
isothermal LCF tests shows on a large difference in
lifetime for the same plastic strain-range at different
temperatures. This makes any interpretation of
cumulated plastic strain on an anisothermal stress-strain
curve hazardous.
Another difficulty can be remarked in the SWT
interpretation of the anisothermal experiments as the
choice of smax is impossible in an anisothermal context

( SWT function = E. σ max . ∆ε ).

An interesting interpretation of the cyclic behavior of the Figure 6. Experimental versus predicted lifetime for the
material is presented by Skelton [12]. His results on LCF nodular cast iron (structure : fatigue results on
tests suggest that the cumulated dissipated energy to the manifolds)
plastic hardening or softening stabilization can be used
as a crack initiation criterion in low cycle fatigue. RESULTS
Denoting by Ns the number of cycles to stabilization,
assumed to correspond to crack initiation, Skelton In the discussion of the result, we shall essentially
showed that the cumulated dissipated energy to present the predictions related to the two versions of the
stabilization : N S × ∆WS is relatively constant for a given exhaust manifold. A detailed discussion of other results
can be found in [7, 13-14].
material.
At this point, we shall assume that the failure of the NUMERICAL ASPECTS – The first remark concerns the
representative elementary volume on real structures number of cycles needed to achieve a stabilized limit
(which corresponds to Nf – number of cycles to failure - cycle for the mechanical field, conducting to a
on specimen) conducts within a few cycles to a bad stabilization of the cyclic dissipated energy. It appeared
functioning of the structure in accordance with the fact that three cycles were enough to get a representative
that the engineering design should not guarantee the value of the dissipated energy. Therefore, the next lifetime
integrity of the component. Therefore, with both results are computed using the third loading cycle. It is
information – Nf and ∆Ws – a thermo-mechanical fatigue important to notice that this is a pure observation and
that to our knowledge no mathematical results are
criterion can be established in the form :
available to predict the existence of the limit cycle and the
N f = f (∆WS ) . number of cycles needed to attain it.
As a consequence of the previous considerations, the The thermomechanical analyses for both manifold
criterion has been established using classical isothermal versions were completed within approximately 6 hours of
LCF and anisothermal TMF tests [13-14]. The LCF tests CPU time on a HP-V class computer. This CPU time is
have been carried out on classical specimens at different only an indication since it is very sensitive to the amount

5
of inelastic deformations experienced by the structure as Figure 9 presents a second predicted zone on the rear of
well as to the non-linearities induced by frictional contact. this version. The validity of this zone is also confirmed by
experimental data as shown on figure 10. Both zones
FATIGUE RESULTS – Figure 7 presents a plot of the present approximately the same level of cracking danger,
distribution of the criterion function on the inner surface of and the lifetime estimation falls within the standard
version 1. deviation of the criterion as shown on figure 6 (points
quoted « structure »).

Figure 7. Version 1 ; main predicted crack zone ;

visualized on the inner surface of the manifold

Figure 10. Version 1 ; second experimentally observed

crack zone on the rear of the Manifold

Figure 8. Version 1 ; first experimentally observed crack Figure 11. Version 2 ; Von Mises stress distribution
(Mpa) at full load
The computed critical zone on the shortest tube of the
manifold is in good agreement with one of the Figure 11 presents an example of von Mises equivalent
experimental cracked zone presented on figure 8. stress distribution at full load on version 2. It appears
clearly that the highest von Mises equivalent stresses are
located in the coldest parts of the manifold (see figure 2).
This is coherent with the yield limit evolution sketched on
figure 5, but gives no hint on the location of the critical
zone.

Figure 9. Version 1 ; second predicted crack zone at

the back of the manifold

6
 The results of these computations, associated with a
reasonable simple anisothermal low cycle fatigue
criterion show a good agreement between predicted and
observed critical zones for two versions of a given
manifold. Numerical and experimental lifetime increase
between the two versions is also satisfactorily correlated.
This approach is not limited to exhaust manifolds or cast
iron: further applications could concern steel or aluminum
alloys and other parts in the automotive industry or
elsewhere.

ACKNOWLEDGMENTS

The authors would like to thank Mr K. Dang Van, Ecole

Polytechnique, and Mr A. Bignonnet, PSA Peugeot
Citröen, for fruitfull discussion.

REFERENCES

1. AVL LIST GmbH, FIRE Menu reference, Graz (1996)

2. Taine J., Petit J.P., Transferts thermiques et
mécanique des fluides anisothermes, Dunod, Paris
(1989).
3. McAdams W.H, Heat Transmission, 3rd ed. Mc Graw
Hill, (1954)
4. Hibbitt, Karlson & Soerensen Inc, ABAQUS/Standard
User’s Manual, (1998)
5. Anderson D. H., Bisaro D.R., Haan D.M., Olree M.,
Figure 12. Criterion decrease (lifetime increase) SAE paper N° 980697, (1998).
between a, version 1 and b, version 2 for the 6. Simo J.C. , Hughes T.J.R., Computationnal
first failure zone. Inelasticity, Springer Verlag, (1998).
7. Charkaluk E., Verger L., Constantinescu A., Lederer
Computations performed on version 2 of this manifold G., Stolz C., in Actes du 4ème Colloque de Calcul de
exhibited an important increase in lifetime : The rear zone Structure, Teknea ed., Giens, (1999)
is not critical anymore, and the criterion shows a 8. Coffin L. F, ASME paper 53-A76, pp 931-950, .
significant lifetime increase for the first zone, as detailed (1953)
on figure 12. This predicted lifetime increase of 150%
9. .Manson S., NACA TN 2933, . (1953).
between version 1 and 2 can be satisfactorily compared
with the experimental lifetime increase of 110%. 10. .Halford, G. R., and Manson, S. , In : Thermal fatigue
of materials and components, ASTM STP 612, pp
239-254, (1976).
CONCLUSION
11. Smith, K.N., Watson P. and Topper T. H., J. Mater.
A complete lifetime estimation approach has been Vol. 5., No 4, pp 767-778, (1970).
derived for structures undergoing thermomechanical 12. Skelton R. P., Mat. Sc. Tech. Vol. 7, pp 427-439,
fatigue. It has been applied to the design of cast iron (1991).
exhaust manifolds. 13. Charkaluk E., PhD Dissertation, Ecole
Polythechnique, Palaiseau, France (1999).
The structural approach integrates a simple elasto-visco-
plastic description of the high temperature behavior of 14. Charkaluk E., Constantinescu A., Bignonnet A.,
Dang Van K., in Low cycle fatigue and elasto-plastic
the material. This material law allows sufficient precision
behavior of materials pp 119-130, K.T. Rie & R. D.
without major CPU time increase implying that this Portella eds., Elsevier (1998).
approach can be included in a classical development
schedule of such parts.