Download as pdf or txt
Download as pdf or txt
You are on page 1of 14

This article was downloaded by: [UNAM Ciudad Universitaria]

On: 23 December 2014, At: 21:50


Publisher: Taylor & Francis
Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House,
37-41 Mortimer Street, London W1T 3JH, UK

Structure and Infrastructure Engineering:


Maintenance, Management, Life-Cycle Design and
Performance
Publication details, including instructions for authors and subscription information:
http://www.tandfonline.com/loi/nsie20

Reliability assessment of innovative designs in


automotive industry
a a a b
Guillaume Morin , Jean-Marc Cardona , Olivier Prince , Maurice Pendola & Frederic
b
Perrin
a
Renault SAS, Powertrain Engineering Division , 67 rue des Bons Raisins, F92508, Rueil-
Malmaison Cedex, France
b
Phimeca Engineering SA , 1 allée Alan Turing, F63170, Aubière, France
Published online: 09 Jun 2009.

To cite this article: Guillaume Morin , Jean-Marc Cardona , Olivier Prince , Maurice Pendola & Frederic Perrin (2009)
Reliability assessment of innovative designs in automotive industry, Structure and Infrastructure Engineering: Maintenance,
Management, Life-Cycle Design and Performance, 5:5, 367-379, DOI: 10.1080/15732470701388736

To link to this article: http://dx.doi.org/10.1080/15732470701388736

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained
in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no
representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the
Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and
are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and
should be independently verified with primary sources of information. Taylor and Francis shall not be liable for
any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever
or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of
the Content.

This article may be used for research, teaching, and private study purposes. Any substantial or systematic
reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any
form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http://
www.tandfonline.com/page/terms-and-conditions
Structure and Infrastructure Engineering
Vol. 5, No. 5, October 2009, 367–379

Reliability assessment of innovative designs in automotive industry


Guillaume Morina*, Jean-Marc Cardonaa, Olivier Princea, Maurice Pendolab and Frederic Perrinb
a
Renault SAS, Powertrain Engineering Division, 67 rue des Bons Raisins, F92508, Rueil-Malmaison Cedex, France; bPhimeca
Engineering SA, 1 alle´e Alan Turing, F63170, Aubie`re, France
(Received 1 June 2006; final version received 4 April 2007)

Car manufacturers have to introduce innovative designs during new product development. Field reliability of these
new products must be achieved and demonstrated before the start of production. As the design is innovative, this
demonstration cannot be based on field and warranty results of existing engines. To overcome this difficulty, the
present paper presents a method based on numerical simulations, and two aspects are studied. The reliability of
Downloaded by [UNAM Ciudad Universitaria] at 21:50 23 December 2014

mechanical components is assessed, taking into account the manufacturing processes variability. This is performed
with numerical reliability based design (RBD) methods such as the first or second order reliability methods (FORM/
SORM). Loads applied by customers are also evaluated according to typical car usage and the specificities of engine
behaviour generated by the new concept. The combination of these two types of information provides an evaluation
of the future field reliability of the new concept. Results such as parameter sensitivities also help to improve
reliability during development by optimizing the design or the manufacturing process. This numerical approach is
illustrated in an example of a manufactured steel exhaust manifold.
Keywords: reliability based design; innovation; exhaust manifold; polynomial chaos; stress-strength; field reliability

1. Introduction performance. It also reduces warranty costs and


Innovation is a constant and necessary aspect of car increases profitability. For incrementally modified
design and development. At the global vehicle level, designs, reliability can be evaluated on the basis of
innovations lead to offering a high level of perfor- information obtained from the field for previous and
mance and service to the customer. They increase the comparable products. This is, of course, impossible for
attractiveness of new models and the motivation of completely new designs for which there is no existing
customers for buying a new car. At the engine level, experience.
innovations are often necessary to satisfy increasingly This paper presents a specific method dedicated to
stringent emission standards. These standards apply to assessing and optimizing reliability of innovative
greenhouse effect gas emission, which is directly designs of engine components during a new engine
connected to fuel consumption, and to exhaust development. This method is based on numerical
pollutant emissions such as hydrocarbons, carbon simulations and is divided into two steps:
monoxide, nitrogen oxides, or particulates. Such
innovations can concern combustion chamber design . The first one is the analysis of the new com-
(cylinder head, piston), air intake system design, fuel ponent reliability connected to its design and
injection system design or exhaust system design. In an manufacturing process (reliability based design).
exhaust system, the design of existing parts can be . The second one is the evaluation of the
modified and improved and some new components, innovative component field reliability, in terms
such as cooled exhaust gas recirculation, diesel of failure rate at a given mileage, associated with
particulate filters (DPF) or nitrogen oxides (NOx) use by real customers.
traps, can be added.
In parallel with this innovation process, reliability These two steps are described in the two main parts of
must be guaranteed because it is an important factor of this paper and illustrated in an example involving steel
competitiveness. A good reliability has a very positive fabricated exhaust manifolds for high specific power
impact on brand image and contributes to commercial diesel engines.

*Corresponding author. Email: guillaume.morin@renault.com

ISSN 1573-2479 print/ISSN 1744-8980 online


Ó 2009 Taylor & Francis
DOI: 10.1080/15732470701388736
http://www.informaworld.com
368 G. Morin et al.

In this paper, only the initiation and propagation


2. Innovative component reliability based design
of a crack in the manifold wall are considered. These
2.1. Context cracks can be initiated by mechanical loadings (e.g.
It has been seen that there is a strong necessity to reduce engine vibration) or thermal loadings. Some studies
fuel consumption in order to limit greenhouse effect gas have shown that the dominant loadings in this case
emissions. In Europe, a popular and cost effective way to were thermal cyclic loadings, resulting in low cycle
reach this target is the development of direct injection fatigue and creep (Bignonnet and Charkaluk 2001).
diesel engines associated with a reduction in the engine The most important failure factor is then the tempera-
displacement (‘downsizing’). To also meet new pollutant ture history, which is connected to the variations of
emission standards, specific after treatment systems, such power and torque demand on the engine.
as diesel particulate filters and NOx traps, must be To reproduce this failure mode, the validation of
added. These devices, combined with downsizing, an exhaust manifold is based on a specific endurance
severely increase thermal loads on the engine exhaust test performed on an engine test bed. The hot/cold
system. The exhaust manifold is the first component of cycle is traditionally used with the objective of creating
the exhaust system, as it connects the cylinder head the greatest temperature amplitude between the max-
Downloaded by [UNAM Ciudad Universitaria] at 21:50 23 December 2014

exhaust face to the turbo charger. It is therefore directly imum temperature (at maximum power) and the
affected by an increase in the exhaust gas temperature minimum temperature (at idle for instance). The
and thermal loads. typical temperature evolution of this hot/cold cycle is
Exhaust manifolds were traditionally made of cast shown in Figure 1.
material (mainly cast iron) for cost reasons. This kind Due to these loadings, a high temperature gradient
of design can no longer sustain the exhaust gas on the exhaust manifold is observed between the exit
temperature (greater than 8508C) encountered in high flange (high area temperature) and the inner flange,
specific output diesel engines equipped with diesel which is cooler because it is in contact with the cylinder
particulate filters and NOx traps. It is therefore head. Therefore, this differential of temperature, and
necessary to develop fabricated steel exhaust manifolds the friction between the inner flange and the cylinder
that can resist to this range of temperatures. They also head, create cyclic thermo mechanical stresses and
present the advantage of a lower thermal inertia, then low cycle fatigue cracks and leakages. The critical
reducing the emissions during the engine cold start. areas are often localized on the inter-ducts or near the
Although this type of exhaust manifold has been exit flange where the temperature is maximal (see
used for a few years on gasoline engines, it is still Figure 2).
innovative for high specific output diesel engines
equipped with diesel particulate filters and NOx traps.
The reliability assessment of such a novel component 2.3. Deterministic simulation of failure
design cannot be based on field results. For these The first step of the global approach is to develop a
reasons, the new method developed to assess the method that is able to predict the crack initiation in an
reliability of innovative designs has been first applied exhaust manifold subjected to simple thermo mechan-
to steel fabricated exhaust manifolds of modern diesel ical fatigue cycles such as the hot/cold cycle on the
engines.

2.2. Failure mode analysis


Two mechanical failure modes can typically be
encountered on an exhaust manifold:

. Initiation and propagation of a crack on the


manifold wall, generating a leakage of exhaust gas
from the inside to the outside of the manifold.
. Thermo mechanical deformation of the manifold
at the interface with the other parts (cylinder
head or turbo charger), also generating a leakage
of exhaust gas from the inside to the outside of
the manifold.

These exhaust gas leakages can generate some noise or Figure 1. Exhaust manifold wall temperature evolution
loss of performance. during a hot/cold cycle on an engine test bed.
Structure and Infrastructure Engineering 369

engine test bed. For this first step, the simulation is Lemaitre and Chaboche formulation to describe an
deterministic. All parameters (geometry, materials elastoplastic cyclic behaviour (Chaboche 1989) is used
characteristics, loadings) are considered to be at their here.
nominal values without any dispersion. The load criterion is defined by:
During tests on an engine test bed, the histories of
engine speed and torque are perfectly defined by the f ¼ fy ðr  XÞ  R; ð1Þ
engine designer and can be controlled during the test.
Therefore, the cyclic thermal solicitations (shown in where f is the plasticity threshold, fy is the Von
Figure 1) are well known and can be computed using Mises criteria, s is the stress tensor, X is the non-
finite element analysis. The boundary conditions are linear kinematic tensor and R is the yield stress. Both
the assembly with the adjacent parts (cylinder head, non-linear kinematic and isotropic hardening are used
turbo charger, catalyst, including gaskets and brack- here:
ets). Thermal loadings include convective exchange
2
with the exhaust gas, and convective and radiative dX ¼ C dep  DX dp; ð2Þ
exchanges between the exhaust manifold and the 3
Downloaded by [UNAM Ciudad Universitaria] at 21:50 23 December 2014

engine compartment. dR ¼ bðQ  RÞdp; ð3Þ


Knowing the temperature history on the exhaust
system, the next step is to compute the stress-strain C  
field and the history in the exhaust manifold (e.g. X¼ þ 1  exp ðDep Þ ; ð4Þ
D
Azzouz et al. 2002). For thermal fatigue evaluation, it
is important to select and use an accurate model to R ¼ R0 þ Q½1  exp ðbpÞ; ð5Þ
describe the material cyclic mechanical behaviour. The
s ¼ X þ R; ð6Þ

C
su ¼ þ Q þ R; ð7Þ
D
where ep is the plastic strains tensor, p is the
accumulated plastic strain, s is the uniaxial stress, su
is the ultimate uniaxial stress and R0, C, D, Q and b are
material parameters.
One difficulty with such a model is the identifica-
tion of the material parameters. This work needs many
experimental data at different strain rates and levels
in isothermal low cycle fatigue (LCF) conditions.
Figure 3 presents some results of this identification,

Figure 3. LCF at 6008C: experimental and numerical


Figure 2. Example of cracks on the inter-ducts. results.
370 G. Morin et al.

and shows stress versus imposed strain during a computed number of hot/cold cycles before crack
uniaxial traction/compression test. This shows a good initiation . It can be noted that the model sensitivity to
correlation in term of stress level between numerical engine type and engine maximum power is consistent
simulation and experiments. with the experimental results. After the deterministic
The fatigue life prediction under anisothermal low model has been validated, it can be extended in a
cycle fatigue conditions is still a problem under probabilistic way in order to perform reliability
discussion. One difficulty is the choice and the analyses.
definition of the critical variable governing the fatigue
life. Strain based models (Manson 1954, Taira 1973),
stress based models (Lemaitre et al. 1978) and mixed
2.4. Reliability Based Design methods
models using both stress and strain (Skelton 1991) can
be found in the literature. 2.4.1. Hypotheses and principles
To compute the lifetime of steel exhaust manifolds, Consider, the structural reliability problem defined by:
a plastic strain based model (Manson – Coffin model)
is used (Manson 1954), and which can be expressed . A random vector X whose marginal densities and
Downloaded by [UNAM Ciudad Universitaria] at 21:50 23 December 2014

as: correlations, or better, the joint density function


fX(x) have been estimated.
K
Nf ¼   ; ð8Þ . A performance function G(x), which separates
Dep a
2
the physical admissible realizations from the
non-admissible: G(x) 4 0 is a success realiza-
where Nf is the number of cycles to crack initiation, tion, G(x) 5 0 is a failure realization and
Dep is the variation of plastic strain during one G(x) ¼ 0 defines the limit state function.
cycle, and K and a are material parameters. The tem- . A mechanical model with associated numerical
perature dependency of the damage law complicates solution that allows realizations of G(x) to be
the identification. Table 1 presents typical material evaluated over all the physical definition domain
parameter values for a standard exhaust manifold of x.
stainless steel.
The model identified on LCF tests is validated with The evaluation of structural reliability has been
engine hot/cold test results for two types and different described by several authors (e.g. Ditlevsen and
versions of diesel engines. Engine 1 is a 2.2 litre engine Madsen 1996). If the joint density is known, the
and Engine 2 is a 2.0 litre engine. Engine 2 is tested failure probability Pf is given by:
with different power rates. For each engine, the hot/ Z
cold cycle definition is specific and related to the engine Pf ¼ fx ðXÞ dx1 . . . dxn : ð9Þ
maximum power and idle running conditions. Table 2 GðxÞ<0
shows the comparison between experimental and
In most cases, this integral cannot be directly evaluated
for at least two reasons: the joint density function is
Table 1. Material parameter values. not known and incomplete information constituted by
Room temperature Maximum temperature the marginal distributions and the second order
moments has to be used, and the calculation of the
R0 (MPa) 200 30 performance function can only be obtained by a
C (MPa) 3606103 1506103
D (MPa) 2200 2000 numerical procedure given by the finite element
Q (MPa) 60 710 method.
b 20 12 Traditionally, the analysis is performed in three
K 48 5.5 steps: a probabilistic transformation T associates with
a 1.2 0.9
the vector X of any physical random variables, the
vector U of normally uncorrelated and standardized
Table 2. Fatigue lives. Gaussian variables N(0,1). For this purpose, Nataf,
Rosenblatt or Hermite transformations can be used
Simulation using (Ditlevsen and Madsen 1996). Therefore:
Experimental equation (8)
 
Engine 1 4100 cycles 3950 cycles U ¼ TðXÞ and GðXÞ ¼ G T1 ðUÞ  HðUÞ: ð10Þ
Engine 2: low power 5000 cycles 4850 cycles
Engine 2: medium power 2400 cycles 2250 cycles The Hasofer – Lind reliability index b is then
Engine 2: high power 2150 cycles 2200 cycles
evaluated by solving the optimization problem
Structure and Infrastructure Engineering 371
pffiffiffiffiffiffiffiffiffiffi
b ¼ min ð Ut UÞ under the constraint H(U)  0. The When using FORM or SORM, such combinations
failure probability is then obtained by using the first can be made to obtain reliability results with a
order reliability method (FORM), by second order reasonable number of finite element calculations (e.g.
approximation (SORM) or by conditional simulations see Pendola et al. 2003). If the gradients of the
(see Figure 4). performance function need to be computed analyti-
cally 4(n þ 1) is a good approximation of the required
number of FEA if n represents the number of random
2.4.2. Coupling of finite element analysis and variables
reliability algorithms In principle, this method can also be used with
The direct coupling method between the finite element Monte-Carlo simulations or directional simulations.
analysis (FEA) and the reliability algorithms is But, to obtain a low probability level, the CPU time
proposed here. It solves the reliability problem without can be very large because of the realization of a large
more hypotheses than a probabilistic representation of number of finite element calls. This is why the FORM
the correlated random vector X given by the marginal approximation has been used in conjunction with
distributions, instead of using the joint probability importance sampling in order to validate the estimated
Downloaded by [UNAM Ciudad Universitaria] at 21:50 23 December 2014

function that is seldom available in practice. A good probability.


efficiency indicator of these methods is the number of
FEA calls.
In this method, the optimization algorithm used to 2.5. Influence of design and manufacturing process on
solve the problem (Rackwitz and Fiessler 1979, Abdo component reliability
and Rackwitz 1990) directly calls the finite element The previous reliability methods have been applied
model (FEM) code each time a realization of the here on a steel exhaust manifold considering uncer-
performance function G(X) and its derivatives have to tainties in:
be evaluated. The gradients calculation is performed
by finite differences. The number of FEA depends on . Geometry: four thicknesses of the manifold are
the performance of the optimization algorithm and on considered random due to the manufacturing
the required accuracy for the iterative schemes. process (Thickness numbers 3 to 6 on Figure 6).
Supplementary calls are necessary if curvatures of the
limit state (SORM approximations) are used. The
Table 3. FORM method attributes.
advantages and drawbacks of this method are sum-
marized in Table 3. Advantages Drawbacks
To minimize these drawbacks, special software
Accuracy of the FEA Expensive development
programs such as PHIMECAÓ (PHIMECA 2004) are preserved. of link between FEA
now available to couple reliability algorithms and FEA Algorithms used are efficient. and reliability analyses.
(see Figure 5), and non-based gradient algorithms have Automatic procedure Numerical evaluation of
been developed to consider some discontinuities in the (no intervention from gradients (discontinuities
the analyst). may occur).
limit state function.

Figure 4. FORM/SORM approximations. Figure 5. PHIMECA softwareÓ.


372 G. Morin et al.

Figure 6. Variable thickness areas on the manifold.

. Material: the following parameters of the stress-


strain curve (see x2.3) are considered uncertain Figure 7. Global influence of the random variables given by
Downloaded by [UNAM Ciudad Universitaria] at 21:50 23 December 2014

and temperature dependant. Young’s modulus direction cosines.


E, yield strength SE, hardening coefficients C, D
due to material supplier scattering.
. Lifetime: damage coefficient K involved in the consistent with the large scatter of material
Manson – Coffin law (a is considered to be fatigue tests results on which this law is based.
totally correlated with K).
. Loadings: minimal (Tral) and maximal (Tmax) To go deeper into the reliability analysis, interest-
exhaust temperatures during the hot/cold test are ing results are given by sensitivities or elasticities of the
considered random due to fluctuations asso- failure probability (e.g. reliability index) to the mean or
ciated with the engine control and running. standard deviation of the random variables. A high
sensitivity to the mean value of a variable suggests that
A critical area on the structure, where the plastic a new design with a different mean value can have
strain amplitude is maximal, is considered. At this considerably higher reliability (e.g. a new material with
point, the probability that the number of life cycles higher strength). If it is sensitive to the standard
during the hot/cold test (Nfailure) becomes lower than deviation, it suggests that the knowledge of the
the required number (Nrequired) is evaluated by: parameter variability has to be improved, or that
quality control has to be intensified for this parameter
Prob½GðXi Þ  0 ¼ ProbðNfailure  Nrequired Þ: ð11Þ (as for process variability sources). For example, the
sensitivity of the reliability index b with respect to mX,
Introducing the 11 previous random variables, the the mean of the parameter X, is given by:
presented reliability method requires 37 finite element
analyses for a FORM reliability assessment. This @b
: ð12Þ
FORM approximation has been validated both by a @mX
SORM calculation and by some importance sampling
around the design point found by FORM. The typical The elasticity, which is a normalized measure of the
results are: sensitivity given above, is given by:

. Estimation of the failure probability Pf and @b mX


: ð13Þ
reliability index b as well as the design point P*, @mX b
i.e. the point on the limit state function where the
joint density is maximal in the standard space. Typical results in the manifold are given in Figures
. Global influence of the uncertainties on the 8 and 9.
failure probability given by direction cosines of Figure 8 shows the elasticities of the failure
the direction P*O (see Figure 4), O being the probability with respect to the mean values of the
origin in the standard space (typical results are random variables. For example, an increase of 1% of
shown in Figure 7). Here, the parameter K in the the exhaust gas maximum temperature will induce an
Manson – Coffin law contributes to more than increase of 11% of the probability of failure (i.e. a
93% to the reliability followed by the thickness reduction of 11% of the exhaust manifold intrinsic
ep5, with a contribution of 4%. This importance reliability). It shows that the nominal values of the
of the parameter of the fatigue life law is maximal temperature Tmax, the yield stress SE and the
Structure and Infrastructure Engineering 373

thickness ep5 have a great influence on the failure to reduce fatigue strength scatter by material process
probability whereas their global uncertainty measure control, if possible, than to change it for a material
given by the direction cosines in Figure 7 are not with a higher fatigue strength.
important. This can be explained by the elasticities
with respect to the standard deviation in Figure 9. The
3. Field reliability analysis
high importance of the mean values is reduced by the
low importance of the standard deviations. Inversely, 3.1. General methodology
the mean value of the Manson – Coffin law K is not The goal of the field reliability analysis is to predict the
important, whereas its standard deviation is of great future field cumulative failure rate of the component at
importance, due to the large scatter of material fatigue a given mileage. This failure is compared to the
tests results, as mentioned before. reliability target assigned to the component associated
All the reliability results are important in providing to a given failure mode. It is clear that if this analysis
design guidance to improve the reliability of the can be performed during engine development, several
component. In this example, it would be interesting months before beginning production, countermeasures
to modify the design by reducing the nominal thickness could be applied if necessary, at much lower cost and
Downloaded by [UNAM Ciudad Universitaria] at 21:50 23 December 2014

of area number 5 of the manifold. It would be also time than if first results from the field are awaited.
valuable to decrease exhaust gas maximum tempera- The global approach to reach this goal is based on
ture by optimizing combustion and engine control. As the strength-stress method (Thoft-Cristensen 1986)
far as material is concerned, it would be more efficient and has already been presented in detail by Morin
et al. (2005). The basis of the method is to compare,
for a given critical area of the exhaust manifold, the
statistical distribution of thermo mechanical damage
applied by the whole population of customers at a
given mileage (the stress), with the statistical distri-
bution of thermo mechanical damage that the com-
ponent can sustain without failure (the strength).
Typical probability densities of these two distribu-
tions are plotted in Figure 10. The cumulative field
failure rate is the probability for all customers to
apply a level of damage that will exceed the damage
that their exhaust manifold can sustain without
failure (strength).
Using the known analytic forms of the probability
density function of the stress distribution fStress and of
the cumulative distribution function of the strength
Figure 8. Elasticities of the failure probability with respect
to the means.

Figure 9. Elasticities of the failure probability with respect Figure 10. Probability densities of stress and strength of
to the standard deviations. inter-ducts area.
374 G. Morin et al.

distribution FStrength, the failure rate Pf can be that are very fast (a few minutes). The overall thermal
computed as: analysis method is presented in Figure 11.
A typical example of the exhaust manifold max-
Zþ1 imum temperature history computed under road test
Pf ¼ fStress ðxÞFStrength ðxÞdx: ð14Þ conditions is presented in Figure 12. On this graph, the
0 influence of the after treatment system can also be
seen. When the vehicle is equipped with a diesel
This integral can be computed by standard numerical particulate filter, an increase of the temperature when
methods such as the Simpson rule. To take into particulates are burning for regeneration is observed.
account errors in deterministic models for predicting
stress and strength and probabilistic models, a
confidence interval is associated with the probability 3.2.3. Thermo mechanical analysis
of failure by assuming some variations in the mean and With the temperature time histories, it is now possible
standard deviation for stress and strength distribu- to determine the corresponding thermo mechanical
tions. This aspect is presented in detail in Prince et al.
Downloaded by [UNAM Ciudad Universitaria] at 21:50 23 December 2014

(2005).

3.2. Evaluation of damage applied by customers


3.2.1. Engine loads
One difficulty of this approach is to obtain histories
of engine speed and torque for customer use mainly
due to the wide scatter and variety of customer
behaviour. The method is based on data recorded on
instrumented vehicles that are lent, for a few weeks, to
customers with typical every day use: urban, road,
highway, mountain, etc. Based on these records,
typical histories of vehicle speed and acceleration are
defined for each type of use. With records coming
from different customers with the same type of
journey, the scatter of driving styles can be evaluated.
They are grouped into three types: fuel efficient,
standard and sport. These histories can be trans-
Figure 11. Synoptic of thermal analysis method.
formed in engine speed and torque histories, as men-
tioned previously, by computation based on vehicle
and engine characteristics. They can be later adapted
for new vehicles of the same category based on vehicle
performance evolution.

3.2.2. Thermal analysis


The next step of the method is to compute the transient
evolution of temperature on critical points of the
exhaust manifold. At each time step, stationary tem-
perature for corresponding engine speed and torque is
interpolated from a table assessed by experimental
measurement. Then, depending on the value of the
time step, this temperature is corrected to take into
account the thermal inertia of the part. The thermal
inertia has been primarily characterized by computing,
with the FEA model, the part temperature time
response to typical thermal loads evolution (step,
ramp, etc.). This response is correlated with the Figure 12. Time histories of exhaust manifold during road
corresponding measurements on an engine test bed test with and without a diesel particulate filter (DPF).
Structure and Infrastructure Engineering 375

behaviour using a specific local model developed for


inter-ducts. This model is a local FEA model based on 3.2.5. Damage distribution
a classical sub-modelling technique, and driven by This method permits damage values for typical
results from the global model described in x2.3. The customer use characterized as previously described to
evolution of the plastic strain during the road test is be computed. With information of whole customer
computed and presented in Figure 13. distribution around these types, it is possible to
correlate a probability law for damage inflicted by
real customers with a classical statistical correlation
3.2.4. Damage computation method, for example, under the form of a normal,
For a lifetime evaluation of real and complex stresses lognormal or Weibull law. This law provides the stress
and strains histories, methods for damage accumula- distribution for the stress-strength method.
tion computation become a critical issue. The first step
of damage computation is to decompose complex
plastic strain histories obtained from thermo mechan- 3.3. Component strength
ical simulations into a set of simple fatigue cycles. Now that an evaluation of the stress distribution exists,
Downloaded by [UNAM Ciudad Universitaria] at 21:50 23 December 2014

After a critical review of existing models, the rainflow the strength distribution must be obtained in order to
method (Masuishi 1968) has been selected and apply the stress-strength method. As exposed pre-
implemented for this purpose. viously, the strength distribution in the case here is the
The decomposition results in a set of NB fatigue statistical distribution of thermo mechanical damage
blocks defined by equivalent plastic strain amplitude that generates the failure in the critical area of the
Deip and number of cycles applied ni. The Manson – manifold. This distribution is connected to the scatter
Coffin model (see equation (8)) provides a number of of manifold or engine characteristics: wall thicknesses,
cycles to failure NiF for each elementary block obtained material properties, exhaust gas temperature, etc.
from the rainflow decomposition by: The strength distribution assessment can be based
on endurance tests results. But, very often, during a
K
NiF ¼ Dei a : ð15Þ new engine development, for innovative components,
p
2
the number of these results is not sufficient to have a
good statistical confidence in the probability density
Damage accumulated on the entire plastic strain function evaluation. Another approach is to use
history is then computed based on the Palgreem – numerical reliability methods as presented in the
Miner method: previous chapter. It is obvious that this estimation
can be performed by more or less improved simulation
X
NB
ni procedures such as crude Monte-Carlo simulations.
D¼ i : ð16Þ
i¼1
NF Nevertheless, in cases where realizations of the
strength can only be obtained through complicated
numerical procedures such as finite element models,
estimating the probability density function (PDF) with
sufficient accuracy requires huge numerical efforts.
This is why some numerical procedures have been
developed in the field of probabilistic mechanics in
order to access the PDF of variables with less
numerical effort compared to simulations procedures.
One that is described here is derived from stochastic
finite element procedures and is non-intrusive, i.e. it
does not require developments in the finite element
code (Berveiller et al. 2004).
In this study, the main random quantity of interest
is the lifetime of an exhaust manifold subjected to both
mechanical and thermal loadings. The lifetime is
calculated on a special part of the manifold using a
random fatigue model. A finite element model that is
coupled with a stochastic model composed of several
random variables, including the fatigue randomness is
Figure 13. Equivalent plastic strain history during the road also defined. It was shown in x2.4.2 that the proposed
test. reliability methodology can predict accurately a
376 G. Morin et al.

quantile of the lifetime of the manifold (i.e. the polynomials of fxk gM


k¼1 whose degree is less than or
probability that the lifetime of the manifold is less equal to a fixed value p. Note that the following
than a given value) and the parameters that are the relationship holds:
most important on the reliability.
In order to establish the PDF of the lifetime of the ðM þ pÞ!
P¼ : ð19Þ
manifold, a non-intrusive finite element scheme to M!p!
compute the PDF of the plastic strain Dep (o) at a
special location on the manifold is proposed. Then, the Berveiller et al. (2005) suggest taking p ¼ 3 in order to
PDF of the number of cycles to failure Nf(o) is get an accurate estimation of the response, this implies
obtained by Monte-Carlo samplings of Dep (o) made that P ¼ 35.
into the random fatigue model. The stochastic The purpose of the non-intrusive SFEM is to
finite element method (SFEM) allows the evaluation determine the values of the 35 deterministic coefficients
of the statistical moments of the response and its PDF sj computed by finite element calculations. In this
for a reasonable number of finite element model study, a method of regression is adopted. Indeed, the
calculations. studies carried out show that this type of approach gets
Downloaded by [UNAM Ciudad Universitaria] at 21:50 23 December 2014

very good results for a reasonable number of calcula-


tions (Berveiller et al. 2005).
3.3.1. Principles and hypothesis (2) Non-intrusive regression method. Given n
ðiÞ
(1) Expansion of the response onto the polynomial chaos. realizations of fxk gM k¼1 ; i ¼ 1; . . . ; n, which produce a
Let X be the input random vector whose joint PDF is benchmark, the n realizations of the random vector
prescribed. Then, using an isoprobability transforma- X(1), . . . , X(n) are made using the transformation T.
tion T, the four independent normal standard random For each realization, S(i) ¼ S(X(i)) is evaluated, where
variables of the new stochastic model can be defined S is the corresponding value computed with the finite
as: element code. Moreover, the approximated response is
8 then computed by equation (2). The coefficients sj,
>
> x ¼ TðTmaxÞ j ¼ 0, . . . , P – 1 are determined by minimizing the
< 1
x2 ¼ TðuSE Þ squared error defined by:
ð17Þ
>
> x ¼ TðuE Þ
: 3
x4 ¼ TðuC Þ: n 
X 2
DS ¼ SðiÞ  S~ðiÞ : ð20Þ
Let S(o) ¼ Dep (o) be the response to be approxi- i¼1

mated. This random variable is expressed onto the so-


called polynomial chaos as: Or, equivalently:

  X
n X
P1 n o !2
X
P1
ðiÞ ðiÞ
M
S  S~ ¼ Sj Cj fxk gM
k¼1 ; ð18Þ DS ¼ S  Sj Cj xk : ð21Þ
k¼1
j¼0 i¼1 j¼0

where M ¼ 4 (number of random variables) and Cj, These unknown coefficients are solutions of the
j ¼ 0, . . . , P – 1 are P multi-dimensional Hermite following linear system:

0 n oM  n oM  n oM  n o  1
P
n
ðiÞ ðiÞ P
n
ðiÞ ðiÞ M
B C0 xk C 0 xk  C0 xk CP1 xk C
B i¼1 k¼1 k¼1 i¼1 k¼1 k¼1 C
B C
B .. .. .. C
B C
B n n o .  n o  . n o
.
 n o  C
@P ðiÞ
M
ðiÞ
M P
n
ðiÞ
M
ðiÞ
M A
CP1 xk C 0 xk  CP1 xk CP1 xk
i¼1 k¼1 k¼1 i¼1 k¼1 k¼1
0 n n o  1 ð22Þ
P ðiÞ ðiÞ
M
0 1 S C 0 xk
S0 B k¼1 C
B i¼1 C
B . C B C
B . C¼B .. C
@ . A B . C
B n  n oM  C
SP1 @ P ðiÞ A
SðiÞ CP1 xk
i¼1 k¼1
Structure and Infrastructure Engineering 377

An important point in this approach is to properly


ðiÞ
select the benchmarking points fxk gM k¼1 ; i ¼ 1; . . . ; n.
Note that n  P is required so that a solution of
equation (6) exists. Based on the early work made by
Berveiller et al. (2005), (p þ 1) roots of the (p þ 1)th
order Hermite polynomial are chosen here, and then
(p þ 1)M vectors of length M using all possible
combinations were built. Then, N outcomes out of
these (p þ 1)M possibilities were selected. In this study,
N ¼ 105 allows a satisfactory approximation of the
response to be obtained. For further details on
convergence criterion for the approximated response,
see Berveiller et al. (2005).
(3) Computation of the statistical moments of the
response quantities. From equation (2), all statistical
Figure 14. PDF of plastic strain amplitude.
Downloaded by [UNAM Ciudad Universitaria] at 21:50 23 December 2014

moments of the response can be computed. The mean


of S is then:

E½S ¼ s0 : ð23Þ

The variance of S is:

X
P1  
Var½S ¼ s2s ¼ E C2i s2i : ð24Þ
i¼1

The skewness and the kurtosis coefficients of S are:

1 X P1 X
P1 X
P1
dS ¼ dijk Si Sj Sk ; ð25Þ
s3S i¼1 j¼1 k¼1

1 X P1 X
P1 X
P1 X
P1
Figure 15. PDF of lifetime.
KS ¼ E½Ci Cj Ck Cl Si Sj Sk Sl : ð26Þ
s4S i¼1 j¼1 k¼1 l¼1

after endurance tests on an engine test bed on several


identical engines.
3.3.2. Application to the manifold
The histogram of the PDF of Dep(o) obtained by the
non-intrusive SFEM is given in Figure 14 with 10 000 3.3.3. Field reliability evaluation
Monte-Carlo samplings. It can be seen that the strain Now, a global numerical stress-strength approach
plastic amplitude PDF is rather close to a Gaussian exists, and this can be used to study the field reliability
PDF. of a new steel fabricated exhaust manifold, installed on
The previous samplings are now integrated in the a four-cylinder diesel direct injection engine. Figure 16
random fatigue model. Then the plot of the PDF of Nf presents levels of damage undergone by typical
at the same node of the FE manifold model is shown in customers. Logically, the sport driving style is more
Figure 15 with 10 000 Monte-Carlo samplings. damaging than the others, as the power demand is
As a result of these simulations, the failure higher. The damage is also higher for road use because
probability computed (e.g. the value of the quantile cyclic variations of power, exhaust gas temperature
Nrequired) is very close to the value obtained from the and plastic strains per kilometre (i.e. number of cycles
previous FORM calculation, confirming the choice for fatigue) are more important than for highway use.
made on the random variables retained for the Then, the field cumulative failure rate can be
development of the strain plastic amplitude on poly- computed and compared to the reliability target. If
nomial chaos. These results can be validated by the target is not reached, the sensitivity analysis
comparison with the scatter of fatigue life, obtained described earlier gives information about the design
378 G. Morin et al.

and process parameters whose modification will be the work under specific combustion conditions in order to
most efficient to increase product reliability and be maintained in good conditions. During current
decrease the field failure rate. customer use, exhaust gas temperature must be
The method can be validated on engines under artificially increased in order to burn particulates in
serial production, using information obtained from the filter. For nitrogen oxides reduction in NOx traps,
engines under warranty. Failed parts and elements the fuel/air ratio must be frequently increased. For a
about mileage and customer use can be brought back given customer use, these modifications of combustion
through dealers. Detailed analysis of these data can will increase thermal loads on the exhaust manifold
confirm general and numerical results obtained by the and then decrease its reliability. Figure 12 shows the
method presented in this paper. increase of temperature due to diesel particulates filter
regeneration during a road test. Before the start of
production of a new engine, the impact of innovative
3.3.4. Endurance test calibration after treatment systems on field reliability can be
As well as the evaluation of damage undergone by evaluated with the numerical method given in this
customers, this method can be easily used to compute paper. Figure 18 shows the level of damage undergone
Downloaded by [UNAM Ciudad Universitaria] at 21:50 23 December 2014

the damage applied during various endurance tests, by the most demanding customers, for the same kind
performed on engine test beds, or on vehicles. The
comparison of the two categories of damage is shown
in Figure 17.
With the probability density function of damage
undergone by customers and the maximum rate of
failure defined by a reliability target, it is possible to
define the optimal level of damage to apply during
validation tests (‘Test target’ in Figure 17). If this
target is too low (e.g. ‘Test 1’ in Figure 17), too many
customers will not be protected by validation tests, and
failure rate in service will be too high. If this target is
too high (e.g. ‘Test 2’ in Figure 17), validation tests will
be too long and expensive and will penalize develop-
ment schedule. In addition, it will lead to oversize
engines and thus an increase in its mass and cost. The
global method, taking into account scatter of tests
results, is described in detail in Prince et al. (2005).

3.3.5. Influence of after treatment systems Figure 17. Comparison between damage undergone by
customers and applied during tests.
Diesel engine after treatment systems, such as diesel
particulate filters and NOx traps need regularly to

Figure 18. Exhaust manifold damage for Euro 3, Euro 4


Figure 16. Damage indicator for various customer types. and Euro 5 engine versions.
Structure and Infrastructure Engineering 379

of engine, on a Euro 3 version, a Euro 4 version with Berveiller, M., Sudret, B., and Lemaire, M., 2005. Construc-
diesel particulate filter and a Euro 5 version with diesel tion de la réponse paramétrique déterministe d’un
système mécanique par éléments finis stochastiques. In:
particulate filter and a NOx trap. The increase of Proceedings of 7th National Colloquium on Structural
severity with exhaust emissions standards can be Assessment.
clearly seen. Bignonet, A. and Charkaluk, E., 2001. Thermo-mechanical
fatigue in automotive industry. In: International Con-
ference on Temperature-fatigue Interaction, SF2M
4. Conclusion congress.
Chaboche, J.L., 1989. Constitutive equations for cyclic
The general method presented in this paper permits the plasticity and cyclic viscoplasticity. International Journal
field reliability of an engine part to be evaluated using of Plasticity, 5, 247–302.
a completely numerical approach. The method is Ditlevsen, O. and Madsen, H.O., 1996. Structural Reliability
important for the development of innovative compo- Methods. John Wiley & Sons.
Lemaitre, J. and Chaboche, J.L., 1978. Phenomenological
nents and new engines for which there is no experience approach of damage rupture. Journal de Me´canique
or field data. It also gives access to design and Applique´e, 2 (3).
manufacturing process parameters that will have the Manson, F.W., 1954. Transactions. ASME, 76, 931.
Downloaded by [UNAM Ciudad Universitaria] at 21:50 23 December 2014

most important influence on reliability, in terms of Masuishi, M. and Endo, T., 1968. Fatigue of Metals
nominal value and tolerance intervals. It is then Subjected to Varying Stresses. Japan Society of Mechan-
ical Engineers.
possible to optimize reliability at a minimum cost Morin, G., Nicouleau-Bourles, E., Simon, F., and Prince, O.,
and development time. 2005. Reliable diesel engine design based on a new
During the development, endurance tests results numerical method. SAE paper 2005-01-1762.
can be taken into account regularly in order to increase Pendola, M., Morin, G., Mohamed, A., and Ragot, P., 2003.
the confidence in the reliability evaluation up to the Reliability based analysis of exhaust manifold lifetime.
In: Simulation: an essential tool for risk management in
start of production. Specific methods that are able to industrial product development, SIA congress.
integrate numerical and experimental results closely PHIMECA SA, PHIMECA SoftwareÓ User’s manual, 2004.
are now under development and application. They will Prince, O., Morin, G., and Jouzeau, C., 2005. Validation test
contribute to maximizing the efficiency of the present optimization based on a statistical approach for diesel
global numerical approach. engine cylinder heads reliability. SAE paper 2005-01-
1780.
Rackwitz, R. and Fiessler, B., 1979. Structural reliability
References under combined random load sequences. Computers and
Abdo, T. and Rackwitz, R., 1990. A new beta-point Structures, 9, 489–494.
algorithm for large time-invariant and time variant Skelton, R.P., 1991. Energy criterion for high temperature
reliability problems. In: 3rd WG IFIP Conference. low cycle failure. Materials Science and Technology, 7,
Azzouz, F., Cailletaud, G., Foerch, R., Morin, G., Quilici, 427–438.
S., and Ragot, P., 2002. Identification of visco-plastic Taira, S., 1973. Relation between thermal fatigue and low-
constitutive and creep-fatigue damage coefficients to use cycle fatigue at elevated temperature. Fatigue at elevated
in ABAQUS automotive structural calculations with the temperatures, ASTM STP 520, 80–101.
Zmat library. In: ABAQUS Users’ Conference. Thoft-Cristensen, P., 1986. Application of Structural Systems
Berveiller, M., Sudret, B., and Lemaire, M., 2004. Presenta- Reliability Theory. Springer-Verlag.
tion of two methods for computing the response .
coefficients in stochastic finite element analysis. In:
Proceedings of 9th ASCE Specialty Conference on
Probabilistic Mechanics and Structural Reliability.

You might also like