You are on page 1of 6

A New Methodology to Design a Reliable Product Based on Warranty

Financial Data
Aziz Moudoub, VALEO JAPAN Co., Ltd.
David DeLaux, VALEO Thermal Systems
Abdelkhalak El Hami, PhD, INSA Rouen

Key Words: Automotive warranty cost, Automotive reliability, field data failure, and lifetime

SUMMARY & CONCLUSIONS assessment, warranty cost and its percentage of sale for
financial impact and failure rate. Time and mileage are largely
This paper proposes a methodology to verify products
used to monitor the reliability of a product but it requires good
reliability through its life time using financial information
knowledge of field claim, issues in order to treat the data to
from warranty cost for a long period at 8 and 15 years. It will
get fitting lifetime data and distribution laws, we will present
present the challenge that automotive industry faces and study
it later. This paper will present a reliability model based on the
the potential improvement to be considered in order to
financial warranty information to simplify the pre-treatment of
improve the reliability during the design and validation phases.
data.
The formulas developed in this research permit one to define
the reliability of the product without detailed warranty 2 AUTOMOTIVE WARRANTY VS. RELIABILITY
information from the field and as a consequence simplify the
Automotive industry is in constant evolution and vehicles
treatment of the data. In addition, this approach allows to
are getting more and more sophisticated system to respond to
assess a financial risk for warranty extension.
consumer and market demands. Manufacturers have less and
1 INTRODUCTION less time to develop and validate a vehicle and its 30,000[15]
components in order to stay competitive by restyling and
Many consumer durable goods such as automotives,
upgrading the vehicles every years.
electronic appliances, clothing and even services include a
manufacturer warranty to insure the reliability and quality of
the product or service. By providing a warranty, the maker or
seller makes a promise to the customer that a product or
service will perform its required function for at least the
duration of the warranty contract. The resulting claims contain
actual performance data from field obtained under the product
real operating conditions. Manufacturer could use those
warranty information to compare field performances with
bench test reliability, life time data. Reliability
In automotive industry, the manufacturers provide Warranty
Wa
arranty
customers a warranty using two metric to quantify a vehicle
Quality
Q
Qu allit
ity
lifetime: time and mileage. The manufacturers prefer to use
time dimension to model or assess field performance to avoid
the uncertainty associated with driving patterns and calculate
Fig 1- Bathtub curve vs. reliability expression
the reliability of their products and expected failure rate at the
end of the warranty period. Consumers want products which answer their usage in
In case of failure during the warranty period, the maker their daily environment. Manufacturers need to define the
seller commits to fix or replace the faulty product. The required mission, the conditions and duration of usage to
activities triggered by a warranty claim will generate costs that design and validate that the product has the ability to perform
will be bear by the manufacturer of the faulty product. Those its required function for the warranty period which is getting
costs are monitored as performance indicator of a product on longer.
the field using the percentage of sales that it represents. As illustrated in Fig. 1, the warranty is part of the
In automotive industry, one can use three different vehicle’s reliability. Indeed, we often talk about reliability to
metrics in order to monitor the field performance of a product: cover the first 15 years of the vehicle (or 300,000 km)
time and mileage for failure rate, reliability and lifetime associated with a tolerance of defective parts (for example

‹,(((
10% failure at 15 years or 90% reliability). A three-year must consider the sophistication of vehicles leading to more
warranty for example on a car corresponds to the first three complex systems to be studied and validated. When one was
years of vehicle life when the vehicle is reliable. The considering only mechanical issues when studying the
difference being that the guarantee is a financial commitment, reliability of a vehicle in the 70’s, the same person must
while reliability is a “moral” commitment on the part of the consider today mechanical, electronic and electromechanical
designer. Of course, the design of a product or system is based failures.
on reliability commitments (i.e. 15 years or 300,000 Km) and
2.2 Quantitative analysis
not only over the warranty period. The study of products
failing during the guarantee provides interesting data on The author will consider two approaches for the
reliability performance even though one must keep in mind quantitative analysis of automotive reliability: during the
that this is but a restricted vision of only few years (often one warranty period when data from field are abundant and easy to
to three years) in the field. collect, and after the warranty period.
Over the past years, we have witnessed several warranty During the warranty period the manufacturers have access
crises in the automotive industry leading to huge warranty cost to very detailed information related to the vehicle usage from
amount and teaching some lessons concerning the reliability periodic maintenance and warranty claim. Thus it becomes
of the product in the field to the makers concerning the design possible to make lifetime estimation by fitting the Hazard
at the right cost, reliability in environmental testing and function of the claim with an estimation model. By using
through the product lifecycle. So it is necessary to evaluate warranty data, one has access to the usage duration in time or
monitor and improve the reliability of the products through its mileage plus the delay between the failure and detection
lifecycle. The first indicator that could be considered to occurrence, which will be use to estimate the reliability of the
monitor the reliability is warranty cost since its represent the product. Several models exist to estimate the reliability or
quantity of parts failing during the warranty period or lifetime of a product but a common tool used is the standard
unreliability to ensure their function. Following the Weibull distribution.
sangenshugi mindset [1] one monitoring the warranty cost ౐షಋ ஒ
ஒ ୘ିஓ ஒିଵ ିቀ ቁ
could detect any variation of cost or quantity of claim (id est ˆሺሻ ൌ ቀ ቁ ‡ ಏ (1)
஗ ஗
the reliability of product) and use this indicator to
The standard Weibull distribution is a very versatile
communicate internally to trigger analysis.
equation which can be simplified or modified to fit the user
2.1 Qualitative analysis needs. Its reliability function can be described as following:
౐షಋ ஒ
A qualitative analysis of issues in automotive industry ିቀ ቁ
ሺሻ ൌ ‡ ಏ (2)
which led to recall can help us to understand what the main
root causes are and define the required improvement actions
for the sector.
According to Japan Ministry of Land, Infrastructure and
Transport done on the recalls made between 2006 and 2010
[2] 54% of all the claims were due to reliability design issues:
z Insufficient test settings (47%)
z Defect of development evaluation (5%)
z Lack of vehicle testing (2%)
In addition, a study from Wu [3] states that more than
10% of all the reported warranty claims are linked to various
human factors and misuses of consumers. Those studies point
out directly at the lack of testing validation and in fine
Fig 2 - Reliability function of a system
reliability engineering of the products. From this statement
and a study of several recalls impacting all types of vehicle The reliability defined using warranty information can
range, the author can define several axis of improvement: provide reliable estimation based on the known observable
z Reliability throughout the product life cycle in order to period. After the warranty period, it becomes more
prevent critical failure from occurring with the aging of complicated to collect vehicle failure information and we can’t
the part for its lifetime integrate in the estimation model the impact of failure
z Reliability in environmental testing to consider the occurring out of warranty period. There are two types of
components failure due to all the interactions from its failures to be considered: the non-incapacitating one having a
environment (e.g.: surrounding parts wearing, climatic small impact on comfort (LED screen, roof light...) which will
condition, human interactions) be ignored by consumer and the incapacitating (brake failure,
z Reliability in robustness engineering or design at the right gear box…) for which the consumer will decide based on his
cost considering the warranty period increasing tendency budget and difficulty how to repair it (manufacturer dealers,
in the automotive industry third part agent, by oneself). It becomes impossible to survey
In order to improve those three axes, automotive manufactures
field reliability or failure rate if the manufacturer is not dimensional warranty by fitting the parameter Ȝ with various
controlling the flux of spare parts which could indicate the functions of time and mileage. In the same register, Hu and
failure rate in the market. Lawless [6] suggest a technique for modeling warranty claims
However, the market is attentive to vehicle issues and the as truncated data that assumes warranty claims follow a
evolution of a vehicle aftermarket price versus its age or Poisson process. Oh and Bai [7] present a method for
mileage can be the reflection of its reliability. The augmenting parametric warranty data models with selected
depreciation price of used car observed in the Fig. 3. identify observations from products whose lifetime exceeded the
three parts : warranty period.
z First years of usage (0-60,000 Km) – assumed as the Another model suggesting to use the time domain and
“early phase” of reliability in the bathtub curve cumulative quantity of claim per month in service developed
z Second phase (60,000-150,000 Km) – assumed as by Wasserman [8] and Robinson and McDonald [9] using a
the “random phase” of reliability in the bathtub non-parametric model to define the statistic R(t)-claims per
curve thousand vehicles reported cumulatively by month in service
z Third phase (over 150,000 Km) – assumed as the that automobile manufacturers can use to track warranty
“wear-out” phase of reliability in the bathtub curve performance. Wasserman [8] developed a dynamic linear
predictive model for R(t) using data from multiple model
years of a given vehicle. By using a log-log paper, Robinson
and McDonald [9] suggest to plot R(t) and fitting a line to the
observed data. Singpurwalla and Wilson [10] develop a bi-
variate failure model for automobile warranty data indexed by
time and mileage. They derive the two marginal failure
distributions and present a method for predicting R(t) using a
log–log model.
3.2 Mixture models
Automotive products, be it a vehicle or a component, are
a mixture of several parts which operates together influencing,
impacting each others like the figure 4 expose.

Fig 3 - Aftermarket price for Honda vehicles in Spain


It becomes complicated to separate the hazard functions
(source : 16)
described by field data and as a result what we can observe is
A vehicle will have a normal drop of value after its first the sum of all hazard functions as we can see in the figure 5.
registration and when a new model is released. Aside from
those two events considering a state where failures will occur
Early failures

randomly the price should decrease proportionally to its Wear out


mileage until reaching the state of wearing out marked by a
sudden drop of price.
Of course this interpretation has got some limits but
mainly illustrate the assumption of the link “cost in the field”
versus “Reliability”.
Random failures
3 ESTIMATION MODELS
Automotives manufacturers and scholars have developed
several models in order to treat the warranty data, estimate
product lifetime and track warranty performance. Hereunder,
we will present few models with their advantage and
inconvenient. Fig 4 - Example of reliability distribution
3.1 Parametric, non parametric & other models We can propose to decompose field data to identify the
different failure types (early, random, wear out) allowing to
Several research propose to use as basement the Poisson
calculate Weibull distribution (Fig. 6) fitting the different
distribution as presented by Kalbfleisch, Lawless and
portion of the product lifetime.
Robinson [4] who estimate automotive warranty claims by
using the time domain or Moskowitz and Chun [5] suggesting
to use a bi-variate Poisson model to predict claims for a two
’Ʌǡ– ൌ Ͳ
‫ۓ‬ ୲ ಊ
ି஑୲
ሺ–ሻ ൌ ’Ʌ ൅ ’ሺͳ െ Ʌሻ ୲‫ כ‬൅ ሺͳ െ ’ሻ ቀͳ െ ‡ ቁǡ Ͳ ൏ ‫ ݐ‬൑ – ‫כ‬
‫۔‬ ି஑୲ಊ
‫ ’ ە‬൅ ሺͳ െ ’ሻ ቀͳ െ ‡ ቁ ǡ– ൐ – ‫כ‬
(5)
3.3 Advantages and inconvenient
The warranty modeling techniques mentioned hereabove
fit warranty data. Depending on the complexity of the
selected model one can assess a product warranty integrating
Fig 5 - example of hazard function of a system z Pre-delivery claims
z Several failure modes
z Different design improvements
Such analyses require detailed field data in quantity with
enough history to have a full picture of the product behavior
all along its lifetime. We might be able to provide feedback or
assessment if field failures are a result of inadequate design,
assembly error…
However, it requires to the practitioner to have experience
knowledge of the product to identify the parameters to be
considered or cluster the field data. Special application or
software is necessary in order to use complex models and
mastering the model used.

Fig 6 - example of density function of a system 4 FROM WARRANTY COST TO RELIABILITY

The Hjorth distribution [11] can describe the entire This paper develops a model which allows to define the
bathtub curve thanks to its three parameters, including two reliability of product and its design by using warranty
shape parameters. Thus, if the failure behavior changes during financial information thus limiting the treatment of data and
usage we can describe the bathtub curve using only one simplifying the combination of data.
distribution. The Hjorth distribution might describe the field Table 1 - List of parameters
reality better than the classic Weibull distribution.
ಌ౪మ
ష మ

ሺ–ሻ ൌ ͳ െ ಐ ǡͲ ൑ ‫ ݐ‬൏ λ (3) Variable Description
ሺଵାஒ୲ሻಊ t Year
In a similar way, we could also mention the modified Ȝ Failure rate (% failure/year)
Weibull extension [12] or the Dhillon distribution model[13] N Number of part produced in warranty period
which are able to describe a bathtub shaped failure function, n Number of part failed in warranty
increasing and decreasing hazard rates. Chung, Wu, and Ф Warranty cost in % of sales
Herrin [14] and Meeker [15] proposed mixed Weibull models Cw Cost of part failed in warranty
equivalent to Cs Cost of part sold

’ ቀͳ െ  ‡ିሺ஑୲ሻ ቁ ǡ Ͳ ൑ ‫ ݐ‬൏ λ Cl Cost of labor
ሺ–ሻ ൌ ቊ (4) Į Coefficient of warranty contract (i.e.: 1.1)
ͳǡ– ൌ λ
which fit occupational injury data and integrated circuit data
Depending on the instant t where we are in the lifetime of
respectively. These authors enhance the need for the mixtures
a product, we can describe its reliability by two different
by providing intuitive arguments regarding properties of
formulas by period:
lifetime. A similar approach can be considered for the
z Random failure mode:
automotive warranty data. It is common in automotive
industry to have a product impacted by two failure modes, the ሺ–ሻ ൌ ‡ሺି஛୲ሻ ǡ – ൑ – ୧ (6)
lifetime data can be model following Mori, Arai, Kaneko and z Wear out failure mode
Yoshikawa [16] demonstration. In their example, the multiple ౪ ஒ
ିቀ ቁ
failure modes result in higher early failure rates than the single ሺ–ሻ ൌ ‡ ಏ ǡ – ൒ – ୧ (7)
failure mode case. Majeske [17] proposes a Uniform/weibull Combining those two equations with have:
mixture model able to separate two failure modes and ౪ ஒ
ିቀ ቁ
including the pre-delivery claims ሺ–ሻ ൌ ‡ሺି஛୲ሻ ൌ ‡ ಏ (8)
୲౟
>Ʉ ൌ భ (9) R(t) = 99.9% and F(t) = 930ppm per year
ሾ஛‫୲כ‬౟ ሿಊ For the period deemed as wearing out of the system, we
The warranty cost in percentage of sale is expressed as will be interested to know the reliability of the system after 15
followed years of usage to compare the evolution of the reliability after
୬‫כ‬େ౭ 7 additional years of usage:
ɘΨ ൌ (10) ஒ
୒‫כ‬େ౩ ଵ
୛ ൌ  ୱ ൅  ሺ୪ ‫ כ‬Ƚሻ(11) ୲ ஒ
ିቀ ቁ – തതതୱ
 ஒ
ሺ–ሻ ൌ ‡ ஗ ൌ ‡š’ ൞െ ‫  כ‬൭ሺɘΨ š– ୧ ሻ ൱ ൢ
େ౩ –୧ തതത
ୱ ൅  ሺഥ୪ šȽሻ
Ɂൌ (12)
େ౩ ାሺେౢ ‫כ‬஑ሻ

With this, the failure rate formula can be revised as: at t • 8years (16)
୬ େ౩ ାሺେౢ ୶஑ሻ
ɉሺ–ሻ ൌ ‫כ‬ ‫ כ‬Ɂ(13) R(t) = 95%
୒ େ౩
The demonstration suggests that the system undergoes a
>ɉሺ–ሻ ൌ ɘΨ ‫ כ‬Ɂ(14) quicker degradation during the second period of usage (8 to 15
We can now redefine the reliability by integrating the years). Manufacturers can now judge if the current design
warranty percentage in cost: reliability is acceptable to provide longer warranty condition
ሺ–ሻ ൌ ͳ െ ሺ–ሻ ൌ ͳ െ ‡ሺି஛୲ሻ  to final user. On the other hand, manufactures can investigate
തതതത౩

if changes are required in order to increase the lifespan of the
ൌ ͳ െ ‡š’ ቄെɘΨ ‫  כ‬ቀതതതത തതതౢ ୶஑ሻ
ቁቅ ሺ–ሻat t ” ti (15) system and the effort required to reach acceptable level of
େ౩ ାሺେ
ଵ ஒ reliability depending on the criticality of the failure exposed to
୲ ஒ – തതതୱ
 ஒ the final user for longer usage (i.e.: improve design,
ିቀ ቁ
ሺ–ሻ ൌ ‡ ஗ ൌ ‡š’ ൞െ ‫  כ‬൭ሺɘΨ š– ୧ ሻ ൱ ൢ recommend maintenance of system).
–୧ തതത
ୱ ൅  ሺഥ୪ šȽሻ
6 CONCLUSION AND RECOMMENDATIONS
at t • ti (16)
This research suggests that manufacturers can assess the
5 APPLICATION TO AUTOMOTIVE WARRANTY DATA reliability of a system past its warranty period by using
Taking the example of an automotive system, we are able financial information. For example a supplier simplify the
to observe a change of the aftermarket price of the vehicle treatment of the data allowing to combine several application
presented in the figure [3] after 150,000km. According to using the same product without consideration of the system
current warranty contract proposed by automotive integration and select the improvement required the product
manufacturers and the author experience, we will make the life time. But it also presents a great benefice for suppliers
assumption that a vehicle is running on average 20,000km per who might not be able to collect detailed information from
year (small, medium or high vehicle for Europe application, field claim from its customers. We also suggest three axis of
gasoline or diesel, estimation based on Log-Normal law of the improvement concerning the reliability improvement link to
driving pattern). So the drop down of price observed at testing and validation of the product linked to the
160,000km can be assimilated to 7,5 years of usage in the field. sophistication of the systems.
The period of 60,000 Km (equivalent to 3 years) usage until 8 REFERENCES
years (equivalent to 150,000 Km) the system is in a state of
random failure, past the 8 years the wearing out of the system 1. Testa. Q, Aoudia.H. Perfect QRQC: the basics
and number of failure will increase with the time. 2. Japan domestics cars, notification sample from HS.15
(2003) to HS 19 (2007)
Table 2 – Test data for automotive system 3. Wu. Warranty Claim Analysis Considering Human
Variable Description Value Factors. Reliability Engineering and System Safety, 96
(1). pp. 131-138. ISSN 09518320
t i Failure mode changing point 8 years
4. Kalbfleisch JD, Lawless JF, Robinson JA. Methods for
Ф Warranty cost in % of sales 0.50%
the analysis and prediction of warranty claims.
Cs Cost of part sold 30.00 €
Technometrics 1991;33:273–85.
Cl Cost of labor 120.00 €
5. Moskowitz H, Chun YH. A poisson regression model for
Į Coefficient of warranty contract 1.1 two-attribute warranty policies. Naval Res Logistics
ɴ Weibull parameter 3 1994;41:355–76.
6. Hu XJ, Lawless JF. Estimations of rate and mean
During the period of random failure occurring for a usage functions from truncated recurrent event data. J Am Stat
between 3 and 8 years of usage the reliability of the system Assoc 1996;91(433):300–10.
will be defined as followed: 7. Oh YS, Bai DS. Field data analysis with additional after
ሺ–ሻ ൌ ͳ െ ሺ–ሻ ൌ ͳ െ ‡ሺି஛୲ሻ  warranty failure data. Reliab Engng Syst Safety
തതതത౩
େ 2001;72(1):1–8.
ൌ ͳ െ ‡š’ ቄെɘΨ ‫  כ‬ቀതതതത തതത ቁቅ ሺ–ሻat t ” 8 years (15)
େ౩ ାሺେౢ ୶஑ሻ 8. Wasserman GS. An application of dynamic linear models
for predicting warranty claims. Comput Indus Engng Kumagaya-shi, Saitama 360-0193, JAPAN
1992;22(1):37–47.
e-mail: aziz.moudoub@valeo.com
9. Robinson JA, McDonald GC. Issues related to field
reliability and warranty data. Data Quality Control: David DELAUX PhD
Theory and Pramatics, New York: Marcel Dekker; 1991. VALEO Thermal Systems
10. Singpurwalla ND, Wilson S. The warranty problem: its 8 Rue Louis Lormand,
statistical and game theoretic aspects. SIAM Rev 78321 La Verrière, FRANCE
1993;35:17–42.
e-mail: david.delaux@valeo.com
11. Hjorth U. "A reliability distribution with increasing,
decreasing, constant and bathtub䇲㼟㼔㼍㼜㼑㼐㻌 㼒㼍㼕㼘㼡㼞㼑 rates." Head of worldwide Reliability Engineering at VALEO and
Technometrics 22.1 pp. 99䇲㻝 07, 1980 Reliability Senior Expert. Lecturer at Rouen University and
12. Xie M., Tang Y., Goh T., "A modified Weibull extension Engineer School ENSMA Poitiers, he is also a National
with bathtub ̺ shaped failure rate function."Reliability auditor at COFRAC (ISO 17025). President of the European
Engineering & System Safety 76.3 pp. 279䇲㻞㻤㻡㻘㻌㻞㻜㻜㻞㻚 Reliability Environmental Commission (CEEES)
13. Dhillon B. S. "A hazard rate model." IEEE transactions
Abdelkhalak El Hami, PhD
on Reliability 28.2, 1979
INSA Rouen Normandie,
14. Chung MK, Wu SH, Herrin GD. The use of a mixed
Saint Etienne du Rouvray, 76130 FRANCE
Weibull Model in occupational injury analysis. J
Occupational Accidents 1986;7:239–50. e-mail: aelhami@insa-rouen.fr
15. Meeker WQ. Limited failure population life tests: Abdelkhalak El Hami, PhD, is a Full Professor at INSA
application to integrated circuit reliability. Technometrics (National Institute of Applied Sciences) of Rouen,- Normandy
1987;29(1):51–65. University, France, as well as Deputy Director of Laboratoire
16. Mori S, Arai N, Kaneko Y, Yoshikawa K. Polyoxide d’Optimisation et FIabilité en Mécanique des Structures
thinning limitation and superior ONO interpoly dielectric (LOFIMS laboratory and Head of mechanical chair of the
of nonvolatile memory device. IEEE Trans Electron Dev National Conservatory of Arts and Crafts in Normandy,
1991;38:270–6. France. He has been training director of mechanical engineers
17. Majeske KD. A mixture model for automobile warranty at INSA-Rouen. Professor El Hami’s research activities
data. Reliability Engineering and System Safety 81 (2003) include reliability-optimization systems. He is Operator
71–77 manager INSA Group in Maghreb (Algeria, Morocco and
18. How many parts is each car made of ? – Toyota Motor Tunisia)). He has supervised 38 PhD theses. He also is the
Corporation. source available at (seen Sept. 21, 2017): author and co-author of more than a twenty books and more
http://www.toyota.co.jp/en/kids/faq/d/01/04/ than 550 papers published in international journals and
19. Honda Civic for sale in Torrevieja conferences. He has a doctorate in engineering sciences from
http://www.vehiclus.com/Torrevieja/Honda/Civic (seen the University of Franche-Comté in France (1992). He
Sept. 21, 2017) received his Habilitation diploma to supervise research (HDR)
in 2000. He's Editor in chief of 3 Set of international Book,
BIOGRAPHIES ISTE, Wiley and Son and Elsivier. He has been awarded:
Aziz MOUDOUB 1992: Prize, Micronora of Piezoelectric actuator 2013: Prize,
VALEO JAPAN Co., Ltd. Pepite Noramandy Aerospace 2015: Prize, Success story of
Moveo, "Pôle de compétitivité international"
39, Sendai,

You might also like